Enable to have more control over serializing or its result than
`xkb_keymap_get_as_string()` or `xkb_keymap_get_as_string2()`.
In particular, it allows to choose which layouts are serialized.
Added:
- `xkb_keymap_serialize()`
- `struct xkb_keymap_serialize_config`
- `struct xkb_keymap_serialize_result`
Use `uint32_t` instead of enums in public structs, because the enum
size is not reliable for size-controlled ABI.
Fixed:
- `struct xkb_state_components_update`
- `struct xkb_layout_policy_update`
Duplicating the options of compile-keymap in the various dump-keymap
tools is time-consuming with little benefit: it suffices to pipe the
output to xkbcli-compile-keymap to get the full set of options.
From now, do not duplicate these options and remove the one that were
not yet published, for consistency.
Before this commit, key modifier map with multiple modifiers and/or
with extended real modifiers could not be serialized properly,
resulting in parsing/serializing rountrip failure.
Fixed by properly serializing the `modifier_map` entries as allowed by
the corresponding keymap format.
Due to backward compatibility with the X11 ecosystem, serializing to the
v1 keymap format is subject to the following restrictions:
- Non-core modifiers are not supported.
- Multiple modifiers per key is not supported, unless using keysyms in the
original `modifier_map` entry.
Before this commit, `modifier_map` entries were limited to use a
*single*, *core* real modifier. However:
- XOrg xserver does support multiple modifiers per key modmap.
- XOrg xkbcomp can produce such keymap when different `modifier_map`
entries resolve to the same key but at least one keysym is involved.
- xkbcommon nor XOrg xkbcomp can roundtrip parsing/serializing
if some keys have a modmap with multiple modifiers: in such case,
the keys appear in multiple single modifier entries and only the
latest entry is kept, resulting in an incomplete modmap after
roundtrip.
- xkbcommon supports extended real modifiers but cannot use them
in `modifier_map` entries.
This commit enables to use real modifier *mask expression* when
parsing in the v2 keymap format:
modifier_map Shift { … };
// New: numeric equivalent of the previous entry
modifier_map 0x001 { … };
// New: modifier mask expression
modifier_map Mod1+Mod2 { … };
// New: numeric equivalent of the previous entry
modifier_map 0x018 { … };
// New: non-X11 core modifiers
modifier_map 0x100 { … };
// New: multiple non-X11 core modifiers
modifier_map 0xf00 { … };
// New: Mix of X11 core and non core modifiers
modifier_map Mod2+0x100 { … };
// New: Numeric equivalent of the previous entry
modifier_map 0x110 { … };
Serializing will be done in a later commit.
The `TwoKeys` X11 tag alters the `StickyKeys` feature and is
described as:
If the XkbAX_TwoKeys flag is set, XKB automatically turns
StickyKeys off if the user presses two or more keys at once.
This serves to automatically disable StickyKeys when a user
who does not require sticky keys is using the keyboard.
Source: https://www.x.org/releases/current/doc/kbproto/xkbproto.html#:~:text=XkbAX_TwoKeys
The implementation in xkbcommon follow the one in XOrg, albeit some
renaming:
- Rename `XKB_A11Y_LATCH_TO_LOCK` to `XKB_A11Y_STICKY_KEYS_LATCH_TO_LOCK`.
- Added `XKB_A11Y_STICKY_KEYS_NO_SIMULTANEOUS_KEYS` to the
`xkb_a11y_flags` enumeration.
- As in XOrg, the detection is limited to the case that at least one
modifier key is pressed. While it does not strictly implement the
XKB specification, neither does XOrg. The pros is that it does not
require to track the state of the individual keys.
- A future implementation may fully implement the spec by tracking
individual keys; such implementation would enable a new API to
query keys states.
Replace `atoi` with `parse_dec_to_uint32_t` so that parsing is faster,
stricter and handle errors instead of silently discarding them.
NOTE: Some tests keymaps required and update because the XKB context
is shared between tests and atoms have new encodings.
Mention in the public API that possible padding bytes MUST be set to
0 with `memset()`.
Ensure there is no *implicit* padding so that we can avoid using
`memset()` internally.
Before this commit, detection of the `default` tag of XKB section
was broken: it wrongly matched the first section with at least one tag,
independently of the tag value. This resulted in possibly including the
wrong section when the include statement does not use an explicit
section name, e.g. `include "fr"` instead of `include "fr(basic)"`.
Thanksfully, xkeyboard-config has always the `default` section at the
top of the files, so only custom configurations could be affected.
- Fixed typo: `&&` → `&` introduced by
9b0b8c6863.
- Added corresponding test.
- Rename `user-configuration.md` to `custom-configuration.md`, since
it also applies to system-wide configuration.
- Add a quick start
- Add further internal links
- Security first
- Fix implicit definitions
- Fix typos
Before this commit, only user-configuration was documented, while in
fact the instructions can be applied for system-wide configuration
by modifying `/etc/xkb` directory (or similar) rather than
`~/.config/xkb` (or similar).
- Ensure pre-generated parser files work everywhere.
- Ensure pre-generated parser files update works on Linux.
For some reason the Windows CI fails on the same task and I could
not figure out why, so I removed it.
- Ensure pre-generated files are up-to-date.
This change introduces pre-generated `parser.c` and `parser.h` files
into the source tree and adds a manual mechanism to regenerate them.
- Added the initial version of the generated `parser.c` and `parser.h`
to allow out-of-the-box compilation without a yacc dependency.
- Added the meson option `enable-parser-auto-generation` (enabled by
default) to choose between auto-generated and pre-generated parser
files.
- Implemented the `src/xkbcomp/update-parser` custom target (not built
by default) to allow developers to manually sync the pre-generated
parser files after a `parser.y` update.
- Added a `yacc-wrapper` script to ensure Bison runs from the project
root and adds a license header. This also guarantees that `#line`
directives in generated files use relative paths, keeping the source
tree sanitized and portable.
This approach prevents “Bison lock-in” while maintaining the ability
to update the grammar as needed.
The libxml2 header probing did not specify the libxml2 dependency,
resulting in failure or success depending of the system.
The system include path is `/usr/include/libxml2/libxml`. Some systems
like openSUSE Tumbleweed have a symlink `/usr/include/libxml` →
`/usr/include/libxml2/libxml`, but some other do not (e.g. ArchLinux).
Fixed by specifying the libxml2 dependency in the header probing.
Before this commit, `xkb_machine_options` did not depend on a keymap,
postponing various checks to the call of `xkb_machine_new()`.
The issue is that non-trivial errors of `xkb_machine_options_*()`
functions are not triggered at call site.
- Refactored `xkb_machine_options` into `xkb_machine_builder` with an
explicit dependency on the corresponding keymap.
- Fixed tests & bench.
⚠️ Tools will be fixed in next commits.
- Created `xkb_state_mod_internal` enum for state mode to handle
client + legacy + `xkb_machine` use cases.
- Added `XKB_ERROR_UNEXPECTED_STATE_MODE`.
- Refactored `xkb_state` to be a minimal common state.
- Added private `xkb_client_state` as a wrapper to `xkb_state`
for strong typing.
- Added `xkb_server_state` to handle legacy server state.
Inherits from `xkb_state`.
- Refactored `xkb_machine` to inherits from `xkb_server_state`.
- Added tests.
Curently `xkb_state` is used for both client and server applications
for historical reasons. However, `xkb_state` contains much more data
than necessary for clients and may confuse users with the mixed
client/server API.
This commit introduces state *modes* to select the appropriate API:
- The `xkb_state_mode` enumeration provides 3 modes:
- `XKB_STATE_MODE_CLIENT`: the client-only API mode, with a reduced
memory footprint, updatable only via `xkb_state_update_mask()`.
- `XKB_STATE_MODE_SERVER_QUERY`: the state companion to `xkb_machine`
API mode, with a reduced memory footprint, updatable only via
`xkb_state_update_event()`.
- `XKB_STATE_MODE_SERVER`: the server-only API mode, updatable only
via `xkb_state_update_key()`, `xkb_state_update_synthetic()` and
the legacy `xkb_state_update_latched_locked()`.
Each mode restricts the usable API to its use case.
- The new constructor `xkb_state_new_with_mode()` enables to select
explicitly the state mode.
NOTE: The implementation will be done in the next commits.
```
--- before
+++ after
@@ -43,7 +43,7 @@
Dependency xkeyboard-config-2 found: YES 2.47 (cached)
Has header "unistd.h" : YES
Has header "dirent.h" : YES
-Checking if "__builtin_expect" links: NO
+Checking if "__builtin_expect" links: YES
Compiler for C supports arguments -Werror=attributes: YES
Checking if "counted_by attribute for pointers" compiles: NO
Header "unistd.h" has symbol "eaccess" : YES
@@ -63,7 +63,7 @@
Checking for function "gnu_get_libc_version" : YES
Header "locale.h" has symbol "newlocale" : YES
Checking if "newlocale fails on invalid locales" runs: YES
-Checking if "-Wl,--version-script" links: NO
+Checking if "-Wl,--version-script" links: YES
Program scripts/map-to-def found: YES (/home/soap/git/libxkbcommon/scripts/map-to-def)
Program bison found: YES 3.8.2 3.8.2 (/usr/bin/bison)
Found pkg-config: YES (/usr/bin/pkg-config) 2.5.1
```
Keep checking EXIT_INVALID_USAGE for invalid options, but stop requiring
the exact GNU-style stderr text "unrecognized option".
On Solaris the same failure may report "illegal option", "unknown option",
or only usage/help text, while still returning the expected exit code.
Change the name to follow the same pattern of the rest of the API and
the signature to use `uint32_t` instead of `int`:
```diff
bool
- xkb_has_feature(enum xkb_feature feature, int value);
+ xkb_feature_supported(enum xkb_feature feature, uint32_t value);
```
Enumeration values are always non-negative in practice, and flags-style
enumerations can have members with the high bit set, which causes
undefined behavior with signed `int`. Negative sentinel values such as
-1 are accommodated by the well-defined unsigned wraparound cast.
`uint32_t` is thus both semantically more accurate and safer than `int`.
Assign values in the `xkb_feature` enumeration according to a
segmented decimal namespace that organizes enumerators by module
and sub-module. The namespace is purely organizational: values are
opaque constants and are never meant to be parsed or decomposed at
runtime.
Values follow the pattern `AABCD`:
| Numeric pattern | Segment | Description |
| --------------- | ---------- | ------------------------------------------------------------ |
| AA000 | Module | Top-level object; unallocated slots are independent spares |
| 00B00 | Sub-module | Functional category; primary slots at ×200, companions +100 |
| 000CD | Feature | Fine-grained index within a sub-module; related pairs at +20 |
Current module assignments:
Slot Module
──────────────────────────
0 Core / global
1 Error
2 (reserved for error or context-related object)
3 Context
4 (reserved for context-related object)
5 Logging
6-8 (reserved for logging or core keymap-related objects)
9 Keysym
10-17 (reserved for core keymap-related objects and keymap builders)
18 RMLVO
19-20 (reserved for core keymap-related objects and keymap builders)
21 Keymap
22-23 (reserved for keymap and state -related objects)
24 Keyboard state components
25 Keyboard state machine (reserved)
26 (reserved for state-related object)
27 Keyboard events / input
28-29 (spare)
30 Compose table
31 Compose state
32 (spare)
Cross sub-modules:
Slot Sub-module
──────────────────────────
0 General / components / properties
100 General / components / properties
200 Creation / parsing
300 Update
400 Serialization
500 (spare / module-specific)
600 Iterator
700 (spare / module-specific)
800 Query
900 (spare / module-specific)
The current values are meant to leave enough room for future additions.
Compose is much more independent than the rest and has no foreseen
developments, so it is put at the very end to leave room for keymap and
state handling.
The DLLs are versioned with the major version, so the DEF files
must point to the suffixed DLLs (e.g. `LIBRARY xkbcommon-0`).
This ensures the linker resolves symbols to the correct filename.
Note that this issue pre-dated the build files refactor starting
at 32de22deff.
Relocates fuzzing definitions to the `fuzz` subdirectory.
This makes the meson.build files easier to navigate and keeps
target definitions close to their respective source files.
Relocates tools definitions to the `tools` subdirectory.
This makes the meson.build files easier to navigate and keeps
target definitions close to their respective source files.
Relocates test suite definitions to the `test` subdirectory.
This makes the meson.build files easier to navigate and keeps
target definitions close to their respective source files as
the number of test cases grows.
Encapsulates benchmark targets using subdir().
This is the first of four commits that splits the monolithic
`meson.build` into subdirectory files using `subdir()`. Each
subdirectory will own its own build definitions, making each
meson.build file easier to navigate and reducing coupling
between unrelated sections.
`configh_data.set('HAVE_X', 1)` was used for some features while
`set10()` was used for others. set10() is the correct choice for
boolean probes: it always emits either `#define HAVE_X 1` or
`#define HAVE_X 0`, making `config.h` self-documenting and safe
for use with `#if` regardless of whether the flag was detected.
Also refactors the `has_extensions_directories` logic to use local
boolean variables instead of reading back from `configh_data`.
Meson does not compile header files — listing them in sources: has no
effect on the build. Header dependencies are tracked automatically via
the compiler's dependency files (-MMD/-MF), so any .c file that
includes a changed header will be recompiled correctly without the
header being listed explicitly.
Removing them makes the source lists shorter and unambiguous: every
entry is a file that actually gets compiled. It also avoids the false
impression that listing a header somewhere "registers" it with the build
system in a meaningful way.
Change the return type from `int` to `enum xkb_error_code` of:
- xkb_machine_options_update_a11y_flags()
- xkb_machine_options_remap_mods()
- xkb_machine_options_update_shortcut_mods()
- xkb_machine_options_remap_shortcut_layout()
Also rename some errors:
- `XKB_ERROR_UNSUPPORTED_GROUP_INDEX` to
`XKB_ERROR_UNSUPPORTED_LAYOUT_INDEX`, because the public API use
“layout” instead of “group”.
- `XKB_ERROR_INVALID_A11Y_FLAG` to
`XKB_ERROR_UNSUPPORTED_A11Y_FLAGS`, because the issue may be a
lib version mismatch.
Created the enumeration `xkb_error_code` that will be used as a richer
return type in the 1.14 API. Since we already have error codes in the
`xkb_message_code` enumeration, the relevant codes are now shared
with the `xkb_error_code` codes.
`xkb_message_code` is not made public because:
- Only few of its codes are currently used as return values.
- Although the codes and their corresponding ID are meant to be
immutable, we are not quite there yet, because some messages
have semantic issues or lack a proper namespace. Such entries should
be kept private until they stabilize.
So the safest path is to have two enumerations:
- `xkb_error_code` is public and does follow ABI/API stability rules.
- `xkb_message_code` is internal and is a superset of `xkb_error_code`,
so that we have a unique type for use cases such as the
`check-messages` tool. `xkb_message_code` is modified so that
the public codes are now suffixed with `_`.
xkb_state_update() and xkb_machine_process_state_update() make some
functions redundant. Clean up the unreleased one.
Removed:
- xkb_state_update_enabled_controls()
- xkb_machine_update_latched_locked()
- xkb_machine_update_enabled_controls()
- xkb_machine_state_update_control()
Merge all out-of-band state updates into a single function
to get a real atomic update.
Build the update request from public types, designed to make the
ABI forward-compatible:
- `enum xkb_layout_out_of_range_policy`
- `struct xkb_layout_policy_update`
- `struct xkb_state_components_update`
- `struct xkb_state_update`
- `xkb_machine_process_synthetic()`
Rename to `xkb_events_new_batch()` in order to clarify that the
intended use is sync. We may implement an async `xkb_events_new_queue()`
in a latter release.
`xkb_server_state` name is problematic.
Both `xkb_server_state` and (historically) `xkb_state` are in fact
Mealy machines. Their inputs are raw key events and their outputs
are `xkb_event`. `xkb_state` is additionally an observable state
of both machines, with associated query functions.
`xkb_server_state` obsoleted the server API of `xkb_state`, i.e.
its state machine API. The modern API is therefore:
- state machine: `xkb_server_state`
- observable state: `xkb_state`
- input: pair (`xkb_keycode_t`, `enum xkb_key_direction`)
- output: `xkb_event`
`xkb_server_state` / `xkb_state` is just confusing since there is no
`xkb_client_state`. Also keeping “state” as a suffix for the new
server API does not help to claim that the server-side API of
`xkb_state` is deprecated.
This commit renames `xkb_server_state` and its corresponding functions
to `xkb_machine`:
- It makes its role obvious: a state machine, with all the event
processing logic.
- It pairs well with the existing `xkb_state`, which primary goal
is for querying state in clients.
- It is shorter.
- While generic, its meaning is clear in the XKB context.
Use the verb “process” instead of “update” in the API to denote
transition updates of the state machine, while the “update” verb
will be used for state machine *configuration* changes.
Thus `xkb_server_state_update_key()` is renamed to
`xkb_machine_process_key` for transitions based only on input changes,
while a future `xkb_machine_process_state_update()` will deal with
out-of-bound state changes.
Note that we will not update the Mealy machine with *explicitly* both
(input, state) entries.
xorgproto commit: 04482cdee458445eab7c6a0b6d4ea64b74387401
Relevant MR: https://gitlab.freedesktop.org/xorg/proto/xorgproto/-/merge_requests/110
Added:
- `XKB_KEY_dead_apostrophe`
- `XKB_KEY_SSHARP`
- `XKB_KEY_leftsingleanglequotemark`
- `XKB_KEY_rightsingleanglequotemark`
Changed:
- `ISO_Group_Shift` is now the canonical name of the corresponding keysym.
Previously it was `Mode_switch`, which refers to a core X group
mechanism obsoleted by XKB.
Updated the tests accordingly.
Only 2 overlays may be limitating. Let’s increase the maximum to
*8*, which seems reasonable. Note that the maximum is limited
by `KeyInfo` size; currently we could go up to 16 overlays without
incresing its size, but that seems unnessary for an already niche
feature. So keep the last free byte available for future features.
- Added keymap format-specific overlay limit.
- Added `XKB_KEYBOARD_CONTROL_OVERLAY{3,8}` to the
`xkb_keyboard_control_flags` enumeration.
X11 limits at most 1 overlay per key. However, it is possible to
handle *overlapping* overlays by setting the effective key overlay
to the *latest* corresponding activated overlay control.
E.g. if a key has both overlays 1 and 2 defined, then:
- Activate overlay 1: effective overlay is 1
- Activate overlay 2: both overlays are now active, latest
is 2 so effective key overlay is 2.
- Deactivate overlay 1: effective key overlay is still 2.
- Activate overlay 1: effective key overlay is now 1.
- Deactivate overlay 1: effective key overlay is now 2.
This commit only deals with *parsing* and *serializing*.
Overlapping overlays are store in a sparse array.
Keyboard overlays allow a key to report a different keycode when an
overlay control is active, a feature present in the X11 XKB
implementation.
This commit only implements key overlay *parsing*. It also fixes an
issue in the Macintosh US layout.
Next commits will implement the features required to bring libxkbcommon
closer to feature parity with X11 XKB implementation.
Enable the configuration of out-of-range layout handling using the
following new API:
- `state_machine::xkb_state_machine_update_control()`
- `enum xkb_keyboard_control_param`, with values:
- `XKB_KEYBOARD_CONTROL_OUT_OF_RANGE_LAYOUT_POLICY`
- `XKB_KEYBOARD_CONTROL_OUT_OF_RANGE_LAYOUT_REDIRECT`
- `enum xkb_out_of_range_layout_policy`, with values:
- `XKB_OUT_OF_RANGE_LAYOUT_WRAP`: wrap into range using integer
modulus (default, as before).
- `XKB_OUT_OF_RANGE_LAYOUT_SATURATE`: clamp into range, i.e.
invalid indices are corrected to the closest valid bound (0 or
highest layout index).
- `XKB_OUT_OF_RANGE_LAYOUT_REDIRECT`: redirect to a specific
layout index.
- Install missing setxkbmap using MacPorts.
- Update renamed formula.
- Deactivate x11comp test for now, because it has a strange error
with a duplicated "SHIFT+ALT" key type.
Simply ignore them. Note that it only supports the pattern
`STMT_TYPE ID = …;`, where `ID` is a terminal and not a general
expression. Otherwise it would be more complex to handle due to
overlapping patterns.
Simply ignore them. Note that it only supports the pattern
`STMT_TYPE [ID] { […] };`, where `ID` is a terminal and not a general
expression. Otherwise it would be more complex to handle due to
overlapping patterns.
It enables handling:
- unknown actions: turn it into a `NoAction()`;
- unknown arguments: ignore them;
- invalid argument values: invalidate the whole action,
converting it to `NoAction()`.
Make the XKB parser *lenient by default* to enable *forward*-compatibility.
The primary motivation is the evolution of the keymap format v2, which
has a living specification.
Added `XKB_KEYMAP_COMPILE_STRICT_MODE` flag to the `xkb_keymap_compile_flags`
enumeration.
Added `xkb_key_direction::XKB_KEY_REPEATED` to improve state handling
when the compositor is in charge of key repetition. This mirrors the
`wl_keyboard::key_state::repeated` Wayland value.
When a key is repeated, all filters will consume the event if it is
the same key, or ignore the event. The key actions are not triggered
again. All filters will produce only a single key repeat event if it
is the key that triggered it, except `RedirectKey()` which may also
produce surrounding state change events (e.g. when using `mods` or
`clearMods` parameters). Tweaks (mods, shortcuts) also need to be
applied again.
Also improved the tooling to handle/display key repetition correctly.
While seemingly dummy, it is useful:
- to write the explicit default value;
- to enable valid result from a computation;
- to be consistent with the handling of relative values from other
variables.
Before this commit, `RedirectKey()` without the `keycode` parameter was
accepted but obviously ineffectual, except for secondary effects such as
breaking latches.
A better default is to use the keycode of the key where the action is
located, e.g.:
```c
key <1> { [RedirectKey()] };
```
is *automatically* converted to:
```c
key <1> { [RedirectKey(keycode=<1>)] };
```
Note that the previous example is for illustration only: real use of
`RedirectKey()` with an implicit keycode would also involve `mods` or
`clearMods` parameters.
The default keycode parameter value can be set explicitly using the new
`auto` special value:
```c
key <1> { [RedirectKey(keycode=auto)] };
```
To spare parsing YAML output in zsh code, these completers rely on yq to parse
the xkbcli output.
If yq is not installed, the completion will fallback to a more rudimentary
version using only sed for parsing.
- Added `xkb_v1` and `xkb_v2` as canonical labels, in order to match
Wayland’s `wl_keyboard::keymap_format` enumeration.
- Enumerations values are not user-friendly: prefer canonical labels.
Added `xkb_state_machine_options_mods_set_mapping()` to enable
remapping modifiers combos, e.g. to make `Control+Alt` map to
`LevelThree` (`AltGr`).
This helps improving *compatibility* across platforms.
The iterator is just a container and should not be linked to
a particular state machine, so that it can be reused when a
state machine is destroyed.
Also, add a `flags` parameter to enable to configure the iterator
in the future.
Currently each toolkit (Gtk, Qt, etc.) implements its strategy to deal
with keyboard shortcuts on non-Latin layouts. There are some edge
cases where this does not work:
- Punctuation is usually not remapped. E.g. the standard Israeli
layout cannot remap its key <AD01> “slash” to the US layout “q”.
- If one have multiple Latin layouts, shorcuts may be positioned
differently. E.g. US Qwerty and French Azerty have different
positions for Ctrl+Q, while the user might want all shortcuts to
be positioned independently of the layout.
This commit enables tweaking keyboard shortcuts by adding the option
to switch to a different layout when some modifiers are active, typically
`Control`, `Alt` or `Super`. The target layout can be set individually
for each source layout, although the most common use case would probably
be the same target for all layouts.
The tweak consists of computing a *base* layout such that the *effective*
layout is the corresponding target layout.
Added to the keymap compile options API:
- `xkb_state_machine_options_shortcuts_update_mods()`
- `xkb_state_machine_options_shortcuts_set_mapping()`
Since glibc-2.43:
For ISO C23, the functions bsearch, memchr, strchr, strpbrk, strrchr,
strstr, wcschr, wcspbrk, wcsrchr, wcsstr and wmemchr that return pointers
into their input arrays now have definitions as macros that return a
pointer to a const-qualified type when the in put argument is a pointer
to a const-qualified type.
https://lists.gnu.org/archive/html/info-gnu/2026-01/msg00005.html
Resolves the following warnings:
../test/utils-text.c: In function 'uncomment':
../test/utils-text.c:82:18: warning: initialization discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]
82 | char *next = strstr(start, prefix);
| ^~~~~~
../test/utils-text.c:94:14: warning: assignment discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]
94 | next = strchr(start, 0x0a);
| ^
../test/utils-text.c: In function 'split_lines':
../test/utils-text.c:122:14: warning: assignment discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]
122 | next = strchr(start, 0x0a);
| ^
Signed-off-by: Rudi Heitbaum <rudi@heitbaum.com>
`First` is always equal to 1 and `Last` depends on the layouts count.
It enables to define e.g. a proper interpretation entry of the keysym
`ISO_Last_Group`:
```c
interpret ISO_Last_Group {
action= LockGroup(group=Last);
};
```
Note that contrary to `First`, `Last` cannot be used as an *array*
index, except in the corner case when using the RMLVO API with only
*one* group – with more groups the index gets rejected because multiple
groups per key are disallowed in xkb_symbols sections.
E.g. `symbols[Last]` is discarded (RMLVO API) or raises an error (buffer
API) in most cases. Indeed, only the RMLVO API enables to know the
number of groups *before* parsing *all* the symbols, so enabling this
feature for other APIs would require parsing the symbols twice, which is
overkill.
These constants are parsed but never used for serialization, in order to
maintain compatibility with xkbcomp and older libxkbcommon versions.
Keymap access in actions parsers enables:
- upcoming `RedirectKey()` implementation
- `Last` group index implementation
Instead of just passing the keymap, we use a keymap *augmented* with
compilation-specific data, which is intended to be extended in further
commits.
- `xkb_has_feature()`: test feature availability. This is useful when
the library is dynamically linked.
- `enum xkb_feature`: enumerate all the testable feature. Only
enumerations are implemented for now.
- `xkbcommon-features.h`: new public header.
While `xkb_mod_mask_t` is used for both real and virtual modifiers
throughout the code base, the modifiers *encoding* is considered opaque.
Thus the public API only uses `xkb_mod_mask_t` as a mask of *real* modifiers,
i.e. a mask of modifier *encodings*.
Fixed the documentation of `xkb_mod_mask_t` to clarify this point.
Rename keys and aliases with names longer than 4 characters when
serializing using `XKB_KEYMAP_FORMAT_TEXT_V1`, to ensure compatibility
with the X11 ecosystem.
The new name is one of the following 2 forms:
- If key name and keycode ≤ 0xfff: prefix character + hexadecimal keycode
- If alias or keycode > 0xfff: prefix character + hexadecimal incrementing
counter.
Prefixes are chosen in the set “!#$%&'()*+,-./0” and avoid any name
conflict.
In the very unlikely case that renaming fails due to unavailable prefixes,
the original name is kept so that Wayland clients can still parse the
keymap.
Doing a single loop is a bit faster.
This speeds up bench/compose.c by ~4% for me (compose is a heavy user of
`xkb_keysym_from_name`, the benchmark spends about ~40% in its fast
path).
Signed-off-by: Ran Benita <ran@unusedvar.com>
- The default is `debug` which is not realistic.
- I find that compilers tend to give more warnings in an optimized
build.
- In an upcoming commit I'm going to set `_FORTIFY_SOURCE=3` in tests
which wants optimizations according to glibc docs.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Before this commit there was no proper derived state initialization: its
values were simply zeroes. It works in most cases, since only LEDs would
be affected and xkeyboard-config does not ship configurations[^1] that
would activate LEDs in such conditions:
- effective group = 1;
- match some keyboard controls.
Fixed by always calling `xkb_state_update_derived()` at state creation,
in order to properly initialize the derived state.
[^1]: Well, there is only one: the `Mouse Keys` indicator. But it was not
supported by previous version of libxkbcommon.
The `xkb_state_machine` API has more potential than the `xkb_state` API
for *server* applications.
So remove the `xkb_state_options` API in favor of the corresponding
`xkb_state_machine_options` API, so that we can focus on a single
server-side API.
Prepare for the removal of the state options API:
- do not use the API
- make `--controls` requires `--legacy-state-api=false`
- keep the state *machine* options API
Enable to use `--legacy-state-api` in the following forms:
- `--legacy-state-api=true|false`
- `--legacy-state-api`: equivalent to `--legacy-state-api=true`
It avoids introducing another flag, e.g. `--no-legacy-state-api` or
`--event-state-api`.
Added documentation to illustrate how to query if a key is defined or
bound using `xkb_keymap_key_get_name()` and `xkb_keymap_num_layouts_for_key()`.
These tricks were not visible because of the missing documentation for
their return types.
Added an iterator API over keymaps keys:
- `enum xkb_keymap_key_iterator_flags`
- `xkb_keymap_key_iterator_new()`
- `xkb_keymap_key_iterator_destroy()`
- `xkb_keymap_key_iterator_next()`
It has the following pros over `xkb_keymap_key_for_each()`:
- lazy (i.e. on-demand);
- enable early termination;
- flags to control the iterator’s behavior, e.g. ascending/descending
order.
This enables to configure keyboard controls. The argument value is a
comma-separated list of the following optional values:
- `sticky-keys`
- `latch-to-lock`
- `latch-simultaneous`
Each entry can be preceded by +/- to respectively enable or disable the
corresponding option, e.g. `--controls=+sticky-keys,-latch-to-lock`. It
ensures that defaults can be explicitly overridden.
- Added `XKB_STATE_A11Y_FLAG_LATCH_TO_LOCK` member to the
`xkb_state_accessibility_flags` enum
- Added `XKB_KEYBOARD_CONTROL_A11Y_STICKY_KEYS` member to the
`xkb_keyboard_controls` enum
- Added support in X11 API
This enables to configure keyboard controls. The argument value is a
comma-separated list of an optional list of controls.
Each entry can be preceeded by +/- to respectively enable or disable the
corresponding option, e.g. `--controls=+xxx,-yyy`. It ensures that
defaults can be explicitly overriden.
This is a preparatory work: no option is currently implemented.
Added API to change global controls of the keyboard state:
- new enumeration `xkb_keyboard_controls`; we do not expose the
existing `xkb_action_controls` because it features controls whose
effects we do not (and probably will not) support
- new function `xkb_state_update_controls()`
- new function `xkb_state_serialize_controls()`
- new members `XKB_STATE_CONTROLS` in the `xkb_state_component`
enumeration.
Introduce the following public API to enable state configuration:
- `enum xkb_state_accessibility_flags`
- `xkb_state_options_update_a11y_flags()`
This only adds the initial API but does not bring any option yet.
This API is intended for *server* applications only.
Introduce the following public API to enable state configuration:
- `struct xkb_state_options`
- `xkb_state_options_new()`
- `xkb_state_options_destroy()`
- `xkb_state_new2()`
This only adds the initial API but does not bring any option yet.
This API is intended for *server* applications only.
libxkbcommon introduced flexibility by enabling multiple XKB lookup
paths. However, there is no easy way to distribute packages that provide
XKB data: such packages usually resort to modify the root XKB directory,
which is very fragile.
This can be solved by adopting the `.d` directory approach used in e.g.
`/etc`.
Added another XKB path lookup mechanism: the XKB extensions directories.
There are 2 root extensions directories, both optional:
- `XKB_CONFIG_UNVERSIONED_EXTENSIONS_PATH`, defaulting to
`<xkeyboard-config-root>` - version suffix + `.d`, e.g.
`/usr/share/xkeyboard-config.d`. It is intended as the default
extensions path.
- `XKB_CONFIG_VERSIONED_EXTENSIONS_PATH`, defaulting to
`<xkeyboard-config-root>` + `.d`, e.g. `/usr/share/xkeyboard-config-2.d`.
It is intended for backward compatibility.
`XKB_CONFIG_VERSIONED_EXTENSIONS_PATH` has higher priority than
`XKB_CONFIG_UNVERSIONED_EXTENSIONS_PATH`.
When appending the default include paths, after the extra path but before
the canonical XKB root, for each directory above: its subdirectories are
appended in lexicographic order to the include paths; unversioned
subdirectories are appended only if there is no corresponding versioned
subdirectory.
Example: given the following setup:
- `XKEYBOARD_CONFIG_ROOT` = `/usr/share/xkeyboard-config-2`
- `/usr/share/xkeyboard-config-2.d` tree:
- `p1` (installed by package p1)
- `rules/evdev.xml`
- `symbols/a`
- `p2` (installed by package p2)
- `rules/evdev.xml`
- `symbols/b`
- `/usr/share/xkeyboard-config.d` tree:
- `p1` (installed by package p1)
- `rules/evdev.xml`
- `symbols/a`
- `p3` (installed by package p3)
- `rules/evdev.xml`
- `symbols/c`
then the include path list may be:
1. `~/.config/xkb`
2. `~/.xkb`
3. `/etc/xkb`
4. `/usr/share/xkeyboard-config-2.d/p1`
5. `/usr/share/xkeyboard-config-2.d/p2`
6. `/usr/share/xkeyboard-config.d/p3`
7. `/usr/share/xkeyboard-config`
In this example:
- Package p1 distributes a layout with specific compatibility files for
xkeyboard-config v2 and fallback files for v3+. In this case the
versioned files have priority, so the unversioned
`/usr/share/xkeyboard-config.d/p1` path is not included.
- Package p2 distributes only a layout specific to xkeyboard-config v2.
- Package p3 distribute a layout for any xkeyboard-config version.
Before this commit, the only way to extend rules was to override the
default rules files by creating a file with the same name in an include
directory with a higher priority than the default rules file. However,
this prevented to *compose* any rules extension by other mean than
explicit includes/inlining all rules in the higher priority rules files.
This commit introduces a new lookup mechanism that enables to extend
rules without altering the main rules file and facilitate *rules
composition*. The main motivation is to support composing custom rules
with the upcoming XKB *extensions mechanism*. This is done by enabling
*partial* rules files `<rules>.pre` and `<rules>.post`.
The previous lookup mechanism:
1. Find the rules file `<rules>`
2. Fail if not found
3. Otherwise resolves the rules
4. Fail if any required component is missing
5. Otherwise return the resolved KcCGST names
is replaced with:
1. Find the main rules file `<rules>`
2. Fail if not found
3. For each include directory:
1. Try to find the *partial* rules file `<rules>.pre` in the directory
2. If none is found, continue to the next directory
3. Otherwise resolve the rules from `<rules>.pre` by appending the
resolved components to the previous result
This may combine *multiple* `<rules>.pre` files.
4. resolve the rules from `<rules>` by appending to the previous result;
5. same as step 3, but with the rules file `<rules>.post`.
Thus the final rule set is equivalent to:
! include <include path 1>/rules/<rules>.pre // only if defined
…
! include <include path n>/rules/<rules>.pre // only if defined
! include <rules> // main rules file
! include <include path 1>/rules/<rules>.post // only if defined
…
! include <include path n>/rules/<rules>.post // only if defined
Example: given the following tree:
- `~/.config/xkb/rules`
- `evdev.post`
- `/etc/xkb/rules`
- `evdev.pre`
- `evdev.post`
- `/usr/share/xkeyboard-config-2/rules`
- `evdev`
the new mechanism is equivalent to parsing the following rules file:
// .pre rules
include %E/evdev.pre // only .pre file
// main rules
include %S/evdev
// .post rules
include %H/.config/xkb/evdev.post // higher priority than /etc/xkb
include %E/evdev.post
This fixes the lookup of the following entry from the standard `en_US`
Compose file:
<Multi_key> <minus> <minus> <minus> : "—" U2014 # EM DASH
Before this commit the sequence was found with `U2014` but not with
“—” nor `emdash`.
Fixed by looking at the entry’s UTF-8 string as well.
Default includes paths are delayed and added only if necessary.
It is more efficient for most clients, which only get the keymap from
the server: it avoids unnecessary allocations and file system queries.
It also avoid the corner case where a containerized app lack access to
the XKB directories.
There might be an issue in case the environment variables relevant to
the include paths change between the context initialization and the
call to `xkb_context_include_path_append_default()`. However this seems
very unlikely and would have already triggered issues when the `%`-
expansion is used.
Let's make sure we destroy our wayland resource objects in a nice order to avoid
any errors at shutdown time. Fixes errors like:
$ xkbcli interactive-wayland
[...]
[destroyed object]: error 6: surface was destroyed before its role object
Previously we would try to commit an incomplete xdg_surface. This is illegal in
xdg_shell and results in a "not constructed" xdg_wm_base error. Shift the
ifdef's around a bit so that only the actually relevant code for the keymap-dump
utility is compiled.
Do not perform the test if the build machine cannot run host binaries.
Instead fallback to detect if the `newlocale` lib is *known* to fail on
missing locales, as we expect.
Do not use `newlocale()` if it accepts missing locales. It happens with
muslc, while glibc works as expected.
It fixes the incorrect behavior of the fallback introduced in
135b3204a1.
Some setups use the assumption that the canonical XKB root is always
the legacy X11 one, but this is no longer true since xkeyboard-config
2.45, where the X11 path is now a mere symlink to a dedicated data
directory for xkeyboard-config.
Fixed by using a fallback to the legacy X11 path for misconfigured
setups where the canonical XKB root is not available.
This fallback can still be skipped if the environment variable
`XKB_CONFIG_ROOT` is deliberately set to an empty string.
This information is quite important to investigate bugs and should not
require the bloat of the debug log level.
Also report include paths additions attempts in registry.
There are 4 mandatory *canonical key types* in the XKB protocol:
- `ONE_LEVEL`
- `TWO_LEVEL`
- `ALPHABETIC`
- `KEYPAD`
They are always present in the keymap generated from xkeyboard-config.
But since 31900860c6 we drop unused key
types by default, which may happen for the types hereinabove with e.g.
4+ level layouts like `es`.
In theory these types are automatically filled by libX11 if missing,
but there are some bugs in the X11 ecosystem that prevents the keymap
to be properly uploaded in the X server, leading to errors when
retrieving it with libxkbcommon-x11. See:
https://gitlab.archlinux.org/archlinux/packaging/packages/libxkbcommon/-/issues/3
The following fixes were filed to fix the issues:
- https://gitlab.freedesktop.org/xorg/lib/libx11/-/merge_requests/292
- https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/2082
- https://github.com/xkbcommon/libxkbcommon/pull/871
However it’s not clear when new versions of libX11 and xserver will be
released. So this commit is a hack to ensure that we do not drop the
XKB canonical key types, as an effort to reduce breakage.
WARNING: contrary to `xkbcomp`, we do not supply these types if they
are missing, because a keymap that uses them (explicitly `type="…"` or
implicitly with automatic types) without providing them is considered
buggy. The only exception is if no key type is provided, a default one-
level type `ONE_LEVEL` is provided and assigned to all keys.
Before this commit there were 2 issues in action comparison:
- Private actions comparison wrongly compared pointers;
- Custom action types were mistakenly interpreted as errors and thus
always failed comparison.
Also added assertions on the action type enum so that the relevant code
must be updated when the enum changes.
a83a482c71 introduced the new tools
`xkbcli-{interactive,dump-keymap}`, along with their corresponding tests.
However, these tools cannot be tested because they hardcode the paths of
the `xkbcli-{interactive,dump-keymap}-{wayland,x11}` tools they call to
the *install* directory, while we need to use those of the *build*
directory.
It seems that it got unnoticed thanks to:
- the special settings in the CI: unset `*DISPLAY` environment variables;
- a local installation of the development version in the developer’
setup.
The *default* XKB root directory is now set from the *most recent*
xkeyboard-config installed package, in case [multiple versions] are
installed in parallel. It also try to use the package-specific XKB root
(if available) rather than the X11 root, respectively defined by the
pkg-config variables `xkb_root` (xkeyboard-config ≥ 2.45) and `xkb_base`.
If no such package is found, it fallbacks to the historical X11
directory, as previously.
[multiple versions]: https://xkeyboard-config.freedesktop.org/doc/versioning/
When compiling a keymap from text, some items may be unnecessary in the
final keymap, i.e. they do not affect the keymap behavior:
- unused key types;
- unused keysym interpretations.
Deactivate the serialization of these items *by default* and add a new
flag to enable it for debugging.
This greatly improves the keymap serialization: 1.22× speedup and
about 5% less allocations. The resulting keymap is also a bit faster
to parse.
Another improvement is that it eases keysym names migrations (removal
and additions) by using only keysym numeric values. This requires some
care, i.e. `NoSymbol` must be serialized with its name and not its
value 0x0, because xkbcomp and libxkbcommon < 1.12 would interpret the
numeric value as `XKB_KEY_0`.
Keysyms written as single decimal digits are interpreted in the range
`XKB_KEY_0`..`XKB_KEY_9`, consistent with the general interpretation
`<name>` -> `XKB_KEY_<name>`, e.g.:
- `1` → `XKB_KEY_1`
- `a` → `XKB_KEY_a`
However, before this commit integers in the range 0..9 in *any format*
were treated as digit keysyms, which is wrong if the number is written
with 2+ characters. E.g. the following were wrongly treated as the
keysym digit `XKB_KEY_1`: `01`, `0x1`.
Fixed by introducing a new token type to handle this corner case.
This is a preparatory work to enable serializing keysyms as numbers.
Since it is not usual to lookup for keys by their names, we can drop it
to save allocations (about 2KiB on usual keymaps).
We use an union of the LUT with the aliases array and try to reuse the
memory allocated by the LUT. We only do so if the space is trivially
available: either before the first alias entry or after the last entry,
which is possible in practice, so that we get the best performance. Else
we allocate a new array.
Replace linear search with O(1).
On my setup I get 1.05x speedup compared to previous commit and
1.09x speedup compared to 44fad8a067
(no keycode name LUT).
Contrary to Xorg’s xkbcomp, keys and aliases share the same
namespace. So we need to resolve name conflicts as they arise,
while xkbcomp will resolve them just before copying aliases into
the keymap.
The following example:
xkb_keycodes {
<A> = 8;
<B> = 9;
alias <B> = <A>;
};
will resolve in libxkbcommon to:
xkb_keycodes {
<A> = 8;
alias <B> = <A>;
};
while in Xorg’s xkbcomp:
xkb_keycodes {
<A> = 8;
<B> = 9;
};
The former does result in the intended mapping.
In practice, there is no such conflict in xkeyboard-config.
Another corner case is that now an alias can override a key even if
proved invalid aftwerwards, e.g.:
xkb_keycodes {
<A> = 1;
<B> = 2;
alias <B> = <C>; /* Override key, even if invalid entry */
};
results in:
xkb_keycodes {
<A> = 1;
};
This should be considered a bug in either the keycodes or rules files,
not libxkbcommon.
Replace linear search of keycode names with O(1).
Given that:
- the number of atoms is usually < 1k;
- the keycode section usually appears first;
then the first atoms will be mostly the keycodes and their aliases,
so we can achieve O(1) lookup of the key names by using a simple
atom → keycode array.
The LUT will be used also to process `xkb_symbols` for further speedup.
On my setup I get a 1.039x speedup at the costs of less than 1%
additional allocations.
In 502e9e5bff we restricted the supported
keycode range in order to avoid memory exhaustion and inefficient storage
in sparse arrays. This solution enabled keycodes up to 0xfff, which
seemed good enough at the time.
However there are huge keycodes in use in the wild, e.g. in WebOS.
So let’s enable the whole keycode range by using 2 methods of storage:
- “Low” keycodes (≤ 0xfff): stored contiguously as before at indexes
[0..num_keys_low); fast O(1) access.
- “High” keycodes (> 0xfff): stored noncontiguously at indexes
[num_keys_low..num_keys); slow access via binary search.
This tool enables simple analysis of XKB files with a YAML or DOT output:
- resolve XKB paths;
- list sections of a file;
- list the includes of each section;
- optionally process each include recursively.
Additionally, the RDF Turtle output enables to query using the powerful
SPARQL language.
The tool is for internal use only for now, so that we can test it in
various use cases before deciding if making it public.
This enables to use a local Unicode Character Database instead of ICU via
Python.
Useful when ICU is not available or when it does not support the latest
Unicode version.
Data are available at: https://www.unicode.org/Public/<UNICODE_VERSION>/ucd
This enables to provide semantics and to ensure we use the values
uniformly in the code base.
While one could expect that verbosity `0` silences the logging, it
is actually our default verbosity level for a long time. So this
commit does not change that in order to avoid possible breakage.
Silencing the logging is achieved by using a negative value for the
verbosity level.
Before this commit, loading a Compose file based on the locale would fail
if the locale is not in the X11 Compose locale registry. While there are
workarounds (e.g. `~/.XCompose` file or `$XCOMPOSEFILE`), it does not
work if the corresponding file has `include "%L"`.
This commit adds the fallback `en_US.UTF-8` in case the locale is
installed but not registered in the X11 Compose locale registry. The
choice is motivated by the fact that most locales use `en_US.UTF-8`
anyway.
Ideally we should have a mechanism to extend the Compose locale registry
at the *system* level. Mechanisms at the user level (e.g. custom Compose
file, environment variable) are deemed sufficient.
We could still improve it by first trying to fallback to the locale
without the country bits, but there is no function to do such function
in the stdlib and we do not want to mess with locales manually.
Unfortunately is not possible to test it in our test suite. One can
still check it works following these instructions:
1. Create a custom locale, e.g. `en_XX.UTF-8`. `glibc-i18ndata` or
similar package may be required to get the required files in
`/usr/share/i18n/`.
`sudo localedef -i en_US -f UTF-8 en_XX.UTF-8`
2. `xkbcli compile-compose --verbose --locale en_XX.UTF-8`
Retrospectively, `xkb_keymap_mod_get_mask()` should have used a
modifier index rather that a modifier name in its type. Since we
already published a version with this API, it’s too late to change
that, so instead add a new function `xkb_keymap_mod_get_mask2()`.
Added the `--keymap` option to enable providing the keymap to use
instead of the one from the display server. Implies `--local-state`.
Note that the other RMLVO options do not need to be implemented, since
one can simply pipe a keymap from `xkbcli compile-keymap`.
The `--keymap` option is also useful for users that do not have the
input permissions to access evdev API.
- Fixed incorrect numeric keysym parsing due to `errno` not being
reset.
- Added parsing keysym as fallback when `--keysym` is omitted.
- Added arguments tests.
- Fixed tools test Python types.
- Updated documentation.
Improved the output of events for `xkbcli interactive-*` tools:
- Added `--multiline` to enable multiline event output, which provides
more details and does not have formatting limitations.
- Added `--uniline` to enable uniline event output. While this is the
current default, future versions may switch to multiline display.
`xkbcli interactive-*` tools:
- Print key release events. This is particularly useful when analyzing
the output sent by another person, in order to know the exact key
sequence.
This commit enables to use the pattern `Group<INDEX>` for any valid
group index `<INDEX>`.
