Windows is a ridiculous operating system designed by toddlers, and so
requires us to close all file handles in the `tmp/xxxxxxx` cache dir
before renaming it into `o/xxxxxxx`. We have a hack in place to handle
this for the main output file, but the MachO linker also outputs a file
with debug symbols, and we weren't closing it! This led to a fuckton of
CI failures when we enabled `.whole` cache mode by default for
self-hosted backends.
thanks jacob for figuring this out while i sat there
These were previously incremental tests, so weren't running. They didn't
*need* to be incremental. They worked under the old runner because of
how it directly integrated with the compiler so tracked error messages
differently.
Unfortunately, I can't easily add a test for this, because the repro
depends on some details of DWARF layout; but I've confirmed that it
fixes a bug repro on another branch.
Clearing the analysis roots was very clever and all, but not actually
valid. We need to avoid *any* reference to the analysis errors if there
were any fatal files, and that includes sorting the errors!
Resolves: #22774
Sometimes we emit runtime instructions in comptime scopes. These
instructions will be discarded, but they allow comptime blocks to
contain intermediate runtime-known values, which is necessary for
expressions like `runtime_array.len` to work.
Since we will always throw away these runtime instructions, including
safety checks is a time waste at best and trips an assertion at worst!
Resolves: #20064
The changes from a few commits earlier, where semantic analysis no
longer occurs if any Zig files failed to lower to ZIR, mean `file`
dependencies are no longer necessary! However, we now need them for ZON
files, to be invalidated whenever a ZON file changes.
This came with a big cleanup to `Zcu.PerThread.updateFile` (formerly
`astGenFile`).
Also, change how the cache manifest works for files in the import table.
Instead of being added to the manifest when we call `semaFile` on them,
we iterate the import table after running the AstGen workers and add all
the files to the cache manifest then.
The downside is that this is a bit more eager to include files in the
manifest; in particular, files which are imported but not actually
referenced are now included in analysis. So, for instance, modifying any
standard library file will invalidate all Zig compilations using that
standard library, even if they don't use that file.
The original motivation here was simply that the old logic in `semaFile`
didn't translate nicely to ZON. However, it turns out to actually be
necessary for correctness. Because `@import("foo.zig")` is an
AstGen-level error if `foo.zig` does not exist, we need to invalidate
the cache when an imported but unreferenced file is removed to make sure
this error is triggered when it needs to be.
Resolves: #22746
This is mainly in preparation for integrating ZonGen into the pipeline
properly, although these names are better because `astGenFile` isn't
*necessarily* running AstGen; it may determine that the current ZIR is
up-to-date, or load cached ZIR.
Instead, `source`, `tree`, and `zir` should all be optional. This is
precisely what we're actually trying to model here; and `File` isn't
optimized for memory consumption or serializability anyway, so it's fine
to use a couple of extra bytes on actual optionals here.
The ZON PR (#20271) is causing these tests to inexplicably fail. It
doesn't seem like that PR is what's breaking GPA, so these tests are now
disabled. This is tracked by #22731.
This commit effectively reverts 9e683f0, and hence un-accepts #19777.
While nice in theory, this proposal turned out to have a few problems.
Firstly, supplying a result type implicitly coerces the operand to this
type -- that's the main point of result types! But for `try`, this is
actually a bad idea; we want a redundant `try` to be a compile error,
not to silently coerce the non-error value to an error union. In
practice, this didn't always happen, because the implementation was
buggy anyway; but when it did, it was really quite silly. For instance,
`try try ... try .{ ... }` was an accepted expression, with the inner
initializer being initially coerced to `E!E!...E!T`.
Secondly, the result type inference here didn't play nicely with
`return`. If you write `return try`, the operand would actually receive
a result type of `E!E!T`, since the `return` gave a result type of `E!T`
and the `try` wrapped it in *another* error union. More generally, the
problem here is that `try` doesn't know when it should or shouldn't
nest error unions. This occasionally broke code which looked like it
should work.
So, this commit prevents `try` from propagating result types through to
its operand. A key motivation for the original proposal here was decl
literals; so, as a special case, `try .foo(...)` is still an allowed
syntax form, caught by AstGen and specially lowered. This does open the
doors to allowing other special cases for decl literals in future, such
as `.foo(...) catch ...`, but those proposals are for another time.
Resolves: #21991Resolves: #22633
This check isn't valid in such cases, because the source and destination
pointers both refer to zero bits of memory, meaning they effectively
never alias.
Resolves: #21655
This instruction is like `intcast`, but includes two safety checks:
* Checks that the int is in range of the destination type
* If the destination type is an exhaustive enum, checks that the int
is a named enum value
This instruction is locked behind the `safety_checked_instructions`
backend feature; if unsupported, Sema will emit a fallback, as with
other safety-checked instructions.
This instruction is used to add a missing safety check for `@enumFromInt`
truncating bits. This check also has a fallback for backends which do
not yet support `safety_checked_instructions`.
