`.loop` is also a block, so the block_depth must be stored *after* block
creation, ensuring a correct block_depth to jump back to when receiving
`.repeat`.
This also un-regresses `switch_br` which now correctly handles ranges
within cases. It supports it for both jump tables as well as regular
conditional branches.
The parse of `fn foo(a: switch (...) { ... })` was previously handled
incorrectly; `a` was treated as both the parameter name and a label.
The same issue exists for `for` and `while` expressions -- they should
be fixed too, and the grammar amended appropriately. This commit does
not do this: it only aims to avoid introducing regressions from labeled
switch syntax.
This commit introduces a new AIR instruction, `repeat`, which causes
control flow to move back to the start of a given AIR loop. `loop`
instructions will no longer automatically perform this operation after
control flow reaches the end of the body.
The motivation for making this change now was really just consistency
with the upcoming implementation of #8220: it wouldn't make sense to
have this feature work significantly differently. However, there were
already some TODOs kicking around which wanted this feature. It's useful
for two key reasons:
* It allows loops over AIR instruction bodies to loop precisely until
they reach a `noreturn` instruction. This allows for tail calling a
few things, and avoiding a range check on each iteration of a hot
path, plus gives a nice assertion that validates AIR structure a
little. This is a very minor benefit, which this commit does apply to
the LLVM and C backends.
* It should allow for more compact ZIR and AIR to be emitted by having
AstGen emit `repeat` instructions more often rather than having
`continue` statements `break` to a `block` which is *followed* by a
`repeat`. This is done in status quo because `repeat` instructions
only ever cause the direct parent block to repeat. Now that AIR is
more flexible, this flexibility can be pretty trivially extended to
ZIR, and we can then emit better ZIR. This commit does not implement
this.
Support for this feature is currently regressed on all self-hosted
native backends, including x86_64. This support will be added where
necessary before this branch is merged.
This commit modifies the representation of the AIR `switch_br`
instruction to represent ranges in cases. Previously, Sema emitted
different AIR in the case of a range, where the `else` branch of the
`switch_br` contained a simple `cond_br` for each such case which did a
simple range check (`x > a and x < b`). Not only does this add
complexity to Sema, which we would like to minimize, but it also gets in
the way of the implementation of #8220. That proposal turns certain
`switch` statements into a looping construct, and for optimization
purposes, we want to lower this to AIR fairly directly (i.e. without
involving a `loop` instruction). That means we would ideally like a
single instruction to represent the entire `switch` statement, so that
we can dispatch back to it with a different operand as in #8220. This is
not really possible to do correctly under the status quo system.
This commit implements lowering of this new `switch_br` usage in the
LLVM and C backends. The C backend just turns any case containing ranges
entirely into conditionals, as before. The LLVM backend is a little
smarter, and puts scalar items into the `switch` instruction, only using
conditionals for the range cases (which direct to the same bb). All
remaining self-hosted backends are temporarily regressed in the presence
of switch range cases. This functionality will be restored for at least
the x86_64 backend before merge.
Most of the required renames here are net wins for readaibility, I'd
say. The ones in `arch` are a little more verbose, but I think better. I
didn't bother renaming the non-conflicting functions in
`arch/arm/bits.zig` and `arch/aarch64/bits.zig`, since these backends
are pretty bit-rotted anyway AIUI.
These names aren't matching any formal specification; they're mostly
just ripped from LLVM code. Therefore, we should definitely follow Zig
naming conventions here.
Most of these changes seem like improvements. The PDB thing had a TODO
saying it used to crash; I anticipate it works now, we'll see what CI
does.
The `std.os.uefi` field renames are a notable breaking change.
These won't live in the parent namespace as decls which causes problems
later in this function, and tests are guaranteed not to be referenced at
comptime anyway, so there's actually no need to run this logic.
The compiler actually doesn't need any functional changes for this: Sema
does reification based on the tag indices of `std.builtin.Type` already!
So, no zig1.wasm update is necessary.
This change is necessary to disallow name clashes between fields and
decls on a type, which is a prerequisite of #9938.