Note that the original code in xkbcomp allows constants up to
`Group8`, but then will fail if the resulting group is > 4.
There does not seem to be any use case for this for such “feature”;
it seems rather to be a relic from times were the 4-groups limit
was not hopelessly fixed in X.
So for consistency in our code base, starting with this commit we now
disallow `Group5`..`Group8` for keymap format v1, since it is limited
to 4 groups.
Also fixed a regression in the serialization of group action, when
the group is relative.
Before this commit there was a mix between the two forms.
While “indexes” is correct, “indices” is more usual and also
the historical form used in this project.
Before this commit, some code points could be converted to deprecated
keysym. This is incorrect, because the relevant keysyms are all
deprecated because their mapping to Unicode is uncertain!
Ensure that `xkb_utf32_to_keysym()` never returns deprecated keysyms,
because there is either another non-deprecated keysym or in last resort
we always have the correct keysym available in the Unicode keysym range.
Consider the following use cases:
1. If `Caps_Lock` is on the second level of some key, and `Shift` is
latched, pressing the key locks `Caps` while also breaking the `Shift`
latch, ensuring that the next character is properly uppercase.
2. On the German E1 layout, `ISO_Level5_Latch` is on the third level
of `<AC04>`. So if a level 3 latch is used to access it, the level 5
must break the previous level 3 latch, else both latches would be
active: the effective level would be 7 instead of the intended 5.
Both uses cases can be implemented using existing features:
- multiple actions per level;
- `VoidAction()`: to break latches.
Prior to this commit, the sequences:
- 1. latch A ↓
2. latch B ↓
3. latch B ↑
4. latch A ↑
- 1. latch A ↓
2. latch B ↓
3. latch A ↑
4. latch B ↑
would result in only B being latched, because the XKB protocol specifies
that latches are triggered only if keys are *sequentially* tapped, i.e.
a strict sequence of press and release of each key.
It seems an unnecessary limitation:
- `SlowKeys` and `XkbAX_TwoKeys` are the proper accessibility features
to control accidental key presses, not latches nor `StickyKeys`.
- Latches are also used outside their original accessibility role.
A user may activate multiple latch keys simultaneously:
- same hand: two latch keys being close to each other;
- different hand: two keys being activated independently.
Changed the latching behavior so that the rules used to break a latch
are the same than those used to prevent it. Depressing and releasing two
latching keys simultaneously will now activate both latches, as expected.
Since this is a breaking change, it is enabled only by the keymap format
`XKB_KEYMAP_FORMAT_TEXT_V2`.
Previously we use inlined version of the corresponding filter functions
of the `SetMods()` and `LockMods()` actions, but they were incomplete
and did not set some fields (`priv`, `refcnt`) properly. Also, it is
error-prone: it requires discipline to keep it in sync.
E.g. before this commit, converting to `LockMods()` would always try to
unlock `CapsLock` due to the wrong value of the `priv` field.
Fixed by using the corresponding filter functions directly, so that we
always mutate the filter properly, as in `xkb_filter_group_latch_func`.
Traditionally, *virtual* modifiers were merely name aliases for *real*
modifiers (X *core* modifiers), e.g. `NumLock` was usually mapped to
`Mod2` (see `modifier_map` statement). Virtual modifiers that were never
mapped to a real ones had no effect on the keymap state.
xkbcommon already supports the concept of “pure” virtual modifiers, i.e.
virtual modifiers that are *encoded* using the full 32-bit range, not
just the first 8 bits corresponding to the real modifiers.
But until this commit, one had to declare such mapping *explicitly*:
e.g. `virtual_modifiers M = 0x100;`. This has at least two drawbacks:
- Numerical values may look quite arbitrary and are not user-friendly.
It’s OK in the resulting compiled keymap, but it requires careful sync
between sections when developing KcCGST files.
- If the modifier is *also* mapped *implicitly* using the traditional
`vmodmap`/`modifier_map`, then both mappings are OR-combined.
This patch enables to automatically map unmapped virtual modifiers to
their *canonical* mapping, i.e. themselves: their corresponding virtual
and real modifier masks are identical: `1u << mod_index`.
Since this feature is incompatible with X11, this is guarded by requiring
at least keymap text format **v2**.
Note that for now, canonical virtual modifiers cannot be used in an
interpret action’s `AnyOf()`. An interpret action for a canonical virtual
modifier must be `AnyOfOrNone()` to take effect:
virtual_modifiers APureMod, …;
interpret a+AnyOfOrNone(all) {
virtualModifier= APureMod;
action= SetMods(modifiers=APureMod);
};
The above adds a virtual modifier `APureMod` for keysym `a`. It will be
canonical iff it is not mapped implicitly.
Some keyboard layouts use `ISO_Level3_Latch` or `ISO_Level5_Latch` to
define “built-in” dead keys:
- they do not rely on the installation of custom Compose file;
- they do not clash with other layouts.
However, layout projects usually want the exact same behavior on all OS,
but the XKB latch behavior (often misunderstood) also acts as a *set*
modifier, which is not expected.
The usual behavior of a dead key on Linux, macOS and Windows is:
- latch on press;
- deactivate as soon as another (non-modifier) key is pressed.
Added the parameter `latchOnPress` to `LatchMods()` to enable the
aforementioned behavior.
As it is incompatible with X11, this feature is available only using the
keymap text format v2.
[XKB protocol key actions]: https://www.x.org/releases/current/doc/kbproto/xkbproto.html#Key_Actions
When using `XKB_KEYMAP_FORMAT_TEXT_V1`, multiple actions per level are now
serialized using `VoidAction()`, in order to maintain compatibility with X11.
It enables e.g. to deactivate CapsLock on press rather than
on release, as in other platforms such as Windows.
The specification of `LockMods()` is changed to:
- On key *press*:
- If `unlockOnPress` is true and some of the target modifiers were
*locked* before the key press, then unlock them if `noUnlock` false.
- Otherwise:
- add target modifiers to *depressed* modifiers;
- if `noLock` is false, add target modifiers to the *locked*
modifiers.
- On key *release*:
- If `unlockOnPress` is true and triggered unlocking on key press, do
nothing.
- Otherwise:
- remove modifiers from the *depressed* modifiers, if no
other key that affect the same modifiers is down;
- if `noUnlock` is false and if any target modifiers was locked before
the key press, *unlock* them.
It fixes a [12-year old issue] inherited from the X11 ecosystem,
by extending the [XKB protocol key actions].
As it is incompatible with X11, this feature is available only using the keymap
text format v2.
[12-year old issue]: https://gitlab.freedesktop.org/xorg/xserver/-/issues/312
[XKB protocol key actions]: https://www.x.org/releases/current/doc/kbproto/xkbproto.html#Key_Actions
It enables to use e.g. the combination `Control + Shift` *alone* to
switch layouts, while keeping the use of `Control + Shift + other key`
(typically for keyboard shortcuts).
The specification of `LockGroup()` is changed to:
- On key *press*:
- If `lockOnRelease` is set, then key press has no effect.
- Otherwise:
- if the `group` is absolute, key press sets the *locked* keyboard group to
`group`;
- otherwise, key press adds `group` to the *locked* keyboard group.
In either case, the resulting *locked* and *effective* group is brought back
into range depending on the value of the `GroupsWrap` control for the keyboard.
- On key *release*:
- If `lockOnRelease` is not set, then key release has no effect.
- Otherwise, if any other key was *pressed* after the locking key, then
key release has no effect.
- Otherwise, it has the same effect than a key press *without* `lockOnRelease`
set.
This is really useful for people coming from other platforms, such as
Windows.
It fixes a [20-year old issue] inherited from the X11 ecosystem, by
extending the [XKB protocol key actions].
As it is incompatible with X11, this feature is available only using the keymap
text format v2.
[20-year old issue]: https://gitlab.freedesktop.org/xorg/xserver/-/issues/258
[XKB protocol key actions]: https://www.x.org/releases/current/doc/kbproto/xkbproto.html#Key_Actions
Enabled specifying a layout index for each option, so that it applies
only if the layout matches. The layout index is specified by appending
immediately after the option name the `!` specifier delimiter and then
the layout index, in decimal form and 1-indexed.
Note that `!` was chosen instead of the usual `:` specifier delimiter,
because some options contains `:`, e.g. `grp:menu_toggle`. `!` *cannot*
clash with component names, because `!` is a token in the rules files
and thus cannot be used as in component names. It is also vaguely similar
to `:`, compared to e.g. `@` or `#`.
Example: given the following rules:
! layout[any] option = symbol
* opt1 = +s1:%i
l2 opt2 = +s2:%i
it may result in the following configurations:
| Layout | Option | Symbols |
| -------- | -------- | ------------ |
| `l1` | `opt1` | `+s1:1` |
| `l2` | `opt1` | `+s1:1` |
| `l1` | `opt2` | `` |
| `l2` | `opt2` | `+s2:1` |
| `l1,l2` | `opt1` | `+s1:1+s1:2` |
| `l1,l2` | `opt1!1` | `+s1:1` |
| `l1,l2` | `opt1!2` | `+s1:2` |
| `l1,l2` | `opt2` | `+s2:2` |
| `l1,l2` | `opt2!1` | `` |
| `l1,l2` | `opt2!2` | `+s2:2` |
Before this commit, the following rule would always match:
! model = symbols
* = s:%i
and set symbols to `s:1`, but the `:%i` is aimed to be used only when
the rules header specifies the layout or the variant.
Let’s be strict and disallow matching this kind of buggy rule. Emit
an error message so that we can detect it.
Before this commit, the API to work with RMLVO was quite minimal: it
only uses raw strings from the `xkb_rule_names` struct. However:
- it forces the users to deal with error-prone string formatting;
- it does not enforce tying together layouts and variants;
- it limits adding new features by requiring defining delimiter
characters and the corresponding parsing.
Added the following API:
- `xkb_rmlvo_builder_new()`
- `xkb_rmlvo_builder_append_layout()`
- `xkb_rmlvo_builder_append_option()`
- `xkb_rmlvo_builder_unref()`
There is no intermediate `layout` nor `option` object, in order to
to keep the API simple. The only foreseen extension is enabling
configuring layout-specific options.
The upcoming raise of the maximum groups count will require to use
numeric group indexes instead of the syntax `GroupN` if groups > 8.
Let’s not bother with handling two cases (group count ≤ 8 or > 8) and
always serialize group indexes as numeric values.
We need to track the compatibility with X11 with as much details as
possible:
- Transparency;
- Facilitate answering issues by just pointing to the relevant page item;
- Efficient reference documentation for development.
- Added `XKB_KEYMAP_FORMAT_TEXT_V2`.
- Made `xkb_keymap_new_from_names()` use the new keymap format.
- Made the tools default to the new keymap format for input.
This is in preparation for changes in the parsing & state handling.
For now it changes nothing.
- Deprecate `xkb_keymap_new_from_names()` in favor of
`xkb_keymap_new_from_names2()`
- Add new changelog fragment type `deprecated`.
- Change documentation to use the new function.
Enable to use the same `include` features than *rules* files in
*keymap components*:
- *`%`-expansion*: `%H` home directory, `%S` sytem root and `%E` extra.
- absolute file paths.
This is useful if one wants to overwrite the system file with a user
config (i.e. same name, but in `~/.config/xkb`), but still include the
system file:
```
// File: ~/.config/xkb/symbols/de
xkb_symbols "basic" {
include "%S/de(basic)"
key <AB01> { [z, Z] };
key <AD06> { [y, Y] };
}
````
Without the commit, using a mere `include "de(basic)"` would result in
an include loop.
Refactored by using the same code for rules and keymap components.
Since xkeyboard-config 2.45 uses a symlink for:
/usr/share/pkgconfig/xkeyboard-config.pc
/usr/share/X11/xkb
but meson cannot overwrite < 2.45 which used a regular file and directory.
Fixed by first removing the files. This is still a hack and we should
use a proper package update.
Use a polyfill for C23 checked arithmetic. This is a bit paranoid, as we
expect the user to use only 32 bit integers, so the signed 64 bit integer
we use to store the result should be more than enough.
Use jtckdint v1.0:
- repository: https://github.com/jart/jtckdint
- commit: 339450d13d8636f05dcb71ba36efddb226db481e
- removed all C++-specific code
This ensures that it is always possible to override previous definitions,
for example when `include`ing the system Compose file.
Signed-off-by: Jules Bertholet <julesbertholet@quoi.xyz>
Previously if a new sequence did not produce a keysym or a string, the
corresponding property was not overriden, possibly leaking the previous
entry.
- Fixed by always writting all the properties.
- Also try to reuse the previous string entry, if possible, so that we
avoid allocating.
Previous solution with `sed` is both overkilled and unreliable.
Prefer a hack using Doxygen’s buit-ins, although we are not sure
these are much more stable… 😓
- Remove the `ENABLE_PRIVATE_APIS` guards on modmaps printing
functions, since we use no private symbols.
- Copy the `--print-modmaps` options from `interactive-evdev` into
`compile-keymap` and rename it to `--modmaps`. This enables a
feature similar to `xmodmap -pm`.
Added `VoidAction()` action to match the keysym pair
`NoSymbol` / `VoidSymbol`.
It enables overriding a previous action and breaks latches.
This is a libxkbcommon extension. When serializing it will be converted to
`LockControls(controls=none,affect=neither)` for backward compatibility.
We cannot serialize it to `NoAction()`, as it would be dropped in e.g.
the context of multiple actions.
Up to now, the “server state” `xkb_state` API only offered one entry
point to update the server state – `xkb_state_update_key`, which reflects
the direct keyboard keys state. But some updates come out-of-band from
keyboard input events stream, for example, a GUI layout switcher.
The X11 XKB protocol has a request which allows for such updates,
`XkbLatchLockState`[^1], but xkbcommon does not have similar
functionality. So server applications ended up using
`xkb_state_update_state` for this, but that’s a function intended for
client applications, not servers.
Add support for updating the latched & locked state of the mods and
layout. Note that the depressed states cannot be updated in this way --
XKB does not expect them to be updated out of band.
[^1]: https://www.x.org/releases/X11R7.7/doc/kbproto/xkbproto.html#Querying_and_Changing_Keyboard_StateFixes: #310
Signed-off-by: Ran Benita <ran@unusedvar.com>
Co-authored-by: Ran Benita <ran@unusedvar.com>
Co-authored-by: Pierre Le Marre <dev@wismill.eu>
Added a dedicated API to query modifier masks rather than relying on
a hack using `xkb_state_update_mask` and `xkb_state_serialize_mods`.
Furthermore, this hack may not work in the future if we remove virtual
mods resolution in `xkb_state_update_mask` to avoid corner-cases issues.
The following modifiers names in `xkbcommon/xkbcommon-names.h` are now deprecated:
- `XKB_MOD_NAME_ALT`: use `XKB_VMOD_NAME_ALT` instead.
- `XKB_MOD_NAME_LOGO`: use `XKB_VMOD_NAME_SUPER` instead.
- `XKB_MOD_NAME_NUM`: use `XKB_VMOD_NAME_NUM` instead.
It's not much, but instead of
xkbcommon: ERROR: [XKB-769] (unknown file):5:25: syntax error
we get
xkbcommon: ERROR: [XKB-769] (unknown file):5:25: syntax error, unexpected +, expecting INTEGER
which is at least a little helpful.
Signed-off-by: Ran Benita <ran@unusedvar.com>
It doesn't support `%define parse.error detailed` which we want to use.
If needed, we can probably bring back support using some macro hackery.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Since we accept numeric values for the vmod mask in the keymap, we may
have extra bits set that encode *no* real/virtual modifier. Keep them
unchanged for consistency.
E.g. the following keymap:
xkb_keymap {
xkb_keycodes { <a> = 38; };
xkb_symbols {
virtual_modifiers X = 0xf0000000;
key <a> { [ SetMods(mods = 0x00001100) ] };
};
};
would compile to:
xkb_keymap {
xkb_keycodes { <a> = 38; };
xkb_symbols {
virtual_modifiers X = 0xf0000000;
// Internal state
key <a> { [ SetMods(mods = 0xf0001000) ] };
// Serialization
key <a> { [ SetMods(mods = 0x00001100) ] };
};
};
Currently there are 2 issues with the handling of virtual modifiers
in the keyboard state:
1. We assume that the input modifiers masks encode the indexes of all
the modifiers of the keymap, but this is true only for the *real*
modifiers (at least in xkbcommon and X11). Indeed, since the virtual
modifiers *indexes* are implementation-specific, the input modifier
masks merely *encode* the modifiers via their *mapping*.
Consider the following keymap:
```c
xkb_keymap {
xkb_compat { virtual_modifiers M1 = 0x100; };
xkb_types { virtual_modifiers M2 = 0x200; };
};
```
Now to illustrate, consider the following 2 implementation variants
of libxkbcommon (assuming indexes 0-7 are the usual real modifiers):
1. Process `xkb_compat` then `xkb_types`.
M1 and M2 have the respective indexes 8 and 9 and map to
themselves (with the current assumption about mask denotation).
2. Process `xkb_types` then `xkb_compat`.
M1 and M2 have the respective indexes 9 and 8 and map to each
other.
With the current `xkb_state_update_mask`, implementation 2 will swap
M1 and M2 (compared to impl. 1) at each update! Indeed, we can see that
`xkb_state_serialize_mods` doesn’t roundtrip via `xkb_state_update_mask`.
2. We assume that modifier masks use only bits denoting modifiers in
the keymap, but when parsing the keymap we accept explicit virtual
modifiers mapping of arbitrary values.
E.g. if `M1` is the only virtual modifier and it is defined by:
```c
virtual_modifiers M1 = 0x80000000; // 1 << (32 - 1)
```
then the 32th bit of a modifier mask input does *not* denote the
32th virtual modifier of the keymap, but merely the encoding of the
mapping of `M1`.
So when calling `xkb_state_update_mask`, we may discard some bits of
the modifiers masks and end up with an incorrect state.
These 2 issues may break interoperability with other implementations of
XKB (e.g. kbvm) and make pure virtual modifiers handling fragile.
We introduce the notion of *canonical state modifier mask*: the mask
with the smallest population count that denotes all bits used to encode
the modifiers in the keyboard state. It is equal to the bitwise OR of
real modifiers mask and all the virtual modifiers mappings.
This commit fixes the 2 issues by making *weaker* assumptions about the
input modifier masks:
1. Modifiers may map to arbitrary values, not only real modifiers.
2. Input modifier masks merely encode modifiers via their *mapping*:
- *real* modifiers map to themselves;
- *virtual* modifiers map to the bitwise OR of their *explicit*
mapping (via `virtual_modifiers`) and their *implicit* mapping (via
keys’ real and virtual modmaps).
- modifiers indexes are implementation-specific.
Since the implementation before this commit also resolved virtual
modifiers to their mappings, we continue doing so, but using only the
bits that are *not* set in the canonical state modifier mask, so that
we enable roundtrip of `xkb_state_serialize_mods` via
`xkb_state_update_mask`.
3. Input modifier masks do not denote modifiers indexes (apart from real
modifiers), so it is safe to discard only the bits that are not set
in the canonical state modifier mask.
Currently we only ignore key type map entries with non-zero mods and
with a zero modifier mask. However, the XKB protocol states ([source]):
> Map entries which specify unbound virtual modifiers are not considered.
So we currently handle `map[Unbound]` key type map entries (all modifiers
unbound) but not `map[Bound+Unbound]` entries (mix of bound and unbound
modifiers).
Fixed by properly checking unbound modifiers on each key type map entry.
This also fixes a test that was accidentally passing.
[source]: https://www.x.org/releases/X11R7.7/doc/kbproto/xkbproto.html#:~:text=Map%20entries%20which%20specify%20unbound%20virtual%20modifiers,not%20considered
When querying for a modifier mapping, we should treat all modifiers
equally. So simply store real modifier mapping as we do for the virtual
ones.
Also fixed useless boolean conversions.
- Initialize with multiple recurrent questions.
- Document how to get the vmod → rmod mapping.
Rational: Some applications may need it to interface with legacy
code.
Fix compilation error when building for Windows with clang-cl:
../subprojects/libxkbcommon-xkbcommon-1.9.0/test/keysym-unicode.c:11:1: error: return type defaults to 'int' [-Wimplicit-int]
../subprojects/libxkbcommon-xkbcommon-1.9.0/test/keysym-unicode.c:11:1: warning: function declaration isn't a prototype [-Wstrict-prototypes]
`LevelThree` and `LevelFive` are mapped respectively to `Mod5` and
`Mod3`. This is counter-intuitive but set so for ages and probably hard-
coded in some places.
Before this commit, including a *default* map, i.e. without an explicit
section name (e.g. `include "au"` vs `include "au(basic)"`) would match
the first section of the first matching file in the XKB include paths,
even if this section is not an *explicit* default map (i.e. tagged with
`default`) but an *implicit* default map (i.e. the first map of the
file, i.e. a weak match).
It makes user configuration risky: say a user wants to create a custom
version `au(custom)` of the `au` layout:
- `./config/xkb/symbols/au`: custom layout in section “custom”.
- `/usr/share/X11/xkb/symbols/au`: system layout, with *default* section
“basic”.
In this setup *any* layout that imports the default map from `au` would
in fact import the *implicit* default map `au(custom)` instead of the
*explicit* default map `au(basic)`.
This incorrect behavior may thus break setups with multiple layouts.
This is especially true for symbols files such as: `pc`, `us` or `latin`.
Fixed by trying harder to found the exact default map, defaulting to the
old behavior (weak match) only if no *explicit* default map (exact match)
has been found in the XKB include paths.
Previously `xkb_state_key_get_syms()` did not perform capitalization
tranformation, while `xkb_state_key_get_one_sym()` does perform it.
This is unfortunate if we want to promote the use of multiple keysyms
per levels.
The API make it difficult to change now though: we return a pointer to
an immutable array rather than filling a buffer. While we could use an
internal buffer in `xkb_state`, this option would limit the API to
*sequential* calls of `xkb_state_key_get_syms()` or require some buffer
handling (e.g. rotation).
Instead we now store the capitalization directly in `xkb_level`. We
modified `xkb_level` like so (see below for discussion about the size):
```diff
struct xkb_level {
- unsigned int num_syms;
+ uint16_t num_syms;
- unsigned int num_actions;
+ uint16_t num_actions;
+ union {
+ /** num_syms == 1: Upper keysym */
+ xkb_keysym_t upper;
+ /** num_syms > 1: Indicate if `syms` contains the upper case
+ * keysyms after the lower ones. */
+ bool has_upper;
+ };
union {
xkb_keysym_t sym; /* num_syms == 1 */
xkb_keysym_t *syms; /* num_syms > 1 */
} s;
union {
union xkb_action action; /* num_actions == 1 */
union xkb_action *actions; /* num_actions > 1 */
} a;
};
```
- When `level.num_syms` <= 1, we store the upper keysym in `level.upper`.
- Else if there no cased syms, we set `level.has_upper` to false.
- Else if there are some cased syms, we set `level.has_upper`` to `true`
and we double the original size of `level.s.syms`, but *without*
modifying `level.num_syms`. We then append the transformed keysyms
right after the original ones, so that we can access them by a simple
pointer operation: `level.s.syms + level.num_syms`.
The memory footprint is *unchanged*, thanks to the reduced fields for
actions and keysyms counts.
In preparation to support capitalization in `xkb_state_key_get_syms()`,
this commit reduces the number of supported actions and keysyms per
level, going from UINT_MAX to UINT16_MAX. This is most likely still more
than enough and could be even reduced further, but deemed unnecessary
at the moment: alignment of `struct xkb_level` is driven by the fields
`a` and `s`.
- Switched the item count type from `unsigned int` to `uint16_t`.
- Introduced `xkb_{action,keysym}_count_t` type for the respective item
count for exact typing.
- Added relevant bounds checks.
The merge modes tests C file is now only generated locally, because it
is too large. The generator Python script requires Jinja2, so the test
is optional and depends on Jinja22 availability.
The test aim to be exhaustive with various combinations of a base and
an update:
- plain base + plain update, for every mode
- plain base + include (for every mode) update (every mode)
- single include (base +| update)
- Enable setting the XKB root directory. By far the most important
change.
- Enable to pass registry XML files relative to the target rules
directory (with or without the `.XML` extension).
- Enable to set the rules set to use.
- Better JSON/YAML escaping.
- Better error logging.
When using layout index ranges (e.g. special indexes “any” or “later”),
the rules still match following the order in the rules file, so layout
indexes may match without following their natural order. So the resulting
KcCGST value should not be merged with the output until reaching the end
of the rule set.
Because the rule set may also involve options, it may match multiple
times for the *same* layout index. So these multiple matches should not
be merged together either, until reaching the end of the rule set.
When reaching the end of the rule set, for each KcCGST component the
pending values are then merged: for each layout, for each KcCGST value
in the corresponding sequence, merge with the output.
---
Example:
! model = symbols
* = pc
! layout[any] option = symbols
C 1 = +c1:%i
C 2 = +c2:%i
B 3 = skip
B 4 = +b:%i
The result of RMLVO
{layout: "A,B,C", options: "4,3,2,1"}
is:
symbols = pc+b:2+c1:3+c2:3
- `skip` was dropped because it has no explicit merge mode;
- although every rule was matched in order, the resulting order of the
symbols follows the order of the layouts, so `+b` appears before `+c1`
and `+c2`.
- the relative order of the options for layout C follows the order
within the rule set, not the order of RMLVO.
Before this commit, the result would have been:
symbols = pc+c1:3+c2:3+b:2
Some of our test files (e.g. keysym strings) use UTF-8 encoded
strings that require proper handling by the compiler. Let’s favor
readability using modern compiler settings instead of escape
sequences.
Each Unicode code point of the string will be translated to their
respective keysym, if possible. An empty string denotes `NoSymbol`.
When such conversion is not possible, this will raise a syntax error.
This introduces the following syntax:
```c
// Empty string = `NoSymbol`
key <1> {[""]}; // NoSymbol
// Single code point = single keysym
key <2> {["é"]}; // eacute
// String = translate each code point to their respective keysym
key <3> {["sßξك🎺"]}; // {s, ssharp, Greek_xi, Arabic_kaf, U1F3BA}
// Mix string and keysyms
key <4> {[{"ξ", Greek_kappa, "β"}]}; // { Greek_xi, Greek_kappa, Greek_beta}
```
It can also be used wherever a keysym is required, e.g. in `interpret`
and `modifier_map` statements. In these cases a single keysym is expected,
so the string should contain *exactly one* Unicode code point.
- `xkb_utf32_to_keysym`: Allow [Unicode noncharacters]. There is no
requirement to drop them and this would be the only function of our
API doing so.
From the Unicode Standard 16.0, section 23.7 “Noncharacters”:
> Applications are free to use any of these noncharacter code points
> internally. They have no standard interpretation when exchanged
> outside the context of internal use. However, they are not illegal
> in interchange, nor does their presence cause Unicode text to be
> ill-formed.
> If a noncharacter is received in open interchange, an application is
> not required to interpret it in any way. It is good practice,
> however, to recognize it as a noncharacter and to take appropriate
> action, such as replacing it with `U+FFFD` REPLACEMENT CHARACTER,
> to indicate the problem in the text.
The key part is:
> an application is not required to interpret it in any way
Since we handle the reverse conversion with `xkb_keysym_to_utf32` just
fine, I do not see a good motivation to keep this asymmetry. This is
the only function with a special case for these code points.
- `xkb_keysym_from_name`:
- Unicode format `UNNNN`: allow control characters C0 and C1 and use
`xkb_utf32_to_keysym` for the conversion when `NNNN < 0x100`, for
backward compatibility.
- Numeric hexadecimal format `0xNNNN`: *unchanged*. Contrary to the
Unicode format, it does not normalize any keysym values in order to
enable roundtrip with `xkb_keysym_get_name`.
Also added tests to ensure various properties and consistency.
Note about *surrogates*: they are valid valid *code points* but invalid
Unicode *scalar values*, i.e. they cannot be encoded in any Unicode
encoding form (UTF-8, UTF-16, UTF-32). So their corresponding Unicode
keysyms are valid, but:
- cannot be used as input of `xkb_keysym_to_utf32` nor `xkb_keysym_to_utf8`
- cannot result as output of `xkb_utf32_to_keysym`.
Otherwise they are valid e.g. in the Unicode keysym notation.
[Unicode noncharacters]: https://en.wikipedia.org/wiki/Universal_Character_Set_characters#Noncharacters
While the previous code correctly rejected malformed sequences such as
`\u{` (incomplete) or `\u{123x}`, it should try to consume as much input
as possible until reaching the corresponding closing `}` within the
string. Else we can get leftovers and the error message does not
reference the whole malformed sequence.
Also added further tests with surrogates and noncharacters.
Previously there was a distinction between keysyms with code points in
BMP and the others: the former used a 4-padding while the latter used
a 8-padding: e.g `U0001` vs `U00010000`. This is unnecessary and makes
the reading harder.
Let’s use the same padding for all: `U0001` and `U10000`.
Parsing remains unchanged and would parse both paddings.
Also added a test to check no explicit name can clash with Unicode
notation.
Before this commit we supported multiple actions per level, but not in
*interpret* statements. Let’s fix this asymmetry, so we can equivalently
assign all actions sets either implicitly or explicitly.
Unicode code point escape sequences `\u{NNNN}` are replaced with the
UTF-8 encoding of their corresponding code point `U+NNNN`, if legal.
Supported Unicode code points are in the range `1‥0x10ffff`.
Note that we will reject the `U+0000` NULL code point, as we reject it
in the octal escape sequence `\0`.
This is intended mainly for the upcoming feature to write keysyms as
UTF-8 encoded strings. It can be used for various reasons:
- avoid encoding issues;
- avoid issue with font rendering (e.g. Asian scripts);
- make white space or zero-width characters more readable.
Xorg xkbcomp only parses octal sequences with `\0`, while xkbcommon
does not force the `0` prefix of the numeric part. However, we only
parsed up to to 3 digits, which does not allow to parse e.g. `\0377`
while `\377` parses fine.
Fixed by parsing up to 4 octal digits, while checking the result fits
into a byte.
Previously we would write characters without any escaping in some
cases (e.g.: names of indicators, types and groups). E.g. the string
"new\nline"
would be serialized as:
"new
line"
which would raise a syntax error if parsed.
Fixed by escaping any string that was not escaped after parsing (e.g.
the section names are safe already).
The use of `strtol*` functions was already restricted due to its
slowness and its capacity to parse other stuff than digits (e.g.
signs and spaces).
There is also another *big* limitation: it requires a NULL-terminated
string. This is incompatible with our functions that work on buffers,
because we cannot guarantee this. This may lead to a memory violation
if the last token is a number.
We now roll out our own parsers, which are more efficients and
compatible with buffers.
Motivations:
- Follow the principle of least astonishment;
- Ensure consistency;
- Enhance the use of custom defaults;
- Facilitate the tests.
There is some ambiguity because we use `{}` to denote both an empty list of
keysyms and an empty list of actions. But as soon as we get a keysym or an
action, we know whether it is a `MultiKeySymList` or a `MultiActionList`.
So we just count the `{}` at the *beginning* using `NoSymbolOrActionList`,
then replace it by the relevant count of `NoSymbol` or `NoAction()` once the
ambiguity is solved. If not, this is a list of empties of *some* type: we
drop those empties and delegate the type resolution using `ExprEmptyList()`.
Before this commit, the following symbols:
```c
xkb_symbols {
virtual_modifiers M1, M2;
key <A> {};
key <B> { [] };
key.vmods = M1;
key <C> {};
key <D> { vmods = M2 };
};
```
would be equivalent to:
```c
xkb_symbols {
virtual_modifiers M1,M2;
key <B> { [ NoSymbol ] };
};
```
`<B>` entry could be skipped but is harmless. However, `<C>` and `<D>`
are missing, which would lead to the mapping resolution of `M1` and
`M2` failing.
After this commit, it is equivalent to:
```c
virtual_modifiers M1,M2;
key <C> { vmods = M1 };
key <D> { vmods = M2 };
```
Empty keys are skipped entirely, but any explicit field:
- is taken into account: previously they would be skipped if there
were no group;
- forces the key to be printed at serialization.
- Empty `interpret`
- Empty key `type`
- Empty `indicator`
Motivations:
- Follow the principle of least astonishment;
- Ensure consistency;
- Enhance the use of custom defaults;
- Facilitate the tests.
This brings us closer to what `xkbcomp` outputs. One should use
the explicit `VoidSymbol` instead of `NoSymbol`, in order to avoid
dropping empty levels.
This may affect keys that rely on an *implicit* key type. Example:
- Input:
```c
key <> { [a, A, NoSymbol] };
```
- Compilation with xkbcommon \< 1.9.0:
```c
key <> {
type= "FOUR_LEVEL_SEMIALPHABETIC",
[a, A, NoSymbol, NoSymbol]
};
```
- Compilation with xkbcommon ≥ 1.9.0:
```c
key <> {
type= "ALPHABETIC",
[a, A]
};
```
Before this commit the following keymap:
```c
xkb_keymap {
xkb_keycode {};
};
```
would result in (boilerplate removed):
```c
xkb_keymap {
xkb_keycode "(unnamed)" {};
};
```
This is both useless and wasting allocation: section names are optional,
so we should just remove this default name altogether and keep it
undefined, as in the original keymap.
The situation is a bit different if there is an include, as for keymaps
created from RMLVO names. Before this commit, the following keymap:
```c
xkb_keymap {
xkb_keycode { include "evdev+aliases(qwerty)" };
};
```
would result in (boilerplate removed):
```c
xkb_keymap {
xkb_keycode "(unnamed)" { … };
};
```
With this commit we now follow the Xorg xkbcomp style by using the
section reference (the include string) as the *default* section name. So
the previous example would now result in:
```c
xkb_keymap {
xkb_keycode "evdev_aliases(qwerty)" { … };
};
```
which is useful to give a hint of the original include.
Note that if the original section had a name, it would preserve it:
```c
xkb_keymap {
xkb_keycode "test" { include "evdev+aliases(qwerty)" };
};
```
would compile to:
```c
xkb_keymap {
xkb_keycode "test" { … };
};
```
We already accept *empty* components, such as: `xkb_compat {};`. Let’s
accept missing components as well, so that we can reduce the boilerplate
in our tests.
Note that we will still explicitly serialize empty components for
compatibility with previous xkbcommon versions and Xorg xkbcomp.
Previously only the merge modes “override” and “augment” were available
in include statements, using the prefix ‘+’ and ‘|’ respectively. While
on one hand `replace` include statement can be used in keymap files, on
the other hand *rules* files have no way to express the *replace* mode.
This commit enables the merge mode “replace” using the prefix `^`. This
prefix was chosen due to its similarity with the `XOR` bit operator,
which convey *mutual exclusion*.
Other candidates:
- `!` conveys some kind of higher precedence, akin to CSS `!important`.
But it conflicts with the section header `!`, which is a token in the
current parser. It would require special handling, not worth it. It
also convey the meaning of negation, which is confusing.
- `&` has the advantage of not corresponding to a token in the rules
parser. `^` seems however to stand out more and it is less likely to
trigger erroneous comparison with `|` and `&` bit operators.
Added the following wild cards to the rules file syntax, in addition
to the current `*` legacy wild card:
- `<none>`: Match *empty* value.
- `<some>`: Match *non-empty* value.
- `<any>`: Match *any* (optionally empty) value. Its behavior does not
depend on the context, contrary to the legacy wild card `*`.
This will enable writing much simpler rules, see [!764] for an example
of tricky rules in the `xkeyboard-config` project, that would benefit
from the new wild cards.
[!764]: https://gitlab.freedesktop.org/xkeyboard-config/xkeyboard-config/-/merge_requests/764
The verbose wild cards are preferred to single characters:
- More intuitive: self-explanatory.
- Does not steal syntax from other token.
- Extensible syntax, should we need it.
A previous proposal used the characters (`!`, `+`, `?`) for their
similarity with the corresponding syntax of regular expressions
(negative assertion & quantifiers), in line with `*`. But `!` is not
that intuitive after all and conflict with its role as section header.
Furthermore, `+` is also used as a merge mode. Finally, nothing beats
whole short words for readability.
Before this commit we used `strtold`, which depends on the locale. But
the XKB syntax is fixed and uses a period as decimal separator. So ensure
the syntax is correct without relying on `strtold` and truncate the
result, as the parser does not use floating-point numbers.
The new public option `--kccgst` enables to display the result of RMLVO
resolution to KcCGST components.
This option has the same function than `setxkbmap -print`. This is particularly
useful for debugging issues with the rules.
Before this commit it was a private API. This commit enables us to remove
the *internal* version of `xkbcli-compile-keymap`.
Parse, dont’t validate: ensure *at parsing* that `modifier_map`
definitions use a list of keys and keysyms.
This enables to remove the redundant `ExprResolveKeySym` and have keysym
parsing exclusively in handled in `parser.y`.
- Refactor to check conflicts first for the key names and then for the
keycodes. This seems more useful for the user and enable further
memory optimizations.
- Do not allocate until we are sure to add the keycode. The bounds are
only updated afterwards, so the call to `FindKeyByName` should be
more efficient.
- Fixed keycodes bounds not shrunk correctly when an existing keycode
is overridden.
- Do not prepare keyname strings for logging if we are not going to
use them.
gitlab.freedesktop.org is currently migrated to a new infrastructure,
so use a release tarball of xkeyboard-config instead of cloning its repo.
NOTE: This commit should be reverted once the migration is completed.
When logging about virtual modifier *explicit* mappings, we should
always use only real modifiers or hexadecimal numbers to print the mask.
Consider:
```
virtual_modifiers M1, M2=0x200, M2=0x400;
```
Before this commit we would get the following warning:
```
WARNING: Virtual modifier M2 defined multiple times; Using M2, ignoring M1
```
while we would prefer the less confusing:
```
WARNING: Virtual modifier M2 defined multiple times; Using 0x400, ignoring 0x200
```
The expression AST constructors all return `ExprDef *`. `ExprDef` is a
union of all expr types. As a memory optimization, instead of allocating
`sizeof(ExprDef)`, we only allocate the size of the actual type (e.g.
`sizeof(ExprBinary)`) which is sometimes smaller than `sizeof(ExprDef)`.
This is probably undefined behavior, and gcc (with optimization turned
on) complains about it, for example:
src/xkbcomp/ast-build.c:69:23: warning: array subscript ‘ExprDef[0]’ is partly outside array bounds of ‘unsigned char[24]’ [-Warray-bounds=]
Since it doesn't save that much memory, drop this optimization.
Fix#292.
Signed-off-by: Ran Benita <ran@unusedvar.com>
The modular arithmetic is incorrect for negative values, e.g. for
num_groups = 1. It triggers a segfault for the following settings:
- layouts count (per key or total) N: `N > 0`, and
- layout index n: `n = - k * N` (`k > 0`)
% returns the *remainder* of the division, not the modulus (see C11
standard 6.5.5 “Multiplicative operators”: a % b = a - (a/b)*b. While
both operators return the same result for positive operands, they do
not for e.g. a negative dividend: remainder may be negative (in the
open interval ]-num_groups, num_groups[) while the modulus is always
positive. So if the remainder is negative, we must add `num_groups` to
get the modulus.
Fixes: 67b03cea ("state: correctly wrap state->locked_group and ->group")
Signed-off-by: Julian Orth <ju.orth@gmail.com>
Co-authored-by: Julian Orth <ju.orth@gmail.com>
Co-authored-by: Pierre Le Marre <dev@wismill.eu>
Options are always applied depending of their order in the relevant
rules files, independently of their order in the relevant RMLVO config.
Since this is counter-intuitive (e.g. when using `xkbcli`), let’s
highlight this in the documentation.
Our tools are debugging tools and as such we need to have complete
control to be able to reproduce setups. This is not currently the case,
as we do not use `XKB_CONTEXT_NO_ENVIRONMENT_NAMES` by default nor can
we set it.
So it is very easy to forget about the various `XKB_DEFAULT_*`
environement variables for the default RMLVO values, then to get puzzled
by unexpected results.
Added to that, these environment variables do not work correctly in
`xkbcli-compile-xeymap`: calling the tool without RMLVO values will
use these variables only if the RMLVO values are set explicitly empty or
if the various *constants* `DEFAULT_XKB_*` are empty. This is unexpected,
as the environment variables should *always* be used unless:
- `XKB_CONTEXT_NO_ENVIRONMENT_NAMES` is used (not the case here);
- the variable is empty; in this case the constants `DEFAULT_XKB_*` are
used.
Fixed by the following *breaking change*: make the tools use
`XKB_CONTEXT_NO_ENVIRONMENT_NAMES` *by default*, unless the new
`--enable-environment-names` option is used.
We also make `rmlvo` incompatible with `--enable-environment-names` for
now in the public tool, as else it requires a private API.
Spotted by clang-tidy:
```
../src/keymap-priv.c:146:16: warning: comparing object representation of
type 'union xkb_action' which does not have
a unique object representation; consider
comparing the members of the object manually
[bugprone-suspicious-memory-comparison]
````
Indeed, from “EXP42-C. Do not compare padding data”[^1]:
> unnamed members of objects of structure and union type do not participate
> in initialization. Unnamed members of structure objects have indeterminate
> representation even after initialization.
Fixed by using a dedicated comparison function for `union xkb_action`.
[^1]: https://wiki.sei.cmu.edu/confluence/display/c/EXP42-C.+Do+not+compare+padding+data
It mostly works by default, as long as the meson build directory is
called "build". Just needs a couple of project-specific tweaks.
Signed-off-by: Ran Benita <ran@unusedvar.com>
This function was added in commit
4e22851141. But that commit also changed
the grammar:
-KeyNameDecl : KeyName EQUALS Expr SEMI
+KeyNameDecl : KeyName EQUALS KeyCode SEMI
i.e. while before you could write
<AE01> = 9+1;
now this is a syntax error, an integer literal is expected. I'm not sure
if it was intended to remove this ability. In any case, this rendered
`ExprResolveKeyCode` useless since there's no longer an expression to
evaluate, and after some refactoring it went unused.
Even if we choose to restore Expr here, I don't see a reason for the
specialized function over `ExprResolveInteger` except the type (which we
should probably widen from int to int64_t...). So remove it.
Signed-off-by: Ran Benita <ran@unusedvar.com>
This field is a funky attempt at type inference, or perhaps some
optimization? Anyway, after careful examination I conclude it serves no
purpose except specifying the type of a literal
(string/integer/float/boolean/keyname) when `STMT_EXPR_VALUE` (i.e.
literal).
Remove it and replace `STMT_EXPR_VALUE` with specific statement types
for each literal type.
Signed-off-by: Ran Benita <ran@unusedvar.com>
This makes the tests, and especially benches, more realistic, since
xkbcommon is almost always used as a shared library.
Also significantly reduces the build time with LTO enabled (for me, from
90s to 30s).
Signed-off-by: Ran Benita <ran@unusedvar.com>
Without this, the test-internal libraries (which don't use the .def file
because they contain additional private symbols) can't be made shared.
But it also makes sense for consistency with GCC.
Signed-off-by: Ran Benita <ran@unusedvar.com>
- Use classical layout without a side bar, which is the default in
recent Doxygen versions.
- Use Github-style heading IDs: more user-friendly and stable.
- Enable the Javascript-base search engine.
- Generate `sitemaps.xml`.
- Fix CSS width incompatible with small devices.
We currently have a mix of include headers, pragma once and some
missing.
pragma once is not standard but is widely supported, and we already use
it with no issues, so I'd say it's not a problem.
Let's convert all headers to pragma once to avoid the annoying include
guards.
The public headers are *not* converted.
Signed-off-by: Ran Benita <ran@unusedvar.com>
The AST is heavily based on intrusive lists for representing lists, but
actions are an exception, instead using darray. I don't see any reason
for this; it ends up allocating more, and we don't especially need a
flat array for this.
Change it to use the familiar linked-list style.
Signed-off-by: Ran Benita <ran@unusedvar.com>
The scanner functions are hot, and the line/column location tracking is
quite expensive. We only use it for errors, which don't need to be fast,
because we bail if there are too many; and for warnings, which are
usually not shown by default. So only keep the token start pos, and
compute the line/column lazily from that. This will also allow some
further improvements ahead.
bench/rulescomp
before: compiled 1000 keymaps in 1.669028s
after: compiled 1000 keymaps in 1.550411s
bench/compose:
before: compiled 1000 compose tables in 2.145217s
after: compiled 1000 compose tables in 2.016044s
Signed-off-by: Ran Benita <ran@unusedvar.com>
Our current implementation uses continuous arrays indexed by keycodes
and levels. This is simple and good enough for realistic keymaps.
However, they are allowed to have big values that will lead to either
memory exhaustion or a waste of memory (sparse arrays).
Added the much stricter upper bounds `0xfff` for keycodes[^1] and 2048
for levels[^2], which should still be plenty enough and provides stronger
memory security.
[^1]: Current max keycode is 0x2ff in Linux.
[^2]: Should be big enough to satisfy automatically generated keymaps.
The data set is big enough so we can just drop most sequences with
deprecated keysyms and fix most keysyms an alternative non-deprecated
name.
We keep a handful of them for testing purpose.
Add `--keymap` as a more intuitive alias to `--from-xkb`; it also makes
it consistent with `interactive-evdev`. It optionally accepts a keymap
file path; if the argument is empty or `-` then the keymap is read from
`stdin`.
Previously when using `--keymap` we mistakenly returned a boolean so
that `true` (resp. false) translated to `EXIT_FAILURE` (resp.
`EXIT_SUCCESS`).
Fixed by returning the proper return code integer.
In essence empty levels are levels with just a `NoSymbol` keysym and
a `NoAction()`, which breaks latches.
Fixed regression introduced in fdf2c52597.
Added tests also for the case where the keycode is unknown.
- Added a big bunch of tests for modifier latch. No yet exhaustive,
but should cover the most usual use cases.
- Added missing test cases for breaking the group latch. Ideally, more
tests should be added to match the coverage of modifiers latches.
WARNING: it is ambiguous what prevents a latch when multiple keys are
*simultenously* operated together. We currently assuming that
any action that is not identical to the considered latch
prevents it.
Due to how our parser is implemented, keysyms names that are also valid
keywords require special handling.
Added a check for these clashes in the keysym generator. The only
current clash, `section`, is already handled. Note that it means that
e.g. `section`, `Section` and `sEcTiOn` all parse to the same keysym.
This side effect is fine here, because *currently* there is no other
keysym that clashes with any possible of the case variation of `section`.
But in order to be extra cautious, we now test thoses clashes too.
Hopefully we will never have a clash again, but while it is unlikely
that we modify the keywords, the keysyms are not a frozen set.
Currently the following keymap triggers a critical error (invalid
`vmods`) only for the second key statement, while it should handle both
equally.
```
xkb_keymap {
xkb_keycodes { <> = 9; };
xkb_types { };
xkb_compat { };
xkb_symbols {
key <> { vmods = [], repeats = false };
key <> { repeats = false, vmods = [] };
};
};
```
Fixed by parsing the whole symbols body and failing if any error was found.