Resolves: #21946
I recently saw a user hit the "comptime call of extern function" error,
and get confused because they didn't know why the scope was `comptime`.
So, use `explainWhyBlockIsComptime` on this and related errors to add
all the relevant notes.
The added test case shows the motivating situation.
* The langspec definition of `@memcpy` has been changed so that the
source and destination element types must be in-memory coercible,
allowing all such calls to be raw copying operations, not actually
applying any coercions.
* Implement aliasing check for comptime `@memcpy`; a compile error will
now be emitted if the arguments alias.
* Implement more efficient comptime `@memcpy` by loading and storing a
whole array at once, similar to how `@memset` is implemented.
The real problem here is that Git for Windows has horrendous defaults
which convert LF to CRLF. However, rather than changing this
configuration on the CI runners, it's worth supporting inexplicable CRLF
in these files so that anyone else cloning Zig on Windows doesn't get
unexpected test failures.
Uses of `@embedFile` register dependencies on the corresponding
`Zcu.EmbedFile`. At the start of every update, we iterate all embedded
files and update them if necessary, and invalidate the dependencies if
they changed.
In order to properly integrate with the lazy analysis model, failed
embed files are now reported by the `AnalUnit` which actually used
`@embedFile`; the filesystem error is stored in the `Zcu.EmbedFile`.
An incremental test is added covering incremental updates to embedded
files, and I have verified locally that dependency invalidation is
working correctly.
* `std.builtin.Panic` -> `std.builtin.panic`, because it is a namespace.
* `root.Panic` -> `root.panic` for the same reason. There are type
checks so that we still allow the legacy `pub fn panic` strategy in
the 0.14.0 release.
* `std.debug.SimplePanic` -> `std.debug.simple_panic`, same reason.
* `std.debug.NoPanic` -> `std.debug.no_panic`, same reason.
* `std.debug.FormattedPanic` is now a function `std.debug.FullPanic`
which takes as input a `panicFn` and returns a namespace with all the
panic functions. This handles the incredibly common case of just
wanting to override how the message is printed, whilst keeping nice
formatted panics.
* Remove `std.builtin.panic.messages`; now, every safety panic has its
own function. This reduces binary bloat, as calls to these functions
no longer need to prepare any arguments (aside from the error return
trace).
* Remove some legacy declarations, since a zig1.wasm update has
happened. Most of these were related to the panic handler, but a quick
grep for "zig1" brought up a couple more results too.
Also, add some missing type checks to Sema.
Resolves: #22584
formatted -> full
Allows the stack trace tests to be additionally compiled and run with
`.use_llvm = false, .use_lld = false` depending on the host target. This
is currently enabled for x86_64 targets emitting ELF.
Self-hosted backends emit slightly different DWARF info to the LLVM
backend, so the checking logic (and the tests themselves) had to be
tweaked slightly to support both backends at once.
It's now unnecessary to explicitly pass this, because it receives an
implicit error trace parameter anyway, so can just use
`@errorReturnTrace()`. The previous commit updated Sema to expect this
new interface. This saves an AIR instruction at all `returnError` call
sites.
This check doesn't make sense with the modern Allocator API; it's left over
from when realloc could change alignment. It's statically known (but not
comptime-known) to be true always. This check was one of the things
blocking Allocator from being used at comptime (related: #1291).
The original motivation here was to fix regressions caused by #22414.
However, while working on this, I ended up discussing a language
simplification with Andrew, which changes things a little from how they
worked before #22414.
The main user-facing change here is that any reference to a prior
function parameter, even if potentially comptime-known at the usage
site or even not analyzed, now makes a function generic. This applies
even if the parameter being referenced is not a `comptime` parameter,
since it could still be populated when performing an inline call. This
is a breaking language change.
The detection of this is done in AstGen; when evaluating a parameter
type or return type, we track whether it referenced any prior parameter,
and if so, we mark this type as being "generic" in ZIR. This will cause
Sema to not evaluate it until the time of instantiation or inline call.
A lovely consequence of this from an implementation perspective is that
it eliminates the need for most of the "generic poison" system. In
particular, `error.GenericPoison` is now completely unnecessary, because
we identify generic expressions earlier in the pipeline; this simplifies
the compiler and avoids redundant work. This also entirely eliminates
the concept of the "generic poison value". The only remnant of this
system is the "generic poison type" (`Type.generic_poison` and
`InternPool.Index.generic_poison_type`). This type is used in two
places:
* During semantic analysis, to represent an unknown result type.
* When storing generic function types, to represent a generic parameter/return type.
It's possible that these use cases should instead use `.none`, but I
leave that investigation to a future adventurer.
One last thing. Prior to #22414, inline calls were a little inefficient,
because they re-evaluated even non-generic parameter types whenever they
were called. Changing this behavior is what ultimately led to #22538.
Well, because the new logic will mark a type expression as generic if
there is any change its resolved type could differ in an inline call,
this redundant work is unnecessary! So, this is another way in which the
new design reduces redundant work and complexity.