Pointer subtraction on `void *` or function pointers is UB by the C
spec, but is permitted by GCC and Clang as an extension. So, avoid
crashing translate-c in such cases, and follow the extension behavior --
there's nothing else that could really be intended.
Implements the accepted proposal to introduce `@branchHint`. This
builtin is permitted as the first statement of a block if that block is
the direct body of any of the following:
* a function (*not* a `test`)
* either branch of an `if`
* the RHS of a `catch` or `orelse`
* a `switch` prong
* an `or` or `and` expression
It lowers to the ZIR instruction `extended(branch_hint(...))`. When Sema
encounters this instruction, it sets `sema.branch_hint` appropriately,
and `zirCondBr` etc are expected to reset this value as necessary. The
state is on `Sema` rather than `Block` to make it automatically
propagate up non-conditional blocks without special handling. If
`@panic` is reached, the branch hint is set to `.cold` if none was
already set; similarly, error branches get a hint of `.unlikely` if no
hint is explicitly provided. If a condition is comptime-known, `cold`
hints from the taken branch are allowed to propagate up, but other hints
are discarded. This is because a `likely`/`unlikely` hint just indicates
the direction this branch is likely to go, which is redundant
information when the branch is known at comptime; but `cold` hints
indicate that control flow is unlikely to ever reach this branch,
meaning if the branch is always taken from its parent, then the parent
is also unlikely to ever be reached.
This branch information is stored in AIR `cond_br` and `switch_br`. In
addition, `try` and `try_ptr` instructions have variants `try_cold` and
`try_ptr_cold` which indicate that the error case is cold (rather than
just unlikely); this is reachable through e.g. `errdefer unreachable` or
`errdefer @panic("")`.
A new API `unwrapSwitch` is introduced to `Air` to make it more
convenient to access `switch_br` instructions. In time, I plan to update
all AIR instructions to be accessed via an `unwrap` method which returns
a convenient tagged union a la `InternPool.indexToKey`.
The LLVM backend lowers branch hints for conditional branches and
switches as follows:
* If any branch is marked `unpredictable`, the instruction is marked
`!unpredictable`.
* Any branch which is marked as `cold` gets a
`llvm.assume(i1 true) [ "cold"() ]` call to mark the code path cold.
* If any branch is marked `likely` or `unlikely`, branch weight metadata
is attached with `!prof`. Likely branches get a weight of 2000, and
unlikely branches a weight of 1. In `switch` statements, un-annotated
branches get a weight of 1000 as a "middle ground" hint, since there
could be likely *and* unlikely *and* un-annotated branches.
For functions, a `cold` hint corresponds to the `cold` function
attribute, and other hints are currently ignored -- as far as I can tell
LLVM doesn't really have a way to lower them. (Ideally, we would want
the branch hint given in the function to propagate to call sites.)
The compiler and standard library do not yet use this new builtin.
Resolves: #21148
I don't recall why I put these checks here -- they aren't correct. We
can freely recreate a type even if its fields have changed, because we
are going to re-do all type resolution.
The only conditions for recreations are (a) the ZIR index must not be
lost and (b) the number of captures must be the same. These conditions
are permissible because if either is violated, we can guarantee that
analysis of a valid `zirStructDecl` (etc) will never reference this
type (since the ZIR index has just been tracked, and the captures have
just been created based on the ZIR).
Adds a corresponding test case.
Resolves: #21185
* Indices of referenced captures
* Line and column of `@src()`
The second point aligns with a reversal of the "incremental compilation"
section of https://github.com/ziglang/zig/issues/2029#issuecomment-645793168.
This reversal was already done as #17688 (46a6d50), with the idea to
push incremental compilation down the line. My proposal is to keep it as
comptime-known, and simply re-analyze uses of `@src()` whenever their
line/column change.
I think this decision is reasonable for a few reasons:
* The Zig compiler is quite fast. Occasionally re-analyzing a few
functions containing `@src()` calls is perfectly acceptable and won't
noticably impact update times.
* The system described by Andrew in #2029 is currently vaporware.
* The system described by Andrew in #2029 is non-trivial to implement.
In particular, it requires some way to have backends update a single
global in certain cases, without re-doing semantic analysis. There is
no other part of incremental compilation which requires this.