Introduce a new Expression type, `EXPR_EMPTY_LIST`, to avoid the
ambiguity between action and keysym empty lists. Two alternatives were
rejected to keep the semantics clear:
- Using `EXPR_KEYSYM_LIST`: because we would end up accepting an empty
keysym list while processing actions.
- Using NULL: convey no info and is hazardous.
The enum casts can possibly lead to unaligned access.
The warnings trigger in the Windows CI but not on Linux. One may use
`-fshort-enums` with gcc in order to trigger the errors.
Currently we accept values of -255..255 while parsing the screen value
of `SwitchScreen` actions`, but then we silently cast it against `int8_t`,
i.e. in range -128..127.
We actually do as xkbcomp, but this seems a bug because the target storage
is a `char`. Let’s simply raise a parse error if the value does not fit
in our type.
```
libxkbcommon.so.0.7.0.p/parser.c:1632:9: warning: variable '_xkbcommon_nerrs' set but not used [-Wunused-but-set-variable]
1632 | int yynerrs = 0;
```
Signed-off-by: Ran Benita <ran@unusedvar.com>
When dumping the keymap, we first try to write to the buffer and get
the min size requirement. Then we reallocate if necessary. However, we
should not write it again if we got enough space previously!
Various problems can occur when using the full `xkbcommon.map` file when
checking whether the `--version-script` linker option works in
`meson.build`.
For instance, newer linkers typically complain about non-existing
symbols, so in commit ebe4157 `--undefined-version` was added to work
around that. But then in commit 1d8a25d6 another workaround needed to be
added for older linkers.
It is better to use a minimalistic linker script for testing instead.
The other workarounds can then be removed.
The modifiers filters should ensure minimal interaction between them,
but currently the Latch mod filters are overzealous and mess with the
mods from other filters set to be cleared, resulting in some modifiers
permanently set.
Fixed by clearing mods properly with `OR` rather than direct setting
of `state::clear_mods`.
While we are at it, `state::set_mods` should be `OR`ed as well. This
should not have any impact for now, but this is more future-proof.
Fixes#583
Co-authored-by: Jules Bertholet <julesbertholet@quoi.xyz>
Co-authored-by: Pierre Le Marre <dev@wismill.eu>
This commit adds tests for merging various key configurations:
- With/without keysyms/actions
- Single/multiple keysyms/actions per level
We test all the merge modes for including a map (global) as well as
directly on the keys (local):
- default (global: include, local: implicit)
- augment
- override
- replace
The tests data are generated with:
- A Python script `scripts/update-merge-modes-tests.py` for keycodes
and symbols data. Use `--debug` for extra comments to help debugging.
The script can optionally generate C headers for alternative key
sequence tests, that were used before implementing golden tests.
The latter tests are not used anymore (duplicate with golden tests)
but their generator is kept for now, as they can still be useful for
debugging or writing similar tests.
- The `merge-modes` test generates its own keymap files for golden
tests, using: `build/test-merge-modes update`. It can also replace
them with the obtained output rather than the expected one using
`build/test-merge-modes update-obtained`, which is very useful for
debugging.
When the indicator field `whichGroupState` is set to `Base` or `Latched`,
the group masks should be considered only as boolean values as per the
XKB specification.
When getting the keymap from X11, the following:
```
key <AD01> { actions=[SetGroup(2)] };
```
is currently converted to:
```
key <AD01> { };
```
This commit fixes dropping a defined action when the keysym is undefined
`$HOME/.xkb` was supported before the addition of the XDG directories.
Removal of the functionality is just going to cause potential breakage
for now good technical reason.
See https://specifications.freedesktop.org/basedir-spec/latest/ for
further details.
The following syntax does not parse in xkbcommon, but it does in xkbcomp:
```
xkb_symbols "x" {
key <AD01> { };
};
```
While the usefulness of such statement is debatable, the fact that it
does parse in xkbcomp and that tools may generate such keymap entry make
it relevant to handle.
We use the xkbcommon/website pages for the website, so here it's just
for previewing and we only need to the docs, not the website.
Fix#561.
Signed-off-by: Ran Benita <ran@unusedvar.com>
The case mapping `ssharp` ß ↔ `U1E9E` ẞ was added in
13b30f4f0d but was broken:
- For the lower case mapping it returned the keysym `0x10000df`, which
is an invalid Unicode keysym.
- For the upper case mapping it returned the upper Unicode code point
rather than the corresponding keysym.
It did accidentally enable the detection of alphabetic key type for the
pair (ß, ẞ) though. However this detection was accidentally removed in
5c7c79970a (v1.7) with an attempt to fix
the wrong keysym case mapping. Finally both the *lower* case mapping
and the key type detection were fixed for good when we implemented the
complete Unicode simple case mappings and corresponding tests in
e83d08ddbc.
However, the *upper* case mapping `ssharp` → `U1E9E` remained disabled.
Indeed, ẞ is a relatively recent addition to Unicode (2008) and had no
official recommendation, until recently. So while the lower mapping ẞ→ß
exists in Unicode, its converse upper mapping does not. Yet since 2017
the Council for German Orthography (Rat für deutsche Rechtschreibung)
recommends[^1] ẞ as the capitalization of ß.
Due to its stability policies, the Unicode Character Database (UCD)
that we use to generate our keysym case mappings (via ICU) cannot update
the simple case mapping of ß. Discussions are currently ongoing in the
Unicode mailing list[^2] and CLDR[^3] about how to deal with the new
recommended case mapping. However, the discussions are oriented on
text-processing and compatibility mappings, while libxkbcommon is
on a rather lower level.
It seems that the slow adoption of ẞ is partly due to the difficulty
to type it. Since ẞ is used only for ALL CAPS casing, the expectation
is to type it using CapsLock. While our detection of alphabetic key
types works well[^4] for the pair (ß,ẞ), the *internal capitalization*
currently does not work and is fixed by this commit.
Added the ß → ẞ upper mapping:
- Added an exception in the generation script
- Fixed tests
- Added documentation of the exceptions in `xkbcommon.h`
- Added/updated log entries
[^1]: https://www.rechtschreibrat.com/regeln-und-woerterverzeichnis/
[^2]: https://corp.unicode.org/pipermail/unicode/2024-November/011162.html
[^3]: https://unicode-org.atlassian.net/browse/CLDR-17624
[^4]: Except libxkbcommon 1.7, see the second paragraph.
Previously it was not possible to query the status of virtual modifiers
with the following functions:
- `xkb_state_mod_index_is_consumed`
- `xkb_state_mod_index_is_consumed2`
Note that it may *overmatch* if some modifier mappings overlap. For
example, the default “us” PC layout maps both “Alt” and “Meta” to the
real modifier “Mod1”; thus “Mod1”, “Alt” and “Meta” modifiers will
return the same result with these functions.
Previously it was not possible to query the status of virtual modifiers
with the following functions:
- `xkb_state_mod_index_is_active`
- `xkb_state_mod_indices_are_active`
- `xkb_state_mod_name_is_active`
- `xkb_state_mod_names_are_active`
Note that it may *overmatch* if some modifier mappings overlap. For
example, the default “us” PC layout maps both “Alt” and “Meta” to the
real modifier “Mod1”; thus “Mod1”, “Alt” and “Meta” modifiers will
return the same result with these functions.
Respect the claim that real modifiers names are built-in:
- Add `XKB_MOD_NAME_MOD[1-5]` constants.
- Replace modifiers names with their corresponding constants.
i don't have access to github but this patch enabled me to compile
libxkbcommon with -std=gnu23
otherwise the following compilation error will result :
../src/compose/paths.c:70:16: error: incompatible types when returning type ‘_Bool’ but ‘char *’ was expected
( gcc (GCC) 15.0.0 20240509 (experimental) )
Signed-off-by: Ran Benita <ran@unusedvar.com>
- Refactored to write files within the workers, else we have only one
worker writing, slowly consuming the result queue. In some scenarios
it may even result in out-of-memory as the resulting keymaps piles up.
- Added the option `--compress` to allow compressing keymaps files with
gunzip.
- Use the recently added `--test` option for `xkbcli-compile-keymap`
when we only want to test compilation, not the resulting keymap file.
This speeds up the test suite of `xkeyboard-config`.
The new flag `--test` enables to only test if compilation succeeds,
without printing the corresponding keymap or Compose file.
This is useful for quick feedback and to speedup some tests suites, e.g.
for `xkeyboard-config`.
- Refactor to more modern Python
- Allow iterating over models too
- Add filter for models and options
- Add “short output” option
- Fix layout variant
- Use models subfolders for keymaps
- Allow to load multiple registries simultaneously
There is currently no easy way to dump a keymap from a Wayland compositor,
such as `xkbcomp -xkb $DISPLAY -` could do for X servers.
As `xkbcomp` may not be intuitive, a corresponding tool for X servers
would also be useful.
Add the tools `xkbcli-dump-keymap-{wayland,x11}` by tweaking the existing
`xkbcli-interactive-{wayland,x11}` tools.
This makes 1 keysym == 1 action holds also for multiple keysyms per level.
The motivation of this new feature are:
- Make multiple keysyms per level more intuitive.
- Explore how to fix the issue with shortcuts in multi-layout settings
(see the xkeyboard-config issue[^1]). The idea is to use e.g.:
```c
key <LCTL> {
symbols[1] = [ {Control_L, ISO_First_Group } ],
actions[1] = [ {SetMods(modifiers=Control), SetGroup(group=-4) } ]
};
```
in order to switch temporarily to a reference layout in order to get
the same shortcuts on every layout.
When no action is specified, `interpret` statements are used to find
an action corresponding for *each* keysym, as expected.
For an interpretation matching Any keysym, we may get the same
interpretation for multiple keysyms. This may result in unwanted
duplicate actions. So set this interpretation only if no previous
keysym was matched with this interpret at this level, else set the
default interpretation.
For now, at most one action of each following categories is allowed
per level:
- modifier actions: `SetMods`, `LatchMods`, `LockMods`;
- group actions: `SetGroup`, `LatchGroup`, `LockGroup`.
Some examples:
- `SetMods` + `SetGroup`: ok
- `SetMods` + `SetMods`: error
- `SetMods` + `LockMods`: error
- `SetMods` + `LockGroup`: ok
[^1]: https://gitlab.freedesktop.org/xkeyboard-config/xkeyboard-config/-/issues/416
The current field `u` (short for “union”) is not very descriptive.
Next commit will add multiple actions per level, so let’s rename the
keysym field to `s` (short for “symmbols”).
Currently `xkb_keymap_num_layouts` may return a greater number than the
number of layouts configured using RMLVO, because we allow symbols
sections to define various groups per key.
This is unintuitive and kind of buggy: groups should be added via rules
by setting an explicit `:n` modifier.
Fix: when parsing a keymap using RMLVO resolution:
- Get the expected layouts count from the resulting KcCGST.
- Drop the groups after the first one in included symbols sections. This
will ensure that a symbol section can only define one group per key.
Notes:
- Compiling a keymap string directly is unaffected.
- RMLVO resolution may still produce more groups than the input layouts.
Indeed, some legacy rules in xkeyboard-config rely on this to insert
automatically a US layout before the given non-Latin one, resulting
in two layouts while only one was given.
Previously setting explicit actions for a group in symbols files made
the parser skip compatibility interpretations for the corresponding
*whole* key, so the other groups with *no* explicit actions could result
broken on some levels.
In the following example, `<RALT>` would have an action on group 2,
because it is explicit, but none on group 1 because interpretation are
also skipped there as a side effect:
```c
key <RALT> {
symbols[1]= [ ISO_Level3_Shift ],
symbols[2]= [ ISO_Level3_Shift ],
actions[2]= [ SetMods(modifiers=LevelThree) ]
};
```
Fixed by skipping interpretations *only* for groups with explicit actions.
We still set `key->explicit |= EXPLICIT_INTERP` if at least one group
has explicit actions. In such case, when dumping a keymap, we will
write explicit actions for *all* groups, in order to ensure that X11 and
previous versions of libxkbcommon can parse the keymap as intended. One
side effect is that no interpretation will be run on this key anymore,
so we may have to set some extra fields explicitly: repeat, virtualMods.
Thus the previous example would be bumped as:
```c
key <RALT> {
repeat= No,
symbols[1]= [ ISO_Level3_Shift ],
actions[1]= [ SetMods(modifiers=LevelThree,clearLocks) ],
symbols[2]= [ ISO_Level3_Shift ],
actions[2]= [ SetMods(modifiers=LevelThree) ]
};
```
These allow us to store comments in files (especially keymaps), that
can then be removed (e.g. to compare with output) or uncommented (e.g.
to activate an optional line).
Some layout options require to be applied to every group to maintain
consistency (e.g. a group switcher). Currently this must be done manually
for all layout indexes. This is error prone and prevents the increase of
the maximum group count.
This commit introduces the `:all` qualifier for KcCGST values. When a
rule with this qualifier is matched, it will expands the qualified
value (and its optional merge mode) for every layout, e.g.
`+group(toggle):all` (respectively `|group(toggle)`) would expand to
`+group(toggle):1+group(toggle):2` (respectively
`|group(toggle):1|group(toggle):2`) if there are 2 layouts, etc.
If there is no merge mode, it defaults to *override* `+`, e.g. `x:all`
expands to `x:1+x:2+x:3` for 3 layouts.
Note that only the qualified *value* is expanded, e.g. `x+y:all` expands
to `x+y:1+y:2` for 2 layouts.
`:all` can be used in combination with special layout indexes. Since
this can lead to an unexpected behaviour, a warning will be raised.
There is a lot of repetition in the current rules files provided by
xkeyboard-config, because the MLVO mappings need to match on the exact
layout/variant index. This also prevents to increase the number of
maximum groups, because it does not scale.
We introduces the following new special layout/variant indexes:
- “single”: matches a single layout; same as with no index.
- “first”: matches the first layout/variant, no matter how many layouts
are in the RMLVO configuration. It allows to merge `layout` and
`layout[1]` patterns.
- “later”: matches all but the first layout. This is an index range.
- “any”: matches layout at any position. This is an index range.
We also introduces the new `%i` expansion, which correspond to the index
of the matched layout in a mapping with an index range. Example:
layout[later] = symbols
my_layout = +my_symbols:%i
* = +%l[%i]:%i
Let’s have a look at concrete examples from xkeyboard-config:
! model layout = symbols
* * = pc+%l%(v)
! model layout[1] = symbols
* * = pc+%l[1]%(v[1])
! model layout[2] = symbols
* * = +%l[2]%(v[2])
! model layout[3] = symbols
* * = +%l[3]%(v[3])
! model layout[4] = symbols
* * = +%l[4]%(v[4])
! layout option = symbols
* grp:toggle = +group(toggle)
! layout[1] option = symbols
* grp:toggle = +group(toggle):1
! layout[2] option = symbols
* grp:toggle = +group(toggle):2
! layout[3] option = symbols
* grp:toggle = +group(toggle):3
! layout[4] option = symbols
* grp:toggle = +group(toggle):4
With this commit we can now simplify it into:
! model layout[first] = symbols
* * = pc+%l[%i]%(v[%i])
! model layout[later] = symbols
* * = +%l[%i]%(v[%i]):%i
! layout[any] option = symbols
* grp:toggle = +group(toggle):%i
The latter rule will work even if we increase XKB_MAX_GROUPS, whereas
the former would require to add the missing entries for the new groups.
In order to maintain consistent rules, we now disallow layout and
variant to have different indexes. For example, the following mapping
are now invalid:
- layout variant[1]
- layout[1] variant[2]
- layout[1] variant
- layout[first] variant[1]
- layout[first] variant
- layout variant[any]
- etc.
Contrary to keymap files, the `! include` statement in rules does not
lookup include paths added to `xkb_context`. So it is not possible e.g.
to import another file in the same folder without using an absolute path.
- Added path utils: `is_absolute(path)`.
- Added XKB paths lookup to enable e.g. `! include evdev` to work.
- Added test.
Currently the positional parameter of the CLI is either a Unicode code
point or a keysym. However their respective format is not documented.
It turns out that there are multiple issues due to the use of `strtol`:
- Code points can be parsed as octal, decimal and hexadecimal, while
keysyms can only be parsed as hexadecimal. Some programs outputs
keysyms in their decimal form (e.g. `wev`) so it is worth to bring
symmetry with code points.
- Octal format is unusual for both and is triggered by leading zeros,
which is unintuitive in this context.
- `U+NNNN` format is the standard format for Unicode code points but is
not supported.
- Plain characters are not supported, e.g.: a, é, ß, Æ, γ, 🦆, etc.
Although this is probably the easiest format for most users.
Fixed the issues above:
- Allow the code point to be passed exactly in the following formats:
- Literal character (requires UTF-8 character encoding of the terminal);
- Decimal number;
- Hexadecimal number: either `0xNNNN` or `U+NNNN` (any digit count)
- Allow the keysym to be passed exactly in the following formats:
- Decimal number;
- Hexadecimal number: `0xNNNN` (any digit count);
- Name.
- Improve both `--help` message and manual page.
Add 2 new warnings:
- Deprecated keysym name (typo, historical alias, etc.);
- Deprecated keysym (all names and forms).
Guard deprecated keysym tests with verbosity level ≥2, so they are
run only when actually needed.
This function allow to check whether a keysym is deprecated, based
on the keysym and optionally its name.
The generation of the table of deprecated keysyms relies on the
rules described in `xkbcommon-keysyms.h`.
The `ks_table.h` is now generated deterministically by setting
explicitly the random seed to a constant. This will avoid noisy
diffs in the future.
Implement `tasty-bench` method:
1. Set n = 1.
2. Measure execution time t_n of n iterations and execution time t_2n
of 2n iterations.
3. Find t which minimizes deviation of (n·t, 2·n·t) from (t_n, t_2n),
namely t = (t_n + 2·t_2n)/5n.
4. If deviation is small enough (see --stdev CLI parameter), return t
as a mean execution time.
5. Otherwise set n = 2·n and jump back to Step 2.
See: https://hackage.haskell.org/package/tasty-bench
The constant did not account for the terminating `NULL` byte and this was
sadly not caught by the tests.
Fixed the invalid value, the corresponding script and the tests.
- Move `print_compose_table_entry` to own file and add a `file`
argument to select output.
- Add `xkb_compose_table_dump` to dump a Compose table.
This change is needed in order to share the feature “dump a Compose
table” between tests and tools.
Rational (from: https://gitlab.freedesktop.org/wayland/wayland/-/issues/175):
The ABI of a shared library on Linux is given by a major version, which
is part of the SONAME and is incremented (rarely) on incompatible changes,
and a minor version, which is part of the basename of the regular file to
which the SONAME provides a symlink. It's conventional to manage the ABI
version in the style of semantic versioning (major.minor.micro where the
major version goes up for incompatible changes, the minor version goes up
for new ABI, or the micro version goes up for internal fixes), although
this is not strictly required.
The minor version is used by `ldconfig(8)` and by some projects to choose
which of potentially several copies of a library is the newest; keeping
it the same means we can't tell, and will potentially choose an outdated
version.
---
In xkbcommon we did not have an API break for a long time, so in order
to avoid an unnecessary SONAME bump, let’s keep the major version version
at 0, e.g. xkbcommon 1.x.y produces `libxkbcommon.so.0.x.y`.
The same goes for `libxkbcommon-x11` and `libxkbregistry`.
test/registry.c:532:9: overrun-local: Overrunning array "tl->iso639" of 3
8-byte elements at element index 23 (byte offset 191) using index "i"
(which evaluates to 23).
This ensures that our string termination is handled by utf32_to_utf8()
as opposed to false positives due to previously set null terminators.
Fixes:
../../../test/compose.c:912:34: warning: The left operand of '==' is a
garbage value [core.UndefinedBinaryOperatorResult]
912 | assert_printf(buf[c] == '\0', "NULL-terminated string\n");
../../../test/xvfb-wrapper.c:166:5: warning: Undefined or garbage value
returned to caller [core.uninitialized.UndefReturn]
166 | return rc;
../../../tools/interactive-wayland.c:820:14: warning: The left operand
of '>=' is a garbage value [core.UndefinedBinaryOperatorResult]
820 | exit(ret >= 0 ? EXIT_SUCCESS : EXIT_FAILURE);
(hopefully) silences an analyzer warning:
test/compose.c:955:16: store_truncates_time_t: A "time_t" value is stored
in an integer with too few bits to accommodate it. The expression
"time(NULL)" is cast to "int".
If xkb_compose_table_new_from_locale() fails we end up de-initializing
the keyboard state before it was ever initialized.
../../../tools/interactive-x11.c:196:5: warning: 1st function call
argument is an uninitialized value [core.CallAndMessage]
196 | xkb_state_unref(kbd->state);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~
src/compose/table.c:66:5: uninit_use: Using uninitialized value "dummy".
Field "dummy.lokid" is uninitialized.
This looks like a false positive but it's easy enough to shut up:
../../../src/compose/dump.h:48:16: warning: Result of 'calloc' is
converted to a pointer of type 'char', which is incompatible with
sizeof operand type 'char *' [unix.MallocSizeof]
48 | char* to = calloc(4 * length + 1, sizeof(to));
| ~~~~~ ^~~~~~ ~~~~~~~~~~
No security impact as sizeof(char*) is always >1 so we were just overallocating
here.
The current code to handle keysym case mappings is quite complex and
slow. It is also incomplete, as it does not cover recent Unicode
database. Finally, it does not handle title case correctly.
It would be easier if we were to use only a lookup table, but a trivial
implementation would lead to a huge array: the cased characters range
from `U+0041` to `U+`1F189, i.e. a span of 127 304 elements.
Thus we need some tricks to compress the lookup table. We based our
work on the post:
https://github.com/apankrat/notes/blob/3c551cb028595fd34046c5761fd12d1692576003/fast-case-conversion/README.md
The compression algorithm is roughly:
1. Compute the delta between the characters and their mappings.
2. Split the delta array in chunk of a given size.
3. Rearrange the order of the chunks in order to optimize consecutive
chunks overlap.
4. Create a data table with the reordered chunks and an index table
that maps the original chunk index to its offset in the data table.
The compression algorithm is then applied a second time to the previous
index table.
The complete algorithm optimizes the two chunk sizes in order to get
the lowest total data size.
The mappings were generated using CPython 3.12.4, PyICU 2.13, PyYaml
6.0.1 and ICU 75.1.
Also:
- Added explicit list of named keysyms and their case mappings.
- Added benchmark for case mappings.
- Rework ICU tests.
Note: 13b30f4f0d introduced a fix for
sharp S `U+00DF`. With the new implementation, the *conversion*
functions `xkb_keysym_to_{lower,upper}` leave it *unchanged*, while
the *predicate* functions `xkb_keysym_is_{lower,upper_or_title}`
produce the expected results:
```c
xkb_keysym_to_upper(XKB_KEY_ssharp) == XKB_KEY_ssharp;
xkb_keysym_to_lower(XKB_KEY_ssharp) == XKB_KEY_ssharp;
xkb_keysym_to_lower(XKB_KEY_Ssharp) == XKB_KEY_ssharp;
xkb_keysym_is_lower (XKB_KEY_ssharp) == true;
xkb_keysym_is_upper_or_title(XKB_KEY_Ssharp) == true;
```
There are several mismatches with the Unicode database.
Fix the case mappings in `data/keysyms.yaml` for the following keysyms:
- `Iabovedot`
- `idotless`
- `Greek_finalsmallsigma`
- `function`
Added a tool to export explicitly named keysyms to a yaml file. This
includes the following keysyms properties:
- Value
- Name
- Corresponding Unicode code point, if relevant
- Corresponding case mapping, if relevant
This will allow us to check the case mappings easier and to refactor the
related functions.
Require only 5 bytes for the buffer of `xkb_keysym_to_utf8`, as UTF-8
encodes code points on up to 4 bytes + 1 byte for the NULL-terminating
byte.
Previous standard [RFC 2279] (1998) required up to 6 bytes per code
point, but has been superseded by [RFC 3629] (2003).
[RFC 2279]: https://datatracker.ietf.org/doc/html/rfc2279
[RFC 3629]: https://datatracker.ietf.org/doc/html/rfc3629
Previously on FreeBSD/mandoc (probably affecting others), everything
except xkbcli was being listed twice and wrapping on standard console
in apropos because mdoc(7) names should not have spaces or be different
than the title.
Further, xkb is already parsed in search results by being in the name,
so expand xkb to "X keyboard" so that they match for apropos keyboard.
We did this already in xkbutils, which matches setxkbmap.
These don't need to be quoted literals, and - doesn't need to be escaped,
but I left them because there may be environments which are not to spec,
but I'm happy to ammend this commit if that cleanup seems desirable.
Co-authored-by: Ran Benita <ran@unusedvar.com>
Xvfb is required for *running* X11 tests. We do not make it mandatory
in the meson setup though, because we did not find a way to require a
*run* time dependency.
- Add a diagram to illustrate how key types work.
- Add examples to explain the key types mappings.
- Add note for using `preserve` to tweak shortcuts.
- Improve the key statement section.
On macOS, it seems to be responsible for the following error in the CI:
```
_XSERVTransmkdir: ERROR: euid != 0,directory /tmp/.X11-unix will not be created.
```
With the current setup, the tests passes even if there is an issue with
the setup. X11 tests are already guarded with `enable-x11=true`, so
there is no reason to skip these tests if we built the X11 support.
Meson supports the GNU convention for exit code in tests:
- 77: test skipped; already used in xkbcommon: `SKIP_TEST`.
- 99: hard error/setup failure; introduced in this commit:
`TEST_SETUP_FAILURE`.
Meson reports setup failure as `ERROR`, while a test failure is reported
as `FAIL`. In both cases it will make the CI fail, allowing us to
correctly detect any errors.
Fixed:
- Use `TEST_SETUP_FAILURE` instead of `SKIP_TEST` for X11 setup
failures.
- Xvfb wrapper: Add verbosity for debugging.
Install meson and pyyaml via brew, as this the standard way on macOS.
Remove Doxygen and Graphviz dependencies, as we do not build
documentation by default anymore.
Signed-off-by: Rui Chen <rui@chenrui.dev>
The only docs we build is the doxygen API and I reckon there's a near-zero
amount of users who actually want these built locally instead of just browsing
the online doc.
Let's not build this by default, dropping the requirement on doxygen and
speeding up the build by quite a fair bit.
Towncrier is a utility to produce useful, summarized news files.
See:
- https://pypi.org/project/towncrier/
- https://towncrier.readthedocs.io
Custom configuration for xkbcommon:
- New fragments are located in the `changes` directory.
- 3 sections:
- API: `changes/api`
- Tools: `changes/tools`
- Build System: `changes/build`
- 3 news fragments:
- Breaking changes: `.breaking`
- New: `.feature`
- Fixes: `.bugfix`
`NEWS` is renamed to `NEWS.md` because the tool requires `.md`
extension to write in markdown format.
`xkb_keysym_from_name` has inconsistent behavior when used with the flag
`XKB_KEYSYM_CASE_INSENSITIVE`:
```c
xkb_keysym_from_name("a", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_a;
xkb_keysym_from_name("A", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_a;
xkb_keysym_from_name("dead_a", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_dead_A;
xkb_keysym_from_name("dead_A", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_dead_A;
xkb_keysym_from_name("dead_o", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_dead_o;
xkb_keysym_from_name("dead_O", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_dead_o;
xkb_keysym_from_name("KANA_tsu", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_kana_tsu;
xkb_keysym_from_name("KANA_TSU", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_kana_tsu;
xkb_keysym_from_name("KANA_ya", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_kana_YA;
xkb_keysym_from_name("KANA_YA", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_kana_YA;
xkb_keysym_from_name("XF86Screensaver", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_XF86ScreenSaver;
xkb_keysym_from_name("XF86ScreenSaver", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_XF86ScreenSaver;
```
So currently, if two keysym names differ only by case, then the
lower-case *keysym* is returned, not the keysym corresponding to the
lower-case keysym *name*. Indeed, `xkb_keysym_from_name` uses
`xkb_keysym_is_lower` to test if a keysym is a lower-case keysym.
Let’s look at the example for keysyms `a` and `A`: we get the keysym `a`
not because its name is lower case, but because `xkb_keysym_is_lower(XKB_KEY_a)`
returns true and `xkb_keysym_is_lower(XKB_KEY_A)` returns false.
So the results are correct according to the doc:
- Katakana is not a bicameral script, so e.g. `kana_ya` is *not* the lower
case of `kana_YA`.
- As for the `dead_*` keysyms, they are not cased either because they do
not correspond to characters.
- `XF86ScreenSaver` and `XF86Screensaver` are two different keysyms.
But this is also very counter-intuitive: `xkb_keysym_is_lower` is not
the right function to use in this case, because one would expect to check
only the name, not the corresponding character case:
```c
xkb_keysym_from_name("KANA_YA", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_kana_ya;
xkb_keysym_from_name("XF86ScreenSaver", XKB_KEYSYM_CASE_INSENSITIVE) == XKB_KEY_XF86Screensaver;
```
Fixed by making the order of the keysyms names consistent in `src/ks_tables.h`:
1. Sort by the casefolded name: e.g. `kana_ya` < `kana_YO`.
2. If same casefolded name, then sort by cased name, i.e for
ASCII: upper before lower: e.g `kana_YA` < `kana_ya`.
Thus we now have e.g. `kana_YA` < `kana_ya` < `kana_YO` < `kana_yo`.
The lookup logic has also been simplified.
Added exhaustive test for ambiguous case-insensitive names.
Do not accept statements like garbage.level_name in types files
Fix parser accepting clearly nonsensical type definitions like
type "ONE_LEVEL" {
garbage.modifiers = None;
garbage.map[None] = Level1;
garbage.level_name[Level1] = "Any";
};
and ignoring the garbage part.
Co-authored-by: Mikhail Gusarov <dottedmag@dottedmag.net>
Co-authored-by: Pierre Le Marre <dev@wismill.eu>
- Simplify error handling.
- Improve error message: add message ID and relevant
quotes and try to standardize a bit.
- Add proper doc for in the message registry.
Note: Instead of testing the value of `expr.op`, we test if the argument
`elem` of `ExprResolveLhs` is set: this allows us to catch also the
error with `x.y[z]` rather than just `x.y` as previously.
- Add introduction to keymap and its components
- Add a diagram to explain the relationships between RMLVO and KcCGST.
- Add keywords & comments sections.
- Improve Quick guide section in the index page.
- Add links to User-configuration page.
- Fix typos.
with lld 1.17, linking with a symbol map which contains symbols which actually do not exists now result in an error, which means the test in meson.build to check if -Wl,--version-script works always fails. because it tries to use the general map file with "int main(void) {}".
Fixes#424
Print the op too to make it easier to find which part in a long test_key_seq
triggered an assertion and add some whitespace for nicer debugging output.
* interactive-evdev: reduce the printed space used for keysyms
In commit 8cca3a7bfb the buffer for the keysym
was extended to accommodate for up to XKB_COMPOSE_MAX_STRING_SIZE bytes. This
caused the printf to expand to the same width for the keysym alone, making the
output less useful. Drop this back down to the same width it was before.
Currently we do not check that the Compose files we try successively are
*regular* files. This may result in an error, while we should just really
just skip the corresponding path.
Fixed by adding the new utily function `open_file`.
For the sake of compatibility, this reintroduce some deleted keysyms and
postpone the effective deprecation of others.
xorgproto commit: fe12c5102762afcbf852e50dcbbdea2ef625570c
Also added tests for some canonical names.
An empty element is allowed in SymbolsBody definition, so the following
keymap is gramatically correct.
```
xkb_keymap {
...
xkb_symbols "sym" {
key <SPC> {, [Space] };
};
};
```
However, the current parser crashes with the keymap due to null pointer
access.
This change fixes it by changing the parser not to allow it.
The current `xkb_compose_table_iterator_next` segfaults when used with an
empty table. Indeed, in this case we initialize cursors in
`xkb_compose_table_iterator_new` with the dummy node and the direction
`NODE_LEFT`, but the dummy node is a leaf!
Fixed by initializing with no cursors when the table is has no non-dummy
nodes.
Fixed the upper case mappings for:
- XKB_KEY_ydiaeresis
- XKB_KEY_mu
- XKB_KEY_ssharp
Note: In Unicode, the upper case of “ß” (U+00DF) is *not* “ẞ” (U+1E9E)
but “SS”. “ẞ” is reserved for text in capitals.
Added tests of the simple case mappings when the ICU library is
available. A warning is raised for missing mappings.
Note: `xkb_keysym_is_upper` is interpreted as matching the disjunction of the
Unicode character properties “Uppercase” or “Titlecase”.
Currently `xkb_keysym_is_modifier` does not detect the following keysyms:
- `XKB_KEY_ISO_Level5_Shift`
- `XKB_KEY_ISO_Level5_Latch`
- `XKB_KEY_ISO_Level5_Lock`
Indeed, there is a mistake in the keysym interval that the code checks.
The reason seems a confusing order of the keysyms in
`xkbcommon-keysyms.h`: the current code has a comment “libX11 only goes
up to XKB_KEY_ISO_Level5_Lock”, but in fact the modifiers keysyms are
listed in a _semantic_ order in `xkbcommon-keysyms.h`, not in the
increasing keysym _value_ order.
Fixed by using the same (correct) code as libX11 and added some tests.
Currently there is no indication of the maximum length of keysym names.
This is statically known, so add the new *internal* following API:
`XKB_KEYSYM_NAME_MAX_SIZE`.
Add an efficient way to iterate over the assigned keysyms.
Currently only provided for testing, so we guard it by
`ENABLE_PRIVATE_APIS` in order to reduce the installed library.
Currently there is no direct way to know the minimum and maximum keysym
values that are assigned, i.e. that have an explicit name or are
Unicode keysyms.
Introduce the new following internal API:
- XKB_KEYSYM_MIN_ASSIGNED
- XKB_KEYSYM_MAX_ASSIGNED
- XKB_KEYSYM_MIN_EXPLICIT
- XKB_KEYSYM_MAX_EXPLICIT
- XKB_KEYSYM_COUNT_EXPLICIT
Also add a bunch of tests to ensure consistant keysyms bounds.
I couldn't find any reference to *how* the keymap format actually needs to look
like if you want multiple keysyms per level. So let's add a test for it and a
minimal documentation entry.
Previously the attribute “popularity” was completely ignored. It also
did not respect the modified DTD, because its default value depends if
we are currently parsing an “extras” rules file.
Fixed:
- Always parse the popularity attribute.
- Change the DTD to reflect that the default value is implied.
Current options to set the locale are convoluted:
- An explicit locale *must* be given, while a sane default would be
to use the user environment.
- Then there are two options that were useful while testing locale
handling: read environment variables or use `setlocale`. But the
program has already called:
```
setlocale(LC_ALL, "");
```
so it turns out the two options lead to the same results.
Remove options `--locale-from-env` and `--locale-from-setlocale`
and make the locale default to the user environment.
- Add check for recursive includes of keymap components. It relies on
limiting the include depth. The threshold is currently to 15, which
seems reasonable with plenty of margin for keymaps in the wild.
- Add corresponding new log message `recursive-include`.
- Add tests for recursive includes.
The string `buf` was not freed after each call to `asprintf_safe`.
Avoid allocating and introduce the new message: `XKB_ERROR_INSUFFICIENT_BUFFER_SIZE`.
When there is no error the types are “stolen” and copied to the keymap.
But when there is an error, `MergeIncludedKeyTypes` just return without
“stealing” nor freeing the types.
Fixed by explicitly freeing the key types.
Fixed another leak in `HandleKeyTypeDef` that may occur if there is an
error in parsing a type definition.
GCC 14 introduces a new -Walloc-size included in -Wextra which gives:
```
src/state.c:589:9: warning: allocation of insufficient size ‘1’ for type ‘struct xkb_state’ with size ‘128’ [-Walloc-size]
```
The calloc prototype is:
```
void *calloc(size_t nmemb, size_t size);
```
So, just swap the number of members and size arguments to match the prototype, as
we're initialising 1 struct of size `sizeof(struct xkb_state)`. GCC then sees we're not
doing anything wrong.
Signed-off-by: Sam James <sam@gentoo.org>
Leading BOM is legal and is used as a signature — an indication that
an otherwise unmarked text file is in UTF-8.
See: https://www.unicode.org/faq/utf_bom.html#bom5 for further details.
Leading BOM is legal and is used as a signature — an indication that
an otherwise unmarked text file is in UTF-8.
See: https://www.unicode.org/faq/utf_bom.html#bom5 for further
details.
Leading BOM is legal and is used as a signature — an indication that
an otherwise unmarked text file is in UTF-8.
See: https://www.unicode.org/faq/utf_bom.html#bom5 for further details.
The handling of keysym name guards (e.g. `#ifndef XK_Ydiaeresis`) was
incomplete and led to a missing keysym.
Make `sripts/makeheader` more robust to C macros handling.
- Add bash completion script. It parses the commands help messages to
provide the completions, thus any new subcommand or option will be
supported, as long as it has its entry in the help messages. This
should result in low maintenancei effort.
- Add installation entry in Meson. The path can be configured using
the following options:
- `enable-bash-completion` to enable the installation;
- `bash-completion-path` to control the installation path. It will
default to: `share/bash-completion/completions`.
TODO: completion for other shells, such as zsh?
See https://pre-commit.com for further info. Current checks:
- Whitespaces
- Yaml
- Large files
- Python linting/formatting
TODO: investigate options for C linter and formatter
- Update keymap to use reference keysym names.
- Fix x11comp test by handling old x11proto.
We need xkbcomp to be compiled with at least x11proto-dev 2023.2.
So we replace the unsupported keysyms with supported ones not
already in the keymap. This is kind of ugly, but it works. If we
ever want to restore the original keysyms with their supported names,
the substitute keysyms will be easy to spot.
The octal parser accepts the range `\1..\777`. The result is cast to
`char` which will silently overflow.
This commit prevents overlow and will treat `\400..\777` as invalid
escape sequences.
NULL usually terminates the strings; allowing to produce it via escape
sequences may lead to undefined behaviour.
- Make NULL escape sequences (e.g. `\0` and `\x0`) invalid.
- Add corresponding test.
- Introduce the new message: XKB_WARNING_INVALID_ESCAPE_SEQUENCE.
Allow users to iterate the entries in a compose table. This is useful
for other projects which want programmable access to the sequences,
without having to write their own parser.
- New API:
- `xkb_compose_table_entry_sequence`;
- `xkb_compose_table_entry_keysym`;
- `xkb_compose_table_entry_utf8`;
- `xkb_compose_table_iterator_new`;
- `xkb_compose_table_iterator_free`;
- `xkb_compose_table_iterator_next`.
- Add tests in `test/compose.c`.
- Add benchmark for compose traversal.
- `tools/compose.c`:
- Print entries instead of just validating them.
- Add `--file` option.
- TODO: make this tool part of the xkbcli commands.
Co-authored-by: Pierre Le Marre <dev@wismill.eu>
Co-authored-by: Ran Benita <ran@unusedvar.com>
Signed-off-by: Ran Benita <ran@unusedvar.com>
Previously we had two types of macros for logging: with and without
message code. They were intended to be merged afterwards.
The idea is to use a special code – `XKB_LOG_MESSAGE_NO_ID = 0` – that
should *not* be displayed. But we would like to avoid checking this
special code at run time. This is achieved using macro tricks; they
are detailed in the code (see: `PREPEND_MESSAGE_ID`).
Now it is also easier to spot the remaining undocumented log entries:
just search `XKB_LOG_MESSAGE_NO_ID`.
Motivation: normalization of keysyms header files in `xorgproto`. See:
https://gitlab.freedesktop.org/xorg/proto/xorgproto/-/merge_requests/80
Improve `scripts/makeheader`:
- Simplify `evdev` and `XK_` substitution and improve alignment. Also, perform
some additional `XK_` substitutions in comments.
- Format with `black`.
Usually it is better to use the corresponding human-friendly keysym
names. If there is none, then the keysym is most probably not
supported in the ecosystem. The only use case I see is similar to the
PUA in Unicode (see: https://en.wikipedia.org/wiki/Private_Use_Areas).
I am not aware of examples of this kind of use.
This tool checks whether messages codes are supported.
This is useful e.g. for CI, where one may want to grep for some XKB
error codes and ensure that these are still supported.
Currently there is little structure in the log messages, making
difficult to use them for the following use cases:
- A user looking for help about a log message: the user probably
uses a search engine, thus the results will depend on the proper
indexing of our documentation and the various forums. It relies
only on the wording of the message, which may change with time.
- A user wants to filter the logs resulting of the use of one of the
components of xkbcommon. A typical example would be testing
xkeyboard-config against libxkbcommon. It requires the use of a
pattern (simple words detection or regex). The issue is that the
pattern may become silently out-of-sync with xkbcommon.
A common practice (e.g. in compilers) is to assign unique error codes
to reference theses messages, along with an error index for
documentation.
Thus this commit implements the following features:
- Create a message registry (message-registry.yaml) that defines the
log messages produced by xkbcommon. This is a simple YAML file that
provides, for each message:
- A unique numeric code as a short identifier. It is used in the
output message and thus can be easily be filtered to spot errors
or searched in the internet. It must not change: if the
semantics of message changes, it is better to introduce a new
message for clarity.
- A unique text identifier, meant for two uses:
1. Generate constants dealing with log information in our code
base.
2. Generate human-friendly names for the documentation.
- A type: currently warning or error. Used to prefix the constants
(see hereinabove) and for basic classification in documentation.
- A short description, used as concise and mandatory documentation.
- An optionnal detailed description.
- Optional examples, intended to help the user to fix issues
themself.
- Version of xkbcommon it was added. For old entries this often
unknown, so they will default to 1.0.0.
- Version of xkbcommon it was removed (optional)
No entry should ever be deleted from this index, even if the message
is not used anymore: it ensures we have unique identifiers along the
history of xkbcommon, and that users can refer to the documentation
even for older versions.
- Add the script update-message-registry.py to generate the following
files:
- messages.h: message code enumeration for the messages currently
used in the code base. Currently a private API.
- message.registry.md: the error index documentation page.
- Modify the logging functions to use structured messages. This is a
work in progress.
Add an option to print modmap and vmodmap of relevant keys, as well as
virtual modifiers mapping to real modifier. This is useful for debugging.
It uses private API, so we compile it separately in the fashion of
`xkbcli-compile-keymap/compile-keymap`.
Currently the interactive tools print the string result of key strokes
as it is, without any escape. This is especially annoying for trivial
keysyms such as: Return, BackSpace and Escape.
Fix this by displaying the Unicode code point notation (e.g U+000D for
Return) for single control characters from the C0 set and DEL.
This is a hack: ideally we would like to escape any non-printable
character in the utf-8 string.
Based on the work done by Peter Hutterer. Original commit message:
If SIGUSR1 is set to SIG_IGN, X servers (all of them, including Xvfb)
will send that signal to the parent process when they're ready to accept
connections. We can use that instead of a hardcoded sleep which brings
the wait down to ~37ms on my box.
The x11 test is currently silently skipped in CI, because it requires a
running X server.
Create a xvfb wrapper to run the test. We do not use `xvfb-run`, because
it is a shell script and it causes valgrind to detect unrelated memory
issues in the shell (dash, bash).
Improve wrapper using a special ELF section
TODO: The wrapper is intended to be used with the x11comp test as well.
```
libxkbcommon/doc/introduction-to-xkb.md:67: warning: unable to resolve reference to 'rule-file-format' for \ref command
libxkbcommon/doc/introduction-to-xkb.md:181: warning: unable to resolve reference to 'keymap-text-format-v1' for \ref command
libxkbcommon/doc/rules-format.md:10: warning: unable to resolve reference to 'xkb-intro' for \ref command
```
Signed-off-by: Ran Benita <ran@unusedvar.com>
- Add a keymap test with decimal and hexadecimal keysyms.
- Reorganize code in `test/keysym.c` by parsing type: name, Unicode and
hexadecimal.
- Add more tests for edge cases. In particular:
- test decimal format (currently not supported);
- test the Unicode and hexadecimal ranges more thoroughly;
- test with wrong case without the XKB_KEYSYM_CASE_INSENSITIVE flag;
- test surrounding spaces.
- Document the tests.
In `parser.y`, a numeric keysym is parsed by formatting it in its
hexadecimal form then parsed as a keysym name. This is convoluted.
Fixed by checking directly the upper bound.
When parsing hexadecimal keysym using `xkb_keysym_from_name`,
the result is limited by `parse_keysym_hex` to 0xffffffff, but the
maximum keysym is XKB_MAX_KEYSYM, i.e. 0x1fffffff.
Fixed by adding an upper bound.
Keysyms are 32-bit integers with the 3 most significant bits always set
to zero. See: Appendix A “KEYSYM Encoding” of the X Window System
Protocol at https://www.x.org/releases/current/doc/xproto/x11protocol.html#keysym_encoding.
Add a new constants XKB_KEYSYM_MIN and XKB_KEYSYM_MAX to make the
interval of valid keysyms more obvious in the code.
Doxygen 1.9.7 breaks our urls, see issue #347. Let's put a check for the
doxygen version into our CI build so that if our base distro updates beyond
that, the CI fails and we know we have to build doxygen from scratch or
update to some other version that's supported.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Upload the doxygen output as artifact from the linux build and use that from
the pages job where we combine the static website with our newly build HTML
docs. The GitHub actions/download-artefact doesn't work across workflows so
we use the other popular one that can do this. The rest of the job is
basically copy/paste from the "Static HTML" example GitHub provides.
To make this useful as drop-in replacement, replace the one fixed link to the
API docs a relative one.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The script `ensure-stable-doc-urls.py` relies on the Doxygen output files
names. These may change between Doxygen versions, although the Doxygen
developers intend stability.
Since the script is useful mainly for the online documentation of
xkbcommon, make the target `doc-cool-uris` optional.
The test `test_from_locale` fails when there is a user-defined compose
file, e.g. ~/.XCompose.
Indeed, the function `xkb_compose_table_new_from_locale` use various
environment variables to determine the location of the compose file.
Ensure no environment variables but the required ones are set, in order
to have robust tests.
Currently there is no interactive tool allowing to set the include
paths of the context, such as in "compile-keymap". Note that only
"interactive-evdev" makes sense, because it does not rely on a
compositor.
Add --include and --include-defaults to "interactive-evdev" tool.
The code is adapted from "compile-keymap".
- Force Doxygen version 1.9.6, because 1.9.7 breaks our doc cool URIs,
i.e. our system that maintain stable doc pages via redirection.
Associated issue in Doxygen:
https://github.com/doxygen/doxygen/issues/10146
- Install graphviz to build the .dot files.
In commit 1638409b22, the number of
compose nodes was limited to 65535 to enable "future optimizations",
which apparently means slightly reduced memory usage due to fitting in
a uint16_t. At this time, it was mentioned that the author was not
aware of "any compose files which come close".
However, I'm one of the users that actually do require a larger number
of nodes for their compose file. Thus, use a uint32_t again and raise
the limit significantly.
Doc URLs may change with time because they depend on Doxygen machinery.
This is unfortunate because it is good practice to keep valid URLs
(see: https://www.w3.org/Provider/Style/URI.html).