Resolves: #22494Resolves: #22532Resolves: #22538
We can still often determine a comptime result based on the type, even
if the pointer is runtime-known.
Also, we previously used load -> is non null instead of AIR
`is_non_null_ptr` if the pointer is comptime-known, but that's a bad
heuristic. Instead, we should check for the pointer to be
comptime-known, *and* for the load to be comptime-known, and only in
that case should we call `Sema.analyzeIsNonNull`.
Resolves: #22556
`std.Build.Step.Run` makes the very reasonable assumption that
`error.InvalidExe` will be reported on `spawn` if it will happen.
However, this property does not currently hold on POSIX targets. This
is, through a slightly convoluted series of events, partially
responsible for the sporadic `BrokenPipe` errors we've been seeing more
and more in CI runs.
Making `spawn` wait for the child to exec in the POSIX path introduces
a block of up to 400us. So, instead of doing that, we add a new API for
this particular case: `waitForSpawn`. This function is a nop on Windows,
but on POSIX it blocks until the child successfully (or otherwise) calls
`execvpe`, and reports the error if necessary. `std.Build.Step.Run`
calls this function, so that it can get `error.InvalidExe` when it wants
it.
I'm not convinced that this API is optimal. However, I think this entire
API needs to be either heavily refactored or straight-up redesigned
(related: #22504), so I'm not too worried about hitting the perfect API:
I'd rather just fix this bug for now, and figure out the long-term goal
a bit later.
The previous logic here was trying to assume that custom test runners
never used `std.zig.Server` to communicate with the build runner;
however, it was flawed, because modifying the `test_runner` field on
`Step.Compile` would not update this flag. That might have been
intentional (allowing a way for the user to specify a custom test runner
which *does* use the compiler server protocol), but if so, it was a
flawed API, since it was too easy to update one field without updating
the other.
Instead, bundle these two pieces of state into a new type
`std.Build.Step.Compile.TestRunner`. When passing a custom test runner,
you are now *provided* to specify whether it is a "simple" runner, or
whether it uses the compiler server protocol.
This is a breaking change, but is unlikely to affect many people, since
custom test runners are seldom used in the wild.
For representing struct field default values and array/pointer type
sentinel values, we use `*const anyopaque`, since there is no way for
`std.builtin.Type.StructField` etc to refer back to its `type` field.
However, when introspecting a type, this is quite awkward due to the
pointer casts necessary.
As such, this commit renames the `sentinel` fields to `sentinel_ptr`,
and the `default_value` field to `default_value_ptr`, and introduces
helper methods `sentinel()` and `defaultValue()` to load the values.
These methods are marked as `inline` because their return value, which
is always comptime-known, is very often required at comptime by use
sites, so this avoids having to annotate such calls with `comptime`.
This is a breaking change, although note that 0.14.0 is already a
breaking release for all users of `std.builtin.Type` due to the union
fields being renamed.
Implementing the changes from the prior commit, to prepare for the
following commit.
This also means that zig1 now uses the new value interpret mode, so
that adding and removing fields from `std.builtin` types is easier.
Signed-off-by: mlugg <mlugg@mlugg.co.uk>
This was done by regex substitution with `sed`. I then manually went
over the entire diff and fixed any incorrect changes.
This diff also changes a lot of `callconv(.C)` to `callconv(.c)`, since
my regex happened to also trigger here. I opted to leave these changes
in, since they *are* a correct migration, even if they're not the one I
was trying to do!
This matches established naming conventions. Now is an opportune time to
make this change, since we're already performing breaking changes to
`std.builtin.Type`.
`Type.hasWellDefinedLayout` was in disagreement with pointer loading
logic about auto-layout structs with zero fields, `struct {}`. For
consistency, these types should not have a well-defined layout.
This is technically a breaking change.
`Sema.explainWhyValueContainsReferenceToComptimeVar` (concise name!)
adds notes to an error explaining how to get from a given `Value` to a
pointer to some `comptime var` (or a comptime field). Previously, this
error could be very opaque in any case where it wasn't obvious where the
comptime var pointer came from; particularly for type captures. Now, the
error notes explain this to the user.
Before the prior commit, the maximum comptime recursion depth on my
system was 4062. After the prior commit, it decreased to 2854. This
commit increases the compiler's stack size enough so that the recursion
depth limit is no less than it was before the `Sema.analyzeCall`
rewrite, preventing this from being a breaking change. Specifically,
this stack size increases my observed maximum comptime recursion depth
to 4105.
This rewrite improves some error messages, hugely simplifies the logic,
and fixes several bugs. One of these bugs is technically a new rule
which Andrew and I agreed on: if a parameter has a comptime-only type
but is not declared `comptime`, then the corresponding call argument
should not be *evaluated* at comptime; only resolved. Implementing this
required changing how function types work a little, which in turn
required allowing a new kind of function coercion for some generic use
cases: function coercions are now allowed to implicitly *remove*
`comptime` annotations from parameters with comptime-only types. This is
okay because removing the annotation affects only the call site.
Resolves: #22262
Matthew Lugg