* Having `@src().line` be comptime-known is useful. For instance, #17688
was justified by broken Tracy integration because the source line
couldn't be comptime-known.
In a `memoized_call`, store how many backwards braches the call
performs. Add this to `sema.branch_count` when using a memoized call. If
this exceeds the quota, perform a non-memoized call to get a correct
"exceeded X backwards branches" error.
Also, do not memoize calls which do `@setEvalBranchQuota` or similar, as
this affects global state which must apply to the caller.
Change some eval branch quotas so that the compiler itself still builds correctly.
This commit manually changes a file in Aro which is automatically
generated. The sources which generate the file are not in this repo.
Upstream Aro should make the suitable changes on their end before the
next sync of Aro sources into the Zig repo.
This replaces the constant `Zir.Inst.Ref` tags (and the analagous tags
in `Air.Inst.Ref`, `InternPool.Index`) referring to types in
`std.builtin` with a ZIR instruction `extended(builtin_type(...))` which
instructs Sema to fetch such a type, effectively as if it were a
shorthand for the ZIR for `@import("std").builtin.xyz`.
Previously, this was achieved through constant tags in `Ref`. The
analagous `InternPool` indices began as `simple_type` values, and were
later rewritten to the correct type information. This system was kind of
brittle, and more importantly, isn't compatible with incremental
compilation of std, since incremental compilation relies on the ability
to recreate types at different indices when they change. Replacing the
old system with this instruction slightly increases the size of ZIR, but
it simplifies logic and allows incremental compilation to work correctly
on the standard library.
This shouldn't have a significant impact on ZIR size or compiler
performance, but I will take measurements in the PR to confirm this.
Without this, incremental updates which would change inferred error sets
fail, since they assume the IES is resolved and equals the old set,
resulting in false positive compile errors when e.g. coercing to an IES.
This function now has to allocate anyway to resolve references, so we
may as well just build the error bundle and check its length.
Also remove some unnecessary calls of this function for efficiency.
The old logic here had bitrotted, largely because there were some
incorrect `else` cases. This is now implemented correctly for all
current ZIR instructions. This prevents instructions being lost in
incremental updates, which is important for updates to be minimal.
Another big commit, sorry! This commit makes all fixes necessary for
incremental updates of the compiler itself (specifically, adding a
breakpoint to `zirCompileLog`) to succeed, at least on the frontend.
The biggest change here is a reform to how types are handled. It works
like this:
* When a type is first created in `zirStructDecl` etc, its namespace is
scanned. If the type requires resolution, an `interned` dependency is
declared for the containing `AnalUnit`.
* `zirThis` also declared an `interned` dependency for its `AnalUnit` on
the namespace's owner type.
* If the type's namespace changes, the surrounding source declaration
changes hash, so `zirStructDecl` etc will be hit again. We check
whether the namespace has been scanned this generation, and re-scan it
if not.
* Namespace lookups also check whether the namespace in question
requires a re-scan based on the generation. This is because there's no
guarantee that the `zirStructDecl` is re-analyzed before the namespace
lookup is re-analyzed.
* If a type's structure (essentially its fields) change, then the type's
`Cau` is considered outdated. When the type is re-analyzed due to
being outdated, or the `zirStructDecl` is re-analyzed by being
transitively outdated, or a corresponding `zirThis` is re-analyzed by
being transitively outdated, the struct type is recreated at a new
`InternPool` index. The namespace's owner is updated (but not
re-scanned, since that is handled by the mechanisms above), and the
old type, while remaining a valid `Index`, is removed from the map
metadata so it will never be found by lookups. `zirStructDecl` and
`zirThis` store an `interned` dependency on the *new* type.
When a type becomes outdated, there will still be lingering references
to the old index -- for instance, any declaration whose value was that
type holds a reference to that index. These references may live for an
arbitrarily long time in some cases. So, we can't just remove the type
from the pool -- the old `Index` must remain valid!
Instead, we want to preserve the old `Index`, but avoid it from ever
appearing in lookups. (It's okay if analysis of something referencing
the old `Index` does weird stuff -- such analysis are guaranteed by the
incremental compilation model to always be unreferenced.) So, we use the
new `InternPool.putKeyReplace` to replace the shard entry for this index
with the newly-created index.