I could not find a built-in solution in Doxygen, so the solution proposed
here is to maintain a registry of all URLs and manage legacy URLs as
redirections to their canonical page.
This commit adds a registry of URLs that has three functions:
- Check no previous URL is now invalid.
- Add aliases for moved pages.
- Generate redirection pages for aliases. The redirection works with
a simple <meta http-equiv="refresh"> HTML tag. See:
https://developer.mozilla.org/en-US/docs/Web/HTML/Element/meta#http-equiv
This commit also initialize the URLs registry with current pages and some
redirections needed after recent documentation refactoring.
Finally, the CI is updated to catch any change that invalidate previous
URLs.
There is no feedback that they are both replaced with default values.
Fix it by adding a warning informing about missing layout and show the
defaults for both.
At least 2.3 (released 2006) which is the version shipped with macos
doesn't work. Reading the changelog I think 2.3a *should* work, so
require that.
Signed-off-by: Ran Benita <ran@unusedvar.com>
As the documentation for xkb_keymap_new_from_buffer() states, the "input string
does not have to be zero-terminated". The actual implementation however failed
with "unrecognized token/syntax error" when it encountered a null byte.
Fix this by allowing a null byte at the last position of the buffer. Anything
else is likely a client error anyway.
Fixes#307
This is backend-agnostic and the recommended way to invoke compilation.
This requires meson 0.55 (July 2020) and we only require 0.51 but at this point
I'm expecting most users who require the README instructions will have a
recent-enough meson anyway.
strdup() is the least likely call to fail here, let's move it to the bottom so
we don't need to worry about the allocated string.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
meson needs to know that the executable tools
link against libxkbcommon.dylib so that the
@rpath references used during the build/test phases
can be rewritten to full path names on install
Unlike current xkbcommon, X11’s xkbcomp allows to remove entries in
the modifiers’ map using “modifier_map None { … }”.
“None” is translated to the special value “XkbNoModifier” defined in
“X11/extensions/XKB.h”. Then it relies on the fact that in "CopyModMapDef",
the following code:
1U << entry->modifier
ends up being zero when “entry->modifier” is “XkbNoModifier” (i.e. 0xFF).
Indeed, it relies on the overflow behaviour of the left shift, which in
practice resolves to use only the 5 low bits of the shift amount, i.e.
0x1F here. Then the result of “1U << 0xFF” is cast to “char”, i.e. 0.
This is a good trick but too magical, so in libxkbcommon we will use
an explicit test against our new constant XKB_MOD_NONE.
- Add table of contents
- Add terminology section
- (WIP) Add Introduction to the format
- Improve the keycode section
- Improve the interpret section
- Add guide to create and use modifiers
- (WIP) Add actions documentation
- Add cross-references
- Add keysyms header to documentation
Meson complains; it's probably not that important anymore when using the
`-Db_sanitize` options.
meson.build:36: WARNING: Consider using the built-in option for sanitizers instead of using "-fsanitize-undefined-trap-on-error".
Signed-off-by: Ran Benita <ran@unusedvar.com>
A test case failure usually also triggers valgrind leaks, sifting through those
to find the actual test failure is painful. So let's separate the tests and run
them separately.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We need to request the lower version of the interface versions we
support and the server supports, not the higher version.
Using the higher version caused crashes due to unbound callbacks on
GNOME, which supports a higher version of `xdg_wm_base`.
This entry maps a non existing special keysym 0x20a9.
The correct mapping for XKB_KEY_Korean_Won (0x0eff) already exists.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Not sure what it's doing here, but converting "€" to a keysym
doesn't work with this entry. 0x13a4 doesn't appear in
xkbcommon-keysyms.h. 0x20ac is the keysym documented in the
header (and it's the last entry in the table).
It's been in the table since it was introduced in e0524296d2
("Add API for getting unicode representation of a keysym").
Co-authored-by: Simon Ser <contact@emersion.fr>
lokid and hikid actually stores the sibling to current node, which
should not be cleared when override. This would break the sequence with
a common prefix when override another.
Fix#286
Signed-off-by: Weng Xuetian <wengxt@gmail.com>
From the documentation:
> It does not clean up parser state, it cleans up memory allocated by the library
> itself. It is a cleanup function for the XML library. It tries to reclaim all
> related global memory allocated for the library processing. [...]
> One should call xmlCleanupParser() only when the process has finished using the library.
http://xmlsoft.org/html/libxml-parser.html#xmlCleanupParser
Since we're a library ourselves we cannot know if something else in the same
proces uses the parser, so we must not call this.
Reported-by: M Hickford
This makes it easier for contributors to check if their code runs correctly
without having to file a PR.
The Mac and Windows workflows are a bit more involved, so let's keep those on
pull requests only.
"main" is a bit non-descriptive, let's name them after the platforms we run them
on. Splitting them up allows us to be less selective on how we run the various
workflows, e.g. always running the linux one.
Because otherwise the 'no' layout is treated as disagreement with whatever is to
be disagreed with. Fixed in YAML 1.2 but that's not universally supported.
Fixes#268
In 1b3a1c277a we changed the error
handling in this code to not bail out immediately but only after
everything has been processed, to simplify the code. But I suspect the
code isn't prepared for this and that's what causing the crash reported
in the issue.
Bring back the short-circuit error handling which would hopefully fix
such crashes.
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/252
Signed-off-by: Ran Benita <ran@unusedvar.com>
Previously, could only check how to type a Unicode codepoint, but
searching for a keysym directly is also occasionally useful.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Don't assume that keycode 9 means Escape. Instead, use the keymap
to check for Esc.
Logic copied from the Wayland version.
Signed-off-by: Simon Ser <contact@emersion.fr>
While the previous 1987-style[0] scheme was fun (and I reasonably
optimized it for a fair comparison), this task is more suited to a hash
table. Even a simple implementation beats the old one.
[0] Seems to have first appeared in X11R1, released September 1987.
See server/dix/atom.c here: https://www.x.org/releases/X11R1/X.V11R1.tar.gz
Signed-off-by: Ran Benita <ran@unusedvar.com>
Specify where to find the headers for libxkbcommon_dep,
libxkbcommon_x11_dep, and libxkbregistry_dep, which allows other
projects to correctly locate the headers when libxkbcommon is being
built as a Meson subproject.
The dep_libxkbregistry variable is renamed to libxkbregistry_dep,
to follow the usual convention for variables which hold declared
dependencies to be used from subproject builds.
Arrange for passing .def files with the lists of symbols to export from
DLLs when building on Windows with MSVC. Without this no symbols were
being exported at all.
The .def files are generated from the .map files at build time using
scripts/map-to-def, which avoids needing to maintain two different sets
of files.
We have a lot of keyboard layouts and the current output format is virtually
useless at searching for a specific one to debug any issues with either the
layout list or the output from libxkbregistry.
Let's use YAML instead because that can easily be post-processed to extract the
specific layouts wanted, e.g. to get the list of all layouts:
xkbcli-list | yq -r ".layouts[].layout"
to get the list of all variants of the "us" layout:
xkbcli-list | yq -r '.layouts[] | select(.layout == "us") | .variant
and the number of option groups:
xkbcli-list | yq -r '.option_groups[] | length'
Note that the top-level nodes have been de-capitalized, so where it was "Models"
before it is now "models" and the "Options" node is now "option_groups".
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is hit when passing an empty string and XKB_KEYSYM_CASE_INSENSITIVE
to xkb_keysym_from_name currently if `(lo + hi) / 2` is 0 and `cmp < 0`,
causing mid to underflow and the the array access into name_to_keysym on
the next iteration of the loop to be out of bounds .
We *would* use ssize_t here as it is the appropriate type, but windows
unfortunately does not define it.
Add a skipError test function that can analyse rc, stdout, stderr to skip a test
even when we expect an error. We use that to skip if we couldn't find a keyboard
in the interactive-evdev test.
Fixes#235
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
tqdm prints to stderr by default but we're using that for failed keymap
compiles (which are the ones that really matter). Plus, whether we're using tqdm
is dependent on isatty(sys.stdout) anyway.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
With --keymap-output-dir, the given directory will contain a list of files named
after the layout + variant ('us', 'us(euro)', ...) that contain the keymaps for
each variant + option combination compiled.
This is still a lot, but better to sift through hundreds of keymaps than tens of
thousands.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The previous output is largely unusable. The result in the CI test runs is a 6GB
file with every compiled keymap in it and while we can grep for ERROR, it's not
particularly useful.
Let's change this and print out YAML instead - that can be machine-processed.
This patch adds a new parent class that prints itself in YAML format,
the tool invocations are child classes of that class. The result looks like this:
Example output:
- rmlvo: ["evdev", "pc105", "us", "haw", "grp:rwin_switch"]
cmd: "xkbcli-compile-keymap --verbose --rules evdev --model pc105 --layout us --variant haw --options grp:rwin_switch"
status: 0
- rmlvo: ["evdev", "pc105", "us", "foo", ""]
cmd: "xkbcli-compile-keymap --verbose --rules evdev --model pc105 --layout us --variant foo"
status: 1
error: "failed to compile keymap"
Special status codes are: 99 for "unrecognized keysym" and 90 for "Cannot open
display" in the setxkbmap case.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Fallback to ftruncate() if the underlying filesystem does not
support posix_fallocate().
Idea by: Jan Beich <jbeich@FreeBSD.org>, Niclas Zeising <zeising@FreeBSD.org>
Inspired by: Wayland cursor/os-compatibility.c
[ran: small adjustments]
Signed-off-by: Ran Benita <ran@unusedvar.com>
The numpad:mac option doesn't specify a name for the first level:
// On Mac keypads, level 1 and 2 are swapped.
partial xkb_types "mac" {
type "KEYPAD" {
modifiers = None;
map[None] = Level2;
level_name[Level2] = "Number";
};
include "extra(keypad)"
};
This means the atom for level name is XCB_ATOM_NONE. We tried to get its
name, which fails. This regressed in 40c00b4721.
Instead, translate it to XKB_ATOM_NONE, same as the previous behavior.
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/229
Signed-off-by: Ran Benita <ran@unusedvar.com>
Make it use a bit operation instead of an expensive modulo.
perf diff:
Baseline Delta Abs Shared Object Symbol
........ ......... ................. ...................................
28.15% -6.57% bench-compose [.] xkb_keysym_from_name
Signed-off-by: Ran Benita <ran@unusedvar.com>
In 7d84809fdc I added a fast path for the
case-sensitive case, but it is still slowing down Compose parsing.
Instead of the binary search, use a perfect hash function, computed with
a simple python module I found (vendored).
It is faster -- perf diff is:
Baseline Delta Abs Shared Object Symbol
........ ......... ................. ...................................
22.35% -14.04% libc-2.33.so [.] __strcmp_avx2
16.75% +10.28% bench-compose [.] xkb_keysym_from_name
20.72% +2.40% bench-compose [.] parse.constprop.0
2.29% -1.97% bench-compose [.] strcmp@plt
2.56% +1.81% bench-compose [.] resolve_name
2.37% +0.92% libc-2.33.so [.] __GI_____strtoull_l_internal
26.19% -0.63% bench-compose [.] lex
1.45% +0.56% libc-2.33.so [.] __memchr_avx2
1.13% -0.31% libc-2.33.so [.] __strcpy_avx2
Also reduces the binary size:
Before:
text data bss dec hex filename
341111 5064 8 346183 54847 build/libxkbcommon.so.0.0.0
After:
text data bss dec hex filename
330215 5064 8 335287 51db7 build/libxkbcommon.so.0.0.0
Note however that it's still larger than before 7d84809fdc:
text data bss dec hex filename
320617 5168 8 325793 4f8a1 build/libxkbcommon.so.0.0.0
Signed-off-by: Ran Benita <ran@unusedvar.com>
Previously we used a simple trie with a linked list for each chain.
Unfortunately most compose files have very long chains which means the
constructions performs an almost quadratic number of comparisons.
Switch to using a ternary search tree instead. This is very similar to a
trie, only the linked list is essentially replaced with a binary tree.
On the en_US/Compose file, the perf diff is the following (the modified
function is `parse`):
Event 'cycles:u'
Baseline Delta Abs Shared Object Symbol
........ ......... ................ .................................
39.91% -17.62% bench-compose [.] parse.constprop.0
20.54% +6.47% bench-compose [.] lex
17.28% +5.55% libc-2.33.so [.] __strcmp_avx2
12.78% +4.01% bench-compose [.] xkb_keysym_from_name
2.30% +0.83% libc-2.33.so [.] __GI_____strtoull_l_internal
3.36% +0.78% bench-compose [.] strcmp@plt
Thanks to some careful packing, the memory usage is pretty much the
same.
Signed-off-by: Ran Benita <ran@unusedvar.com>
I'd really like to use anonymous unions/structs. Supposedly even MSVC
supports it now. Let's try and see.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Relatedly, strtoul allows a lot of unwanted stuff (spaces, +/- sign,
thousand seperators), we really ought not use it. But that's for another
time.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Fits in uint16_t, which enables some future optimizations. But also a
good idea to have some limit. Not aware of any compose files which come
close.
Signed-off-by: Ran Benita <ran@unusedvar.com>
../../../src/xkbcomp/compat.c:693:16: warning: Although the value stored to
'merge' is used in the enclosing expression, the value is never actually read
from 'merge' [deadcode.DeadStores]
si.merge = merge = (def->merge == MERGE_DEFAULT ? merge : def->merge);
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The target buffer is 7 bytes long, null-termination is optional (as the comment
already suggests). Coverity is unhappy about this though so let's use memset and
memcpy instead.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
xkb_keysym_from_name() is called a lot in Compose file parsing. The
lower case handling slows things down a lot (particularly given we can't
use the optimized strcasecmp() due to locale issues). So add separate
handling for the non-case-sensitive case which is used by Compose.
To do this we need to add another version of the ks_tables table. This
adds ~20kb to the shared library binary. We can probably do something
better here but I think it's fine.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Turns out FreeBSD supports evdev, so this toll can work on it; however
it does not support epoll, so switch to poll, which is portable.
Reported-by: Evgeniy Khramtsov <evgeniy@khramtsov.org>
Signed-off-by: Ran Benita <ran@unusedvar.com>
src/x11/keymap.c:980:26: warning: comparison of integer expressions of different signedness: ‘size_t’ {aka ‘long unsigned int’} and ‘int’ [-Wsign-compare]
980 | for (size_t i = 0; i < length; i++) {
| ^
Signed-off-by: Ran Benita <ran@unusedvar.com>
libxkbcommon-1.0.3/src/xkbcomp/ast-build.c:526: leaked_storage: Variable "file"
going out of scope leaks the storage it points to.
Where we exit the loop early, we don't release the various allocated memory.
Make this patch more obvious my moving the declaration for those into the loop
as well, this way we know that they aren't used outside the loop anywhere.
Found by coverity
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If the name is missing in a configItem, we'd fail and leak the memory for
description, brief and vendor.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
False positive because we rely on xkb_components_from_rules() to initalize this
struct, but let's localize the variable anyway to shut coverity up.
libxkbcommon-1.0.3/bench/rules.c:59:9: warning[-Wanalyzer-double-free]:
double-free of kccgst.symbols
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Both get_atom_name() and the new atom interner required a round trip. Move
get_atom_name() into the atom interner to save one more round trip. This brings
xkb_x11_keymap_new_from_device() down to two round trips, which is the minimum
possible number.
(Also, I think the new code in keymap.c is more readable than the mess I
previously created)
With this last commit in the series, this definitely:
Fixes: https://github.com/xkbcommon/libxkbcommon/pull/217
Signed-off-by: Uli Schlachter <psychon@znc.in>
There are a number of XKB requests needed to request all the information from
the X11 server. So far, the code was sending one request and waiting for the
reply. This commit starts batching the request so that we get multiple replies
with one round trip.
This removes three round trips.
Only the simple requests are converted. get_map() and get_names() use some
bitmasks that are needed for both the request and the reply. These will be dealt
with separately.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Instead of asking for an atom name and waiting for the reply four times, this
now sends four GetAtomName requests and waits for all the replies at once. Thus,
this saves three round trips.
Signed-off-by: Uli Schlachter <psychon@znc.in>
On my system, calling xkb_x11_keymap_new_from_device() did 78 round trips to the
X11 server, which seems excessive. This commit brings this number down to about
9 to 10 round trips.
The existing functions adopt_atom() and adopt_atoms() guarantee that the atom
was adopted by the time they return. Thus, each call to these functions must do
a round-trip. However, none of the callers need this guarantee.
This commit makes "atom adopting" asynchronous: Only some time later is the atom
actually adopted. Until then, it is in some pending "limbo" state.
This actually fixes a TODO in the comments.
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/216
Signed-off-by: Uli Schlachter <psychon@znc.in>
A simple script that creates a new layout with the given keysym replacing TLDE.
Then we compile a keymap and search for the keysym being assigned to TLDE and
bail if that fails.
The list of keysyms is manually maintained but we only need to add one or two to
spot-check whenever the xorgproto is updated.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
As of xorgproto commit 5dbb5b76597f [1], the 0x10081XXX keycode range is defined
for direct evdev kernel keycode mapping. For example, KEY_MACRO1 (0x290) is
mapped to 0x10081290. The format of the #define lines for these keys is
stable to allow for parsing:
#define XF86XK_FooBar _EVDEVK(0x123) /* optional comment */
Update our script so we detect these new lines. Our keysym generation is a
two-step process: makeheader and then makekeys. Replacing the key with its full
value in the makeheader script means we don't have to update makekeys to handle
the _EVDEVK macro and our header file is fully resolved.
[1] https://gitlab.freedesktop.org/xorg/proto/xorgproto/-/merge_requests/23
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
rmlvos is the parent list which then fails during a list join because, well,
it's a list of lists.
Fixes#206
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Makes this test easier to run from the commandline. Where either of top_srcdir
or top_builddir isn't set, fill them in from the CWD or fail otherwise.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
On every keymap notify event, the keymap should be refreshed, which
fetches the required X11 atoms. A big keymap might have a few hundred of
atoms.
A profile by a user has shown this *might* be slow when some intensive
amount of keymap activity is occurring. It might also be slow on a
remote X server.
While I'm not really sure this is the actual bottleneck, caching the
atoms is easy enough and only needs a couple kb of memory, so do that.
On the added bench-x11:
Before: retrieved 2500 keymaps from X in 11.233237s
After : retrieved 2500 keymaps from X in 1.592339s
Signed-off-by: Ran Benita <ran@unusedvar.com>
Where resolve_keysym fails we warn but use the otherwise uninitialized variable
as our keysym. That later ends up in the keymap as random garbage hex value.
Simplest test case, set this in the 'us' keymap:
key <TLDE> { [ xyz ] };
And without this patch we get random garbage:
./build/xkbcli-compile-keymap --layout us | grep TLDE:
key <TLDE> { [ 0x018a5cf0 ] };
With this patch, we now get NoSymbol:
./build/xkbcli-compile-keymap --layout us | grep TLDE:
key <TLDE> { [ NoSymbol ] };
For me, installing pytest for libxkbcommon is a bit problematic, so I
end up skipping it which is not great.
Switch to unittest which is built in to Python. It's not as nice as
pytest but good enough in this case.
Note: I was too lazy to switch the plain asserts to unittest
assertions...
Signed-off-by: Ran Benita <ran@unusedvar.com>
MacOS doesn't have eaccess/euidaccess but it does have unistd.h, so let's
include it to silence the R_OK redefinition compiler warnings.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
../src/context.c:57:9: warning: variable 'err' is used uninitialized whenever
'if' condition is true [-Wsometimes-uninitialized]
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
mkdtmp, rmdir and unlink are in unstd.h on MacOS. Since including that it
doesn't hurt us on Linux, let's do it without ifdefs.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
While this tool is useful for users starting with a new keyboard layout, it is a
somewhat bad fit for libxkbcommon. It's the only python tool, we don't even
install it yet (because we're not sure yet what it's supposed to do) and there's
a potential for it to expand into more corner cases.
The only tie it has to libxkbcommon is that it templates the data files that
libxkbcommon reads, but those files are effectively public API.
Let's remove this tool from there and instead move it to a separate git
repository where it can go its own way.
This reverts commit d00cf64dbc
PATH_MAX is not POSIX and can be missing on some systems, notably Windows (which
provides MAX_PATH instead tough) and Hurd. Let's define it to a sane value where
missing, i.e. the one it's defined to in limits.h. Except on Windows where
we're limited to 260.
Fixes https://github.com/xkbcommon/libxkbcommon/issues/180
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
It's a niche use-case but basically the same as adding symbols, so let's go with
a general handwavy explanation.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This tool set ups the required directory structure and template files to add new
keyboard layouts or options. For example, run like this:
xkbcli-scaffold-new-layout --layout 'us(myvariant)' --option 'custom:foo'
This will up the evdev rules file, the evdev.xml file, the symbols/us file and
symbols/custom file in $XDG_CONFIG_HOME so that the user has everything in place
and can start filling in the actual key mappings.
This tool is currently uninstalled until we figure out whether it's useful.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This completes the usual triplet of configuration locations available for most
processes:
- vendor-provided data files in /usr/share/X11/xkb
- system-specific data files in /etc/xkb
- user-specific data files in $XDG_CONFIG_HOME/xkb
The default lookup order user, system, vendor, just like everything else that
uses these conventions.
For include directives in rules files, the '%E' resolves to that path.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Previously, a 'symbols/us' file in path A would shadow the same file in path B.
This is suboptimal, we rarely need to hide the system files - we care mostly
about *extending* them. By continuing to check other lookup paths, we make it
possible for a XDG_CONFIG_HOME/xkb/symbols/us file to have sections including
those from /usr/share/X11/xkb/symbols/us.
Note that this is not possible for rules files which need to be manually
controlled to get the right bits resolved.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Streamline the code a bit - instead of handling all the if (!file) conditions
handle the case of where we have a file and jump to the end.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
'xkbcli compile-keymap' doesn't work unless we ninja install first. But for a
test that's to be run from the test directory, that's not a useful option so
let's call the binary directly. The script adds the meson builddir to the PATH
anyway.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Let's not have our tests fail if the user has an incompatible
$XDG_CONFIG_HOME/xkb directory.
libxkbcommon has fallbacks when XDG_CONFIG_HOME isn't set so we need to override
this with a real directory instead of just unsetting it.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The tools previously linked against a static version (by simply recompiling
everythiong). This isn't necessary, we can link them against libxkbcommon.so.
Only exception: The xbkcli-compile-keymap tool needs a private API for the
--kccgst flag. Avoid this by disabling this flag in the installed tool and
building the same tool, statically linked but not-installed.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is merely to fill in some NULL pointers anyway, we can just use
the #defines we have available at build time.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Drop the ronn source files, check in the generated files instead. This gets rid
of the ruby+gem+ronn toolchain requirement at the cost of having to edit raw man
pages.
ronn files are as-generated but with the preamble and generation date removed.
The latter isn't important enough to keep, it'll just go stale for manually
maintained files and it's not worth setting up a configure_file() just for that
date.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
A pytest wrapper around our xkbcli tool - copied from libinput.
This calls our various xkbcli tools with varying options and check that they
either succeed or return the right error code. The coverage is limited, it
does not (and cannot) test for all possible combinations but it should provide a
good red flag if we have inconsistent behavior or accidentally break some
combination of flags.
Meanwhile, we can at least assume that all our commandline arguments are parsed
without segfaulting or worse.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
On all platforms we build on where getopt.h is available, getopt_long is also
available. Only Windows doesn't have either but that's no reason for us to
differentiate between the two.
If we need to special-case getopt vs getopt_long, it's probably better to
implement our own cross-platform version of it and use that.
Fixes#161
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This provides consistency with the other tools that now all take long options.
Plus, it's more obvious to have the arguments spelled out.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Requiring long options for this tool means it's immediately obvious what an
invocation does, compare e.g.
xkbcli interactive-evdev -gcd
to the equivalent:
xkbcli interactive-evdev --consumed-mode=gtk --enalbe-compose --report-state-changes
This drops the evdev offset argument - that offset should never be anything
other than 8, having this as argument here is more likely to confuse or
produce misleading debugging logs.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The xkbcli tool usage help is ifdef'd out where the tool isn't built but the
man page always includes all tools. Easier that way.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is the base tool, no subtools are currently connected so you only get help
and version for now. The goal here is to have a git-like infrastructure where
/usr/bin/xkbcli is the main tool, anything else will hide in libexec.
The infrastructure for this is copied from libinput. Tools themselves will
will be installed in $prefix/libexec/xkbcommon and the xkbcli tool forks
off whatever argv[1] is after modifying the PATH to include the libexec dir.
libinput has additional code for checking whether we're running this from the
builddir but it's a bit iffy and it's usefulness is limited - if you're in the
builddir anyway you can just run ./builddir/xkbcli-<toolname> directly.
So for this code here, running ./builddir/xkbcli <toolname> will execute the
one in the prefix/libexecdir.
Since we want that tool available everywhere even where some of the subtools
aren't present, we need to ifdef the getopt handling.
man page generation is handled via ronn which is a ruby program but allows
markdown for the sources. It's hidden behind a meson option to disable where
downloading ronn isn't an option. The setup is generic enough that we can add
other man-pages by just appending to the array.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Windows doesn't have getopt.h. This would prevent building the tools but
they are behind other checks that cause them to be disabled. The only tools
that don't need getopt.h are interactive-wayland and interactive-x11 but
neither is particularly useful on Windows. Just hide all tools behind the getopt
check in preparation for the upcoming tool consolidation work.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
meson complains because this requires 0.50.0 and we don't require that. But
since it defaults to false anyway, let's just omit it.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This way we can invoke the expected setup with
meson test --setup=valgrind
And because we don't care about valgrinding python script, mark that test as
part of the "python-tests" suite and skip it during our CI valgrind run.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Ran Benita <ran@unusedvar.com>
Python leaks like crazy when run under valgrind. But if we make the script
executable **and** it has uses the env invocation (i.e. #!/usr/bin/env python3),
the leaks disappear. This is not the case for a shebang of /usr/bin/python3.
Why exactly this is the case I'm not sure but executables we plan to run
should have the exec bit set. So this is a janitor patch with the nice side
effect of fixing our valgrind runs.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Before reading ~/.XCompose, try to read $XDG_CONFIG_HOME/XCompose
(falling back to ~/.config/XCompose).
This helps unclutter the home directory of users who want that.
../meson.build:239: DEPRECATION: Library xkbcommon was passed to the "libraries"
keyword argument of a previous call to generate() method instead of first
positional argument. Adding xkbcommon to "Requires" field, but this is a
deprecated behaviour that will change in a future version of Meson. Please
report the issue if this warning cannot be avoided in your case.
This was introduced in meson 0.49.0
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The CI started installing some wrapper instead of a real bash which is
what gets found.
See:
https://github.com/actions/virtual-environments/pull/1081
Given meson is written in python, it should always be available
hopefully.
Disabled valgrind wrapper for now because it now also applies to the
python interpreter which leaks like a sieve.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Don't do the realloc dance, just asprintf to the buffer and move on. The check
is likely pointless anyway, if we run out of asprintf size, log_error will
probably blow up as well.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We only ever care about whether we error out or not, so let's wrap this into
something more sane.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Since the most common use-case is to provide only some elements of RMLVO, this
makes it possible to show what is actually being used in the background based on
the built-in defaults.
Print this in a format that's mostly JSON-compatible or at least easy to parse,
just in case.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This obsoletes the print-compiled-keymap tool though we now require that the
kccgst components are passed via stdin, there is no file loading ability.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Since we want to log the include paths too we need to split the context init up,
otherwise include paths are added before we can set the verbosity.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Now that we're relying on various different include paths, let's log that ones
we use and the ones we failed to use.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The most common case for this tool is to check what the keymap is for a given
RMLVO. Let's print that by default, the use-cases that just check for
compilation success can discard stdout instead.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This library is the replacement for clients parsing evdev.xml directly.
Instead, they should use the API here so that in the future we may even
be able to swap evdev.xml for a more suitable data format.
The library parses through evdev.xml (using libxml2) and - if requested -
through evdev.extras.xml as well. The merge approach is optimised for
the default case where we have a system-installed rules XML and another file in
$XDG_CONFIG_DIR that adds a few entries. We load the system file first, then
append any custom ones to that. It's not possible to overwrite the MLVO list
provided by the system files - if you want to do that, get the change upstream.
XML validation is handled through the DTD itself which means we only need to
check for a nonempty name, everything else the DTD validation should complain
about.
The logging system is effectively identical to xkbcommon.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Returns true on success or false on error _or_ truncation. Since truncation is
almost always an error anyway, we might as well make this easier to check.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The idea is to make the tools/demos as standalone as possible so that
they may serve as examples as well.
Signed-off-by: Ran Benita <ran@unusedvar.com>
This connects two tools to be useful together:
xkbcommon-rmlvo-to-kccgst | xkbcommon-print-compiled-keymap -
which will result in the full keymap generated by the former tool.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
These were moved to tools/tools-common.c and now that all tools are switched
over, they're no longer needed.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Move (sometimes duplicate) the required bits into new shared files
tools-common.(c|h) that are compiled into the internal tools library. Rename the
test_foo() functions to tools_foo() and in one case just copy the code of the
keymap compile function to the tool.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Passing -errno around and having separate labels depending on failure types is
superfluous here. All the unref calls can handle NULL and nothing cares about
errno once we're out of the immediate scope. So let's simplify this and deal
with 0 and 1 only.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is currently identical to the internal test library, but it's a start to
disentangle the two.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Using test helpers to init the context gives it fairly specific behavior; unless
the user sets the right environment variables and/or calls it from the right
PWD, it may or may not include the test data.
Let's drop this behavior, make it a default tool to compile a keymap. If there
is a specific need to modify the include paths, we can add this later.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Commit 16c84cdd81 removed the getopt handling for
RMLVO arguments, so now this tool only takes a keymap file and compiles it.
Using test helpers to init the context gives it fairly specific behavior; unless
the user sets the right environment variables and/or calls it from the right
PWD, it may or may not include the test data.
Let's drop this behavior, make it a default tool to compile a keymap. If there
is a specific need to modify the include paths, we can add this later.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The unref() functions take NULL as argument, so we don't need different labels
for every possible exit path.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This contained three groups entries:
- autotools stuff which no longer needed
- entries to ignore built artifacts, no longer needed with meson as everyting is
in the builddir
- special files like cscope.out, etc. These should be handled by a user in their
global gitignore file since they're fairly specific to a user's setup.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Most of this is currently hidden in the commit message for ca033a29d2, let's
make it a bit more public so we have a link to point users to.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Useful to have this as part of the documentation. The rendering isn't great but
at least not any worse than pure text. Markdown escapes % so explaining our use
of %S and %H would require a double % - not idea. Let's just wrap it as a code
block and done.
Includes two typo fixes too, yay.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If the active set of modifiers doesn't match any explicit entry of the
key type, the resulting level is 0 (i.e. Level 1). Some key types don't
explicitly map Level 1, taking advantage of this fallback.
Previously, xkb_keymap_key_get_mods_for_level didn't consider this, and
only reported masks for explicit mappings. But this causes some glaring
omissions, like matching "a" in the "us" keymap returning not results.
Since every mask which isn't explicitly mapped falls back to 0, we can't
return the all. Almost always the best choice for this is the empty
mask, so return that, when applicable.
Fixes https://github.com/xkbcommon/libxkbcommon/issues/140.
Reported-by: https://github.com/AliKet
Signed-off-by: Ran Benita <ran@unusedvar.com>
Link libxkbcommon_x11_internal with libxkbcommon_test_internal, rather
than libxkbcommon.
This avoids some tests linking with both libxkbcommon_test_internal and
libxkbcommon, which causes duplicate symbol problems on PE targets (e.g.
Cygwin) (as all the symbols from libxkbcommon are pulled in at link
time, which clash with libxkbcommon_test_internal)
We currently use Azure Pipelines. But it became out of date. Also it
requires a different account and setup than github account itself.
The configuration here is probably not very good and is less featureful
than the Azure one but it's what I managed.
Signed-off-by: Ran Benita <ran@unusedvar.com>
The program takes a unicode codepoint and an RMLVO and prints out all
key + modifier combinations that would result in that codepoint.
The program was written to exercise the new
xkb_keymap_key_get_mods_for_level() function. It's handy and can be
extended in several ways, but enough for now.
Example:
$ ./build/how-to-type -l us,il,ru 0x41 | column -ts $'\t'
keysym: A (0x41)
KEYCODE KEY NAME LAYOUT# LAYOUT NAME LEVEL# MODIFIERS
38 AC01 1 English (US) 2 [ Shift ]
38 AC01 1 English (US) 2 [ Lock ]
38 AC01 2 Hebrew 2 [ Shift ]
38 AC01 2 Hebrew 2 [ Lock ]
$ ./build/how-to-type -l de -v neo 0x3b6 | column -ts $'\t'
keysym: Greek_zeta (0x7e6)
KEYCODE KEY NAME LAYOUT# LAYOUT NAME LEVEL# MODIFIERS
56 AB05 1 German (Neo 2) 4 [ Shift Mod5 ]
56 AB05 1 German (Neo 2) 4 [ Shift Mod2 Mod3 Mod5 ]
56 AB05 1 German (Neo 2) 4 [ Shift Lock Mod5 ]
56 AB05 1 German (Neo 2) 4 [ Lock Mod2 Mod3 Mod5 ]
Signed-off-by: Ran Benita <ran@unusedvar.com>
The new API is useful to implement features like auto-type and
desktop automation. Since the inputs for these features is usually
specified in terms of the symbols that need to be typed, the
implementation needs to be able to invert the keycode->keysym
transformation and produce a sequence of keycodes that can be used
to type the requested character(s).
In order to support features like auto-type and UI automation, the
relevant tools need to be able to invert the keycode->keysym->text
transformation. In order to facilitate that, a new API was added.
It allows querying the keysyms that correspond to particular Unicode
codepoints. For all practical purposes, it can be thought of as an
inverse of xkb_keysym_to_utf32().
We apparently broke byacc support in the switch to meson.
byacc only supports short option names. And to make things fun, bison
only supports long option for `--defines`.
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/133
Signed-off-by: Ran Benita <ran@unusedvar.com>
This provides implementations of the test_enable_stdin_echo and
test_disable_stdin_echo which do not require <termios.h>, which is
not available on Windows.
Only the input/output functions from <unistd.h> options are used, so
using <io.h> when building with MSVC should be enough. The inclusion
of the header in context-priv.c does not seem to be needed (tested
on GNU/Linux) and so it is removed.
Signed-off-by: Ran Benita <ran@unusedvar.com>
This is a stopgap measure to quickly get tests building with MSVC for
now, at some point the tests could be rewritten to avoid using getopt()
and mkdtemp() or to ship an implementation.
The tests stringcomp and buffercomp do binary comparison on some files;
if the files are changed to CRLF on checkout, the tests fail.
Signed-off-by: Ran Benita <ran@unusedvar.com>
From MSVC:
test\atom.c(98): note: consider using '%zu' in the format string
test\atom.c(98): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 1 has type 'size_t'
test\atom.c(100): note: consider using '%zu' in the format string
test\atom.c(100): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 1 has type 'size_t'
test\atom.c(114): note: consider using '%zu' in the format string
test\atom.c(114): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 1 has type 'size_t'
test\atom.c(128): note: consider using '%zu' in the format string
test\atom.c(128): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 1 has type 'size_t'
test\atom.c(130): note: consider using '%zu' in the format string
test\atom.c(130): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 1 has type 'size_t'
test\atom.c(137): note: consider using '%zu' in the format string
test\atom.c(137): warning C4477: 'fprintf' : format string '%lu' requires an argument of type 'unsigned long', but variadic argument 2 has type 'size_t'
Signed-off-by: Ran Benita <ran@unusedvar.com>
Detected by MSVC:
xkbcomp\xkbcomp.c(111): warning C4047: 'return': 'bool' differs in levels of indirection from 'void *'
Signed-off-by: Ran Benita <ran@unusedvar.com>
Don't use int which can have different size on different machines.
Also avoid some warnings from MSVC:
xkbcomp/parser.y(760): warning C4244: '=': conversion from 'int64_t' to 'int', possible loss of data
xkbcomp/parser.y(761): warning C4244: '=': conversion from 'int64_t' to 'int', possible loss of data
xkbcomp/parser.y(767): warning C4244: '=': conversion from 'int64_t' to 'int', possible loss of data
Signed-off-by: Ran Benita <ran@unusedvar.com>
This increases the size of the struct a bit but it's not very important.
Fixes these MSVC warnings:
src\scanner-utils.h(112): warning C4267: '+=': conversion from 'size_t' to 'unsigned int', possible loss of data
src\scanner-utils.h(147): warning C4267: '+=': conversion from 'size_t' to 'unsigned int', possible loss of data
Signed-off-by: Ran Benita <ran@unusedvar.com>
Previously we included it with an `-include` compiler directive. But
that's not portable. And it's better to be explicit anyway.
Every .c file should have `include "config.h"` first thing.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Some environments (e.g. Windows + MSVC) do not provide asprintf() or
vasprintf(). This tries to detect their presence, and provides suitable
fallback implementations when not available.
Some environments (e.g. Windows + MSVC) do not provide strndup(), this
tries to detect its presence and provide a fallback implementation when
not available.
[ran: some tweaks]
This avoids the problem that MSVC does not provide strcasecmp() nor
strncasecmp(), and at the same time avoids potential problems due to
locale configuration by using istrcmp() and istrncmp() which are
already in the source tree and written to cover only ASCII.
This condition happens in xkeyboard-config keymaps and seems hard to
fix. Currently it incessantly spams people's logs who have no idea what
to do about it. So downgrade to "warning" level, so it doesn't show up
by default.
When working on keymaps, set `XKB_LOG_LEVEL=debug XKB_LOG_VERBOSITY=10`
to see all possible messages.
Refs https://github.com/xkbcommon/libxkbcommon/issues/111
Fixes https://github.com/xkbcommon/libxkbcommon/issues/128
Signed-off-by: Ran Benita <ran@unusedvar.com>
FindXkbFileInPath() opens the file so we're guaranteed that the file not only
exists, but that we can read it. Changing that would alter behavior so instead
let's just pass that file handle along and do the same for include files.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The majority use-case for extending XKB on a machine is to override one or a
few keys with custom keycodes, not to define whole layouts.
Previously, we relied on the rules file to be a single file, making it hard to
extend. libxkbcommon parses $XDG_CONFIG_HOME/xkb/ but that only works as long
as there is a rule that matches the user-specified RMLVO. This works for MLV
but not for options which don't have a wildcard defined. Users have to copy
the whole rules file and then work from there - not something easy to extend
and maintain.
This patch adds a new ! include directive to rules files that allows including
another file. The file path must be without quotes and may not start with the
literal "include". Two directives are supported, %H to $HOME and %S for the
system-installed rules directory (usually /usr/share/X11/xkb/rules).
A user would typically use a custom rules file like this:
! option = symbols
custom:foo = +custom(foo)
custom:bar = +custom(baz)
! include %S/evdev
Where the above defines the two options and then includes the system-installed
evdev rule. Since most current implementations default to loading the "evdev"
ruleset, it's best to name this $XDG_CONFIG_HOME/xkb/rules/evdev, but any
valid name is allowed.
The include functionally replaces the line with the content of the included
file which means the behavior of rules files is maintained. Specifically,
custom options must be defined before including another file because the first
match usually wins. In other words, the following ruleset will not assign
my_model as one would expect:
! include %S/evdev
! model = symbols
my_model = +custom(foo)
The default evdev ruleset has wildcards for model and those match before the
my_model is hit.
The actual resolved components need only be in one of the XKB lookup
directories, e.g. for the example above:
$ cat $XDG_CONFIG_HOME/xkb/symbols/custom
partial alphanumeric_keys
xkb_symbols "foo" {
key <TLDE> { [ VoidSymbol ] };
};
partial alphanumeric_keys
xkb_symbols "baz" {
key <AB01> { [ k, K ] };
};
This can then be loaded with the XKB option "custom:foo,custom:bar".
The use of "custom" is just as an example, there are no naming requirements
beyond avoiding already-used ones. Also note the bar/baz above - the option
names don't have to match the component names.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
No functional changes but we'll need that same lookup in the rules file
include handling in a future patch.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This shouldn't be processed in the matcher itself, especially in the glorious
future when we can have nested matchers. Only handle this once in the caller
to the original parsed file.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
No functional changes, this is what the macro expanded to anyway. Prep work
for putting the scanner on the stack and removing it from the matcher struct.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
lowercase: LATIN SMALL LETTER SHARP S (U+00DF)
uppercase: LATIN CAPITAL LETTER SHARP S (U+1E9E)
The uppercase sharp s (XK_ssharp) is a relatively recent addition to unicode
but was added to the relevant keyboard layouts in xkeyboard-config-2.25
(d1411e5e95c)
https://gitlab.freedesktop.org/xkeyboard-config/xkeyboard-config/issues/144
Alas, the CapsLock behavior was broken on the finnish layout (maybe others).
This was due XConvertCase() never returning the uppercase characters.
Let's make this function return the right lower/upper symbols for the sharp s
and hope that the world won't get any worse because of it.
Corresponding Xlib issue:
https://gitlab.freedesktop.org/xorg/lib/libx11/issues/110
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
In the AST, lists (e.g. the list of statements in a file) are kept in
singly-linked lists -- each AST node has a `next` pointer available for
this purpose.
Previously, a node was added to the list by starting from the head,
chasing to the last, and appending. So creating a list of length N would
take ~N^2/2 pointer dereferences.
Now, we always (temporarily) keep the last as well, so appending is O(1)
instead of O(N).
Given a keymap
xkb_keymap {
xkb_keycodes {
minimum = 8;
minimum = 8;
minimum = 8;
minimum = 8;
minimum = 8;
[... repeated N times ...]
};
xkb_types {};
xkb_compat {};
xkb_symbols {};
};
The compilation times are
N | Before | After
--------|----------|-------
10,000 | 0.407s | 0.006s
20,000 | 1.851s | 0.015s
30,000 | 5.737s | 0.021s
40,000 | 12.759s | 0.023s
50,000 | 21.489s | 0.035s
60,000 | 40.473s | 0.041s
70,000 | 53.336s | 0.039s
80,000 | 72.485s | 0.044s
90,000 | 94.703s | 0.048s
100,000 | 118.390s | 0.057s
Another option is to ditch the linked lists and use arrays instead. I
got it to work, but its more involved and allocation heavy so turns out
to be worse without further optimizations.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Given
interpret ISO_Level3_Shift+AnyOf(all,extraneous) { ... };
Previously, extraneous (and further) was ignored. Now it's rejected.
Signed-off-by: Ran Benita <ran@unusedvar.com>
Given
augment virtual_modifiers NumLock,Alt,LevelThree
Previously it was expanded (directly in the parser) to
augment virtual_modifiers NumLock;
virtual_modifiers Alt;
virtual_modifiers LevelThree;
Now it expands to
augment virtual_modifiers NumLock;
augment virtual_modifiers Alt;
augment virtual_modifiers LevelThree;
Signed-off-by: Ran Benita <ran@unusedvar.com>
Fixup of ccab349 - unlike the commit message, hash a byte twice instead
of zero times, which is probably better. This is how it was before.
Signed-off-by: Ran Benita <ran@unusedvar.com>
FNV-1a instead of the djb2-like one from before.
Keep the unrolling since it seems quite beneficial, even though it loses
one byte if the length is odd...
Signed-off-by: Ran Benita <ran@unusedvar.com>
This requires (well, at least implemented by) casting away `const` which
is undefined behavior, and clang started to warn about it.
The micro optimization didn't save too many allocations, anyway.
Signed-off-by: Ran Benita <ran@unusedvar.com>
A few lines above we check path_rel[0], so any null pointer will blow up
before we get here.
Found by coverity
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
python3 is always python3, but python could be python2 in some cases. Or just
missing (e.g. RHEL8).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This way stdin/stdout of the process are opened in text mode and we don't need
manually decode.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
In python multiprocessing, each process needs to handle (and ignore) the
KeyboardInterrupt to avoid exception logging. This is a separate patch for
easier reviewing, the first hunks merely re-indent all of the
xkbcommontool/xkbcomp functions into a try/except KeyboardInterrupt block.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This is a change in behavior and requires any automated callers to adjust
accordingly. Still, much easier to get the errors that way rather than it
being mixed into a thousands-of-lines output file.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Collect all options into a dictionary, then process that as async actions
through a process pool. This of course requires collecting the various print
statements to avoid mangled output.
This dropped the time to completion from around 14 min to 8 min on my local
machine (unscientific single run only for the original timing).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
All tests create a temporary directory, set up the environment for that
directory and then check the include paths for the presence of that directory,
ideally in the right position of the list.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Use $XDG_CONFIG_HOME/xkb as the primary lookup path for XKB rules. Same
motivation as in 3a91788d92, however the XDG directories are more standard
and recommended these days than application-specific dotfiles.
The XDG spec says to fall back to $HOME/.config where XDG_CONFIG_HOME is not
set so we implement that behavior as well.
Fixes#112
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
xkbcomp only accepts the "Level" prefix for a level name for levels 1 to
8, but the keymap dumping code added it always, e.g. "Level15".
The plain integer, e.g. "8", "15" is always accepted, so just use that.
Fixes https://github.com/xkbcommon/libxkbcommon/issues/113
Signed-off-by: Ran Benita <ran@unusedvar.com>
Reported-by: progandy
Previously, the default include path was XKB_CONFIG_ROOT:~/.xkb.
The ~/.xkb include path is intended to allow the local user to customize
their keymaps without having to modify system paths.
But usually, the user only wants to customize specific parts. When
XKB_CONFIG_ROOT is first, the user can only customize through the "entry
point" (the RMLVO). When ~/.xkb is first, the user can drop in a file
and it will override the system one.
The impetus for this change is the rules file. "evdev" is hard-coded
everywhere, so it not often not possible to change to something else.
And the rules files determines how the rest of the RMLVO is interpreted.
So, to enable customization, we have these options:
A: System includes user.
B: User includes system.
C: Library goes over both in one or the other order.
Option A is problematic due to backward compatibility and is also
unnatural.
Option B gives the user control and is backward compatible, so that's
what we choose. This is also how Compose files are handled, and that
seems to work fine in the wild.
Option C is actually less flexible than B, and more complicated.
(The rules file format doesn't have an include statement yet, but it's
planned).
Signed-off-by: Ran Benita <ran@unusedvar.com>
Fix meson build on Solaris by using __EXTENSIONS__ where Linux & other
platforms use _GNU_SOURCE. Without this the build fails due to missing
prototypes for functions like strdup & getopt not defined in the C99
standard. (In autoconf, this was handled by AC_USE_SYSTEM_EXTENSIONS.)
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
This adds the missing graphviz dependency, from which Doxygen uses
"dot" to create diagrams for the documentation. Also, the meson
dependency is removed as it is installed using Pip, and ninja-build
added.
This splits the steps for installing dependencies into a new template
file, as well as the build steps for Autotools. Both gets used to define
two jobs which can run in parallel for both build systems.
This uploads the files saves to meson-build/meson-logs/ and associates
them to the pipeline task, which makes them availabl for download in the
Azure Pipelines web interface.
xdg_shell v6 was pretty close to the finalised stable version of
xdg-shell. We can now just use the stable version, which is supported
everywhere (Enlightenment, KWin, Mutter, Weston, wlroots).
This requires bumping the wayland-protocols dependency.
Signed-off-by: Daniel Stone <daniels@collabora.com>
We don't need to determine the total number of bits set to determine if
exactly one is set.
Additionally, on x86_64 without any -march=* flag, __builtin_popcount
will get compiled to a function call to the compiler runtime (on gcc),
or a long sequence of bit operations (on clang).
Signed-off-by: Michael Forney <mforney@mforney.org>
With older versions of autotools, one needs to add this line to
Makefile.am. From the autoconf docs: "Note that if you use aclocal from
Automake to generate aclocal.m4, you must also set ACLOCAL_AMFLAGS = -I dir
in your top-level Makefile.am". [1]
I couldn't build with autoconf 2.68 without this.
[1] https://www.gnu.org/software/autoconf/manual/autoconf-2.65/html_node/Input.html
Signed-off-by: milloni <milloni@preemptable.org>
Similar to 75ce741ab9, just for the -x11
sublibrary.
This works around meson bug 3937, 'link_whole' arguments don't get added into
the final static library and we end up with a virtually empty 8-byte
libxkbcommon-x11.a file, see https://github.com/mesonbuild/meson/issues/3937
The internal lib is still built for the one test case that requires it.
Fixes#86
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This works around meson bug 3937, 'link_whole' arguments don't get added into
the final static library and we end up with a virtually empty 8-byte
libxkbcommon.a file, see https://github.com/mesonbuild/meson/issues/3937
Workaround is simply to add all sources to both libraries we need them in.
This obviously compiles them twice but this year's winter was cold and
bit of extra warmth will be appreciated.
Fixes#84
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
xproto recently has been extended with 2 new keysyms:
XF86XK_MonBrightnessCycle
XF86XK_RotationLockToggle
This commit is the result of running "scripts/update-keysyms" on a system
with the updated xproto installed.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Found by Oracle's Parfait 2.2 static analyzer:
Error: Buffer overrun
Read outside array bounds [read-outside-array-bounds] (CWE 125):
In array dereference of xkb_file_type_strings[type] with index type
Array size is 56 bytes, index <= 56
at line 734 of src/xkbcomp/ast-build.c in function 'xkb_file_type_to_string'.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
This test contains of two parts:
- a simple program to convert RMLVO commandline arguments into a keymap (and
print that keymap if requested).
- a python script that runs through rules/evdev.xml, and tries to compile a
keymap for sort-of every layout/variant/option combination. Sort-of, because
we can have multiple options and it really only does one per layout(variant)
combination.
Same thing can be done using xkbcomp, but right now it doesn't take that as
argument, it's hard-coded.
This takes quite a while, installing python-tqdm is recommended to see fancy
progress bars instead of just miles of dumps.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The benchmarks don't show any effect, so turn it off to have one less
thing to worry about. The parser does a lot of casting between AST
nodes.
Signed-off-by: Ran Benita <ran234@gmail.com>
Left shift of a negative integer. For some reason the protocol
representation here got really botched (in the spec it is just a nice
and simple INT16).
Signed-off-by: Ran Benita <ran234@gmail.com>
Don't try to divide through a signed char when indexing an array, lest
ye try to index off the start of it.
Signed-off-by: Daniel Stone <daniels@collabora.com>
The only time we could ever hit this was with count == 0, which seems
unnecessarily pedantic. But OK.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Every user of ExprReturnLhs goes on to unconditionally dereference the
field return, which can be NULL if xkb_intern_atom fails. Return false
if this is the case, so we fail safely.
testcase: splice geometry data into interp
Signed-off-by: Daniel Stone <daniels@collabora.com>
If we have an expression of the form 'l1' in an interp section, we
unconditionally try to dereference its args, even if it has none.
Signed-off-by: Daniel Stone <daniels@collabora.com>
If the scanner indicates that we might have something which looks like a
map, but the parser in fact fails to create that map, we will try to
access the map regardless. Stop doing that.
testcase: 'xkb_keymap {' -> '#kb_keymap'
Signed-off-by: Daniel Stone <daniels@collabora.com>
Coverity complains that a 7-byte string may not be null-terminated when copied
into act->data (size 7). This is fine, make a note of it.
All the strings in xkeyboard-config only use 6 bytes + null terminator so this
won't be an issue. The server (the only user of these) uses an 8-byte array
and forcibly null-terminates the string, see XkbDDXPrivate().
Everything else treats it as byte-array size 7 anyway so whether it's
null-terminated doesn't matter.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Move the aliases copy to within the (num_key_aliases > 0) block.
Passing info->aliases into this fuction with invalid aliases will
cause log messages but num_key_aliases stays on 0. The key_aliases array
is never allocated and remains NULL. We then loop through the aliases, causing
a null-pointer dereference.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
There's a (theoretical?) path where we might end up strcpy() buf without ever
writing to it. This happens if the mask is nonzero but specifies a modifier
larger than the one in the xkb_mod_set.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The parser would enter an infinite loop if an unterminated keysym
literal occurs at EOF.
Found with the afl fuzzer.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkb_geometry sections are ignored; previously the had done so by
returning NULL for the section's XkbFile, however some sections of the
code do not expect this. Instead, create an XkbFile for it, it will
never be processes and discarded later.
Caught with the afl fuzzer.
Signed-off-by: Ran Benita <ran234@gmail.com>
In the XKB format, floats and various keywords can only be used in the
xkb_geometry section. xkbcommon removed support xkb_geometry, but still
parses it for backward compatibility. As part of ignoring it, the float
AST node and various keywords were removed, and instead NULL was
returned by their parsing actions. However, the rest of the code does
not handle NULLs, and so when they appear crashes usually ensue.
To fix this, restore the float AST node and the ignored keywords. None
of the evaluating code expects them, so nice error are displayed.
Caught with the afl fuzzer.
Signed-off-by: Ran Benita <ran234@gmail.com>
The expression evaluator would go into an infinite recursion when
evaluating something like this as a boolean: `!True`. Instead of
recursing to just `True` and negating, it recursed to `!True` itself
again.
Bug inherited from xkbcomp.
Caught with the afl fuzzer.
Signed-off-by: Ran Benita <ran234@gmail.com>
Though text formats aren't exactly fuzzer's strong suit, fuzzers can
catch many surface-level bugs.
The fuzz/ directory contains target programs, testcases and dictionaries
to drive the afl fuzzer.
This commit adds a fuzzer for the XKB keymap text format and the Compose
text format. On my slow machine, using a single core, a full cycle of
the XKB fuzzer takes 5 hours. For Compose, it takes a few minutes.
Fuzzing for the other file formats (rules files mostly) will be added
later.
To do some fuzzing, run `./fuzz/fuzz.sh`.
Signed-off-by: Ran Benita <ran234@gmail.com>
Old meson expects an array with one dependency per element. Providing a
string containing multiple deps results in only the first dep getting
its whitespace properly applied. As a result, the output was:
Requires.private: xcb >= 1.10 xcb-xkb>=1.10
And downstream projects failed to find a package named 'xcb-xkb>=1.10'.
Specifying an array of versioned deps results in correct output:
Requires.private: xcb >= 1.10, xcb-xkb >= 1.10
Fixes#64.
Signed-off-by: Jan Alexander Steffens (heftig) <jan.steffens@gmail.com>
The meson-generated pkgconfig file was missing Requires and
Requires.private.
[ran: adjust for older Meson versions.]
Signed-off-by: Jan Alexander Steffens (heftig) <jan.steffens@gmail.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
Trim the API a bit.
Also, just always use gettimeofday(), which is portable. Hopefully the
system clock doesn't change while a benchmark is running.
Signed-off-by: Ran Benita <ran234@gmail.com>
The CI image only has Python 3.4. It's a bit of a hassle to change this,
so let's pin meson to the version before they bumped their requirement
instead.
I suppose it's not that bad in that it will guarantee compatibility with
older meson over time.
Signed-off-by: Ran Benita <ran234@gmail.com>
Will be deprecated in automake 2.0.
Replaced by AC_CONFIG_MACRO_DIR (already present) which is available
since autoconf 2.63.
Signed-off-by: Ran Benita <ran234@gmail.com>
The x11 tests/demos did not depend on xcb and xcb-xkb directly, only
indirectly through link_with: libxkbcommon_x11_internal. So linking
worked, but the xcb and xcb-xkb cflags were *not* included when
compiling them. So when using xcb installed in a non-standard location,
what would happen is:
- Library will link with custom xcb and compile with custom xcb headers.
- Test will link with custom xcb and compile with system xcb headers (if
exist, otherwise fail).
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/57
Reported-by: @remexre
Signed-off-by: Ran Benita <ran234@gmail.com>
We will never remove the deprecated functions and there is no real
reason to annoy users into stop using them.
If there *will* be a reason, *then* we will add the attribute.
Fixes: https://github.com/xkbcommon/libxkbcommon/issues/56
Signed-off-by: Ran Benita <ran234@gmail.com>
The default fallback type uses
type->level_names = NULL
but the keymap-dump code was not checking this case.
Instead of adding more workarounds and possible bugs (e.g. previous
commit), let's just keep the number of level names separately. This has
the additional advantage retains extraneous level name if someone adds
them for some reason.
Signed-off-by: Ran Benita <ran234@gmail.com>
The buggy code assigned the number of levels based on the number of
level names in the definition, instead of the actual number of levels!
This would completely break type definitions which do not give names to
levels.
This was not noticed for so long because xkeyboard-config always gives
names to all levels.
This regressed in 61fed8dab9.
Reported-by: Gatis Paeglis <gatis.paeglis@qt.io>
Signed-off-by: Ran Benita <ran234@gmail.com>
These can be useful in some odd cases.
There is already an implementation (+ tests) for internal use, so all
that's needed is to export them.
If xkbcommon were to provide a way to convert a Unicode codepoint to a
keysym, this could have been implemented externally as follows:
uint32_t codepoint = xkb_keysym_to_utf32(keysym);
uint32_t upper_codepoint = my_unicode_library_to_upper(codepoint);
xkb_keysym_t upper_keysym = theoretical_xkb_keysym_from_utf32(upper_codepoint);
However keysym -> codepoint is not injective so such a function is not
possible strictly speaking.
Signed-off-by: Ran Benita <ran234@gmail.com>
The previous code would cause failures to find the Compose file if one
of the environment variables LC_ALL, LC_CTYPE or LANG are set to the
empty string.
The description of the fallback procedure in loclale(7) talks about
"non-null environment variable"; I interpreted this to mean the
environment variable is unset, but it actually means unset or empty (I
verified this by looking at what glibc and musl do).
A recent bug in systemd https://github.com/systemd/systemd/issues/6407
exposed this issue. It causes these these variables to be set to the
empty string in TTY sessions.
Reported by "doodoo" in https://bbs.archlinux.org/viewtopic.php?id=228658
Signed-off-by: Ran Benita <ran234@gmail.com>
To be consistent with the meson build; also makes more sense as doxygen
can generate more than html (though we currently are not doing that).
Signed-off-by: Ran Benita <ran234@gmail.com>
Meson is easier to maintain, much faster, encourages better practices,
and is not built on a pile of shell scripts.
The autotools build system is kept intact for now, in order to ease the
migration. The intention is to remove it sooner rather than later, if
all goes well.
Run `meson build && mesonconf build` to see the configuration options
for the new system. Conversion should be straightforward. Environment
variables like CFLAGS work the same.
If meson is used, xorg-util-macros is not required.
In terms of functionality the two systems have about the same
capabilities. Here are some differences I noticed:
- Meson uses `-g` by default, autotools uses `-g -O2`.
- In autotools the default behavior is to install both static and shared
versions of the libraries. In meson the user must choose exactly one
(using -Ddefault_library=static/shared).
It is possible to workaround if needed (install twice...), but
hopefully meson will add the option in the future.
- Autotools has builtin ctags/cscope targets, meson doesn't.
Easy to run the tools directly.
- Meson has builtin benchmarks target. Handy.
- Meson has builtin support for sanitizers/clang-analyzer/lto/pgo/
coverage etc. Also handy.
Signed-off-by: Ran Benita <ran234@gmail.com>
The change in d44ba48 removed -I$(top_builddir)/src/xkbcomp, but this is
needed in order to find the generated parser.h file which is put in the
build dir.
I also added -I$(top_builddir)/src in order to match the meson behavior.
Fixes https://github.com/xkbcommon/libxkbcommon/issues/50
Signed-off-by: Ran Benita <ran234@gmail.com>
These scripts generate source code that is committed to git and hence do
not really belong in the build system. A maintainer runs them as needed.
Signed-off-by: Ran Benita <ran234@gmail.com>
A bit more manageable this way, and the other part of the target is
already using python.
The output is the same, except I removed the reference to Makefile.am.
Signed-off-by: Ran Benita <ran234@gmail.com>
xproto recently has been extended with 4 new keysyms:
XF86XK_Keyboard
XF86XK_WWAN
XF86XK_RFKill
XF86XK_AudioPreset
This commit is the result of running "make update-keysyms" on a system
with the updated xproto installed.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Can happen in cases like:
- There was an error between the error check and the call.
- The internal poll() fails.
Signed-off-by: Ran Benita <ran234@gmail.com>
The enum seems large, and we don't handle all of the values in it.
Previously if we got an unrecognized SHM format we would use an
uninitialized `stride`.
Signed-off-by: Ran Benita <ran234@gmail.com>
Mutter only implements v6 now, and Weston also implements that. Port
interactive-wayland to this so people can keep on using it.
Signed-off-by: Daniel Stone <daniels@collabora.com>
When we fall through to another label in a case, add an explicit comment
noting so, to quiet GCC 7's warnings.
Signed-off-by: Daniel Stone <daniels@collabora.com>
See the NOTES section of getenv(3). Somewhat obscure but it doesn't hurt
to reassure the readers who know about this.
Signed-off-by: Ran Benita <ran234@gmail.com>
The hit rate is high, but either the cache is slow or the function is
not fast enough -- the cache no longer holds its weight, leading only to
very modest improvements. If it's the former, it can definitely be
improved, the code is very dumb (though it worked just as well as any
other I tried back then). But instead, let's just kill it.
Signed-off-by: Ran Benita <ran234@gmail.com>
We currently use strcasecmp, which is locale-dependent. In particular,
one well-known surprise even if restricted just ASCII input is found in
the tr_TR (Turkish) locale, see e.g.
https://msdn.microsoft.com/en-us/library/ms973919.aspx#stringsinnet20_topic5
We have known to avoid locale-dependent functions before, but in this
case, we forgot.
Fix it by implementing our own simple ASCII-only strcasecmp/strncasecmp.
Might have been possible to use strcasecmp_l() with the C locale, but
went the easy route.
Side advantage is that even this non-optimized version is faster than
the optimized libc one (__strcasecmp_l_sse42) since it doesn't need to
do the locale stuff. xkb_keysym_from_name(), which uses strcasecmp
heavily, becomes faster, and so for example Compose file parsing, which
uses xkb_keysym_from_name() heavily, becomes ~20% faster.
Resolves https://github.com/xkbcommon/libxkbcommon/issues/42
Signed-off-by: Ran Benita <ran234@gmail.com>
The current functions dealing with consumed modifiers use the
traditional XKB definition of consumed modifiers (see description in the
added documentation). However, for several users of the library (e.g.
GTK) this definition is unsuitable or too eager. This is exacerbated by
some less-than-ideal xkeyboard-config type definitions (CTRL+ALT seems
to cause most grief...).
So, because we
- want to enable alternative interpretations, but
- don't want to expose too much internal details, and
- want to keep things simple for all library users,
we add a high-level "mode" parameter which selects the desired
interpretation. New ones can be added as long as they make some sense.
All of the old consumed-modifiers functions keep using the traditional
("XKB") mode. I mark xkb_state_mod_mask_remove_consumed() and as
deprecated without adding a *2 variant because I don't it is very useful
(or used) in practice.
Alternative modes are added in subsequent commits (this commit only adds
a mode for the existing behavior).
https://github.com/xkbcommon/libxkbcommon/issues/17
Signed-off-by: Ran Benita <ran234@gmail.com>
In particular, highlight the use of configure flags to control locating
X11 keyboard stuff when building for Wayland.
Of particular note, if the locale root is not specified, then xkbcommon
will look for them under $prefix (i.e. /usr/local/share/X11/locale).
But unless the user has specifically installed them there, it is better
to look in the standard system location, /usr/share/X11/locale.
Otherwise, xkbcommon will error when it can't find them, e.g.:
xkbcommon: ERROR: ~/.XCompose:4:9: failed to expand %L to the locale Compose file
xkbcommon: ERROR: ~/.XCompose:4:12: unterminated string literal
Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
Otherwise it can segfault e.g. running ./compose inside the bench
directory.
Signed-off-by: Bryce Harrington <bryce@bryceharrington.org>
Signed-off-by: Ran Benita <ran234@gmail.com>
keymap was defined in the X11 example, but also define it in the wayland
example just to make it a bit more standalone
Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>
xkb_context_new() returns a xkb_context pointer, so change the variable
definition to be consistent.
Signed-off-by: Bryce Harrington <bryce@osg.samsung.com>
Takes care of GCC's annoyingly persistent warn_unused_result warnings.
But it's better to avoid system() I suppose.
Signed-off-by: Ran Benita <ran234@gmail.com>
More meticulously free everything we create, including hooking up the
buffer-release callback so we actually free those when required. Make
sure seats are actually in the display's seat list.
The xkbcommon object-unref functions don't actually require
NULL-checking, so we can elide those.
Signed-off-by: Daniel Stone <daniels@collabora.com>
interactive-wayland is very similar to x11/xev, and dumps out as much
state as possible.
It provides no titlebar and a completely random cursor, but such is
life.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Looks like the CI machines can use Ubuntu 14.04, which has libxcb 1.10,
which is new enough. Lets see if it works.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of giving Doxygen a series of absolute paths to the source
files and a relative path to the output directory, run it from the
source directory with purely relative paths to the source files, and
give it an absolute path to the build directory.
This fixes the parsing of README.md with a separate build directory,
since the relative includes for doc/quick-guide.md and doc/compat.md
don't resolve otherwise. Doxygen's INCLUDE_PATH turns out not to fix
this either, since that's just a set of paths to open and parse, rather
than an analogue to cpp's -I.
Signed-off-by: Daniel Stone <daniels@collabora.com>
For some reason, the dumped keymap started repeating the two key types
FOUR_LEVEL_PLUS_LOCK and FOUR_LEVEL_KEYPAD. I need to investigate, but
let's disable it for now until I do (the problem would likely not be in
the xkbcommon - at most in the test itself).
Signed-off-by: Ran Benita <ran234@gmail.com>
fileno() can fail, if called on e.g. fmemopen() FILEs which are not
backed by a file descriptor. This functions uses mmap to map the entire
file to memory, so using such FILEs will not work.
(There is actually no change of behavior here, since the following fstat
would have already failed with EBADF. But lets make it clear.)
Another possibility is to fall back to the !HAVE_MMAP case; but it
sounds like a better idea to leave it to the programmer to use the
new_from_string/new_from_buffer functions instead, instead of doing
double allocation behind their back.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkb_keymap_key_by_name() allows finding a keycode from a given keyname and
is useful for generating keyboard events to use in regression tests
during CI
xkb_keymap_key_get_name() is the inverse of xkb_keymap_key_by_name()
Signed-off-by: Mike Blumenkrantz <zmike@osg.samsung.com>
[ran: some stylistic tweaks + another test case]
Signed-off-by: Ran Benita <ran234@gmail.com>
map_file() uses PROT_READ, so const seems fitting; however unmap_file
calls munmap/free, which do not take const, so an UNCONSTIFY is needed.
To avoid the UNCONSTIFY hack, which is likely undefined behavior or some
such, just remove the const.
Signed-off-by: Ran Benita <ran234@gmail.com>
- Add the new files bench.c and bench.h to implement a timer module.
- Implement the module with clock_gettime(), mach_absolute_time(), or
gettimeofday(), depending on a given platform.
- Replace the time measurement code of the benchmark programs with the
functions of the module.
The `test/data/keymaps/host.xkb` file contains a duplicate definition of
this type. On my computer (linux, xkbcomp 1.3.0, xserver 1.17.2), the
test passes as is, but if I remove the duplicate definition, the
roundtrip brings it back and the test fails. I can also reproduce it
without relation to the test, by loading `test/data/keymaps/host.xkb`
(without the duplicate) using
xkbcomp -I $(pwd)/test/data/keymaps/host.xkb $DISPLAY
and downloading it again using
xkbcomp $DISPLAY out.xkb
the duplicate is added. On Mac OS X however, the duplicate is removed
(correctly), so the test fails there.
xkbcommon itself, which was forked from xkbcomp, doesn't have this bug;
in fact, doing
./test/print-compiled-keymap -k keymaps/host.xkb
removes the duplicate if it is present.
This is (probably) a regression in xkbcomp or xserver compared to the
versions used in Mac OS X. Since getting a patch for any of these two is
hopeless from my experience, I did not try to investigate further.
I am not sure why, but if I also add a `PC_SUPER_LEVEL2` type, the
duplicate of `FOUR_LEVEL_KEYPAD` doesn't show up. Hopefully the test
will work on all platforms now.
https://github.com/xkbcommon/libxkbcommon/issues/26
Reported-by: @nuko8
Signed-off-by: Ran Benita <ran234@gmail.com>
xkbcomp doesn't need the search-path argument, since we pass an absolute
path. Keep the plain -I which clears the search path just to be sure.
Signed-off-by: Ran Benita <ran234@gmail.com>
It goes something like:
${XLOCALEDIR} ->
${datarootdir}/locale/X11 ->
${prefix}/share/locale/X11 ->
/usr/share/locale/X11
But AC_MSG_RESULT only performs one substitution. Instead of starting to
expand this stuff manually, just print datarootdir so all the pieces are
there.
Signed-off-by: Ran Benita <ran234@gmail.com>
It will be quite confusing to use V_0.4.3 since this was already
released without symbol versioning.
Hopefully this doesn't cause any trouble for people who are using -git.
Signed-off-by: Ran Benita <ran234@gmail.com>
The current test is incorrect, since 'map[None]' is entirely valid. In
most cases this doesn't cause any problems, since the default fallback
is Level1, and it's almost always 'map[None] = Level1' anyway. But in
one case in xkeyboard-config it isn't, in types/numpad(mac):
type "KEYPAD" {
modifiers = None;
map[None] = Level2;
level_name[Level2] = "Number";
};
So before checking if no modifiers were mapped, make sure there *were*
any modifiers at all.
https://bugs.freedesktop.org/show_bug.cgi?id=85092
Reported-by: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
The keysym cache for the new scanner was not initialized.
To avoid such errors also in the future, require passing the priv
argument in scanner_init(), instead of initializing it separately.
Signed-off-by: Ran Benita <ran234@gmail.com>
Some results from the benchmark (compilation of en_US.UTF-8/Compose):
$ grep 'model name' /proc/cpuinfo
model name : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz
model name : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz
$ uname -a
Linux ran 3.16.1-1-ARCH #1 SMP PREEMPT Thu Aug 14 07:40:19 CEST 2014 x86_64 GNU/Linux
$ ./test/compose bench
compiled 1000 compose tables in 7.776488331s
So according to the above benchmark and valgrind --tool=massif, an
xkb_compose_table adds an overhead of about ~8ms time and ~130KB
resident memory.
For contrast, a plain US keymap adds an overhead of ~3ms time and 90KB
resident memory.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkbcommon-compose is a Compose implementation for xkbcommon. It mostly
behaves like libX11's Compose, but the support is somewhat low-level and
is not transparent like in libX11. The user must add some supporting code
in order to utilize it.
The intended audience are users who use xkbcommon but not a full-blown
input method. With this they can add Compose support in a straightforward
manner, so they have a fairly complete keyboard input for Latin-like
languages at least.
See the header documentation for details.
Signed-off-by: Ran Benita <ran234@gmail.com>
If we have a big array which can be finalized, on average we can give
back 1/4 of its size, which the allocator might be able to use.
Signed-off-by: Ran Benita <ran234@gmail.com>
We added a -fvisibilty=default to the libtest.la CFLAGS. When automake
sees that the CFLAGS are different from those of libxkbcommon.la, it
figures it should recompile all of the $(libxkbcommon_la_SOURCES),
instead of reusing the already-compiled object files.
But actually, the convenience library is not a shared object, it's just
used as an archive, thus the -fvisibility is entirely pointless. So
avoid the recompilations by just removing it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Replace the dog-slow unneeded strncasecmp() with an inlineable memcmp().
Before:
compiled 2500 keymaps in 8.348715629s
After:
compiled 2500 keymaps in 7.872640338s
Signed-off-by: Ran Benita <ran234@gmail.com>
The "width" terminology comes from the group*width+level layout of the
keysyms in a key, as used in the old implementations. We don't keep all
the keysyms of a key in one array so change it to a more accurate name.
Signed-off-by: Ran Benita <ran234@gmail.com>
It's easier to quickly go through the functions in a module this way.
The names provide enough "brief description" to get someone to click for
the full description.
Signed-off-by: Ran Benita <ran234@gmail.com>
An mlvo can also be used in an expansion, but we didn't mark them in
this case in commit d8a4f52cb9. This caused wrongful warnings on
something like -l ch -v fr -- the `fr` is only added via expansion.
Signed-off-by: Ran Benita <ran234@gmail.com>
Due to wildcard matches in the rules file, this is only really useful
for misspelled or missing options, e.g.
$ ./test/rmlvo-to-kccgst -o comprose:ralt > /dev/null
xkbcommon: ERROR: Unrecognized RMLVO option "comprose:ralt" was ignored
Although it is more of a warning, it indicates a misconfiguration which
the user probably wants to see. Therefore the log level is ERROR.
Signed-off-by: Ran Benita <ran234@gmail.com>
Remove the deprecated symbols that were used for ABI compatibility
during the transition period to the first stable version, 0.2.0.
The old *names* can still be used, programs which use the old names will
continue to work, as long as they were compiled against a stable
version (as they have been #defined to the new names from the start; see
xkbcommon/xkbcommon-compat.h). Namely, this will break binaries which:
1. Were compiled against a pre-stable version of libxkbcommon, and
2. Are linked against the next version of libxkbcommon, and
3. Expect to work.
This scenario is very unlikely, and will break in many other ways
anyway. Also, retaining support for these means including them in the
new symbol version file, which I would like to avoid.
Signed-off-by: Ran Benita <ran234@gmail.com>
Symbol versions provide a means by which ELF utilities can determine
whether a program is incompatible with a too-old library version so
that package management tools can autodetect version-based
dependencies and suggest upgrade paths.
[ran: swap xkbcommon.map and xkbcommon-x11.map]
Signed-off-by: Jan Engelhardt <jengelh@inai.de>
Signed-off-by: Ran Benita <ran234@gmail.com>
If the keymap doesn't have any key-aliases (which is certainly
possible), the calloc(num_key_aliases, ...) is allowed to return NULL
according to the C standard, but this is not an error.
Signed-off-by: Ran Benita <ran234@gmail.com>
There is really no reason to deny these tests from different platforms
only for a few #defines.
The only linux-only test (or test program, it is not run by make check)
is interactive-evdev, which actually uses evdev.
Signed-off-by: Ran Benita <ran234@gmail.com>
libxkbcommon 0.4.3 introduces a new test, x11comp, which does not build
on non-Linux OSes because of the unconditional <linux/input.h> include.
This seems not needed even on Linux, so attached there is a simple patch
to remove it.
https://bugs.freedesktop.org/show_bug.cgi?id=83551
Signed-off-by: Ran Benita <ran234@gmail.com>
If Xvfb is not present, posix_spawn still forks, but the child fails.
In that case, since we left the write fd of the pipe open in the parent,
we just kept waiting on the read() without noticing that the other side
is dead.
Signed-off-by: Ran Benita <ran234@gmail.com>
The @level argument is restricted by xkb_keymap_num_levels_for_key(). Fix
the description to no longer mention xkb_keymap_num_layouts_for_key().
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Virtual modifiers can have "mappings" to real modifiers, e.g. NumLock
may also set Mod2. In a normal turn of events, the various components
(depressed, latched, locked, and consequently effective) include the
mapped mods, because the masks are pre-resolved everywhere. However,
xkb_state_update_mask() accepts arbitrary mod masks, which may not be
resolved (if it comes from somewhere other than
xkb_state_serialize_mods()). So let's always resolve them ourselves.
Signed-off-by: Ran Benita <ran234@gmail.com>
It is like test/stringcomp, only instead of using
xkb_keymap_new_from_string(), it uses xkbcomp to upload the keymap to a
dummy Xvfb X server and then xkb_x11_keymap_new_from_device().
If any of these components are not present or fails, the test is shown
as skipped.
The test is messy, fragile, limited and depends on external tools, but I
will improve on that later -- it's better to have a test.
Signed-off-by: Ran Benita <ran234@gmail.com>
It'd be nicer to use C11's static_assert(), but it's easier to roll our
own C99 version using a trick I saw in xv6.
Signed-off-by: Ran Benita <ran234@gmail.com>
There can be at most 16 vmods, and we rely on the facts that #vmods +
NUM_REAL_MODS (8) <= XKB_MAX_MODS (32) when accessing keymap->mods.mods.
But msb_pos() can potentially return up to #vmods = 32 if the server is
malicious, so we need to truncate it.
Signed-off-by: Ran Benita <ran234@gmail.com>
The first 8 modifiers in keymap->mods are the real modifiers; the virtual
modifiers are then at slots 8-24. But XkbGetMap's virtualMods mask
starts the virtual modifiers at zero, so we need to add an offset (like
we do correctly in get_vmod_names()).
https://github.com/xkbcommon/libxkbcommon/issues/9
Reported-by: @rtcm
Signed-off-by: Ran Benita <ran234@gmail.com>
With the following two rules:
InterpretDecl : INTERPRET InterpretMatch OBRACE
VarDeclList
CBRACE SEMI
{ $2->def = $4; $$ = $2; }
;
InterpretMatch : KeySym PLUS Expr
{ $$ = InterpCreate($1, $3); }
| KeySym
{ $$ = InterpCreate($1, NULL); }
;
And the fact that InterpCreate doesn't initialize ->def, if the
VarDeclList fails, the %destructor tries to recursively free the
uninitialized ->def VarDef. So always initialize it.
That was the only problematic code in the parser for %destructor (I'm
pretty sure).
Signed-off-by: Ran Benita <ran234@gmail.com>
If the parser has symbols on the stack, and then enters an error, it
discards the symbols and fails. But their actions which allocate AST
nodes had already ran. So we must free these to avoid leaks.
We use %destructor declarations, see
http://www.gnu.org/software/bison/manual/html_node/Destructor-Decl.html
Note: byacc only supports %destructor when compiled with
--enable-btyacc. Also, it doesn't support using the parse-param in the
destructor. So we might revert this commit before the next release, or
forget about byacc.
https://github.com/xkbcommon/libxkbcommon/issues/8
Signed-off-by: Ran Benita <ran234@gmail.com>
We didn't really have any. It also a exposes a memory leak, since the
parser doesn't clean up the AST nodes of the discarded symbols.
Signed-off-by: Ran Benita <ran234@gmail.com>
If count % SIZE == 0 we did a useless iteration where start==stop. It's
harmless but strange, so don't do that.
Signed-off-by: Ran Benita <ran234@gmail.com>
matcher_match() builds up the kccgst's, and we steal the memory on
success. But on error we didn't free it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Two problems:
- `j` can be >= `SIZE`, and needs to be wrapped like in the rest of the
code.
- `cookies[j % SIZE]` is not initialized if there's no atom in `from[j]`.
The is manifested when:
- We've already gone through one batch (>= 128 atoms) (in fact this
cannot happen in call to `adopt_atoms` in the current code).
- An XCB request failed in the middle of a batch.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead just statically allocate the mods array (of size MAX_MOD_SIZE =
32). The limit is not going anywhere, and static allocations are nicer
(nicer code, no OOM, etc.). It's also small and dense enough.
Signed-off-by: Ran Benita <ran234@gmail.com>
The keymap is not removed entirely from the Info (just constified),
since it is still needed in AddKeySymbols() for looking up aliases. This
dependency will be removed in the future.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is the only place where the modifier information is modified. We
will make it local to a given XKB file (after which it will be merged
into the keymap). Currently it changes the keymap directly, which
sidesteps the abstraction and leaves side-effects even if the XkbFile's
compilation fails.
Signed-off-by: Ran Benita <ran234@gmail.com>
The modifier printing functions only need the modifier information, they
don't care about keys or leds, etc.
Signed-off-by: Ran Benita <ran234@gmail.com>
The only thing that the compilation phase needs the keymap for currently
is for access to the modifier information (it also modifies it in
place!). We want to only pass along the neccessary information, to make
it more tractable and testable, so instead of passing the entire keymap
we add a new 'mod_set' object and pass a (const) reference to that.
The new object is just the old array of 'struct xkb_mod'.
Signed-off-by: Ran Benita <ran234@gmail.com>
Separate the ctx object to its own field in CompatInfo, instead of doing
keymap->ctx.
The compilation functions should not have direct access to the keymap;
instead they should process the files with their own independent state
(in the *Info structs) as much as possible, and only at the end should
they be copied (i.e. commited) to the keymap. If the compilation fails,
it leaves no by-products. It's also just good form.
This was seemingly the original author's intention, but I suppose he cut
a few corners (mostly with the handling of virtual modifiers, which are
threaded through types -> compat -> symbols).
This commit is the first step and may look artificial; however the
'keymap' field will be removed shortly.
Signed-off-by: Ran Benita <ran234@gmail.com>
The documentation should be clear about what is happening, even if it's
rather unlikely anyone will really dig into the details.
Signed-off-by: Ran Benita <ran234@gmail.com>
And also add release dates to the NEWS.
We're adding API freely, so this can make life easier for anyone who
wants to stay compatible with an older version.
Signed-off-by: Ran Benita <ran234@gmail.com>
This CI service https://travis-ci.org/ builds the project in several
configurations, runs the tests, and if something fails it sends an email.
Testing on some other systems is always good, and there don't seem to
be any drawbacks to this, so why not.
Signed-off-by: Ran Benita <ran234@gmail.com>
The current calculation is in short:
entry ? (entry->mask & ~entry->preserve) : 0
This changes it be
type->mask & ~(entry ? entry->preserve : 0)
This is what Xlib does. While less intuitive, it is actually more
correct, if you follow this deduction:
- The key group's type->mask defines which modifiers the key even cares
about. The others are completely irrelevant (and in fact they are
masked out from all sided in the level calculation). Example: NumLock
for an alphabetic key.
- The type->mask, the mods which are not masked out, are *all* relevant
(and in fact in the level calculation they must match *exactly* to the
state). These mods affect which level is chosen for the key, whether
they are active or not.
- Because the type->mask mods are all relevant, they must be considered
as consumed by the calculation *even if they are not active*.
Therefore we use type->mask instead of entry->mask.
The second change is what happens when no entry is found: return 0 or
just take preserve to be 0? Let's consider an example, the basic type
type "ALPHABETIC" {
modifiers = Shift+Lock;
map[Shift] = Level2;
map[Lock] = Level2;
level_name[Level1] = "Base";
level_name[Level2] = "Caps";
};
Suppose Shift+Lock is active - it doesn't match any entry, thus it gets
to level 0. The first interpretation would take them both to be
unconsumed, the second (new one) would take them both to be consumed.
This seems much better: Caps is active, and Shift disables it, they both
do something.
This change also fixes a pretty lousy bug (since 0.3.2), where Shift
appears to apparently *not* disable Caps. What actually happens is that
Caps is not consumed (see above) but active, thus the implicit
capitalization in get_one_sym() kicks in and capitalizes it anyway.
Reported-by: Davinder Pal Singh Bhamra
Signed-off-by: Ran Benita <ran234@gmail.com>
These functions generally have the same effect as
xkb_state_key_get_syms() + xkb_keysym_to_utf{8,32}().
So why add them?
- They provide a slightly nicer interface, especially if the string is
the only interest.
- It makes the handling of multiple-keysyms-to-utf8 transparent. For the
designated use-case of multiple-keysyms (unicode combining
characters), this is a must. We also validate the UTF-8, which the
user might not otherwise do.
- We will need to apply some transformation on the resulting string
which depend on the xkb_state. This is not possible with the
xkb_keysym_* functions.
With these functions, the existing xkb_keysym_to_utf{8,32}() are not
expected to be used by a typical user; they are "raw" functions.
Signed-off-by: Ran Benita <ran234@gmail.com>
We need to validate some UTF-8, so this adds an is_valid_utf8()
function, which is probably pretty slow but should work correctly.
Signed-off-by: Ran Benita <ran234@gmail.com>
It is valid for a keymap to not have key aliases, group names and
various other things. But the current test requires all of them to be
present in the reply, which causes us the fail on such keymaps (as the
XQuartz one).
Instead, require only what we really need. The virtual-mods names may
not be strictly required, but it seems safer to leave it in for now.
https://bugs.freedesktop.org/show_bug.cgi?id=75798
Reported-by: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
Especially a mention of the XKB_DEFAULT_* envvars was missing.
Reported-by: Paeglis Gatis <Gatis.Paeglis@digia.com> (thanks!)
Signed-off-by: Ran Benita <ran234@gmail.com>
There are valid reasons to use the other keymap-creation functions, if
one needs them. On the other hand, if one is supposed to use RMLVO, it
is more or less the only choice, so the comment is not needed in this
case as well.
Signed-off-by: Ran Benita <ran234@gmail.com>
- Specify in detail which parts of the events we care about. In theory
the X server should not bother us with things we didn't ask for. In
practice it still does, but oh well.
- Use the _aux version of select_events. This is the correct one to use,
the non-aux version is useless.
Signed-off-by: Ran Benita <ran234@gmail.com>
Nothing bad can come out of it, but for some reason this error didn't
return early (inherited from xkbcomp).
Also promote the log message to an error, as it clearly is.
Signed-off-by: Ran Benita <ran234@gmail.com>
This retrieves the mask of consumed modifiers for a given key and state,
which is helpful for toolkits without having them to do it one modifier
at a time, or pass in 0xFFFFFFFF to xkb_state_remove_consumed_mods to
"reverse-engineer" the consumed mods.
This is a little shorter and follows easier from the spec flag
description table.
Also a few were too permissive (like allowing LatchToLock in SetMods).
Signed-off-by: Ran Benita <ran234@gmail.com>
This used to *unset* a flag called "SwitchApplication"; we changed the
flag to "same" but forgot to switch the cases.
Signed-off-by: Ran Benita <ran234@gmail.com>
It's the same as no flags, so might as well not print it.
(In fact it is slightly harmful, because it actively *clears* the affect
flags, which might have been set in some other manner. But in practice
this cannot happen).
Signed-off-by: Ran Benita <ran234@gmail.com>
Only the "data" field can have them, and every other field needs to
error out if it appears. But some didn't check.
Signed-off-by: Ran Benita <ran234@gmail.com>
Support for setting this field was missing from the LockMods and
LockControls actions.
Based on a xkbcomp patch by Andreas Wettstein.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of using those t1 t2 variables, pass the final destinations
directly (while making sure they are not modified in case of error).
This also ensures the types are right, e.g. in CheckGroupField it should
be int32_t, not xkb_layout_index_t (and indeed it takes a negation!).
Signed-off-by: Ran Benita <ran234@gmail.com>
This brings back the functionality that was removed in
b9c87eb710. Though it is not used in
xkeyboard-config, from our current perspective it can be quite useful to
be able to set the mappings directly, thus sidestepping the ugly and
legacy-ridden modifier_map statement.
Here's an example of how to get rid of modifier_map statements (though
that would break core-X11 applications, since they must have the
mappings through keysyms):
virtual_modifiers NumLock = Mod2;
virtual_modifiers Alt = Mod1;
// Would be nice to map these to Alt, but that would be
// incompatible with xkbcomp and somewhat complicated
virtual_modifiers LAlt = Mod1;
virtual_modifiers RAlt = Mod1;
virtual_modifiers LevelThree = Mod5;
virtual_modifiers RControl = Control;
virtual_modifiers LControl = Control;
virtual_modifiers Super = Mod4;
virtual_modifiers Meta = Mod1;
virtual_modifiers Hyper = Mod4;
virtual_modifiers AltGr = Mod5;
virtual_modifiers LShift = Shift;
virtual_modifiers RShift = Shift;
Signed-off-by: Ran Benita <ran234@gmail.com>
It only works if 'bison' or 'byacc' are provided, but sometimes byacc
is installed as plain 'yacc'. The check fails for that.
I broke this in bdd8c11, restore Daniel's retrospectively clever check.
Signed-off-by: Ran Benita <ran234@gmail.com>
First, make the rules and xkb scanners/parsers use the same logging
functions instead of rolling their own.
Second, use the gcc ##__VA_ARGS extension instead of dealing with C99
stupidity. I hope all relevant compilers support it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Previously the early-exit codepath might have left some values
unexpanded, and we'd go looking for e.g "%l%(v)".
Signed-off-by: Ran Benita <ran234@gmail.com>
We want all the default logic in a test, so encapsulate it in this
function, and make all the get_default_* functions static.
Signed-off-by: Ran Benita <ran234@gmail.com>
There can be multiple include paths. But it's nicer in any case.
This also makes scanner_error actually use log_err instead of log_warn -
oops.
Signed-off-by: Ran Benita <ran234@gmail.com>
It returns XKB_LAYOUT_INVALID in case num_groups == 0. So we shouldn't
just save it in the state.
Note, though, that this condition is generally impossible.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of thinking about signed <-> unsigned an whatnot.
bsearch() is inline in glibc, so gcc optimizes this away anyway.
Signed-off-by: Ran Benita <ran234@gmail.com>
The PLACEHOLDER was not meant to be used, but c++ doesn't like passing 0
to enums, so it was used. For this reason we add all the NO_FLAGS items,
so the PLACEHOLDER shouldn't be used anymore.
Second, XKB_MAP is the prefix we used ages ago, KEYMAP is the expected
prefix here. So deprecate that as well.
The old names may still be used through the xkbcommon-compat.h header,
which is included by default (no need to include directly).
Signed-off-by: Ran Benita <ran234@gmail.com>
Before:
ran@ran:~/src/libxkbcommon$ ./test/bench-key-proc
ran 20000000 iterations in 6.623018074s
After:
ran@ran:~/src/libxkbcommon$ ./test/bench-key-proc
ran 20000000 iterations in 4.762291091s
Not that anyone needs to process millions of keys per second...
Signed-off-by: Ran Benita <ran234@gmail.com>
It doesn't matter (I think), since the implicit conversion doesn't have
any effect (e.g. sign-extension). But it's better to be aware of the
type.
Signed-off-by: Ran Benita <ran234@gmail.com>
This reverts commit 1e6e5669c6.
It's probably safe, but let's not take any chances, as I don't have any
esoteric arch to test on. But keep the ATTR in case it's ever useful.
Signed-off-by: Ran Benita <ran234@gmail.com>
The int ones cannot be signed (they come as such from the scanner, and
NEGATE is never applied to them).
The uint32_t one is really an atom, but presumably the type was never
converted to xkb_atom_t.
Signed-off-by: Ran Benita <ran234@gmail.com>
size_t is too large; if we ever need it, that's the least of our
problems. Besides, when we roll our own (e.g. in keymap.h) it's already
unsigned int. Instead, add some emergency overflow check. So, why?
- It plays nicer with all the other uint32_t's and unsigned int's (no
extensions, etc.).
- Reduces keymap memory usage by 5% or so as a bonus.
Signed-off-by: Ran Benita <ran234@gmail.com>
Integer may be negative, so also need to test >= 0.
Also, $$ was left uninitialized if the keysym wasn't recognized.
Signed-off-by: Ran Benita <ran234@gmail.com>
We have quite diverged from the upstream file, so let's make it at least
easier to look at. Remove some unused macros and rename some for
consistency.
Signed-off-by: Ran Benita <ran234@gmail.com>
This must always hold (but if there are no actions, #actions==0), and
explicitly ensures there won't be a division-by-zero a bit below.
Signed-off-by: Ran Benita <ran234@gmail.com>
We probably don't want to get a privileged process to compile arbitrary
keymaps. So we should be careful about the envvars which control include
paths or default RMLVOs. But then secure_getenv is more sensible for
everything we do.
Signed-off-by: Ran Benita <ran234@gmail.com>
- Use 'test =' instead of 'test ==' in the x11 test. The second one
might not work with e.g. dash, whoops.
- Use AS_IF instead of 'if'. It actually blends in better..
Signed-off-by: Ran Benita <ran234@gmail.com>
- Remove outdated information about API/ABI stability. If we ever break
API or ABI, we'll do a major release. But currently everything is
stable.
- Remove outdated warnings about internal symbols. You simply cannot
access them...
- Briefly mention xkbcommon-x11 existence.
- Update git and bug URLs.
- Add myself as maintainer :)
Signed-off-by: Ran Benita <ran234@gmail.com>
Add two tests:
./test/interactive-x11
which is like test/interactive-evdev, but should behave exactly like your
X keyboard and react to state and keymap changes - in other words, just
like typing in xterm. Press ESC to exit.
./test/x11
which currently should only print out the same keymap as
xkbcomp $DISPLAY out.xkb
(modulo some whitespace and some constructs we do not support.)
Signed-off-by: Ran Benita <ran234@gmail.com>
These are function to create an xkb_keymap directly from XKB requests
to the X server. This opens up the possibility for X clients to use
xcb + xcb-xkb + xkbcommon as a proper replacement for Xlib + xkbfile for
keyboard support.
The X11 support must be enabled with --enable-x11 for now.
The functions are in xkbcommon/xkbcommon-x11.h. It depends on a recent
libxcb with xkb enabled. The functions are in a new libxkbcommon-x11.so,
with a new pkg-config file, etc. so that the packages may be split, and
libxkbcommon.so itself remains dependency-free.
Why not just use the RMLVO that the server puts in the _XKB_RULES_NAMES
property? This does not account for custom keymaps, on-the-fly keymap
modifications, remote clients, etc., so is not a proper solution in
practice. Also, some servers don't even set it. Now, the client just
needs to recreate the keymap in response to a change in the server's
keymap (as Xlib clients do with XRefreshKeyboardMapping() and friends).
Signed-off-by: Ran Benita <ran234@gmail.com>
This helps ensure we don't ship a library with undefined symbols, which
can easily happen with multiple inter-dependent DSOs.
Signed-off-by: Ran Benita <ran234@gmail.com>
This makes it easier to share the private functions in other DSOs
without relying (too much) on dead code elimination, exported symbols,
etc.