An enum type is kind of like a struct or union type, in that field
errors are happening during type resolution. The only difference is that
type resolution happens at the time the type is created. So, errors in
fields should not cause the type to be deleted: we've already added a
reference entry, and incremenetal dependencies which must be invalidated
if the compile error is fixed. Once we call `WipEnumType.prepare`, we
should never call `WipEnumType.cancel`. This is analagous to logic for
enum declarations in `Sema.zirEnumDecl`.
Two fixes here.
* Prevent a crash when sorting the list of analysis errors when some
errors refer to lost source locations. These errors can be sorted
anywhere in the list, because they are (in theory) guaranteed to never
be emitted by the `resolveReferences` logic. This case occurs, for
instance, when a declaration has compile errors in the initial update
and is deleted in the second update.
* Prevent a crash when resolving the source location for `entire_file`
errors for a non-existent file. This is the bug underlying #20954.
Resolves: #20954.
This commit updates `Zcu.resolveReferences` to traverse the graph of
`AnalUnit` references (starting from the 1-3 roots of analysis) in order
to determine which `AnalUnit`s are referenced in an update. Errors for
unreferenced entities are omitted from the error bundle. However, note
that unreferenced `Nav`s are not removed from the binary.
This commit makes more progress towards incremental compilation, fixing
some crashes in the frontend. Notably, it fixes the regressions introduced
by #20964. It also cleans up the "outdated file root" mechanism, by
virtue of deleting it: we now detect outdated file roots just after
updating ZIR refs, and re-scan their namespaces.
Eliding the namespace when a container type has no decls was an
experiment in saving memory, but it ended up causing more trouble than
it was worth in various places. So, take the small memory hit for
reified types, and just give every container type a namespace.
The type `Zcu.Decl` in the compiler is problematic: over time it has
gained many responsibilities. Every source declaration, container type,
generic instantiation, and `@extern` has a `Decl`. The functions of
these `Decl`s are in some cases entirely disjoint.
After careful analysis, I determined that the two main responsibilities
of `Decl` are as follows:
* A `Decl` acts as the "subject" of semantic analysis at comptime. A
single unit of analysis is either a runtime function body, or a
`Decl`. It registers incremental dependencies, tracks analysis errors,
etc.
* A `Decl` acts as a "global variable": a pointer to it is consistent,
and it may be lowered to a specific symbol by the codegen backend.
This commit eliminates `Decl` and introduces new types to model these
responsibilities: `Cau` (Comptime Analysis Unit) and `Nav` (Named
Addressable Value).
Every source declaration, and every container type requiring resolution
(so *not* including `opaque`), has a `Cau`. For a source declaration,
this `Cau` performs the resolution of its value. (When #131 is
implemented, it is unsolved whether type and value resolution will share
a `Cau` or have two distinct `Cau`s.) For a type, this `Cau` is the
context in which type resolution occurs.
Every non-`comptime` source declaration, every generic instantiation,
and every distinct `extern` has a `Nav`. These are sent to codegen/link:
the backends by definition do not care about `Cau`s.
This commit has some minor technically-breaking changes surrounding
`usingnamespace`. I don't think they'll impact anyone, since the changes
are fixes around semantics which were previously inconsistent (the
behavior changed depending on hashmap iteration order!).
Aside from that, this changeset has no significant user-facing changes.
Instead, it is an internal refactor which makes it easier to correctly
model the responsibilities of different objects, particularly regarding
incremental compilation. The performance impact should be negligible,
but I will take measurements before merging this work into `master`.
Co-authored-by: Jacob Young <jacobly0@users.noreply.github.com>
Co-authored-by: Jakub Konka <kubkon@jakubkonka.com>
This eliminates the statically-reachable recursion loop between code
generation backends and Sema. This is beneficial for optimizers
(although I do not measure any performance improvement for this change),
and for profilers.
Note that the `_ = Address` statements in tests previously were a nop,
and now actually check that the type is valid. However, on WASI, the
type is *not* valid.
I'm so sorry.