Signed-off-by: Ran Benita <ran234@gmail.com>
These functions also return -1 on invalid input. The original tests
didn't check that, but used !tests instead. Since then we've changed
them, but some were missed, and for some we forgot to remove the ! (or
you can say they were extra clever).
Signed-off-by: Ran Benita <ran234@gmail.com>
Someone was nice enough to run this for us:
ftp://ftp.sunet.se/pub/Linux/distributions/Debian/debian/pool/main/libx/libxkbcommon/libxkbcommon_0.3.1.orig.tar.gz
[libxkbcommon-0.3.1/src/keymap.c:86]: (style) The scope of the variable 'j' can be reduced.
[libxkbcommon-0.3.1/src/keymap.c:87]: (style) The scope of the variable 'key' can be reduced.
[libxkbcommon-0.3.1/src/keysym-utf.c:843]: (style) The scope of the variable 'mid' can be reduced.
[libxkbcommon-0.3.1/src/state.c:992]: (style) The scope of the variable 'str' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/action.c:467]: (style) The scope of the variable 'absolute' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/rules.c:468]: (style) The scope of the variable 'consumed' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/rules.c:862]: (style) The scope of the variable 'mlvo' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/rules.c:863]: (style) The scope of the variable 'kccgst' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/rules.c:865]: (style) The scope of the variable 'match_type' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/symbols.c:753]: (style) The scope of the variable 'toAct' can be reduced.
[libxkbcommon-0.3.1/src/xkbcomp/symbols.c:1573]: (style) The scope of the variable 'key' can be reduced.
[libxkbcommon-0.3.1/test/common.c:80]: (warning) %d in format string (no. 1) requires 'int' but the argument type is 'unsigned int'.
[libxkbcommon-0.3.1/test/interactive.c:358]: (style) The scope of the variable 'nevs' can be reduced.
[libxkbcommon-0.3.1/test/interactive.c:236]: (style) Checking if unsigned variable 'nsyms' is less than zero.
[libxkbcommon-0.3.1/test/interactive.c:226]: (style) Unused variable: unicode
Signed-off-by: Ran Benita <ran234@gmail.com>
This shows a measurable improvement in memory and performance for free,
on 64bit at least. Packing is (or should be) safe in this case.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of having a darray of pointers to malloc'ed atom_node's, make it
a darray of atom_node's directly.
This makes the code a bit simpler, saves on some malloc's, and the
memory gain/loss even out.
Unfortunately, we are no longer Three Star Programmers ;(
http://c2.com/cgi/wiki?ThreeStarProgrammer
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently, we have one ExprDef type, which contains a tagged union with
the value of all expression types. Turns out, this union is quite
wasteful memory-wise. Instead, create separate types for all expressions
(e.g ExprBinary, ExprInteger) which embed the common fields
(ExprCommon), and malloc them per their size; ExprDef then becomes a
union of all these types, but is just used as a generic pointer.
[Instead of making ExprDef a union, another option is to use
ExprCommon as the generic pointer type and then do up-castings, like we
do with ParseCommon. But this makes the code much uglier.]
The diff is mostly straightforward mechanical adaptations. It could have
been much smaller with the help of C11 anonymous structs (which were
previously a gnu extension). This will have saved all of the 'op' ->
'expr->op', etc changes. But if we can be a bit more portable for a
little effort, we should.
Before (./test/rulescomp, x86 32 bit, -O2):
==12974== total heap usage: 145,217 allocs, 145,217 frees, 10,476,238 bytes allocated
After:
==11145== total heap usage: 145,217 allocs, 145,217 frees, 8,270,358 bytes allocated
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of having the parser passing strings to the AST, and
symbols/compat etc. resolving them themselves. This simplifies the code
a bit, and makes it possible to print where exactly in the file the bad
keysym originates from.
The previous lazy approach had an advantage of not needlessly resolving
keysyms from unrelated maps. However, I think reporting these errors in
*any* map is better, and the parser is also a bit smarter then old
xkbcomp and doesn't parse many useless maps. So there's no discernible
speed/memory difference with this change.
Signed-off-by: Ran Benita <ran234@gmail.com>
Add a NEWS file, with some retroactive entries. Also add 'check-news' to
configure.ac, though this might be a bit annoying.
Signed-off-by: Ran Benita <ran234@gmail.com>
Fixes build failure with Solaris Studio compilers:
"src/xkbcomp/ast-build.c", line 492: identifier redeclared: XkbFileCreate
current : function(..., enum xkb_map_flags)
previous: function(..., unsigned int) : "src/xkbcomp/ast-build.h", line 98
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
ctype.h is locale-dependent, so using it in our scanners is not optimal.
Let's be deterministic with our own simple functions.
Signed-off-by: Ran Benita <ran234@gmail.com>
Older Automakes give this error without this directive:
Makefile.am: C objects in subdir but `AM_PROG_CC_C_O' not in `configure.ac'
In newer autotools this is included under AC_PROG_CC, but it's harmless
to add.
https://github.com/xkbcommon/libxkbcommon/issues/3
Signed-off-by: Ran Benita <ran234@gmail.com>
1000 is a bit too low for statistical significance on this 6 years old
CPU. Since the benchmark is run manually this shouldn't be a problem.
Signed-off-by: Ran Benita <ran234@gmail.com>
I wanted to avoid the strlen, but we'd better keep the scanner a bit
less surprising and encourage people to use xkb_keymap_new_from_buffer()
instead of they do in fact have access to the size.
Signed-off-by: Ran Benita <ran234@gmail.com>
Improve safety with parenthesis, make the matcher macros use the scanner
ones, and make the 1 variant use %s instead of embedding the msg; this
way the compiler can reuse the string in the binary.
Signed-off-by: Ran Benita <ran234@gmail.com>
Change variable names to avoid the name clash. The warning seen is
src/keysym-utf.c: In function 'bin_search':
src/keysym-utf.c:841: warning: declaration of 'min' shadows a global declaration
src/utils.h:109: warning: shadowed declaration is here
src/keysym-utf.c:842: warning: declaration of 'max' shadows a global declaration
src/utils.h:115: warning: shadowed declaration is here
Signed-off-by: Siddharth Heroor <heroor@ti.com>
'tmp' is stack allocated so tmp->merge is used uninitialized by
AddModMapEntry(). The value doesn't matter much, but it used to
make some modmap merging decision (which doesn't have many
conflicts usually).
Bug inherited from xkbcomp.
Signed-off-by: Ran Benita <ran234@gmail.com>
We now also work with byacc (version tested: 20130925) which some people
prefer, perhaps due to its license (public domain) or performance
(haven't compared).
When using byacc, currently the following warning comes up:
src/xkbcomp/parser.c:954:14: warning: declaration shadows a variable in the global scope [-Wshadow]
YYSTYPE yylval;
^
src/xkbcomp/parser.c:37:20: note: expanded from macro 'yylval'
#define yylval _xkbcommon_lval
^
./src/xkbcomp/parser.h:96:16: note: previous declaration is here
extern YYSTYPE _xkbcommon_lval;
This is due to a bug in byacc - it shouldn't output that extern line in
%pure-parser mode. So the warning stays.
Signed-off-by: Ran Benita <ran234@gmail.com>
Unlike bison, byacc outputs its own parser code *after* our own parser.y
code, which includes the #undef. So this fix is needed for the 'scanner'
-> 'param->scanner' translation to work in the parser.c code generated
by byacc.
Signed-off-by: Ran Benita <ran234@gmail.com>
byacc doesn't support this feature.
We print the line/col of the last scanned token instead. This is slightly
less in case of *parser* errors (not syntax errors), but I couldn't make
it point to another line, and this are pretty cryptic anyways. So it's
good enough. Also might be a bit faster, but haven't checked.
Signed-off-by: Ran Benita <ran234@gmail.com>
Even though the %name-prefix is more sensible, byacc doesn't support it,
but both bison and byacc support the -p argument.
Signed-off-by: Ran Benita <ran234@gmail.com>
Both bison and byacc support this syntax. Bison manpage says something
about this giving more or less options, but we don't care.
Signed-off-by: Ran Benita <ran234@gmail.com>
For most functions taking an enum flags parameter, we use 0 value to
indicate that no flags should be applied.
C++ has a stronger type system than C and will not implicitly convert
int's to enum's. Thus, we create valid 0 enum values for enum types
where it makes sense.
Signed-off-by: Wander Lairson Costa <wander.lairson@gmail.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
This was an oversight: even though we ship the pre-built files, it is
still good behavior to include the original generators / source files in
the tarball.
Signed-off-by: Ran Benita <ran234@gmail.com>
The xkbproto spec says:
http://www.x.org/releases/current/doc/kbproto/xkbproto.html#Interpreting_the_Lock_Modifier
If the Lock modifier is not consumed by the symbol lookup process,
routines that determine the symbol and string that correspond to
an event should capitalize the result.
This was not an issue until now, because most xkeyboard-config keymaps
do not utilize this "feature", and specify the keysyms for the Lock
modifier explicitly instead. However, some keymaps do depend on it, e.g.
ch(fr) for eacute and others.
The spec goes on to describe two options for doing this transformation:
locale-sensitive and locale-insensitive. We opt for the latter; it is
less desirable but we don't want *that* headache.
Also, only xkb_state_key_get_one_sym() is changed;
xkb_state_key_get_syms() is left as-is, and always reports the
untransformed keysyms. This is for the following reasons:
- The API doesn't allow it, since we return a const pointer directly to
the keymap keysyms table and we can't transform that.
- The transformation doesn't make sense for multiple-keysyms.
- It can be useful for an application to get the "raw" keysyms if it
wants to (e.g. maybe it wants to do the transformation itself).
Finally, note that xkb_state_mod_index_is_consumed() does *not*
report Lock as consumed even if it was used in the transformation. This
is what Xlib does.
This definitely doesn't fall under the "hard to misuse" API rule but
it's the best we can do.
https://bugs.freedesktop.org/show_bug.cgi?id=67167
Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
Kind of odd, but get_one_sym() will be getting a different behavior.
Real life users *should* pick one or the other.
Signed-off-by: Ran Benita <ran234@gmail.com>
Including the X server is a bit of a borderline case; we should mostly
encourage people to use update_mask() only when xkbcommon itself
serializes the state on the other side. But it's not entirely wrong
either.. So rephrase a bit.
Signed-off-by: Ran Benita <ran234@gmail.com>
src/xkbcomp/scanner.c:158:17: warning: comparison of constant -1 with expression of type 'enum yytokentype' is always true
[-Wtautological-constant-out-of-range-compare]
if (tok != -1) return tok;
~~~ ^ ~~
Signed-off-by: Ran Benita <ran234@gmail.com>
We should handle empty xkb_keycode and xkb_symbol sections, since
xkbcomp handles them, and apparently XQuartz uses it. There are also
files for it in xkeyboard-config (rules=base model=empty layout=empty,
which translate to keycodes/empty and symbols/empty).
https://bugs.freedesktop.org/show_bug.cgi?id=67654
Reported-By: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
For keycodes, groups, levels, etc, which are unsigned. The really proper
inttypes.h ones seem a bit much though.
Signed-off-by: Ran Benita <ran234@gmail.com>
"Error: " is not very informative when intermingled with other logs. The
format that seems most suitable is:
"xkbcommon: ERROR: %s"
Signed-off-by: Ran Benita <ran234@gmail.com>
The file src/xkbcomp/action.c already doesn't handle this action type
and fails if it encounters it. So lets not pretend to do something with
it, and ignore it rather than failing.
If we/someone wants this we can consider implementing it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Which to choose: xkb_state_update_key() / xkb_state_update_mask(). If you
use update_mask(), you don't (and shouldn't) use update_key().
If you use update_key(), be careful not to have unmatched UP/DOWN
calls; this can bring about stuck modifiers etc.
Probably want to use XKB_STATE_{MODS,LAYOUT}_EFFECTIVE. The others are
only useful in very special cases.
Signed-off-by: Ran Benita <ran234@gmail.com>
PATH_MAX is optional in POSIX, so avoid its unconditional usage
allocating and freeing buffers as needed.
To avoid too many malloc/free in the for loop in FindFileInXkbPath,
a buffer is grown according to the size needed at each iteration.
xkb_atom_intern now takes a len parameter. Turns out though that almost
all of our xkb_atom_intern calls are called on string literals, the
length of which we know statically. So we add a macro to micro-optimize
this case.
Signed-off-by: Ran Benita <ran234@gmail.com>
This ensures the names are escaped before having any interaction with
the user.
This was caught by noticing dump(compile(dump())) != dump. Since that's
a nice test we add it to stringcomp.
https://bugs.freedesktop.org/show_bug.cgi?id=67032
Reported-By: Auke Booij
Signed-off-by: Ran Benita <ran234@gmail.com>
It gives:
Warning: Tag `SYMBOL_CACHE_SIZE' at line 346 of file doc/Doxyfile has become obsolete.
To avoid this warning please remove this line from your configuration file or upgrade it using "doxygen -u"
So run 'doxygen -u' as suggested.
Signed-off-by: Ran Benita <ran234@gmail.com>
Add three new pieces of API:
- xkb_keymap_min_keycode does what it says on the tin
- xkb_keymap_max_keycode likewise
- xkb_keymap_key_for_each calls the provided function once for every
valid key in the keymap
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
If the keycode range is smaller than 8 → 255, artifically widen it when
dumping the keymap as not to displease X.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The scanner is very similar in structure to the one in xkbcomp/rules.c.
It avoids copying and has nicer error reporting.
It uses gperf to generate a hashtable for the keywords, which gives a
nice speed boost (compared to the naive strcasecmp method at least). But
since there's hardly a reason to regenerate it every time and require
people to install gperf, the output (keywords.c) is added here as well.
Here are some stats from test/rulescomp:
Before:
compiled 1000 keymaps in 4.052939625s
==22063== total heap usage: 101,101 allocs, 101,101 frees, 11,840,834 bytes allocated
After:
compiled 1000 keymaps in 3.519665434s
==26505== total heap usage: 99,945 allocs, 99,945 frees, 7,033,608 bytes allocated
Signed-off-by: Ran Benita <ran234@gmail.com>
This wraps the current mmap call and adds a fallback implementation for
systems which do not have mmap (e.g. mingw).
Signed-off-by: Ran Benita <ran234@gmail.com>
The current API doesn't allow the caller to create keymaps from mmap()'ed
files. The problem is, xkb_keymap_new_from_string() requires a terminating
0 byte. However, there is no way to guarantee that when using mmap() so a
user currently has to copy the whole file just to get the terminating zero
byte (assuming they cannot use xkb_keymap_new_from_file()).
This adds a new entry xkb_keymap_new_from_buffer() which takes a memory
location and the buffer size in bytes.
Internally, we depend on yy_scan_{string,byte}() helpers. According to
flex documentation these already copy the input string because they are
wrappers around yy_scan_buffer().
yy_scan_buffer() on the other hand has some insane requirements. The
buffer must be writeable and the last two bytes must be ASCII-NUL. But the
buffer may contain other 0 bytes just fine.
Because we don't want these constraints in our public API,
xkb_keymap_new_from_buffer() needs to create a copy of the input memory.
But it then calls yy_scan_buffer() directly. Hence, we have the same
number of buffer-copies as with *_from_string() but without the
terminating 0 requirement.
The explicit yy_scan_buffer() call is preferred over yy_scan_byte() so the
buffer-copy operation is not hidden somewhere in flex.
Maybe some day we no longer depend on flex and can have a zero-copy API. A
user could mmap() a file and it would get parsed right from this buffer.
But until then, we shouldn't expose this limitation in the API but instead
provide an API that some day can work with zero-copy.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
[ran: rebased on top of my branch]
Conflicts:
Makefile.am
src/xkbcomp/xkbcomp.c
src/xkbcomp/keymap.c:127:12: error: 'found' may be used uninitialized in this function [-Werror=maybe-uninitialized]
Not really, but why not.
Signed-off-by: Ran Benita <ran234@gmail.com>
Previously we allowed you to pass a names struct with five NULL members,
but not just pass NULL for the struct itself. This was pretty dumb. :(
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
You can now set default values in the environment, as well as a context
option to ignore the environment, e.g. for tests.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
src/xkbcomp/rules.c:620:36: error: 'idx' may be used uninitialized in this function [-Werror=maybe-uninitialized]
Can't happen but no harm done.
Signed-off-by: Ran Benita <ran234@gmail.com>
The doxygen page looked a bit dead, the README fills it nicely, and is
already written in the markdown format which doxygen uses (I think?).
Unfortunately the USE_MDFILE_AS_MAINPAGE doxygen config doesn't seem to
do anything.. So we just add a {#mainpage} tag at the top of the README
which isn't so bad. BUT we still need some config option (the
no_extension=md part) so that doxygen will accept README instead of
README.md or somesuch. And that requires an even newer release, 1.8.3.1,
released 2013-01. But if an older version is used, it doesn't spew out
warnings but just skips the README, which is fine.
Signed-off-by: Ran Benita <ran234@gmail.com>
We want to use the USE_MDFILE_AS_MAINPAGE option which was introduced in
doxygen 1.8.3 (released 2012-12).
Right now the new options are commented, otherwise older doxygen spews
these these warnings, which can be ignored:
warning: ignoring unsupported tag `USE_MDFILE_AS_MAINPAGE =' at line 794, file doc/Doxyfile
warning: ignoring unsupported tag `MATHJAX_FORMAT =' at line 1210, file doc/Doxyfile
warning: ignoring unsupported tag `EXTERNAL_SEARCH =' at line 1257, file doc/Doxyfile
warning: ignoring unsupported tag `SEARCHENGINE_URL =' at line 1265, file doc/Doxyfile
warning: ignoring unsupported tag `SEARCHDATA_FILE =' at line 1271, file doc/Doxyfile
warning: ignoring unsupported tag `EXTERNAL_SEARCH_ID =' at line 1278, file doc/Doxyfile
warning: ignoring unsupported tag `EXTRA_SEARCH_MAPPINGS =' at line 1287, file doc/Doxyfile
Signed-off-by: Ran Benita <ran234@gmail.com>
The file_id thing is used to identify the XkbFile some statement
originally came from. This is needed to avoid spurious warnings; for
example, if you write the same alias twice in a file, that's redundant,
and you'd want a warning about it. However if intentionally override it
from another file, that's fine, and you shouldn't get a warning. So by
comparing the file_id's the needed log verbosity is changed.
However, the file_id mechanism is really not needed, because we already
have that info! Each KeyNamesInfo corresponds to one XkbFile, so if the
conflict occurred while handling that one file -> same_file = true, and
if it occurs while merging two Info's -> same_file = false.
Signed-off-by: Ran Benita <ran234@gmail.com>
If 'into' in empty we can just steal 'from'.
Also move the alias-merging into the big function, it's nicer this way.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is already checked when adding a new alias and merging aliases, so
it can never happen when we get to copying to the keymap.
Also the log verbosity decision there is quite useless, we should just
warn always and be done with it. So we can remove the file_id from
AliasInfo, and collapse the alias functions together.
Signed-off-by: Ran Benita <ran234@gmail.com>
The 'merge_mode' situation is quite messy, and we've introduced a
regression compared to original xkbcomp: when handling a composite
include statement, such as
replace "foo(bar)+baz(bla)|doo:dee"
and merging the entire resulting *Info back into the including *Info,
we actually use the merge mode that is set by the last part (here it is
"augment" because of the '|'), when we should be using the one set for
the whole statement (here "replace").
We also take the opportunity to clean up a bit.
Signed-off-by: Ran Benita <ran234@gmail.com>
Some tests use linux/input.h (and epoll), but we're building on some
other kernels (e.g. debian freebsd). We could just copy the file but
it's GPL. We could also skip the tests (exit code 77) but it doesn't
really matter.
Signed-off-by: Ran Benita <ran234@gmail.com>
Make it a bit easier to experiment with other formats.
Add a struct xkb_keymap_format_operations, which currently contains the
keymap compilation and _get_as_string functions. Each format can
implement whatever it wants from these.
The current public entry points become wrappers which do some error
reporting, allocation etc., and calling to the specific format. The
wrappers are all moved to src/keymap.c, so there are no XKB_EXPORT's
under src/xkbcomp/ anymore.
The only format available now is normal text_v1.
This is all not very KISS, and adds some indirection, but it is helpful
and somewhat cleaner.
Signed-off-by: Ran Benita <ran234@gmail.com>
The snprintf trick that LedStateText and ControlMaskText do cannot work,
because you can't use the buffer as an argument to write to itself!
(posix at least has 'restrict' there). So those two actually never
worked for more than one value (i.e. with a +).
Fix that, and do the same cleanup to ModMaskText. Now we have 3
functions which look exactly the same, oh well.
Also increase the context text buffer size, you never know.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkbcomp doesn't indent there, so it's easier to diff.
Also saves some horizontal space which is sorely needed when looking at
these files (especially the xkb_symbols).
Signed-off-by: Ran Benita <ran234@gmail.com>
"Value stored to 'stmt' is never read"
"Value stored to 'grp_to_use' is never read"
And change 'grp' to 'group' if we're here.
Signed-off-by: Ran Benita <ran234@gmail.com>
The code currently uses the two names interchangeably.
Settle on 'led', because it is shorter, more recognizable, and what we
use in our API (though of course the parser still uses 'indicator').
In camel case we make it 'Led'.
We change 'xkb_indicator_map' to just 'xkb_led' and the variables of
this type are 'led'. This mimics 'xkb_key' and 'key'.
IndicatorNameInfo and LEDInfo are changed to 'LedNameInfo' and
'LedInfo', and the variables are 'ledi' (like 'keyi' etc.). This is
instead of 'ii' and 'im'.
This might make a few places a bit confusing, but less than before I
think. It's also shorter.
Signed-off-by: Ran Benita <ran234@gmail.com>
Recent xkeyboard-config introduced the following line in symbols/level3:
vmods = LevelThree,
However, the XKM format which xkbcomp produces for the X server can't
handle explicit virtual modifiers such as this:
https://bugs.freedesktop.org/show_bug.cgi?id=4927
So by doing the following, for example:
setxkbmap -layout de (or another 3-level layouts)
xkbcomp $DISPLAY out.xkb
xkbcomp out.xkb $DISPLAY
The modifier is lost and can't be used for switching to Level3 (see the
included test).
We, however, are affected worse by this bug when we load the out.xkb
keymap. First, the FOUR_LEVEL_ALPHABETIC key type has these entries:
map[None] = Level1;
map[Shift] = Level2;
map[Lock] = Level2;
map[LevelThree] = Level3;
[...]
Now, because the LevelThree virtual modifier is not bound to anything,
the effective mask of the "map[LevelThree]" entry is just 0. So when
the modifier state is empty (initial state), this entry is chosen, and
we get Level3, instead of failing to match any entry and getting the
default Level1.
The difference in behavior from the xserver stems from this commit:
acdad6058d
Which removed the entry->active field. Without bugs, this would be
correct; however, it seems in this case we should just follow the
server's behavior.
The server sets the entry->active field like so in XKBMisc.c:
/* entry is active if vmods are bound */
entry->active = (mask != 0);
The xkblib spec explains this field, but does not specify how to
initialize it. This commit does the same as above but more directly.
Signed-off-by: Ran Benita <ran234@gmail.com>
Sync the files again from xkeyboard-config 2.8, since there have been
some changes we should test against.
Also added a script test/data/sync.sh if we want to do it again in the
future.
Signed-off-by: Ran Benita <ran234@gmail.com>
Why not.
Also forgot to update the xorg-utils error message when bumping the
requirement.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Changed to xkbcommon.org.]
We were using uninitialised memory whilst parsing geometry, leaving
random contents as the return for shape/overlay/etc sections. Somehow
this actually worked everywhere but under Java.
https://bugs.freedesktop.org/show_bug.cgi?id=57913
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
'cur' doesn't make sense anymore. 'components' is a bit long for this,
but not too bad, and nothing better comes to mind.
Signed-off-by: Ran Benita <ran234@gmail.com>
There is really no need to keep this in the struct, we can just allocate
it on the stack when we need to.
Don't know why I did it this way.
Signed-off-by: Ran Benita <ran234@gmail.com>
The functions num_levels_for_key() and get_syms_by_level() have a
'layout' parameter. Currently it is expected that this value is always
legal for the key, as determined by num_layouts_for_key(). However,
there are legitimate use cases for passing an out-of-range layout there,
most probably passing the effective layout, and expecting to get the
keysyms/levels for just this layout. So we wrap it just as we do in the
xkb_state_* functions.
This is also useful for stuff like this:
http://developer.gnome.org/gdk/stable/gdk-Keyboard-Handling.html#gdk-keymap-lookup-key
If this behavior is not desired, the user has the option to check
against num_layouts_for_key herself.
https://bugs.freedesktop.org/show_bug.cgi?id=56866
Reported-by: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
We definitely don't need .gz anymore, and .bz2 seems on its way out.
Mirror what Wayland does, and move to .xz exclusively.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Pretty annoying, but C89 doesn't support that (officially), and it might
cause warning with -pedantic, etc. (though you need -Wsystem-headers to
see them usually). Removing them is not a big deal.
Signed-off-by: Ran Benita <ran234@gmail.com>
The keysym2ucs.c file apparently leaves out some keysyms, which libX11
deals with separately (like in _XkbHandleSpecialSym()).
The problematic keysyms are the keypad ones (for which we already added
some support) and keysyms which use 0xff** instead of 0x00** < 0x20.
This code should fix them properly, as much as I could gather from
libX11 and http://www.cl.cam.ac.uk/~mgk25/ucs/keysym2ucs.c and other
sources (which are not aware of locale).
https://bugs.freedesktop.org/show_bug.cgi?id=56780
Reported-by: Gatis Paeglis <gatis.paeglis@digia.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
When running autoreconf, it's possible to give flags to the underlying
aclocal by declaring a ACLOCAL_AMFLAGS variable in the top level
Makefile.am.
Putting ${ACLOCAL_FLAGS} there allows the user to set an environment
variable up before running autogen.sh and pull in the right directories
to look for m4 macros, say an up-to-date version of the xorg-util macros.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Cc: Daniel Stone <daniel@fooishbar.org>
For error handling code, it's nice to be able to pass NULL to these
function without worrying about segfaults ensuing. free() sets the
precedent here.
Also document this fact.
Signed-off-by: Ran Benita <ran234@gmail.com>
The configure test shouldn't touch CFLAGS, because they come last on the
command line and allow to users to override settings if needed.
Signed-off-by: Ran Benita <ran234@gmail.com>
We still use pkg-config to get the xkb_base variable from
xkeyboard-config, but we removed all of the other PKG_ macro calls. This
still works now, because XORG_DEFAULT_OPTIONS runs it somehow. But we
shouldn't rely on it.
Signed-off-by: Ran Benita <ran234@gmail.com>
This runs a bunch of random keys against xkb_state_update_key() and
xkb_state_key_get_one_sym(), in a fairly unintelligent way.
It might be nice to check when modifying this code path, or changing it,
to see things haven't slowed down considerably. However, given the
numbers this benchmark gives, it is pretty clear that we are not going
to be the bottleneck for anything. So this can more-or-less be ignored.
Incidentally, this also turned out to be a poor man's fuzzer, because it
turned up the fix in the previous commit. Maybe we should consider
beefing it up with an actual 'break stuff' intention and running it as
part of 'make check'.
Signed-off-by: Ran Benita <ran234@gmail.com>
The current code assumes that action->type always falls in the range of
the xkb_action_type enum. But keymaps can also have Private actions,
which are allowed to set their own type number.
So with a default xkeyboard-config keymap, keycode 86 at level 4, which
triggers such an action, causes us to crash.
Fix it by always checking the bounds.
Signed-off-by: Ran Benita <ran234@gmail.com>
These are 'special intrest' function, like the logging functions, so
it's nice to have them in their own logical group.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is the more appropriate for a specific key (also considering the
num_layouts() is a bit of a made-up value).
Signed-off-by: Ran Benita <ran234@gmail.com>
--force copies and installs all the autotools support files, rather than
making symlinks, which can sometimes break things when upgrading your
system autotools. This is what xserver does.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
We add a return value to the xkb_state_update_key and
xkb_state_update_mask, which reports to the caller which of the state
components have changed as a result.
This restores the XKB functionality of the XkbStateNotify and
XkbIndicatorsStateNotify events. See:
http://www.x.org/releases/current/doc/kbproto/xkbproto.html#Events
It is quite useful in some situations. For example, it allows an
application to avoid doing some work if nothing of relevance in the
state has changed. Say, a keyboard layout applet. Also useful for
debugging.
The deltas themselves are not provided, because I can't see a use case.
If needed, it should be possible to add some API for that.
In xkbcommon, keymaps are immutable, so all of the other *Notify events
from XKB are irrelevant.
Signed-off-by: Ran Benita <ran234@gmail.com>
This holds all of the state component fields in the state in one struct.
We will later want to keep the previous state components after updates,
so this will allow us to do it without duplicating the fields.
Signed-off-by: Ran Benita <ran234@gmail.com>
Note first:
This commits breaks the ABI somewhat. If an application is run against
this commit without recompiling against the updated header, these break:
- xkb_state_layout_*_is_active always retuns false.
- xkb_state_serialize_mods always returns 0.
So it might break layout switching in some applications. However,
xkbcommon-compat.h provides the necessary fixes, so recompiling should
work (though updating the application is even better).
Split the enum to its individual components, which enables us to refer
to them individually. We will use that later for reporting which
components of the state have changed after update.
Signed-off-by: Ran Benita <ran234@gmail.com>
And make them use context_get_buffer() instead of using a static char
array.
This was the last non-thread-safe piece we had, as far as I can tell.
Signed-off-by: Ran Benita <ran234@gmail.com>
First we split the LEVEL_ONE_ONLY bit off of the 'match' field, which
allows us to turn enum xkb_match_operation to a simple enum and remove
the need for MATCH_OP_MASK.
Next we rename 'act' to 'action', because we've settled on that
everywhere else.
Finally, SIMatchText is changed to not handle illegal values - it
shouldn't get any. This removes one usage of the GetBuffer hack.
Signed-off-by: Ran Benita <ran234@gmail.com>
The ks_tables.h file is generated by makekeys.py from
xkbcommon-keysyms.h, which in turn is generated initially by 'make
update-keysyms'. The xkbcommon-keysyms.h file is commited to git and
distributed in the tarball. Since ks_tables.h should only ever change
when xkbcommon-keysyms.h changes, it is more sensible to update them
together and treat them the same, instead of generating ks_tables.h
every time for every builder with 'make', as we do now.
This means we don't need python as a build dependency (only the one
running update-keysyms, i.e. no one, needs this), and we can be
sure exactly the same file is used by everyone. We also don't need to
run makekeys.py on every build.
Signed-off-by: Ran Benita <ran234@gmail.com>
This adds a flags argument to xkb_keysym_from_name() so we can perform a
case-insensitive search. This should really be supported as many keysyms
have really weird capitalization-rules.
However, as this may produce conflicts, users must be warned to only use
this for fallback paths or error-recovery. This is also the reason why the
internal XKB parsers still use the case-sensitive search.
This also adds some test-cases so the expected results are really
produced. The binary-size does _not_ change with this patch. However,
case-sensitive search may be slightly slower with this patch. But this is
barely measurable.
[ran: use bool instead of int for icase, add a recommendation to the
doc, and test a couple "thorny" cases.]
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
This removes the complicated and undocumented hash-table creation-helper
and replaces it with an autogenerated sorted array. The search uses simple
bsearch() now.
We also tried using gperf but it turned out to generate way to big
hashtables and when reducing the size it isn't really faster than
bsearch() anymore.
There are no users complaining about the speed of keysym lookups and we
have no benchmarks that tell that we are horribly slow. Hence, we can
safely use the simpler approach and drop all that old code.
Signed-off-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Currently xkb_keymap_num_leds() returns a count of valid (settable)
leds. Because the indexes might be non-consecutive, and some leds
might not be settable, it is incorrect to use this function for
iterating over the leds in the keymap. But this is the main use case of
this function, so instead of the current behavior we adapt the function
to the use case by making it return the needed range of iteration.
The caller needs to handle invalid intermittent indexes, though.
Signed-off-by: Ran Benita <ran234@gmail.com>
Using NOCONFIGURE=1 ./autogen.sh can prevent it from running ./configure
on its own, which is sometimes useful.
Signed-off-by: Ran Benita <ran234@gmail.com>
Use a darray instead of a static array of size 32.
We still enforce XKB_MAX_LEDS because of the size of xkb_led_mask_t.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkb_keymap_num_leds() returns the number of leds that have any
possibility of being set. Even if a led is defined but can not be set in
any way, it is not counted.
In a few places currently we assume that led indexes are smaller than
this number, which is wrong both for the above reason and for the fact
that the xkb format actually allows explicitly setting the indicator
index, which means that the indexes might be non-consecutive.
We don't really have good API to iterate on leds, now, because
xkb_keymap_num_leds is pretty useless. To work around that we use
sizeof(xkb_led_mask_t) * 8.
This makes the "Group 2" led work (try switching to a layout other than
the first in test/interactive).
Signed-off-by: Ran Benita <ran234@gmail.com>
Our current code (taken from the xserver) doesn't handle unicode keysyms
at all, and there seem to be some other changes compared to libX11,
which is what xkbcomp uses. So we just copy the code that does that from
libX11.
It would be much better to not have to hardcode unicode tables like
that, but it's probably better than dealing with glibc locale stuff for
now. It also doesn't affect our binary size much.
Signed-off-by: Ran Benita <ran234@gmail.com>
Commit 9984d1d03c changed the type of
interpret->mods to xkb_mod_mask_t, but this bit of code assumes that the
type is uint8_t.
This code is not usually run (for example by our tests), but when it
does keymap-dump would print out all of the modifiers (including the
virtual ones) which causes recompilation of the output to fail
miserably.
https://bugs.freedesktop.org/show_bug.cgi?id=55769
Signed-off-by: Ran Benita <ran234@gmail.com>
These are both real modifier masks, but we keep this information only in
the program logic now so when we change it we don't have to worry about
the type.
Signed-off-by: Ran Benita <ran234@gmail.com>
This again pushes the mod type annotation to the original call site, to
make it easier to grep to see where the real/virtual distinction
matters.
Signed-off-by: Ran Benita <ran234@gmail.com>
Most of the mod type annotations can now be changed to MOD_BOTH, because
if you pass a mask which can only contain real mods in the first place to
e.g. ModMaskText, then MOD_REAL and MOD_BOTH will give the same result.
In the cases where MOD_BOTH is only ever the argument, we just remove
it. What's left is where it really "matters".
Signed-off-by: Ran Benita <ran234@gmail.com>
This essentially "tags" each invocation of the functions with the
modifier type of the argument, which allows for easy grepping for them
(with the aim being, to remove anything but MOD_BOTH).
Signed-off-by: Ran Benita <ran234@gmail.com>
We change the keymap->vmods array into keymap->mods, and change it's
member type from struct xkb_vmod to struct xkb_mod. This table now
includes the real modifiers in the first 8 places. To distinguish
between them, we add an enum mod_type to struct xkb_mod.
Besides being a more reasonable approach, this enables us to share
some code later, remove XKB_NUM_CORE_MODS (though the 0xff mask still
appears in a few places), and prepares us to flat out remove the
distinction in the future. This commit just does the conversion.
Signed-off-by: Ran Benita <ran234@gmail.com>
Modifier masks can be confusing in some places. For example,
key->vmodmap only contains virtual modifiers, where the first is in
position 0, the second in 1 etc., while normally in a xkb_mod_mask_t the
virtual modifiers start from the 8th (XKB_NUM_CORE_MODS) position. This
happens in some other places as well.
Change all of the masks to be in the usual real+virtual format, and when
we need to access e.g. keymap->vmods we just adjust by
XKB_NUM_CORE_MODS. (This also goes for indexes, e.g.
interpret->virtual_modifier).
This makes this stuff easier to reason about.
Signed-off-by: Ran Benita <ran234@gmail.com>
VModInfo currently is only used to track which virtual modifiers were
declared in the file which owns the VModInfo. This, in turn, is only
used in ResolveVirtualModifier, which in turn is only used to resolve
the virtualModifier field in an interpret statement (compat.c). In other
words, it is used to ensure that interprets can only use a vmod which
was declared in the same map.
We remove this now, because it doesn't do much and distracts from other
changes; we will later re-add it properly. Specificly, we will make it
so that virtual modifiers are not the exception in that they modify the
keymap directly, instead of keeping the changes in some *Info struct and
commiting them to the keymap at the end of the compilation. (This is bad
because if a vmod is added to the keymap, and then the compilation of
this specific file fails, the change sticks around nonetheless).
Signed-off-by: Ran Benita <ran234@gmail.com>
LookupModMask handles this before calling LookupModIndex, and the only
other user in symbols.c doesn't handle this return value at all.
Signed-off-by: Ran Benita <ran234@gmail.com>
Turn the virtual modifiers arrays in the keymap to a single darray,
which doesn't use this limit. The number of virtual modifiers is still
limited by the size of xkb_mod_mask_t, so we make sure not to go over
that.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is a regression introduced in ed78fbcb30.
XKB_STATE_EFFECTIVE is just a OR of the other states, so using & here is
completely wrong. So test/state shows for example:
dumping state for LCtrl down:
group English (US) (0): effective depressed latched locked
mod Control (2): depressed latched locked
dumping state for LCtrl + RAlt down:
group English (US) (0): effective depressed latched locked
mod Control (2): depressed latched locked
mod Mod1 (3): depressed latched locked
dumping state for RAlt down:
group English (US) (0): effective depressed latched locked
mod Mod1 (3): depressed latched locked
dumping state for Caps Lock:
group English (US) (0): effective depressed latched locked
mod Lock (1): depressed latched locked
led Caps Lock (0): active
dumping state for Alt-Shift-+
group English (US) (0): effective depressed latched locked
mod Shift (0): depressed latched locked
mod Mod1 (3): depressed latched locked
which is bogus.
Signed-off-by: Ran Benita <ran234@gmail.com>
One physical xkb file may (and usually does) contain multiple maps. For
example, the us symbols file contains a map for every variant.
Currently, when we need a map from a file (specific or default), we
parse the entire file into a list of XkbFile's, find the map we want and
discard the others. This happens for every include statement. This is a lot
of unnecessary work; this commit is a first step at making it better.
What we do now is make yyparse return one map at a time; if we find what
we want, we can stop looking and avoid processing the rest of the file.
This moves some logic from include.c to parser.y (i.e. finding the
correct map, named or default). It also necessarily removes the
CheckDefaultMap check, which warned about a file which contains multiple
default maps. We can live without it.
Some stats with test/rulecomp (under valgrind and the benchmark):
Before:
==2280== total heap usage: 288,665 allocs, 288,665 frees, 13,121,349 bytes allocated
compiled 1000 keymaps in 10.849487353s
After:
==1070== total heap usage: 100,197 allocs, 100,197 frees, 9,329,900 bytes allocated
compiled 1000 keymaps in 5.258960549s
Pretty good.
Note: we still do some unnecessary work, by parsing and discarding the
maps before the one we want. However dealing with this is more
complicated (maybe using bison's push-parser and sniffing the token
stream). Probably not worth it.
Signed-off-by: Ran Benita <ran234@gmail.com>
This rule allows you to put several xkb_keymaps in one file.
This doesn't make any sense: only the default/first can ever be used,
yet the others are fully parsed as well.
Different keymaps should just be put in different files.
Signed-off-by: Ran Benita <ran234@gmail.com>
This rule allows you to write file maps as:
xkb_keycodes
<BLA> = 5;
[...]
instead of the usual format which is:
xkb_keycodes {
<BLA> = 5;
[...]
};
This is not documented, It is also not used in xkeyboard-config, and I
have never run into it otherwise. It also only allows one map per file.
It *might* be used in some obscure place, but probably nothing we should
care about; the simplified grammar is more useful for us now.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is a proof-of-concept for the long key names. The keycodes in the
file evdev-xkbcommon are autogenerated from linux/input.h, and uses the
names given there; all of the previous names are aliased to the new
names, so they continue to work with the symbols files, etc.
You can try it with 'sudo ./test/interactive -r evdev-xkbcommon -n 0'
The -n 0 means that we don't offset the evdev scan codes - just feed
them directly. The -r evdev-xkbcommon just means to use a new rules file
which makes us use the new keycodes file. (The only problem I can see is
with the MENU and LSGT names which has some conflicts).
Maybe some day xkeyboard-config could ship something similar, so that
the 8 offset is unneeded.
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently you can't give a key in xkb_keycodes a name of more than
XKB_KEY_NAME_LENGTH (= 4) chars. This is a pretty annoying and arbitrary
limitation; it leads to names such as <RTSH>, <COMP>, <PRSC>, <KPAD>
etc. which may be hard to decipher, and makes it impossible to give
more standard names (e.g. from linux/input.h) to keycodes.
The purpose of this, as far as I can tell, was to save memory and to
allow encoding a key name directly to a 32 bit value (unsigned long it
was).
We remove this limitation by just storing the names as atoms; this lifts
the limit, allows for easy comparison like the unsigned long thing, and
doesn't use more memory than previous solution. It also relieves us from
doing all of the annoying conversions to/from long.
This has a large diffstat only because KeyNameText, which is used a lot,
now needs to take the context in order to resolve the atom.
Signed-off-by: Ran Benita <ran234@gmail.com>
This makes the code easier to follow and does more explicitly what the
xkblib spec says:
If no matching symbol interpretation is found, the server uses a
default interpretation where:
sym = 0
flags = XkbSI_AutoRepeat
match = XkbSI_AnyOfOrNone
mods = 0
virtual_mod = XkbNoModifier
act = SA_NoAction
If a level doesn't have any keysyms, we don't apply anything to it.
Signed-off-by: Ran Benita <ran234@gmail.com>
The autoType variable is supposed to tell us whether the type was
explicitly specified by the user or was detected automatically according
to the keysyms. It then allows us to know whether to prints the type
when we dump the keymap to a string or not.
Right now it is not always set when we find an automatic type, according
to some apparently legacy rules. We change it to simply this: type
computed automatically? -> don't print.
Signed-off-by: Ran Benita <ran234@gmail.com>
Add struct xkb_group and xkb_level for use in xkb_key, to mirror how
it's done in KeyInfo, GroupInfo, LevelInfo in symbols.c. This
corresponds more nicely to the logical data layout (i.e. a key has
groups which have levels), and also removes a lot of copying and ugly
code due to the index indirections and separate arrays which were used
before.
This uses more memory in some places (e.g. we alloc an action for every
level even if the key doesn't have any) but less in other places (e.g.
we no longer have to pad each group to ->width levels). The numbers say
we use less overall.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of maintaining a syms array in the GroupInfo + sym_index's in
the levels. This simplifies the code somewhat.
In order not to alloc for every level instead of every group, we only do
it if the level has more than one keysym (with a union). Since for now
this is a special case, it actually works out better memory-wise.
Signed-off-by: Ran Benita <ran234@gmail.com>
If we enter this branch, we have 3 <= width <= 4, so if the width is 3
than syms[3] is out of bounds.
Happily inherited from xkbcomp.
Signed-off-by: Ran Benita <ran234@gmail.com>
The levels will be resized to the number of levels of the type anyway,
so removing useless levels from the end here is unneeded.
Signed-off-by: Ran Benita <ran234@gmail.com>
The code that handles group name statements currently does this:
info->group_names[grp - 1 + info->explicit_group] = name;
Other than the fact that this addition makes no sense, it actually can
reach out of the bounds of the array (which is of size XKB_NUM_GROUPS)
in the (non-realistic) case where (grp - 1) is not 0 (i.e. the statement
is not name[Group1] = "foo").
We also change explicit_group to be XKB_LAYOUT_INVALID if not set
otherwise, instead of initializing it to 0; this is clearer and if
someone happens to write 'us:1' for some reason, it will discard the
other groups in the file as it should.
This entire explicit_group thing was clearly bolted on as an
afterthought.
Signed-off-by: Ran Benita <ran234@gmail.com>
This function really needs a format argument, for symmetry with the
keymap creation functions. If we add new formats, we will almost
certainly want to add support for serializing it into a string. It would
also allow to convert from one format to another, etc.
The in the common case, the user would just want to use the format she
used to create the keymap; for that we add a special
XKB_KEYMAP_USE_ORIGINAL_FORMAT value, which will do that (it is defined
to -1 outside of the enum because I have a feeling we might want to use
0 for something else). To support this we need to keep the format inside
the keymap. While we're at it we also initialize keymap flags properly.
This changes the API, but the old xkb_map_get_as_string name works as
expected so this is the best time to do this.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is useful to see whether the function was successful and whether
truncation occurred.
It just changes void -> int so shouldn't break API or ABI.
Signed-off-by: Ran Benita <ran234@gmail.com>
To make it look better and a bit more structured and informative.
Not all of the functions are converted to doxygen format, so this is not
finished.
Signed-off-by: Ran Benita <ran234@gmail.com>
Simple HTML docs generated from the doxygen comments.
After running 'make' or 'make doc', try firefox doc/html/index.html to
see it (if you have doxygen). It's also installed with 'make install'.
You can use --enable-docs or --disable-docs, or specifically
--with-doxygen or --without-doxygen (autodetected, default yes).
The docs are currently not distributed in the tarball, because I
couldn't make it work properly in all cases :/
Signed-off-by: Ran Benita <ran234@gmail.com>
This is to follow the general scheme set by all of the other API
functions.
Since no one is using these functions yet, we don't (actually better
not) add the old names to xkbcommon-compat.h.
Signed-off-by: Ran Benita <ran234@gmail.com>
Before it was a static array of size XKB_NUM_GROUPS.
The previous cleanups made this transition a bit easier. This is a
first step for removing the XKB_NUM_GROUPS hardcoded limit; but for now
we still check that the groups are < XKB_NUM_GROUPS (e.g. in
ResolveGroup and GetGroupIndex) until the keymap, etc. is worked out as
well.
This also makes us alloc quite a bit less (this is just rulescomp):
Before:
==51999== total heap usage: 291,474 allocs, 291,474 frees, 21,458,334 bytes allocated
After:
==31394== total heap usage: 293,595 allocs, 293,595 frees, 18,150,110 bytes allocated
This is because most rmlvo's don't use the full 4 layouts that KeyInfo
had always alloced statically before.
Signed-off-by: Ran Benita <ran234@gmail.com>
These values weren't wrapped before, which caused group_index_is_active
to stop working after a few group switches.
Also, the current group-wrapping function didn't take into consideration
actions such as LockGroup=-1, which need to wrap around, etc.
xkb_layout_index_t is unsigned, but it was used to hold possibly
negative values (e.g. locked_group is 0 and gets a -1 action).
This group wrapping function should now act like the XkbAdjustGroup
function from xserver, and at least ./test/interactive doesn't bring up
any problems with group switching any more.
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently, xkb_state_layout_{index,name}_is_active may report multiple
groups as effective, because at looks at base,latched,locked separately.
But there can only be one effective group, which is computed from the
other three. So if XKB_STATE_EFFECTIVE is requested, just compare to the
effective group we have computed.
We also modify mod_{index,name}_is_active similarly, just for symmetry
(there the effective mask is just an OR of the other three so the
current test is correct).
Signed-off-by: Ran Benita <ran234@gmail.com>
What this code does is, in case someone compile a keymap like -layout
'us,us,us' then only one group would be created. If there is anything
which differentiates between any of the groups (e.g. a variant, another
layout), then this is not done.
This is pretty obscure, only saves a few kbytes in the final keymap, and
if the user asked for it, why not let her?
Signed-off-by: Ran Benita <ran234@gmail.com>
This old rules parser gives the same kccgst here, so in the interest of
staying compatible we shouldn't fix it there. Similarly we shouldn't
touch ParseIncludeMap, so this is the best place to handle this.
Signed-off-by: Ran Benita <ran234@gmail.com>
Rename the functions to get keysyms by key/layout/level to fit with the
recent public API renames, and expose them.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Move xkb_map_* functions to xkb_keymap_*, xkb_key_* functions under
either xkb_keymap or xkb_state, and rename groups to layout in all
user-visible API.
Backwards-compatible hooks are provided, such that old source will
build, but silently mangled to the new names, and old binaries will
also continue to work.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This just prints the compiled keymap string for to the given command
line arguments. This often useful when developing.
Signed-off-by: Ran Benita <ran234@gmail.com>
xkblib spec says:
XkbSA_LockNoLock If set, and the action type is XkbSA_LockMods,
the server only unlocks the action modifiers.
XkbSA_LockNoUnlock If set, and the action is XkbSA_LockMods,
the server only locks the action modifiers.
Signed-off-by: Ran Benita <ran234@gmail.com>
The xkblib spec says:
If XkbSA_UseModMapMods is not set in the flags field, the mask,
real_mods, vmods1, and vmods2 fields are used to determine the
action modifiers. Otherwise they are ignored and the modifiers
bound to the key (client map->modmap[keycode]) are used instead.
So we should just assign the modmap without considering what's there.
Signed-off-by: Ran Benita <ran234@gmail.com>
This change adds range checks based on the lowest keysym and highest keysym in
the table. This allows a quick check to be applied to identify if the keysym
is inside the table.
To really give value to this optimisation the table is split to have a
separate table for the keypad keysyms.
The test suite passes with this change.
Signed-off-by: Rob Bradford <rob@linux.intel.com>
Trying ''./test/interactive -l us:5' causes us to crash.
The <layout>:<N> syntax says to put this layout at the N'th level.
However the code (inherited from xkbcomp) doesn't check that the group
is valid, and then happily indexes keyi->groups with it, which has a
static size of XKB_NUM_GROUPS (the SetExplicitGroup function assumes the
index is valid). So any value a user might put there > 4 makes nice
things happen.
Signed-off-by: Ran Benita <ran234@gmail.com>
e.g. hhhhhHHHHHHHhhhhhh with shift down and up in the middle.
Unfortunately trying a quick test with test/interactive is not possible
because the evdev soft-repeat stops the repeat when another key is
pressed. So you need real soft-repeat for that.
Signed-off-by: Ran Benita <ran234@gmail.com>
- Add context.h and move context-related functions from xkb-priv.h to
it.
- Move xkb_context definition back to context.c.
- Add keysym.h and move keysym upper/lower/keypad from xkb-priv.h to it.
- Rename xkb-priv.h to map.h since it only contains keymap-related
definitions and declarations now.
- Remove unnecessary includes and some and some other small cleanups.
Signed-off-by: Ran Benita <ran234@gmail.com>
These statements are pretty pointless for us; we don't restrict keycodes
like X does, and if someone writes e.g. maximum = 255 but only has 100
keys, we currently happily alloc all those empty keys. xkbcomp already
handles the case when these statements aren't given, and uses a computed
min/max instead. We should just always use that.
(Of course since keycodes/evdev currently uses almost all of the
keycodes in the range it declares, including 255, this doesn't save any
memory for the common user right now).
Signed-off-by: Ran Benita <ran234@gmail.com>
This removes all of the boilerplate from the *_new functions, and leaves
them just as simple functions which perform the effect of the action on state.
Signed-off-by: Ran Benita <ran234@gmail.com>
Pass the new filter as a parameter instead of getting a new one in each
action function, and introducing a failure condition there.
Signed-off-by: Ran Benita <ran234@gmail.com>
The policy is now consistent: every API functions which recieves a
keycode should resolve it to an xkb_key first thing, and all the
internal functions use that instead of the keycode.
To facilitate it a bit, we move the KeycodeInRange check to XkbKey
itself, which returns NULL if the keycode is illegal.
Signed-off-by: Ran Benita <ran234@gmail.com>
- The Clear* functions should just free the memory associated with the
object. If the object is used again, it is Init'd again.
- s/Free/Clear if the actual pointer is not free'd.
- Zeroise object in Init and only initialize non-zero fields.
Signed-off-by: Ran Benita <ran234@gmail.com>
This layout stretches us pretty well, so it's good for testing nothing
breaks. There are a couple of things that need looking into, though
(particularly the level5 issue).
Signed-off-by: Ran Benita <ran234@gmail.com>
This regression was introduced in 93ce9c7d4f. This meant that actions
specified inside key {} statments were always going to the first group.
But actions are almost never specified in xkb_symbols so this wasn't
noticed.
Signed-off-by: Ran Benita <ran234@gmail.com>
Background:
The CopySymbolsDef has a comment on a couple of lines which supposedly
fixed a bug:
/*
* kt_index[i] may have been set by a previous run (if we have two
* layouts specified). Let's not overwrite it with the ONE_LEVEL
* default group if we dont even have keys for this group anyway.
*
* FIXME: There should be a better fix for this.
*/
if (!darray_empty(groupi->levels))
key->kt_index[i] = types[i];
But neither the comment nor the fix make any sense, because the kt_index
is indexed per group, i.e. each group gets its own type.
The original xkbcomp commit which added this (36fecff58) points to this
bug: https://bugzilla.redhat.com/show_bug.cgi?id=436626
which complains about -layout "ru,us" -variant "phonetic," not working
properly. And indeed when we try:
sudo ./test/interactive -l ru,us -v
the first group doesn't get any syms for the main keys.
The problem (Clearly the fix above is useless):
The ru(phonetic) map is specified using aliases, e.g. LatQ, LatW instead
of AD01, AD02, etc. When combined with another layout which uses the
real names (AD01, AD02), the symbols code should recognize they are the
same key and merge them into one KeyInfo. The current code does that,
but it doesn't catch the case where the alias was processes *before* the
real one; so we get two KeyInfo's and the later one wins. So e.g. the
ru(phonetic) symbols are ignored.
The fix:
Before adding a new KeyInfo to the keys array, always replace its name
by the real name, which avoids the entire issue. Luckily this is done
pretty late so most error messages should still show the alias name.
Signed-off-by: Ran Benita <ran234@gmail.com>
Since now we only use the keycode in this function for logging, it's
better not to mention the keycode at all because the XkbKeyGetKeycode
macro is implemented using a dirty hack 0_0
The key name is sufficient to determine uniquely where to look.
Signed-off-by: Ran Benita <ran234@gmail.com>
This function does some funky stuff, which, as far as I can tell, was
needed to support the functionality of giving different keycodes the
same name and thus make them duplicates (MERGE_ALT_FORM). This stuff was
removed as useless in 0765064b3, but this leftover wasn't noticed.
Signed-off-by: Ran Benita <ran234@gmail.com>
Add a LevelInfo to hold a single array of level specific info inside
a GroupInfo, instead of keeping the acts, symsMapIndex and
symsMapNumEntries arrays and the numLevels field separate and in sync.
This simplifies the code, and goes a long way toward making the
key-merging code somewhat understandable. Also uses less memory.
Note that the syms array is still in GroupInfo for now, with the levels
holding offsets into it.
Signed-off-by: Ran Benita <ran234@gmail.com>
GroupInfo keeps all of the info for a specific group in one struct.
This is the old array-of-structures vs. structure-of-arrays, but in this
case readability wins. It would also help with lifting the
XkbNumKbdGroups limit, because we only have to worry about one array
(instead of 6).
Signed-off-by: Ran Benita <ran234@gmail.com>
With Dan Nicholson's permission (via email), update his copyright and
license statements to the standard X.Org boilerplate MIT license, as
both myself and Ran have been using.
Clean up my copyright declarations (in some cases to correct ownership),
and add copyright/license statements from myself and/or Ran where
appropriate.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This is not something I do often, but I have good reason here ...
utils.h has been totally rewritten since import, and now contains no
original DEC content. Everything in here has been added by Ran, and I
do not believe that any lingering content from previous iterations is
substantial enough as to be copyrightable.
Replace DEC's copyright (and license with hostile advertising clause)
with Ran's boilerplate copyright and license statement.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
As there is currently no stable release of xkbcommon, other projects might
want to include a copy of the keysyms so they can be used even though
libxkbcommon may not be available on the machine. However, if xkbcommon.h
is still included, conflicts will occur. Hence, to avoid nasty hacks,
simply include a header protection in xkbcommon upstream.
[daniels: Added protection to Makefile.am's update-keysyms, as well as
XKB_KEY_NoSymbol, and a comment noting that it shouldn't be
updated directly.]
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Before:
text data bss dec hex filename
234422 11288 2304 248014 3c8ce obj-amd64/.libs/libxkbcommon.so.0.0.0
After:
text data bss dec hex filename
240694 5016 2304 248014 3c8ce obj-amd64/.libs/libxkbcommon.so.0.0.0
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
In file included from external/collabora/libxkbcommon/src/xkb-priv.h:87,
from external/collabora/libxkbcommon/src/xkbcomp/xkbcomp.h:30,
from external/collabora/libxkbcommon/src/xkbcomp/xkbcomp-priv.h:30,
from external/collabora/libxkbcommon/src/xkbcomp/action.h:30,
from external/collabora/libxkbcommon/src/xkbcomp/action.c:27:
external/collabora/libxkbcommon/xkbcommon/xkbcommon.h:279: error: expected declaration specifiers or '...' before 'va_list'
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
We didn't do anything with ISO_Lock, ActionMessage, RedirectKey, and the
device-specifying variants of the pointer actions, so remove those.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
There's no noticeable speed difference, but I think it's nicer and more
explicit than the previous code. Some people just don't like goto,
though..
Signed-off-by: Ran Benita <ran234@gmail.com>
We move the LookupEntry struct from expr.h to text.h, along with most of
the lookup tables. This makes them available everywhere.
Looking up a value in the LookupEntry format is slower than direct index
mapping, but it allows multiple names per value (with the canonical one
being first) and "all"- and "none"-type masks. These functions are not
used anywhere efficiency matters.
Signed-off-by: Ran Benita <ran234@gmail.com>
These were kept as atoms, but since the keymap was exposed in the API,
we converted them to strings; no the keymap is no longer exposed, so we
can go back to atoms. They make the keymap smaller (at least on 64-bit
machines) and the comparisons faster.
Signed-off-by: Ran Benita <ran234@gmail.com>
Almost all callers do xkb_atom_intern on the currently returned string,
while ResolveString converts the atom to the string to begin with...
uselss double work.
Signed-off-by: Ran Benita <ran234@gmail.com>
This will only lookup the string and return the atom if found; it will
not intern it if not. This is useful when e.g. getting a string from the
user (which may be arbitrary) and comparing against atoms.
Signed-off-by: Ran Benita <ran234@gmail.com>
There are two ways to separate multiple files in XKB include statements:
'+' will cause the later file to override the first in case of conflict,
while '|' will cause it augment it (this is done by xkbcomp). '!' is
unrelated here.
Since '|' is practically never used, this wasn't noticed.
In the modified test, the '|some_compat' previously was just ignored.
Signed-off-by: Ran Benita <ran234@gmail.com>
The format of the include statment is not explained anywhere, the code
is confusing and the comments misleading. Try to explain it better.
Signed-off-by: Ran Benita <ran234@gmail.com>
Avoids a warning, from xkeyboard-config:
commit 6676053f2c93596c2aaa9905151a5c76355a1540
Author: Peter Hutterer <peter.hutterer@who-t.net>
Date: Fri Jun 29 09:53:45 2012 +1000
symbols: keypad can only have one default section
Warning: Multiple default components in keypad
Using x11, ignoring pointerkeys
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Ran Benita <ran234@gmail.com>
For some reason this piece of code wasn't copied from xkbcomp, which
causes all of the warnings like these:
Warning: No map in include statement, but "pc" contains several; Using first defined map, "pc105"
Signed-off-by: Ran Benita <ran234@gmail.com>
This way the test logs have all the information, but we don't get eye
bleed every time we run them manually. One can always use
TESTS_ENVIRONMENT (we correctly use AM_TESTS_ENVIRONMENT now), or set
the envvars from the shell.
Signed-off-by: Ran Benita <ran234@gmail.com>
Now that we don't use syslog, "level" does sound more commonplace. We
should change it while there is still nobody using it.
Also leave some space between the integers of the xkb_log_level enum
values, if we ever need to shove more in between.
Signed-off-by: Ran Benita <ran234@gmail.com>
The implementation of changing the default properties of actions, e.g. a
statements such as (from test/data/compat/basic):
setMods.clearLocks= True;
latchMods.clearLocks= True;
latchMods.latchToLock= True;
works by keeping a list of ActionInfo's, each containing the neccesary
info from each statement, and then when some action comes up (e.g. in an
interpret statment) it goes through the list, and applies the relevent
ActionInfo's to the newly-constructed xkb_action.
Instead of doing this, we add a struct ActionsInfo, which contains an
array of xkb_actions, one for each type. When a default changing
statement appears, we change the action in the array; when a new action
comes up, we just copy from the array. This is simpler to figure out,
and pretty straightforward.
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently where it is possible to write:
setMods.clearLocks = True;
It's also possible to write:
action.clearLocks = True;
This will set the default value for the clearLocks action field for
*all* action types, as opposed to just setMods in this case. If
subsequently an action is used for which this field does not make sense,
it will error out.
This doesn't make any sense, because any given field is only possible by at
most 3 or 4 action types... which you might as well write explicitly and
avoid the side effect mentioned above.
Needless to say this is one of xkbcomp's "hidden features" and is not
used anywhere; remove it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Add CompileKeymap to do most of what compile_keymap_file does now, and
move UpdateKeymapFromModifiers along with it from (mostly unrelated)
compat.c.
Also rename UpdateKeymapFromModifiers to UpdateDerivedKeymapFields,
because it does more than update the modifiers.
Signed-off-by: Ran Benita <ran234@gmail.com>
We make the xkb_file_type enum sequential instead of masks, and then
we don't have to repeat the file types several times in the function.
Makes the code cleaner.
Signed-off-by: Ran Benita <ran234@gmail.com>
This function was always returning -1.
Adding a test, we see that test/state.c treat the is_active functions as
returning booleans, which would treat -1 as success, so we test for > 0
instead (most users would probably get this wrong as well...).
Also update the documentation for the are_active functions, and add a
ATTR_NULL_SENTINEL for gcc __attribute__((sentinel)).
Signed-off-by: Ran Benita <ran234@gmail.com>
For the indicator to be set, it is sufficient for at least one of the
group, modifier, or control state to match; this is in line with the
xkblib spec, section 8.2 and ComputeAutoState() in xserver/xkb/xkbLEDs.c
(though the xserver implementation differs from the spec on some
points...).
This also adds a tiny optimization to skip the entire check if the mask
is empty.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is only relevant to the virtualModifier= statement in interpret
statements in xkb_compat. The current code allows to write e.g.
virtualModifier = 4
to get the virtual modifier whose index happens to be 4 (possibly
declared in other files or sections, i.e. xkb_types). Doing this is
undeterministic, will break without notice, and not used anywhere.
Don't allow it.
Signed-off-by: Ran Benita <ran234@gmail.com>
The current code supports statements such as:
virtual_modifiers NumLock = Mod2;
This would set the mapping from the NumLock vmod to the Mod2 real mod
directly, without going through the virtualModifier field in an
interpret statement (in xkb_compat) or vmods field in a key statement
(in xkb_symbols).
This is undocumented, unused and complicates things, so remove it.
Signed-off-by: Ran Benita <ran234@gmail.com>
It's currently possible to write something like this:
interpret Num_Lock+Any {
virtualModifier = NumLock;
action = LockMods(modifiers=NumLock);
!indicator.allowExplicit;
};
The final statement has the same effect as writing it in the global file
scope, which changes the default indicator (which all subsequent
indicators start off as). This very strange and also unused; if someone
does it he probably expects it to affect only the local scope, and he
might then get unexpected behavior. So don't allow it.
Also, HandleInterpVar is clearly a misnomer (as it can also change
indicator defaults) so rename it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Using !allowExplicit sets the XkbIM_NoExplicit flag of the indicator,
which means that an XKB client cannot change the state of the indicator
using e.g. XkbSetNamedIndicator().
We do not support changing the state of an indicator; furthermore doing
it is probably only useful in conjunction with led-drives-keyboard
behavior, which we also do not support. This is because setting an
indicator without led-drives-keyboard would make the indicator and the
modifier/group it's bound to to get out of sync.
We can re-add this if we need this info.
Signed-off-by: Ran Benita <ran234@gmail.com>
We don't support it, as mentioned in the README, so we should stop
processing it and print a message about it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Group compatibility statements are like the following:
group 3 = AltGr;
This currently results in:
keymap->groups[2].mask = <real mod mapped from AltGr vmod>
And we don't do any thing with this value later. The reason it exists in
XKB is to support non-XKB clients (i.e. XKB support disabled entirely in
the server), which do not know the concept of "group", and use some
modifier to distinguish between the first and second keyboard layouts
(usually with the AltGr key). We don't care about all of that, so we can
forget about it.
One artifact of this removal is that xkb_map_num_groups no longer
works, because it counted through keymap->groups (this wasn't entirely
correct BTW). Instead we add a new num_groups member to the keymap,
which just hold the maximum among the xkb_key's num_groups. This also
means we don't have to compute anything just to get the number of
groups.
Signed-off-by: Ran Benita <ran234@gmail.com>
Now that 1fba6189e67 removed support for binding indicator maps by index
instead of name, we can remove some more magic which happens now: if an
indicator map specifies an indicator name which was not previously
declared in a 'indicator 5 = "Caps Lock"'-like statement in
xkb_keycodes, we can just look at the next free index and assign it.
This also allows us to use a darray for the LEDInfo's instead of a list.
Signed-off-by: Ran Benita <ran234@gmail.com>
The current code allows to set the "index" field in an indicator
statment's body. This would bind the indicator to the specified index,
instead of by name (which was declared previously in xkb_keycodes).
Doing this is a bad idea, for the same reasons as in 3cd9704, and is
also happily not used anywhere.
Signed-off-by: Ran Benita <ran234@gmail.com>
The xkblib spec, table 7.1 (indicators), says:
XkbIM_NoAutomatic: Xkb does not automatically change the value of the
indicator based upon a change in the keyboard state,
regardless of the values for the other fields of the
indicator map.
xkbcomp (the real one) never actually implemented a way for an indicator
statement to set this flag, so it's just dead unused code. We definitely
don't want to implement it ourselves, so remove any mention of it.
Signed-off-by: Ran Benita <ran234@gmail.com>
This field is used in conjunction with key behaviors, which we don't
support since c1ea23da5. This is also unused in xkeyboard-config.
Signed-off-by: Ran Benita <ran234@gmail.com>
Since we now handle empty model/layout, the last couple of tests should
not fail. The reason they do is bacause they try to use a non-existent
"base" rules file. When the file is brought in these tests do not fail.
Since we already test for non-existent rules file, we can remove them,
and refine the other tests a bit.
Signed-off-by: Ran Benita <ran234@gmail.com>
Various non-functional changes:
- Re-add keycodes.h and move some stuff there.
- Add parser-priv.h for internal bison/flex stuff.
- Don't include headers from other headers, such that file dependencies
are immediate in each file.
- Rename xkbcomp.h -> ast.h, parseutils.{c,h} -> ast-build.{c,h}
- Rename path.{c,h} -> include.{c,h}
- Rename keytypes.c -> types.c
- Make the naming of XkbFile-related functions more consistent.
- Move xkb_map_{new,ref,unref} to map.c.
- Remove most extern keyword from function declarations, it's just
noise (XKB_EXPORT is what's important here).
- Append XKBCOMP_ to include guards.
- Shuffle some code around to make all of this work.
Splitting this would be a headache..
Signed-off-by: Ran Benita <ran234@gmail.com>
And use union xkb_action instead. We add xkb_private_action, which is
the same as xkb_any_action, but only used where the intention is clear.
This should take care of whatever sizing changes the action struct might
have.
Signed-off-by: Ran Benita <ran234@gmail.com>
The KeyName functions are more appropriate in keycodes.c.
The ProcessIncludeFile can go to path.c along with the other functions
dealing with includes.
Signed-off-by: Ran Benita <ran234@gmail.com>
If this keymap flag is set, whenever a key name appears in one of the
sections which does not exist (i.e. has not been declared in keycodes),
it finds the first unused keycode and attaches it that name.
This might have been useful when you could compile the symbols section
or geometry section without a keycodes section, but we don't support
this anymore. It's also pretty useless for any real work, because the
user has no way of knowing the keycode and so it will never be used.
Finally the only obscure way left to set this flag is by including a
keycodes file called "computed".
Just remove it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Make more extensive use of get_entry_for_key_state, and add
key_get_consumed to use in the other consume functions.
There's also a slight change in the consumed mods calculations, where
we use entry->mods.mask instead of type->mods.mask. The original was
copied from what libX11 does but what we do now is more logically
correct. The result is exactly the same though because:
type->mods.mask ⊇ entry->mods.mask ⊇ entry->preserve.mask
Signed-off-by: Ran Benita <ran234@gmail.com>
The group/level types are unsigned, so it's odd to return -1 for them.
Instead use their invalid values (which happen to be == -1).
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently xkb_mods has the following members:
- uint8_t real_mods - 8 X11 core mods
- xkb_mod_mask_t vmods - 16 virtual mods, zero-based index
- xkb_mod_mask_t mask - the computed effective *real* modifier mask,
basically a cache for the first two which is:
real_mods | real mods computed from vmods
Our API acts on masks which combine the real_mods and vmods into a
single value, which is:
8 first bits real mods | 16 next bits virtual mods
(XkbNumModifiers = 8, XkbNumVirtualMods = 16). This is also the format
which ResolveVModMask uses (which is where all the modifier masks really
"come from", e.g. "Shift+Lock+Level5" -> xkb_mod_mask_t).
What the code does now after getting the mask from ResolveVModMask, is
to break it into real part and virtual part and store them seperately,
and then join them back together when the effective mask is calculated.
This is all pretty useless work. We change xkb_mods to the following:
- xkb_mod_mask_t mods - usually what ResolveVModMask returns
- xkb_mod_mask_t mask - the computed mask cache
And try to consistently use the word "mods" for the original,
non-effective mods and mask for the effective mods (which can only
contain real mods for now, because things break otherwise).
The separation is also made clearer. The effective masks are computed by
UpdateModifiersFromCompat after all the sections have been compiled;
before this the mask field is never touched; after this (i.e. map.c and
state.c) the original mods field is never touched. This single execption
to this rule is keymap-dump.c: it needs to print out only the original
modifiers, not computed ones. This is also the reason why we actually
keep two fields instead keeping one and modifying it in place.
The next logical step is probably to turn the real mods into vmods
themselves, and get rid of the distinction entirely (in a compatible
way).
Signed-off-by: Ran Benita <ran234@gmail.com>
After compiling all of the sections, UpdateModifiersFromCompat does all
of the vmod -> real mods translations, including types/kt_entries.
keytypes.c also has code that does that, but it's unneeded:
- Later sections don't look at their effective masks, so doing it later
is fine.
- When this code is executed, the vmods -> real mods mapping is empty
(that is set up later), so VModsToReal has no effect here.
So we can just remove it.
However UpdateModifiersFromCompat didn't update the preserve mask, so do
that.
Signed-off-by: Ran Benita <ran234@gmail.com>
We don't need the indirection. We store the preserve mask directly in
the entry, and create a new one if it doesn't exists (which is exactly
what the current code does in a roundabout way).
Incidentally this fixes a bug where the effective modifier mask of the
entries' preserve[] wasn't calculated, so the virtual modifiers had no
effect there.
Signed-off-by: Ran Benita <ran234@gmail.com>
A fairly simple helper which, given an xkb_mod_mask_t, removes all
modifiers which are consumed during processing of a particular key.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Recreate the old test/dump scenario, where we test the following map:
- rules: evdev
- model: pc104
- layout #1: us
- layout #2: ru
- layout #3: ca(multix)
- layout #4: de(neo)
This is ever so slightly altered from the xkbcomp output; running the
following:
setxkbmap -rules evdev -model pc105 -layout us,ru,ca,de -variant
,,multix,neo -print | xkbcomp -xkb - -
will give you a map with RCTL added to the modifier_map for both Control
and Mod3. Running the output through xkbcomp -xkb - - again, will give
you RCTL only added to Mod3.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
As a map will implicitly go to level one unless explicitly mentioned
otherwise, remove all explicit =Level1 mappings, except for those with
preserve entries.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The ONE_LEVEL definition from xkeyboard-config doesn't specify any
actual levels, but there's an implicit (anything unmatched) -> Level1
rule. Given this, each type actually has at least one level, whether or
not it specifies anything.
Fixes stringcomp.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Bring the input file into line with recent changes to the dump output,
so we're as close as we can get to a round trip.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
So we can print more intelligent debugging messages without needing
helper functions for the failed_includes array.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Right now it just comes from build-time, but eventually this should be
sourced from configuration files at runtime too.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Currently the user has no way of knowing which of the active modifiers
have been used in the translation of a keycode to its keysyms. The use
case is described in the GTK docs: say there's a menu accelerator
activated by "<Alt>+". Some layouts have "+" shifted, and some have it
on the first level. So in keymaps where "+" is shifted, the Shift
modifier is consumed and must be ignored when the user is testing
for "<Alt>+". Otherwise, we may get "<Alt><Shift>+" and the accelerator
should not actually fire.
For this we also use the preserve[] information in the key types, which
can forces us to report modifiers as unconsumed even if they were used
in the translation. Until now we didn't do anything with this
information.
The API tries to match its surronding. It's not very efficient but this
can be fixed.
Signed-off-by: Ran Benita <ran234@gmail.com>
The default type is copied over for each new key type to build on.
Further, it can be modified from within the xkb_types section itself,
with statements such as "type.modifiers = Lock" which affect all
subsequent type definitions.
The default type is (well, by default) just the simplest one level type
possible, with name "default". When no types are defined at all, it is
copied over to the keymap as the single type.
xkeyboard-config never changes the default type. There is also no sane
use case for doing so; changing any thing there doesn't make sense. So
instead of doing all the hard work of maintaining and copying this type,
which is practically never used, just remove it and initialize new types
appropriately.
Signed-off-by: Ran Benita <ran234@gmail.com>
We don't use these strings much, so storing them in the manner they
were compiled saves some copying and space.
Signed-off-by: Ran Benita <ran234@gmail.com>
If there is no map entry for some modifier combination, the default is
to use level 1. The removed code is an optimization to save some space
by removing these entries. But it doesn't actually save any space, and
did not in fact remove all level 1 entries (it walks the array while
modifying it so there's an off-by-one error).
We can instead keep them in the types but just not print them in
keymap-dump.c, to get about the same behavior.
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently each xkb_key_type has a preserve array, which is only allocated
if a preserve[] statement is specified in the type. In this case each map
entry has an element in the array.
The space savings are negligible; put this field where it logically
belongs.
Signed-off-by: Ran Benita <ran234@gmail.com>
The calculations were performed incorrectly in several places,
specifically shifting by 16 instead of 8 (= XkbNumModifiers) and masking
with 0xff instead of 0xffff.
More stuff that probably never worked as intended. This also makes these
more grep-able when we remove the vmods/real_mods separation.
Signed-off-by: Ran Benita <ran234@gmail.com>
Xkb required every keymap to have at least the four following canonical
types: ONE_LEVEL, TWO_LEVEL, ALPHABETIC, KEYPAD. This is specified in
e.g. the kbproto spec and XkbKeyTypesForCoreSymbols(3) man page.
If these types are not specified in the keymap, the code specifically
checks for them and adds them to the 4 first places in the types array,
such that they exist in every keymap. These are also the types (along
with some non-required 4-level ones) that are automatically assigned to
keys which do not explicitly declare a type (see FindAutomaticType in
symbols.c, this commit doesn't touch these heuristics, whcih are also not
very nice but necessary).
The xkeyboard-config does not rely on the builtin xkbcomp definitions of
these types and does specify them explicitly, in types/basic and
types/numpad, which are virtually always included.
This commit removes the special behavior:
- The code is ugly and makes keytypes.c harder to read.
- The code practically never gets run - everyone who uses
xkeyboard-config or a keymap based upon it (i.e. everyone) doesn't need
it. So it doesn't get tested.
- It mixes policy with implementation for not very good reasons, it
seems mostly for default compatibility with X11 core.
- And of course we don't need to remain compatible with Xkb ABI neither.
Instead, if we read a keymap with no types specified at all, we simply
assign all keys a default one-level type (like ONE_LEVEL), and issue
plenty of warnings to make it clear (with verbosity >= 3). Note that
this default can actually be changed from within the keymap, by writing
something like
type.modifier = Shift
type.whatever_field = value
in the top level of the xkb_types section. (This functionality is
completely unused as well today, BTW, but makes some sense).
This change means that if someone writes a keymap from scratch and
doesn't add say ALPHABETIC, then something like <AE11> = { [ q Q ]; }; will
ignore the second level. But as stated above this should never happen.
Signed-off-by: Ran Benita <ran234@gmail.com>
The possible key behaviors are:
KB_RadioGroup, KB_Overlay1, KB_Overlay2: already removed support for
these.
KB_Lock (with or without KB_Permanent): used to ignore key presses or
releases to simulate and deal with some legacy keyboard behaviors
(like keys that physically lock). Not used at all.
We already ignore them while processing key events in state.c, so make
it official.
Signed-off-by: Ran Benita <ran234@gmail.com>
We treat the key names as fixed length, non NUL terminated strings of
length XkbKeyNameLength, and use the appropriate *Text functions to
print them. We also use strncpy everywhere instead of memcpy to copy
the names, because it does some NUL padding and we might as well.
Signed-off-by: Ran Benita <ran234@gmail.com>
The action.c needs to use two constant Expr values, constTrue and
constFalse. To do this is keeps to static globals Expr's of type boolean
and the values "true" and "false" which need to be interned (and thus
context specific). The interning means they can't be made static const,
so there's a global flag and initializer function.
Instead of using this unsafe global state, we can simply use an integer
boolean expression (1 and 0) instead of a string one ("true" and
"false") and make them const.
Signed-off-by: Ran Benita <ran234@gmail.com>
These were repeated 5 times.
Note that this changes the ABI slightly: XKB_MOD_NAME_CAPS is changed
from "Caps Lock" to "Lock", which is the ordinary legacy mod name for
it. Since its hidden behind a #define, it's best to stay compatible with
the old names (as I think was intended, given that "Mod1", etc. are the
same).
Signed-off-by: Ran Benita <ran234@gmail.com>
The program reads key events from evdev input devices, puts them through
the library and prints some information about them. It's nice for
experimenting, quick testing and trying to break it with random stuff
(already found some!).
It is called "interactive" for lack of a better name. It's a bit
hackish, but can easily be extended, made more portable etc, in the
future.
Signed-off-by: Ran Benita <ran234@gmail.com>
Conflicts:
Makefile.am
test/.gitignore
Convert the IdentLookup typedef away from ExprResult, which drags along
everything else. This should also make all of the conversions explicit.
Signed-off-by: Ran Benita <ran234@gmail.com>
As the comment nicely puts it, this is a bit weird. When you try to
evaluate an expression of type string into an integer, what it does is:
"" -> 0
"c" -> (ascii value, i.e. like a char literal)
more than one char -> error
The first one is obviously not very useful; why not just write 0?
The second one might be useful (though I don't see where in a keymap
it would be), but I don't think anyone would consider trying "X" for
that anyway.
A look through xkeyboard-config shows "" only used once as a string, and
"X" also only used as strings (and mostly in geometry which we don't
evaluate anyway). And I seriously doubt it's used (purposely) anywhere
else. So remove it.
Signed-off-by: Ran Benita <ran234@gmail.com>
Clean up the return code handling from
xkb_context_add_include_paths_default, and thus fail context creation if
we can't add any of the default include paths, but were asked to. If
this happens, dump the DFLT_XKB_CONFIG_ROOT out in the log message, so
at least we know what we aren't looking at.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Explicit is better than implicit, and this union makes it hard to follow
what's what, particularly the confusion with ival/uval.
The other Resolve functions will follow.
Signed-off-by: Ran Benita <ran234@gmail.com>
It's more tidy and less error prone, since we use strcasecmp == 0 a lot.
We replace strcmp == 0 by streq, strcasecmp == 0 by istreq,
uStrCasePrefix by istreq_prefix and uDupString by strdup_safe.
Signed-off-by: Ran Benita <ran234@gmail.com>
The key name is always XkbKeyNameLength (= 4) bytes, so we can maintain
it directly in YYSTYPE union and copy when needed, instead of treating
it like a full blown string and then copy. This means the scanner
checks the length itself.
rulescomp under valgrind, before:
==1038== total heap usage: 168,403 allocs, 168,403 frees, 9,732,648 bytes allocated
after:
==9377== total heap usage: 155,643 allocs, 155,643 frees, 9,672,788 bytes allocated
Signed-off-by: Ran Benita <ran234@gmail.com>
We often get a strdup'd string, just to pass it over the atom_intern and
then immediately free it. But atom_intern then strdup's it again (if
it's not interned already); so instead we can have the interning "steal"
the memory instead of allocing a new one and freeing the old one. This
is done by a new xkb_atom_steal function.
It also turns out, that every time we strdup an atom, we don't actually
modify it afterwards. Since we are guaranteed that the atom table will
live as long as the context, we can just use xkb_atom_text instead. This
removes a some more dynamic allocations.
For this change we had to remove the ability to append two strings, e.g.
"foo" + "bar" -> "foobar"
which is only possible with string literals. This is unused and quite
useless for our purposes.
xkb_atom_strdup is left unused, as it may still be useful.
Running rulescomp in valgrind, Before:
==7907== total heap usage: 173,698 allocs, 173,698 frees, 9,775,973 bytes allocated
After:
==6348== total heap usage: 168,403 allocs, 168,403 frees, 9,732,648 bytes allocated
Signed-off-by: Ran Benita <ran234@gmail.com>
Without the re-initialization, the copying fails. This wasn't noticed
because this code practically never gets executed with ordinary keymaps.
Signed-off-by: Ran Benita <ran234@gmail.com>
==7071== Conditional jump or move depends on uninitialised value(s)
==7071== at 0x40B6CB: AddIndicatorName (keycodes.c:148)
==7071== by 0x40C34F: MergeIncludedKeycodes (keycodes.c:420)
==7071== by 0x40C613: HandleIncludeKeycodes (keycodes.c:480)
==7071== by 0x40D022: HandleKeycodesFile (keycodes.c:733)
==7071== by 0x40D79F: CompileKeycodes (keycodes.c:881)
==7071== by 0x401E22: compile_keymap (xkbcomp.c:157)
==7071== by 0x402091: xkb_map_new_from_kccgst (xkbcomp.c:229)
==7071== by 0x40216A: xkb_map_new_from_names (xkbcomp.c:254)
==7071== by 0x4046F5: test_compile_rules (common.c:164)
==7071== by 0x4015C1: test_rmlvo (rulescomp.c:44)
==7071== by 0x40180D: main (rulescomp.c:98)
Signed-off-by: Ran Benita <ran234@gmail.com>
and let the info always be the first argument to the various functions,
just for consistency (and it acting as the contex for this file).
Signed-off-by: Ran Benita <ran234@gmail.com>
Add new public API to provide the library users with some options to
control and customize the logging output from the library. It is based
upon the skeleton from the libabc demo libray:
https://git.kernel.org/?p=linux/kernel/git/kay/libabc.git
which is public domain and works pretty well.
This requires passing in the context object in every logging call, and
thus the conversion is done file by file. We also remove the global
warningLevel variable in favor of a verbosity level in the context,
which can be set by the user and is silent by default.
One issue is the ACTION calls, which, while nice, do not play very well
with line- and priority-based logging, and would require some
line continuation handling or keeping state or some other compromise. So
instead remove these and just inline them with their respective
warning/error. So instead of:
ERROR("Memory allocation failed\n")
ACTION("Removing all files on hardisk\n")
its something like that:
log_err("Memory allocation failed; Removing all files on harddisk\n")
Signed-off-by: Ran Benita <ran234@gmail.com>
The include/ dir is somewhat redundant and makes it just a bit harder to
handle the -I directives from out side of automake; without it the
default $(top_buildir) just works.
Here's also some further justifications I found:
http://smcv.pseudorandom.co.uk/2008/09/pc-uninstalled/
Signed-off-by: Ran Benita <ran234@gmail.com>
When including a file from another file, its possible to do something
like this:
include "+some(other)+files"
with the "+" or "|" in the beginning. What will happen then is that
instead of processing the include files separately and then merging into
the existing info, we instead start with the existing info and merge
into it as we go, as if it was written explicitly before the first "+".
It's not particulary clear what this may be useful for. Since it's not
used by xkeyboard-config, not documented anywhere (and google doesn't
bring up anything), completely untested and kind of ugly, remove this
"feature". It most likely never been used.
Signed-off-by: Ran Benita <ran234@gmail.com>
Taken from xserver/include/list.h. The changes made are:
* Drop the xorg_ prefix and some typedef from the end.
* Rename _for_each_entry macros to just _foreach (like darray).
* Rename list_is_empty to list_empty (like darray).
* Add a list_replace function which we use later.
Signed-off-by: Ran Benita <ran234@gmail.com>
This way we don't need to look up the key every time. We now only deal
with keycodes in the public API and in keycodes.c.
Also adds an xkb_foreach_key macro, which is used a lot.
Signed-off-by: Ran Benita <ran234@gmail.com>
This is 8 bits which hold how many groups the key has, what to do the
key group is out of bound and the group to redirect to if want to. This
may save a few bytes, but is really annoying. So instead, just lay out
the fields separately. We can optimize later in a sane way, with pahole,
bitfields, etc. if we want.
Signed-off-by: Ran Benita <ran234@gmail.com>
Instead of having a million arrays from the keycode to various
key-specific info in the keymap, add a single struct xkb_key to hold all
of the data for the key in one object. This way we can pass it around,
do some refactoring and make the code simpler. It's also nice to see
everything in one place.
The keys array is still indexed by keycode, which is suboptimal because
there may be a lot of holes (i.e. unused keycodes between min_key_code
and max_key_code). By the end of this series it would be abstracted
enough to replace it by a hash table or similar if there's ever a need.
Signed-off-by: Ran Benita <ran234@gmail.com>
We don't make this distinction anymore, and the separate allocations
just make it harder to reason about. Since we require that all of
symbols, types, compat etc. be present, we should just put stuff
directly in the keymap struct.
Signed-off-by: Ran Benita <ran234@gmail.com>
.uncrustify.cfg committed for future reference also, but had to manually
fix up a few things: it really likes justifying struct initialisers.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
A few problems here:
* In e.g. keycodes.c the fileID field of the Info struct was never
initialized to the id of the appropriate file, so it was always 0.
There's some code which uses it, mostly for warnings.
* Some of the fileID fields were unsigned char, which overflows several
times, seeing as the ID in some of our tests can get > 1000 (because
we reuse the context).
* Some sign mismatches.
* fileID vs file_id.
Hopefully this fixes everything. I doubt this stuff had ever worked as
intended, in xkbcomp or otherwise.
Signed-off-by: Ran Benita <ran234@gmail.com>
Just using the fact that we must have all of the components, without
optional ones.
Also fixes a memleak on the way, by making the functions which allocate
the XkbFiles to free them, which is easier to get right.
Signed-off-by: Ran Benita <ran234@gmail.com>
The mode comes from the "alternate" keyword, which is unused in
xkeyboard-config and mostly undocumented. Its purpose is to allow to
assign the same key name to multiple key codes, which is not allowed
otherwise (and doesn't make much sense). The xkblib specification
implies that this was part of the overlay functionality, which we also
no longer support.
If we do encounter this keyword, we just treat it as MERGE_DEFAULT. The
keycodes.c code will detect a collision and will ignore all but the
first key code (and the error count is not incremented).
Some peripheral code is also removed as a result.
Signed-off-by: Ran Benita <ran234@gmail.com>
The ComputeEffectiveMap function is only called from keytypes.c, with
the last argument NULL, so we can move it there and remove some code.
The function XkbcVirtualModsToRealMods, of which the above is the only
user, is already implemented more simply in compat.c, so make this one
non-static and use it. This leaves src/xkb.c empty, so remove it.
Signed-off-by: Ran Benita <ran234@gmail.com>
commit 46441b1184 removed this from the
public API, and we don't need it internally. So send it to the archives.
Signed-off-by: Ran Benita <ran234@gmail.com>
To quote the spec:
XkbSI_AllOf
All of the bits that are on in mods must be set, but others may be
set as well.
Here "mods" refers to interp->mods. This matches xserver/libX11.
Signed-off-by: Ran Benita <ran234@gmail.com>
It doesn't play well with multiple keysyms per level right now. But
that's OK, because no-one really uses them.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
We'd accidentally inverted silent vs. non-silent compilation, which
would skew the benchmark pretty badly, but also forgot to change base to
evdev for the rules here.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Readd the component names to the keymap->names struct. This is used when
printing the component, e.g.
xkb_keymap {
xkb_keycodes "evdev+aliases(qwerty)" {
instead of
xkb_keymap {
xkb_keycodes {
This makes diffing against xkbcomp $DISPLAY a bit easier and is kind of
useful anyway.
Signed-off-by: Ran Benita <ran234@gmail.com>
Evidently good to have on its own, but also fixes a regression from
xkbcomp where we'd identify implicitly-typed Cyrillic keys as TWO_LEVEL
rather than ALPHABETIC.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The code used to match a keysym to a keycode (see added comment)
differed in behavior from xkbcomp, always taking the first key it found.
This caused some incorrect interpretation of the xkeyboard-config data,
for example the one corrected in dump.data (see the diff): since the
de-neo layout sets the both_capslock option, the Left Shift key (LFSH)
has the Caps_Lock keysym in group 4 level 2; now since
keycode(Left Shift) = 50 < keycode(Caps Lock) = 64
the Left Shift one was picked, instead of the Caps Lock one which is
group 1 level 1. The correct behavior is to pick according to group,
level, keycode.
Signed-off-by: Ran Benita <ran234@gmail.com>
Use a self-contained dataset instead of relying on a globally-installed
set. Data taken from xkeyboard-config 2.5.1.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Including creating a context (will come in useful soon), opening and
reading files, and compiling keymaps.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Since BindIndicators was only ever called immediately after
CopyIndicatorMapDefs, move it up in the file and turn it into a static
function, which avoids the need to ever pass the unbound LEDs around.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
When we merge two KeyInfo's (belonging to the same keycode), we may take
a shortcut from copying if we see that the merged keys will be exactly
like those in one of the two KeyInfo's. In the case where we take the
symbols from the KeyInfo we are *not* merging into, we need to copy
the three arrays:
syms[group], symsMapNumEntries[group], symsMapIndex[group]
The code currently only copies the first one, so if there's a merge
conflict some levels may seem to disappear (i.e. have a NoSymbol
keysym).
This fixes the failing test added in c8d6bba.
Signed-off-by: Ran Benita <ran234@gmail.com>
For some reason, with the grp:alt_shift_toggle option, the following
sequence switches a group:
< Left Shift down, Left Alt down >
While the reverse doesn't:
< Left Alt down, Left Shift down >
And it should.
Signed-off-by: Ran Benita <ran234@gmail.com>
This was broken by commit 18d331b86b
(where only the first option out of a comma-separated string was
matched). Do it correctly this time and add a test.
Signed-off-by: Ran Benita <ran234@gmail.com>
This commit fixes the incorrect current behavior, where at the end of the
following key sequence
Left Shift down, Right Shift down, Left Shift up
the Shift modifier is cleared.
Clearly the code is not as nice as before, but it seems like some count
of the depressed modifiers must be kept.
The code is lifted mostly as is from xkbActions.c. [ There they also
assign to setMods and clearMods each time and not OR it. I assume its
correct, although I wouldn't have guessed... ]
Signed-off-by: Ran Benita <ran234@gmail.com>
This commit removes the ability to specify a keymap *in a rules file*,
e.g. in /usr/share/X11/xkb/rules/evdev or somesuch. This is unused in
xkeyboard-data, and the current code has never even supported it,
because xkb_map_new_from_kccgst (which is no longer exposed in the API)
checks to see that one of the usual components (e.g. symbols, types, ..)
has been filled, while the rules parser, on the other hand, doesn't
allow to specify a keymap and other stuff at the same time.
( The idea was to remove xkb_map_new_from_kccgst entirely, but it's used
by a test so it can stay. )
tl;dr: dead code. Of course passing a keymap file to
xkb_map_new_from_file still works.
Signed-off-by: Ran Benita <ran234@gmail.com>
compile_keymap can only be passes a single keymap file now, from all
code paths leading to it. So this function doesn't do anything.
The remaining check is performed inside CompileKeymap, so we can remove
it as well; compile_keymap doesn't do much now.
Signed-off-by: Ran Benita <ran234@gmail.com>
It seems like at some point it was needed to break the abstraction and
perform this piece of code in the context above CompileCompatMap. The
extra argument and the typedef look strange now, and doesn't seem to be
needed any more, so move them back.
Signed-off-by: Ran Benita <ran234@gmail.com>
enums are nice for some type safety and readability. This one also
removes the distinction between file type mask / file type index and
some naming consistency.
Signed-off-by: Ran Benita <ran234@gmail.com>
If we've only derived that a key should repeat, rather than had it
explicitly specified, don't set the explicit member. Fixes the dump
test.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
An entry for a type will only get marked as active if a virtual modifier
can be directly mapped to it, and not if an action indirectly leads to
it (e.g. LevelThree). We don't really need this test since entries which
can never be triggered ... won't be triggered.
The entire map->active thing should probably just go away.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Our early exit in ApplyInterpsToKey meant we weren't hitting the code
that's supposed to set a sensible default autorepeat value for most
keys.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This test verifies the core purpose of this library, which is to
translate the user's keypresses into keysyms according to the keymap and
the XKB specification.
The tests emulate a series of key presses, and checks that the resulting
keysyms are what we expect.
Several of the tests currently fail, and plenty more should be added and
maybe split up.
It also currently uses an RMLVO keymap, which comes from the
xkeyboard-config data set, and whose behaviour may change in the future.
So it should probably be changed to use several files of our own, but
it's OK for now.
Signed-off-by: Ran Benita <ran234@gmail.com>
For the darray we need to specify the explicit struct xkb_filter type
instead of void*, so we move the definition of struct xkb_state into
state.c thus making it opaque even from the rest of the files. It has
enough getters to get going and is otherwise good style.