This commit was just meant to be making all types fully resolve by
queueing resolution at the moment of their creation. Unfortunately, a
lot of dominoes ended up falling. Here's what happened:
* I added a work queue job to fully resolve a type.
* I realised that from here we could eliminate `Sema.types_to_resolve`
if we made function codegen a separate job. This is desirable for
simplicity of both spec and implementation.
* This led to a new AIR traversal to detect whether any required type is
unresolved. If a type in the AIR failed to resolve, then we can't run
codegen.
* Because full type resolution now occurs by the work queue job, a bug
was exposed whereby error messages for type resolution were associated
with the wrong `Decl`, resulting in duplicate error messages when the
type was also resolved "by" its owner `Decl` (which really *all*
resolution should be done on).
* A correct fix for this requires using a different `Sema` when
performing type resolution: we need a `Sema` owned by the type. Also
note that this fix is necessary for incremental compilation.
* This means a whole bunch of functions no longer need to take `Sema`s.
* First-order effects: `resolveTypeFields`, `resolveTypeLayout`, etc
* Second-order effects: `Type.abiAlignmentAdvanced`, `Value.orderAgainstZeroAdvanced`, etc
The end result of this is, in short, a more correct compiler and a
simpler language specification. This regressed a few error notes in the
test cases, but nothing that seems worth blocking this change.
Oh, also, I ripped out the old code in `test/src/Cases.zig` which
introduced a dependency on `Compilation`. This dependency was
problematic at best, and this code has been unused for a while. When we
re-enable incremental test cases, we must rewrite their executor to use
the compiler server protocol.
This change modifies `Zcu.ErrorMsg` to store a `Zcu.LazySrcLoc` rather
than a `Zcu.SrcLoc`. Everything else is dominoes.
The reason for this change is incremental compilation. If a failed
`AnalUnit` is up-to-date on an update, we want to re-use the old error
messages. However, the file containing the error location may have been
modified, and `SrcLoc` cannot survive such a modification. `LazySrcLoc`
is designed to be correct across incremental updates. Therefore, we
defer source location resolution until `Compilation` gathers the compile
errors into the `ErrorBundle`.
Previously, `reference_table` mapped from a `Decl` being referenced to
the `Decl` that performed the reference. This is convenient for
constructing error messages, but problematic for incremental
compilation. This is because on an incremental update, we want to
efficiently remove all references triggered by an `AnalUnit` which is
being re-analyzed.
For this reason, `reference_table` now maps the other way: from the
`AnalUnit` *performing* the reference, to the `AnalUnit` whose analysis
was triggered. As a general rule, any call to any of the following
functions should be preceded by a call to `Sema.addReferenceEntry`:
* `Zcu.ensureDeclAnalyzed`
* `Sema.ensureDeclAnalyzed`
* `Zcu.ensureFuncBodyAnalyzed`
* `Zcu.ensureFuncBodyAnalysisQueued`
This is not just important for error messages, but also more
fundamentally for incremental compilation. When an incremental update
occurs, we must determine whether any `AnalUnit` has become
unreferenced: in this case, we should ignore its associated error
messages, and perhaps even remove it from the binary. For this reason,
we no longer store only one reference to every `AnalUnit`, but every
reference. At the end of an update, `Zcu.resolveReferences` will
construct the reverse mapping, and as such identify which `AnalUnit`s
are still referenced. The current implementation doesn't quite do what
we need for incremental compilation here, but the framework is in place.
Note that `Zcu.resolveReferences` does constitute a non-trivial amount
of work on every incremental update. However, for incremental
compilation, this work -- which will effectively be a graph traversal
over all `AnalUnit` references -- seems strictly necessary. At the
moment, this work is only done if the `Zcu` has any errors, when
collecting them into the final `ErrorBundle`.
An unsolved problem here is how to represent inline function calls in
the reference trace. If `foo` performs an inline call to `bar` which
references `qux`, then ideally, `bar` would be shown on the reference
trace between `foo` and `qux`, but this is not currently the case. The
solution here is probably for `Zcu.Reference` to store information about
the source locations of active inline calls betweeen the referencer and
its reference.
Matthew Lugg