Signed-off-by: Ran Benita <ran234@gmail.com>
The arrays found in KeyInfo are by far the most complicated, so this is
taken one member at a time so as not to break anything.
Signed-off-by: Ran Benita <ran234@gmail.com>
- Make darray_free also initialize the array back to an empty state, and
stop worrying about it everywhere.
- Add darray_mem, to access the underlying memory, which we do manually
now using &darray_item(arr, 0). This makes a bit more clear when we
actually mean to take the address of a specific item.
- Add darray_copy, to make a deep copy of a darray.
- Add darray_same, to test whether two darrays have the same underlying
memory (e.g. if the struct itself was value copied). This should used
where previously two arrays were compared for pointer equality.
Signed-off-by: Ran Benita <ran234@gmail.com>
Here are some quick numbers from valgrind, running rulescomp only with a
simple, common "us,de" rule set:
before darray: cb047bb
total heap usage: 44,924 allocs, 44,924 frees, 3,162,342 bytes allocated
after darray: c87468e
total heap usage: 52,670 allocs, 52,670 frees, 2,844,517 bytes allocated
tweaking specific inital allocation sizes:
total heap usage: 52,652 allocs, 52,652 frees, 2,841,814 bytes allocated
changing initial alloc = 2 globally
total heap usage: 47,802 allocs, 47,802 frees, 2,833,614 bytes allocated
changing initial alloc = 3 globally
total heap usage: 47,346 allocs, 47,346 frees, 3,307,110 bytes allocated
changing initial alloc = 4 globally
total heap usage: 44,643 allocs, 44,643 frees, 2,853,646 bytes allocated
[ Changing the geometric progression constant from 2 only made things
worse. I tried the golden ratio - not so golden :) ]
The last one is obviously the best, so it was chosen, with the specific
tweaks thrown in as well (these were there before but don't make much
difference). Overall it seems to do better than the previous manual
allocations which is a bit surprising.
Signed-off-by: Ran Benita <ran234@gmail.com>
These were made const when the structs were exposed in the API. Now they
are private and we shouldn't mess around with the UNCONSTIFY business.
Signed-off-by: Ran Benita <ran234@gmail.com>
Since every user building the library, even from git, doesn't need these
files anymore, there's no need to check for them (this goes for makekeys
as well).
The only remaining user is the update-keysyms target, but whoever will
run it again (if ever) will probably know what he's doing (at least
enough to run git diff before git commit). And the defaults should be
fine too.
Signed-off-by: Ran Benita <ran234@gmail.com>
This avoids a couple of special cases in the code, and is more
consistent. Since anyone who includes xkbcommon.h also gets
xkbcommon-keysyms.h, and anyone who include xkbcommon-keysyms.h would
want NoSymbol anyway, there's no down side.
Signed-off-by: Ran Benita <ran234@gmail.com>
This code uses a table and code derived from
http://www.cl.cam.ac.uk/~mgk25/ucs/keysym2ucs.c
The added API calls are:
xkb_keysym_to_utf32
xkb_keysym_to_utf8
[daniels: Changed API to be more in line with keysym_get_name, added
test, changed formatting to 4-space.]
If we can merge cleanly (i.e. use the entirety of one entry rather than
having to go level by level), then just reuse the existing symbols array
and skip the entire merge process.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
bison/flex-generated objects, when being run in a VPATH build with
--disable-dependency-tracking (i.e. Gentoo), would fail to be created
because automake didn't bother creating the destination directories
before trying to create the objects.
Fix this by depending on the destination directory stamp, which
according to the automake mailing list, should hopefully remain fairly
stable.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Add a test/dump.data file which contains the result we're expecting from
xkb_map_get_as_string run on a particularly complex set of keymaps, and
assert that the string representations are the same. This means that
any updates to xkb_map_get_as_string will also need to update the test
data, but should also ensure that we don't have any more parser
regressions.
Compared with diff to the output of setxkbmap + xkbcomp for the same
keymap; seems completely solid.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
A symbols file may contain a global, non key specific setting for
the group out-of-range handling method (wrap, clamp, redirect). Only
that:
* Its parsed and kept in the SymbolsInfo, but is not otherwise used in
any way (it's the same in the real xkbcomp).
* It's not used in any of xkeyboard-config files.
* It's not mentioned in the xkb specs (only the per-key ones).
* It doesn't make much sense anyway.
So remove the struct field, and emit an "unsupported, ignored" warning.
We don't increment the error count because of it, just continue (the
radio group warning just below is changed to do the same - there's no
reason to possibly abort the entire thing for it).
Signed-off-by: Ran Benita <ran234@gmail.com>
Conflicts:
src/xkbcomp/symbols.c
Instead of using NoSymbol in the map, we use num_syms == 0 to signify
the non-presence of a symbol. So instead of adding NoSymbol mappings
to the list regardless, detect them and set num_syms == 0.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The failure mode here is a little irritating:
- server loads map with ISO_Lock action
- server dumps keymap to string, including:
interpret ISO_Lock+AnyOfOrAll(None) {
action= NoAction();
};
as we don't (yet) print ISO_Lock actions
- client parses keymap from string
- client dumps keymap to string, including:
interpret ISO_Lock+AnyOfOrAll(None) {
};
- this results in a syntax error
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Some actions could also take relative rather than absolute parameters,
so they really needed to be signed instead of explicitly unsigned.
Oops.
Fixes, e.g., action= MovePtr(x=-1,y=+1), which was reported as
(x=+65535,y=+1).
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Add a non-extensive test to check that some basic things (e.g. rule
matching, var substitution, indexes and groups) work as expected.
Signed-off-by: Ran Benita <ran234@gmail.com>
Currently, if you pass in an rmlvo with an empty string for layout or
variant, it would not match layout and variant rules even with
wildcards. But if the rules file had set an appropriate default, and someone
passes in the empty string, than he should get the default.
NULL in this case signifies not wanting to match against the layout or
variant at all, and so the rule should still fail to match NULLs.
Signed-off-by: Ran Benita <ran234@gmail.com>
Really all we need from this file is a way to get xkb_component_names
from an xkb_rule_names, which is now the only thing being exposed. This
should allow for some much needed refactoring of this code.
Since this is only used by xkbcomp.c and uses xkbcomp functions, also
move rules.{c,h} under the xkbcomp dir.
Signed-off-by: Ran Benita <ran234@gmail.com>
i.e. xkb_map_new_from_file. The reason is that flex only works with
FILE's, so we must use fdopen on the file descriptor; but to avoid a
memory leak, we must also fclose() it, which, in turn, closes the file
descriptor itself.
Either way is not acceptable, so we can either:
* dup() the fd and use fdopen on that, or
* have the user call fdopen on his own, and accept a FILE* instead of an
fd.
The second one seems better, and is standard C, so why not. We must add
stdio.h to xkbcommon.h though, which is regrettable, but not a big deal.
Signed-off-by: Ran Benita <ran234@gmail.com>
This partly reverts commit 8feba630fa.
This seems to fix valgrind errors:
==9581== Invalid read of size 4
==9581== at 0x4E50928: MergeKeyGroups (symbols.c:544)
==9581== by 0x4E510F3: MergeKeys (symbols.c:644)
==9581== by 0x4E514C6: AddKeySymbols (symbols.c:722)
==9581== by 0x4E51A3F: MergeIncludedSymbols (symbols.c:854)
==9581== by 0x4E51E97: HandleIncludeSymbols (symbols.c:952)
==9581== by 0x4E53D75: HandleSymbolsFile (symbols.c:1619)
==9581== by 0x4E55A0B: CompileSymbols (symbols.c:2187)
==9581== by 0x4E4056C: CompileKeymap (keymap.c:160)
==9581== by 0x4E56953: compile_keymap (xkbcomp.c:149)
==9581== by 0x4E56AC5: xkb_map_new_from_kccgst (xkbcomp.c:195)
==9581== by 0x4009D7: test_names (namescomp.c:56)
==9581== by 0x400A55: main (namescomp.c:75)
==9581== Address 0x5729b04 is 0 bytes after a block of size 4 alloc'd
==9581== at 0x4C29024: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==9581== by 0x4E5C37B: recalloc (utils.c:41)
==9581== by 0x4E4FF50: ResizeKeyGroup (symbols.c:356)
==9581== by 0x4E5229E: AddSymbolsToKey (symbols.c:1058)
==9581== by 0x4E52ABB: SetSymbolsField (symbols.c:1214)
==9581== by 0x4E536C7: HandleSymbolsBody (symbols.c:1481)
==9581== by 0x4E53A63: HandleSymbolsDef (symbols.c:1543)
==9581== by 0x4E53DAD: HandleSymbolsFile (symbols.c:1623)
==9581== by 0x4E51CA4: HandleIncludeSymbols (symbols.c:909)
==9581== by 0x4E53D75: HandleSymbolsFile (symbols.c:1619)
==9581== by 0x4E51E74: HandleIncludeSymbols (symbols.c:951)
==9581== by 0x4E53D75: HandleSymbolsFile (symbols.c:1619)
Signed-off-by: Ran Benita <ran234@gmail.com>
Still keep things as 'ctx' internally so we don't have to worry about
typing it too often, but rename the user-visible API back as it was
kinda ugly.
This partially reverts e7bb1e5f.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
It was a pretty pointless check. Also sanitise the _x11 variant to
actually do what it says on the box.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
These are two aggregate file types which are not used anywhere. We
maintain useful-enough backward compatibility in the parser, by treating
them as xkb_keymap. The keymap type allows for all types of components,
so they will still compile fine if they ever come up.
Signed-off-by: Ran Benita <ran234@gmail.com>
(This breaks the API.)
"context" is really annoying to type all the time (and we're going to
type it a lot more :). "ctx" is clear, concise and common in many other
libraries. Use it!
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Fix for xkb -> keymap change.]
Each context gets its own table, i.e. interning a string in one context
does not affect any other context.
The existing xkb_atom_* functions are turned into wrappers around a new
standalone atom_table object.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb -> keymap.]
In preparation of contextualizing atom handling.
Since we touch every function call, we also rename the function to
xkb_atom_strdup to match xkb_atom_intern, and be more descriptive.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb -> keymap.]
In preparation of contextualizing the atom table.
Since we touch every function call, also rename the function to
xkb_atom_intern, to match better with the rest (which will also be
renamed).
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Fixed for 'xkb' -> 'keymap'.]
Currently the IDs are assigned from a static variable inside
CreateXKBFile. This can lead to some unpleasantness with threads, so
maintain the counter in the context instead.
Signed-off-by: Ran Benita <ran234@gmail.com>
We will need the context to remove some global state.
Also make the Parse* function just return bool while wer'e at it.
Signed-off-by: Ran Benita <ran234@gmail.com>
makekeys must be built with the build-native compiler, not with $(CC)
which is the cross-compiler. The only sane way to achieve this seems to
be to use a separate Makefile.am for it.
This patch fixes the problem apparently caused by:
commit b5efe41f19
Author: Ran Benita <ran234@gmail.com>
Date: Sat Mar 24 04:48:31 2012 +0200
Make build non-recursive
There is no such thing as makekeys_makekeys_CC in automake.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
We depend on parser.h which is generated by flex during the build, so
use an explicit $(top_builddir) rather than a relative path.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Change them to refer to the string representation of the keysym's name
as a name rather than a string, since we want to add API to get the
Unicode printable representation as well.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Two new calls allow users to test the exact modifier state, including
verifying that no other modifiers but the ones you wanted are down.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
xkbcommon-names.h right now just contains a set of hardcoded modifier
strings that are most commonly used for the usual modifiers. Provide
definitions of these so people don't have to worry about typoing a
string or mixing up Mod1 and Mod4.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Make the files in the src/* directory use their own header or a
consilidated private header. This makes the file dependencies clearer.
Also drop the pointless "xkb" file name prefix, add split a few
declarations to their own files (atom.h and text.h).
Signed-off-by: Ran Benita <ran234@gmail.com>
The include dependencies were quite convoluted, where you change the
order and get a ton of errors. Instead, change one file to act as the
internal interface for the xkbcomp files, and make every file use it.
Also drop the pointless "xkb" prefix to file names.
Signed-off-by: Ran Benita <ran234@gmail.com>
clang complains with the xorg-macros warning flags:
src/context.c:58:36: error: extension used [-Werror,-pedantic,-Wlanguage-extension-token]
typeof(new_paths));
This was not entirely correct, too. So bring back the casts to the
results of the allocation macros; might as well make them a bit more
type safe.
Signed-off-by: Ran Benita <ran234@gmail.com>
Fixes an 'unused' warning. There seems to be nothing wrong with these
sections though, all the tests pass.
Signed-off-by: Ran Benita <ran234@gmail.com>
This reverts commit d007cd0a1f.
This is in fact more restrictive, because it breaks the (common) case
where the strings are const themselved, e.g. "evdev", "us", etc. As is
you must either duplicate the strings or suppress the warnings.
If the user needs to retain the non-const strings, he should instead
just keep them in some other struct and use xkb_rules_names just as
a temporary parameter for xkb_map_new_from_names. Mildly annoying but
acceptable.
The two files do exactly the same sort of things, without any discernible
reason for splitting them.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb_desc -> xkb_keymap changes.]
'Cause defining your own True and False is so 1990's.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Fixed for xkb_desc -> xkb_keymap changes.]
Since the most common failure mode here is a failure to properly set the
XKB data path, dump the include path so people at least have a clue
where to look.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
POSIX specifies that these functions require <strings.h>, but we were
only including <string.h>. It did work, but still.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
Rewrite all of the current tests in the following ways:
- Instead of the current mix of C and shell, just use single-process
pure C file per test. All of the .sh files are removed, but everything
that was tested is ported.
- Instead of handling the test logs ourselves, use Automake's
"parallel-test" mechanism. This will create a single log file for each
test with it's stdout+stderr, and a top level "test-suite.log" file
for all the failed tests.
- The "parallel-tests" directive also makes the test run in parallel,
so "make check" runs faster.
- Also use the "color-tests" directive to have the "make check" output
colorized. Who doesn't like to see PASS in green?
- All of the test data files are moved into the test/data subdirectory.
That way we can just put the directory in EXTRA_DIST and forget about
it.
- The test/Makefile.am file is consolidated into the main Makefile.am,
for a completely non-recursive build.
Right now the tests are completely independent and just use simple
assert()'s. More sophistication can be added as needed.
It should also be noted that it's still possible to use shell, python,
etc. if a test wants more flexibility than C can provide, just do as
before.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb_keymap changes.]
This is very useful because it avoids redundent pointers in structs
and/or parameter passing in the application.
Signed-off-by: Ran Benita <ran234@gmail.com>
The definitions in config.h should be available in all files an
implementation detail; it can be included through the build system
instead of having each file pull it every time.
This is especially helpful with AC_USE_SYSTEM_EXTENSIONS, as _GNU_SOURCE
and friends can have an effect by merely being defined, which can lead
to some confusion if its effective for only half the files.
And we don't really support a build _without_ config.h; so, one less
thing to worry about.
Signed-off-by: Ran Benita <ran234@gmail.com>
These are no longer needed for using the library, only building it. Most
users would still want xproto though, for the keysym definitions.
Signed-off-by: Ran Benita <ran234@gmail.com>
The kbproto header is already not needed here anymore.
Move the _X_EXPORT's to the corresponding function definitions, and use
straight extern "C" clauses instead of _XFUNCPROTOBEGIN/END.
It also makes more sense to have the EXPORT's in the source files, as it
provides some documentation to the reader, whereas in the header it's
obvious.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb_keymap changes.]
xkb_state_ref was missing.
Also modify the _ref functions to return the object instead of being
void. This is a useful idiom:
struct my_object my_object_new(struct xkb_state *state)
{
[...]
my_object->state = xkb_state_ref(state);
[...]
}
Essentially "taking" a reference, such that you don't forget to
increment it and it's one line less (see example in our own code).
A case could also be made for _unref to return the object or NULL, but
this is quite uncommon.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb_keymap changes.]
When merging group info from two KeyInfo's, the new size of the keysym
array was off. Fix it to match how it is used a few lines below.
There are also some peripheral fixes, and some comments (took me a
few minutes to get what's going on).
Signed-off-by: Ran Benita <ran234@gmail.com>
(They were not reported, see next commit).
The reset function declaration didn't match its name in the definition;
the _defaults variant matches better with the rest.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated to current master.]
Since we never return an xkb_rules_names and it's all user-provided
strings, seems a bit harsh to have it const.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
autotools was warning that AC_USE_SYSTEM_EXTENSIONS was being called too
late, so move it earlier. Also shove BASE_CFLAGS into CFLAGS so we get
all the added warning flags from xorg.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Various one-liners (mostly removing unused variables) to make the code
safe for the full set of warnings used by the xorg macros.
On Debian-based systems, flex generates incorrect code resulting in two
warnings about yy_getcolumn and yy_setcolumn having no previous
declaration despite being non-static. Fedora carries a patch to fix
this, and a bug has been filed on Debian's flex to add the patch:
http://bugs.debian.org/667027
Aside from this, it's now safe for --enable-strict-compilation.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
If a keysym was specified as "U1039andsomeextrastuffontheend", return
NoSymbol rather than 0x10001039; similarly, return NoSymbol for
"0xdeadbeefhitherehowsyourdaybeen" rather than 0xdeadbeef.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This change makes sure that include does not overwrite previous
compatibility modifier settings when the included files does not
explicitly specify them.
Signed-off-by: Andreas Wettstein <wettstein509@solnet.ch>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
[Cross-picked from xkbcomp commit 14470719.]
When xkb_free_keymap is called the atoms are all free'd, but action.c
keeps a global copy of interned "true" and "false", which remains stale.
The correct fix is to remove the need for the ActionsInit function
entirely.
Signed-off-by: Ran Benita <ran234@gmail.com>
These were several initializations that were forgotten in the previous
memory leak fixes.
Now several xkb_desc's can coexist (relatively) peacefully.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Only the atom.c is relevant with the new context API.]
The "makeit" variable is always true. Remove it and de-indent.
(Also change the type of the "len" variable to size_t to avoid some
useless casting).
Signed-off-by: Ran Benita <ran234@gmail.com>
The NULL check is unneeded, and prevented the atoms from being free'd.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated for xkb_map_unref.]
The new glibc (2.15) appear to cause trouble, particularly the sscanf
call, where makekeys will output empty hash tables. Using the appropriate
macros from inttypes.h makes it work again.
Signed-off-by: Ran Benita <ran234@gmail.com>
Such as:
Compiling path: ./test/data/bad.xkb mapName:
==1300== Conditional jump or move depends on uninitialised value(s)
==1300== at 0x4E46166: HandleVModDef (vmod.c:90)
==1300== by 0x4E3FEC9: HandleKeyTypesFile (keytypes.c:1035)
==1300== by 0x4E3FBE1: HandleIncludeKeyTypes.constprop.11 (keytypes.c:387)
==1300== by 0x4E401DD: HandleKeyTypesFile (keytypes.c:1022)
==1300== by 0x4E3FBE1: HandleIncludeKeyTypes.constprop.11 (keytypes.c:387)
==1300== by 0x4E401DD: HandleKeyTypesFile (keytypes.c:1022)
==1300== by 0x4E4026F: CompileKeyTypes (keytypes.c:1150)
==1300== by 0x4E3DF9B: CompileKeymap (keymap.c:169)
==1300== by 0x4E465E9: compile_keymap (xkbcomp.c:205)
==1300== by 0x4E46BE4: xkb_compile_keymap_from_file (xkbcomp.c:290)
==1300== by 0x400B37: test_file (filecomp.c:47)
==1300== by 0x4008E3: main (filecomp.c:90)
==1300== Uninitialised value was created by a stack allocation
==1300== at 0x4E3FB3F: HandleIncludeKeyTypes.constprop.11 (keytypes.c:366)
Signed-off-by: Ran Benita <ran234@gmail.com>
If we can't find the component of the include file we're looking for,
make sure we don't return success when we meant failure, segfault, or
spectacularly leak everything.
Tested with incorrect component includes for keycodes, compat, symbols,
and types.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reported-by: David Herrmann <dh.herrmann@googlemail.com>
Which also involved moving the global symbol map to be per-key instead;
this should probably be split out into a separate commit.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Instead of allowing only one keysym per level per group, do as the
external API indicates and allow multiples. The existing syntax is:
key <AD01> { [ q, Q ] };
where the new syntax is:
key <AD01> { [ q, Q, { H, E, L, L, O },
{ Y, E, S, space, T, H, I, S, space, I, S, space, D, O, G } };
to make the key in the extreme top left of the keyboard do pretty
surprising things in levels 3 and 4.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
All global state is removed from the parser and scanner.
This makes use of the standard facilities in Bison and Flex for
reentrant/pure scanner/lexer and location tracking.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: Updated to current sources.]
Unify all the different Makefile.am into a single short top level one
(the test/Makefile.am file is left intact though).
This makes the build system simpler to look and should encourage
unifying more currently-disparate code.
Some further motivation can be found in this page:
http://www.flameeyes.eu/autotools-mythbuster/automake/nonrecursive.html
Signed-off-by: Ran Benita <ran234@gmail.com>
Since we have our own xkb_keysym_t type, it makes sense to have our own
NoSymbol value instead of the one from X11/X.h.
Signed-off-by: Ran Benita <ran234@gmail.com>
Since we define our own xkb_atom_t type, it makes sense not to use the
X11/X.h None value. This way we can also remove a lot of X11 includes.
Signed-off-by: Ran Benita <ran234@gmail.com>
i.e comparison of signed and unsigned values. These are mostly
harmless but fixing them allows to compile cleanly with -Wextra.
Signed-off-by: Ran Benita <ran234@gmail.com>
There are some cases where we must free a string with a const qualifier.
Add a macro UNCONSTIFY to trick the compiler into silencing the warning
in the cases where we know what we're doing.
Signed-off-by: Ran Benita <ran234@gmail.com>
And merge all the similar ones into the same name.
The u* prefix is chosen over the _Xkb prefix because it has more uses
throughout the codebase. But It should now be simple to choose a nice
prefix and stay consistent.
Signed-off-by: Ran Benita <ran234@gmail.com>
[daniels: fixed for the case where we have strcasecmp]
Some unused defines and geometry-removal leftovers (specifically the
file geom.c and the struct for the keyboard coordinates).
Signed-off-by: Ran Benita <ran234@gmail.com>
The recalloc function should be expressed in terms of bytes to match its
name. However uTypedRecalloc retains its type so nothing is changed.
Signed-off-by: Ran Benita <ran234@gmail.com>
Sorry if your libc doesn't have this, but it's not my problem.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reported-by: Ran Benita <ran234@gmail.com>
GCC complained:
misc.c:235:13: warning: logical 'and' of mutually exclusive tests is always false [-Wlogical-op]
misc.c:248:13: warning: logical 'and' of mutually exclusive tests is always false [-Wlogical-op]
Looking at keysymdef.h shows that the check is indeed wrong. This commit
updates the check to match the latin8 section of keysymdef exactly.
Signed-off-by: Ran Benita <ran234@gmail.com>
Use XKB_KEY_UP instead of 0 and XKB_KEY_DOWN instead of 1.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reported-by: Ran Benita <ran234@gmail.com>
And also convert state.c to use the state API for mods and groups,
rather than testing the state members directly.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This resulted in duplicate sets of modifiers, since we were comparing
pointer equality of two strings, rather than string equality. Oops.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Add new API to deal with xkb_state objects, including
xkb_state_update_key, which runs the XKB action machinery internally to
calculate what exactly happens to the state when a given key is pressed
or released.
The canonical way to deal with keys is now:
struct xkb_state *state = xkb_state_new(xkb);
xkb_keysym_t *syms;
int num_syms;
xkb_state_update_key(state, key, is_down);
num_syms = xkb_key_get_syms(state, key, &syms);
More state handling API, including a way to get at or ignore preserved
modifiers, is on its way.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Previously, we would clear out the real modmask when updating the
modmask for action maps, if not using the key's modmask. The correct
behaviour here is instead to use the key's modmask if using the modmap,
else use the real mods provided with the action originally.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Unfortunately we can't get the actual file it was defined in this far
down, but at least give the human-readable name rather than just a group
index.
Also, groups are not zero-indexed, such that index 0 is group 1; fix
that too.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Instead of generating a fairly droll internal error, generate a warning
also telling us exactly where the bad definition was.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
By ensuring their mask is only the vmods, rather than also potentially
including the key's modmap. Also remove the unnecessary vmodmask
indirection.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Add the dump of my full current X11/XKB keymap as a test for filecomp,
being as it also includes geometry.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The server used to have to go and do this on our own, but we can do
better than that: after we've compiled the keymap, go through and bind
virtual modifiers to everything that needs it.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Right now we just silently ignore overlay controls, which is probably
bad, but it's not the easiest to fix.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
It looks like this could never have worked anyway, what with num_rg
always being 0 everywhere. Remove it.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The XkbKey* macros used to work through XkbCM* and XkbSM* variants, for
the xkb_map and xkb_server_map respectively; the latter versions weren't
used by anyone at all, so just bin them and make the macros work
directly.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
They're no longer needed since we don't expose any atoms in the
published API anymore. As a result, we don't need to support external
atom implementations either. Result!
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Ironically, the test for named.xkb included a call that relied on the
default keymap, without the file naming an explicit default. Go figure.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
On many layouts, the following error appears:
Internal error: Could not resolve keysym 10005b0
(Which is like the trademark of libxkbcommon now, and makes
unicode-heavy symbol files pretty useless).
This occurs when a keysym string (in this case, 10005b0) is passed to
xkb_string_to_keysym, but cannot be resolved.
This in turn happens because the parser passes on hexadecimal keysym
strings without the leading "0x", thus leaving the resolving function
without a way to disambiguate it as a number.
Therefore, make sure to pass on the "0x". The file symbols.c in xkbcomp
project does the same; it probably got lost in translation.
Signed-off-by: Ran Benita <ran234@gmail.com>
This needlessly occupies memory for the lifetime of the library, and
does not make a noticeable difference otherwise.
This rules file won't be loaded more than once in most cases anyway, so
just load it again when it happens.
Signed-off-by: Ran Benita <ran234@gmail.com>
This needlessly occupies memory for the lifetime of the library, and
does not make a noticeable difference otherwise.
Instead, just parse the same file again when it happens.
Signed-off-by: Ran Benita <ran234@gmail.com>
The yacc implementation can generate all the necessary token
definitions itself; there is no need to maintain a hand written
file for that.
Signed-off-by: Ran Benita <ran234@gmail.com>
This integrates two commits from libX11:
ebd6ef0a4db0ddef0ae17ad14571518ccdeea5ba
XStringToKeysym: Special case for XF86 keysyms
Some XFree86 keysyms were in XKeysymDB as XF86_foo, despite really being
XF86foo. So, if we get to the bottom of XStringToKeysym and haven't
found our XF86_foo, try it again as XF86foo.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
00175397480b76d32bf82b0c7c94c91a2a95954e
makekeys: Scan vendor keysyms as well as core
Since we can't really live without vendor keysyms, scan them all in to
generate ks_tables.h, rather than only doing the core ones, and leaving
the vendor syms to be manually synchronised with XKeysymDB.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Notice that the xkey.sh test is changed to match libX11 behavior, i.e.
XKeysymToString(0x1008FE20) -> "XF86Ungrab" as opposed to "XF86_Ungrab".
Signed-off-by: Ran Benita <ran234@gmail.com>
This matches commit 78a894231ae8ec4959007b6d9b8d2a15d6333a1e from
libX11:
Signed-off-by: Alan Coopersmith <alan.coopersmith@sun.com>
Signed-off-by: Ran Benita <ran234@gmail.com>
This matches commit 24283d40b1e4314c6647dda49d2a159833341a8b from
libX11:
Since makekeys is built using build environment's compiler and
runs natively, we have to make sure that the size of the
Signature type is the same on both the native environment
and the target, otherwise we get mismatches upon running X,
and some LSB test failures (xts5).
Use an unsigned 32-bit integer on all platforms.
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Ran Benita <ran234@gmail.com>
Fix all reported null dereferences from clang-analyzer.
There seems to be one false negative (in file indicators.c), but it is
fixed anyway.
Signed-off-by: Ran Benita <ran234@gmail.com>
Every caller did the exact same check on the group bounds after calling
ExprResolveGroup, so might as well do it inside.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
We never want to solely lookup a virtual modifier without also looking
up core modifiers. So, rather than chaining the vmod lookup inside the
core modifier lookup, invert the ordering.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
None of the lookup functions anyone ever used supported field
references, so don't pretend we do in the API.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
groupNames was declared in compat.c as a global to anything which
included compat.h (for which groupNames was its sole reason to exist),
but only ever used in indicators.c.
Which is kind of fortunate, given that e314931e removed identical
definitions of groupNames (as integers, not masks) from both action.c
and symbols.c.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Move the bulk of ExprResolveInteger into an internal function called
ExprResolveIntegerLookup, and introduce ExprResolveInteger as a simple
wrapper which doesn't take priv/lookup arguments.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Because, joke's on you, it wasn't actually looking up radio groups.
Just checking to see if it was a string that was "none", or an integer.
Lord give me strength.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Which is just a slightly more typesafe wrapper around the chained
ExprResolveModMask everyone was using earlier.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Make sure we carry over an explicit minimum/maximum keycode setting,
rather than just using the computed minimum/maximum; this got broken
while changing the keycode range to be unsigned.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Reported-by: Pekka Paalanen <ppaalanen@gmail.com>
Thanks to autotools happily building stale generated sources, I hadn't
actually ever built my xkbparse.y changes. Fix that so it not only
compiles, but works. This seems to parse long keycodes correctly,
although I very much would not recommend testing this by declaring
0x1fffffff as your highest keycode.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Some error paths don't set info->xkb correctly, so just do like most
utility functions and pass the xkb_desc explicitly.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
And use it consistently everywhere, including with a special long-safe
internal keycode type, to ease the transition to large keycodes.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
For some reason, lex decided to reduce a strcpy into an assignment,
leading to entirely justified valgrind warnings about invalid reads,
when scanFile was set to a string which may have only ever lived on the
stack of a now-exited function.
Make it a strdup() instead.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Some routes through HandleGeometryVar do not set a return value. Set a default
value for the return variable to avoid returning an uninitialised value.
Which just calls XkbcFreeKeyboard with the only arguments you'd ever
pass it.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Avoids assigning the global pointer to a value that may only have a stack
lifetime:
Fixes valgrind warnings such as:
==24795== Invalid read of size 1
==24795== at 0x4A06E9A: strcpy (mc_replace_strmem.c:311)
==24795== by 0x4E54D68: ProcessIncludeFile (misc.c:73)
==24795== by 0x4E59726: HandleIncludeSymbols.constprop.3 (symbols.c:829)
==24795== by 0x4E59D8E: HandleSymbolsFile (symbols.c:1673)
==24795== by 0x4E5A068: CompileSymbols (symbols.c:2211)
==24795== by 0x4E51A61: CompileKeymap (keymap.c:155)
==24795== by 0x4E5B410: xkb_compile_keymap_from_components (xkbcomp.c:236)
==24795== by 0x4E5B587: xkb_compile_keymap_from_rules (xkbcomp.c:161)
==24795== by 0x405ED2: display_create (window.c:2007)
==24795== by 0x403732: main (desktop-shell.c:320)
==24795== Address 0x7fefff0a0 is just below the stack ptr. To suppress, use:
--workaround-gcc296-bugs=yes
==24795==
==24795== Source and destination overlap in strcpy(0x7fefff430, 0x7fefff430)
==24795== at 0x4A06F3D: strcpy (mc_replace_strmem.c:311)
==24795== by 0x4E54D68: ProcessIncludeFile (misc.c:73)
==24795== by 0x4E59726: HandleIncludeSymbols.constprop.3 (symbols.c:829)
==24795== by 0x4E59D8E: HandleSymbolsFile (symbols.c:1673)
==24795== by 0x4E5A068: CompileSymbols (symbols.c:2211)
==24795== by 0x4E51A61: CompileKeymap (keymap.c:155)
==24795== by 0x4E5B410: xkb_compile_keymap_from_components (xkbcomp.c:236)
==24795== by 0x4E5B587: xkb_compile_keymap_from_rules (xkbcomp.c:161)
==24795== by 0x405ED2: display_create (window.c:2007)
==24795== by 0x403732: main (desktop-shell.c:320)
Those warnings disappear accordingly:
| CC parseutils.lo
| parseutils.c:742: warning: no previous prototype for ‘CheckDefaultMap’
| CC xkbscan.lo
| xkbscan.l: In function ‘XKBParseString’:
| xkbscan.l:220: warning: implicit declaration of function ‘CheckDefaultMap’
| xkbscan.l:220: warning: nested extern declaration of ‘CheckDefaultMap’
Reviewed-by: Dirk Wallenstein <halsmit@t-online.de>
Signed-off-by: Cyril Brulebois <kibi@debian.org>
This reverts commit bf9fdceef6.
We really only want to expose symbols that we explicitly mark as part of
the API. This may not work with other platforms or compilers, but the
fact that private symbols are not available on Linux+GCC is enough of an
incentive to not use those.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
There's no need for this xlib include:
| YACC xkbparse.c
| CC xkbparse.lo
| xkbparse.y:98:22: error: X11/Xlib.h: No such file or directory
Signed-off-by: Cyril Brulebois <kibi@debian.org>
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
The Copyright statements must appear in full.
When only the year was different, I added it in an existing
Copyright statement.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The default value ${dataroot}/X11/xkb only works if xkeyboard-config
has created the keymaps in that directory. Let's obtain the true final
value of where the keymaps are and use that as a default. In a production
environment this is the only value that can work.
This new default value also has the merit of making the 'check' target
to work in distcheck which does not have a copy of the xkeyboard-config
keymaps in its sandbox based on ${dataroot}/X11/xkb. The test data
cannot find the "include" keymaps it needs.
.../libxkbcommon-0.1.0/_inst/share/X11/xkb
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The test programs and the test data are required in the tarball
and needed for distcheck.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
LDADD is a Makefile wide variable.
Automake matches prog name with .c file by default
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
This would cover the scenario where these headers file are updated,
for example, a new version is installed. Running 'make' again
on libxkbcommon should rebuild ks_tables.h.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
This program is a utility to generated a header file.
The header file it generates should not be located in the
directory where this utility program is compiled.
Move the /makekeys dir as a sibling of /src.
This reduces the number of bi-directional relationships
between directories.
Make corresponding makefiles simplifications.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
These automake variables are not currently used.
The variable KS_HEADERS is not required anymore.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
CFLAGS is a user variable which should never be set by the configuration.
It allows the user to alter the configuration compiler options.
The visibility is only set for GNU compiler, leaving libraries built
with other compilers with the wrong visibility.
All other xorg libraries set visibilty using _X_EXPORT or _X_HIDDEN.
For the time being, all the symbols will have the default visibility
which does not break anything.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
Rather than appending X11 to the include dir.
It should be safe to use as it has been added in 2005.
Use a local variable name matching the pkgconfig name.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
If pkg-config is missing, the configuration fails while initializing
XORG_* macros:
warning: PKG_PROG_PKG_CONFIG is m4_require'd but not m4_defun'd
line 11622: PKG_PROG_PKG_CONFIG: command not found
The output is sufficient to locate the problem.
It is not possible, or very difficult, to not have pkg-config
installed as the whole desktop and countless other software
depends on it.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
BUILT_SOURCES and MAINTAINERCLEAN are not needed for lex and yacc
Note that xkbscan was missing on those lines.
Automake generates all the rules to handle building, distribution
and cleaning.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
Autoconf does not handle this scenario as well as Automake does.
If xkbparse.c is already there, do not abort the configuration
if yacc is missing as it is not needed.
If both xkbparse.c and yacc are missing, abort configuration.
Yacc is a required tool to build the package.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
The -Im4 in Makefile.am will have to wait until the day
there is an m4 macro checked-in git in the m4 directory.
This does not prevent libtool to install its macros in m4.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
This toplevel .gitignore covers all possible generated files
from autotools, compiler, linker, etc...
It has been in use for over a year in all xorg modules and has
prevented a good number of patches.
For example, in this module, src/stamp-h1 was not ignored even
if the .gitignore had just been updated. Files with a .patch
extension were not ignored which sometimes cause a patch to be
checked-in git accidentally.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
XORG_STRICT_OPTION from XORG_DEFAULT_OPTIONS calls
AC_PROG_C_C99. This sets gcc with -std=gnu99.
If AC_PROG_CC macro is called afterwards, it resets CC to gcc.
Acked-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
This will handle AM_SILENT_RULES such that it will not break
on earlier Autoconf versions.
It provides a host of other configuration items
Acked-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
This is the current minimum version required for all X.Org modules.
The module can be configured with any higher version.
Autoconf features from version > 2.60 cannot be used.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
xkbscan.l: In function 'setScanState':
xkbscan.l:201:1: warning: control reaches end of non-void function
I: Program returns random data in a function
E: libxkbcommon no-return-in-nonvoid-function xkbscan.l:201
Change return type of setScanState to void, since a return value is
never used by its callers.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
libtoolize: Consider adding "AC_CONFIG_MACRO_DIR([m4])" to configure.ac and
libtoolize: rerunning libtoolize, to keep the correct libtool macros in-tree.
libtoolize: Consider adding "-I m4" to ACLOCAL_AMFLAGS in Makefile.am.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Use -no-undefined to assure libtool that the libxkbcommon library has
no unresolved symbols at link time, so that libtool will build a shared
library on platforms require that all symbols are resolved when the
library is linked.
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Keep the parsed form of the last-used rules file around, and reuse that
if we get asked for the same ruleset. If not, bin it and cache the
other one.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Use CARD32 instead of Atom/KeySym/et al to avoid type size confusion
between server and non-server code; relatedly, move the geometry headers
in from kbproto, so every non-simple type (i.e. structs containing
nothing more than basic types) is now copied into xkbcommon.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Allow people to plug in an external atom database (e.g. the X server's),
so we don't have to migrate our own atoms over later. We are a bit
over-keen on atoms at the moment, so it does pollute the atom database a
bit though.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
On 64-bit architectures, XID varies in size between the server (always
32 bits), and non-server (always unsigned long) for some inexplicable
reason. Use CARD32 instead to avoid this horrible trap.
This involves dragging in XkbClientMapRec so we don't get stuck in the
KeySym trap.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Use Xkbc* for all our actions that we intend to keep around, and Xkb*
for deprecated ones we can hopefully get rid of, at least internally.
While we're at it, make vmods be a uint32_t.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
So, it turns out that if you're parsing a fairly large amount of data,
using getc() to get all the input rather than, say, read(), is some kind
of remarkably daft and unperformant idea.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Instead of calling XStringToKeysym on every keysym we parse, store it as
a string until we need to store it in an actual keymap.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
XkbAtomGetString() returns a freshly-allocated string, whereas
XkbAtomText() returns the same in a temporary buffer. XkbAtomText used
to call XkbAtomGetString() and then free the result, which seems quite
spectacularly pointless when you think about it. Shuffle the atom code
around so we don't have to allocate for XkbAtomText().
This changes semantics slightly wrt non-printable characters, but I
haven't been able to see any effect so far. And it may well be ever so
slightly quicker.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
This will let us implement XkbAtomGetString on top of XkbAtomText,
instead of having the latter get a duplicated string, dump it into a
temporary buffer, and subsequently free it (sigh).
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
XkbcInternAtom(XkbcAtomGetString(atom)) has to be the most spectacularly
broken antipattern I've yet seen. Just compare the atoms directly.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Canonicalises two sets of components (new and old), e.g.:
new: +bar
old: foo
result: foo+bar
This is required as part of the spec, so clients can reuse part of the
device's old keymap.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Ensure that Unicode string representations are accepted and turned into
numbers, as well as hexadecimal numbers in 0xabcd1234 form; unknown
keysyms are output as 0xabcd1234 in string form.
This also ensures that strings are never returned malloc()ed.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Add NoSymbol into the keysym table, so keysym <-> string conversion works for
that, too; also eliminate special-casing of VoidSymbol.
This will require special-casing in libX11 to preserve its API.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
filecomp would fail because it couldn't find the input files, after
compilation failed due to missing includes.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
The listing code in xkbcomp had been setup to allocate a set of buffers
with file paths and then later parse through them to find which maps were
needed.
All the allocation (with the additional allocation for the components
list) was making it really slow, so this patch makes everything simpler
by just generating the components list as we walk the directory tree.
In xkbcomp, the listing code printed out the xkb files to stdout and the
caller would parse the output. Here, we can just generate a
XkbComponentListPtr and pass it back.
This should be a series of smaller commits, but there was no clean way to
do it since it's basically a complete rewrite except for the core map
matching logic.
A lot of code used for special printing modes in xkbcomp has been
shedded. Callers can massage the output as they please.
This reverts commit c4c9e36fbf. It turns
out that the listing code is used to support the X_kbListComponents
request (via XkbListComponents).
This will have to be refactored into some reasonable interface instead
of the current usage where the server reads xkbcomp stdout. Gross.
Instead of hardcoding the XKB base directory when searching for rules in
the xkbcomp code, we can extend the xkbpath API to cover rules and reuse
it. That will make it more convenient if it's ever exposed so people can
set their XKB search paths in a reasonable way.
We need to support generating a keyboard description from a keymap file
because there are just some cases where RMLVO or ktcsg is not enough.
The map choosing logic has been refactored into its own function and now
supports choosing a named or default keymap.
The noble intention was to expose all the new API and new generic types
in the split out kbproto headers through XKBcommon.h. It turns out that
would be a massive amount of work in the server. Someday, but first just
wedging in XkbCompileKeymap* would be good.
Most of the API is in new internal xkb*.h headers. In order to allow the
XKBcommon.h header to be used from the server, we can't pull in other
headers from kbproto since the server has its own copies. However, types
that are different (XkbDescRec, XkbAction) still have Xkbc equivalents
here, and I think they should be used in the server.
Some of the XkbAction types are defined differently in the server, so we
add those to XKBcommon.h and use them here like XkbcDescPtr. We'll have
to deal with the impedance mismatch on the client side later.
XkbcRF_GetComponents was returning an error but leaving the generated
components alone. This ensures that the broken XkbComponentNamesPtr is
freed and the error is passed up to the caller.
Instead of requiring the user to call XkbInitIncludePath() and
XkbAddDefaultDirectoriesToPath(), all the path entry points now implicitly
initialize the path. When initializing, the default directories are added
so it's useful.
This provides normal operation without exposing the xkbpath API. That
might happen later to allow apps to edit the XKB search path.
Added a test program, rulescomp, which takes a RMLVO set and generates a
XkbcDescPtr. This is essentially what the xserver will do, except that we
still need to access some xkbcomp internal API to make it work.
XkbcCompileKeymapFromRules can be used to generate a XkbDescPtr from XKB
rules instead of using components. The previous XkbcCompileKeymap has
been renamed to XkbcCompileKeymapFromComponents.
Copy over the Xkb_RF* rules parsing functions from xkbfile's maprules.c.
Eventually, this will be tied into xkbcomp's path searching utilities so
you don't need to supply a full path to the rules file. Also, it this
should eventually incorporate the server's RMLVOSet.
Copy over the Xkb_RF* rules parsing functions from xkbfile's maprules.c.
Eventually, this will be tied into xkbcomp's path searching utilities so
you don't need to supply a full path to the rules file. Also, it this
should eventually incorporate the server's RMLVOSet.
Finally, we can generate a XkbcDescPtr from a XkbComponentNamesPtr. This
involves turning the components into a parsed XKB file and then passing
it into the compiler. This first conversion needs more error handling.
xkbcomp was using Xlib's XConvertCase to check upper/lowercase. That's a
lot of code, so hopefully the xkbfile macros using _XkbKSCheckCase are
good enough. This also required that <X11/keysym.h> is included to get
all the XK_* definitions.
Added common variants of XkbComputeEffectiveMap, XkbInitCanonicalKeyTypes
and all their needed functions. A couple functions that were in alloc.c
moved to malloc.c to mirror the libX11 sources better.
Original code in
libX11/src/xkb/XKBMalloc.c
libX11/src/xkb/XKBMisc.c
libX11/src/xkb/XKB.c
This should cover all the usage in xkbcomp. The format arguments were
dropped except for the special case of XkbModMaskText, which needs to
write in XkbCFile format in HandleVModDef. This was just changed to a
Bool to avoid the need for the macros in XKBfile.h.
The function prefixes have been renamed to be unique from xkbfile.
The only real usage was in the frontend to generate a .xkm file. The
rest of the code just operated on the attached XkbDescPtr. Note that
here we've replaced the usage of the defined field in CompileKeymap with
the equivalent field in a XkbcDescPtr.
A copy of the xkbcomp sources (except the frontend) have been copied in
to provide a means to compile a XkbDescPtr. This definitely doesn't
build or do the right thing yet.
This avoids the BUILT_SOURCES weirdness and forcing the build into the
makekeys subdirectory. Added a new make variable KS_HEADERS that lists
all the keysym headers we're going to parse.
Some coding style nits were cleaned up. Additionally, most of the
functions have been collapsed from the libxkbfile version where there's
distinction with the Xlib atom functions when Display was set. Finally,
the InitAtoms function tests whether the table has already been created
by testing the pointer rather than using a static int.
Mostly tab-to-space conversion plus a few style nits. Dropped the
register keywords as I'm pretty sure modern compilers can be trusted to
do the right thing.
Following the kbproto convention, the headers will be named XKBcommon.h
and XKBcommonint.h. Furthermore, they'll be installed in X11/extensions
directory with the rest of the XKB headers.
Copies the code to initialize and destroy an XkbDescRec from libX11. The
original code is in
libX11/src/xkb/XKBAlloc.c
libX11/src/xkb/XKBGAlloc.c
libX11/src/xkb/XKBMAlloc.c
Switched to the common XKB headers which don't contain Xlib pollution.
A new common keyboard description type, XkbcDescRec, has been added.
This is the XkbDescRec definitions in xkbstr.h in the xserver since it
contains the defined field we'll need.
These are used throughout the XKB code, but are defined in XKBsrv.h,
which we'd like to avoid. Internal definitions for True/False have also
been added since they're in Xlib.h
A test program and script have been added for checking the XkbCommon
keysym functions. This has already highlighted an error in handling of
keysyms from XF86keysym.h.
Mostly tab-to-space conversion plus a few style nits. Dropped the
register keywords as I'm pretty sure modern compilers can be trusted to
do the right thing.
Add the xkbcommon implementations of XKeysymToString and XStringToKeysym.
These symbols have the namespace prefix of Xkbc and are declared in
X11/XkbCommon.h.
The implementation is taken directly from Xlib, but does not include the
XKeysymDB parsing and hashing yet (if it ever will). A couple type
conversions were needed to keep from using Xlib.h. See original files:
libX11/src/KeysymStr.c
libX11/src/StrKeysym.c
The makekeys utility is used to generate the keysym hash tables during
the build. We try to detect a build machine native compiler so the
tables can be generated when cross compiling.