[RFC] Support `.comment` section like GCC/Clang (`!llvm.ident`)
Both GCC and Clang write by default a `.comment` section with compiler information:
```txt
$ gcc -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] GCC: (GNU) 11.2.0
$ clang -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] clang version 14.0.1 (https://github.com/llvm/llvm-project.git c62053979489ccb002efe411c3af059addcb5d7d)
```
They also implement the `-Qn` flag to avoid doing so:
```txt
$ gcc -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
$ clang -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
```
So far, `rustc` only does it for WebAssembly targets and only when debug info is enabled:
```txt
$ echo 'fn main(){}' | rustc --target=wasm32-unknown-unknown --emit=llvm-ir -Cdebuginfo=2 - && grep llvm.ident rust_out.ll
!llvm.ident = !{!27}
```
The RFC part of this PR is about which behavior should `rustc` follow:
- Always add it.
- Add it by default, i.e. have an opt-out flag (GCC, Clang).
- Have an opt-in flag.
- Never add it (current).
There is also the question of whether debug info being enabled matters for that decision, given the current behavior of WebAssembly targets.
For instance, adding it by default gets us closer to other popular compilers, but that may surprise some users with an information leak. The most conservative option is to only do so opt-in, even if debug info is enabled (some users may be stripping debug info and not expecting something else to be leaked elsewhere).
Implementation-wise, this covers both `ModuleLlvm::new()` and `ModuleLlvm::new_metadata()` cases by moving the addition to `context::create_module` and adds a few test cases.
ThinLTO also sees the `llvm.ident` named metadata duplicated (in temporary outputs), so this deduplicates it like it is done for `wasm.custom_sections`. The tests also check this duplication does not take place.
Both GCC and Clang write by default a `.comment` section with compiler
information:
```txt
$ gcc -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] GCC: (GNU) 11.2.0
$ clang -c -xc /dev/null && readelf -p '.comment' null.o
String dump of section '.comment':
[ 1] clang version 14.0.1 (https://github.com/llvm/llvm-project.git c62053979489ccb002efe411c3af059addcb5d7d)
```
They also implement the `-Qn` flag to avoid doing so:
```txt
$ gcc -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
$ clang -Qn -c -xc /dev/null && readelf -p '.comment' null.o
readelf: Warning: Section '.comment' was not dumped because it does not exist!
```
So far, `rustc` only does it for WebAssembly targets and only
when debug info is enabled:
```txt
$ echo 'fn main(){}' | rustc --target=wasm32-unknown-unknown --emit=llvm-ir -Cdebuginfo=2 - && grep llvm.ident rust_out.ll
!llvm.ident = !{!27}
```
In the RFC part of this PR it was decided to always add
the information, which gets us closer to other popular compilers.
An opt-out flag like GCC and Clang may be added later on if deemed
necessary.
Implementation-wise, this covers both `ModuleLlvm::new()` and
`ModuleLlvm::new_metadata()` cases by moving the addition to
`context::create_module` and adds a few test cases.
ThinLTO also sees the `llvm.ident` named metadata duplicated (in
temporary outputs), so this deduplicates it like it is done for
`wasm.custom_sections`. The tests also check this duplication does
not take place.
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
clarify MIR uninit vs LLVM undef/poison
In [this LLVM discussion](https://discourse.llvm.org/t/rfc-load-instruction-uninitialized-memory-semantics/67481) I learned that mapping our uninitialized memory in MIR to poison in LLVM would be quite problematic due to the lack of a byte type. I am not sure where to write down this insight but this seems like a reasonable start.
Rollup of 4 pull requests
Successful merges:
- #113887 (new solver: add a separate cache for coherence)
- #113910 (Add FnPtr ty to SMIR)
- #113913 (error/E0691: include alignment in error message)
- #113914 (rustc_target: drop duplicate code)
r? `@ghost`
`@rustbot` modify labels: rollup
make full field retagging the default
The 'scalar' field retagging mode is clearly a hack -- it mirrors details of the codegen backend and how various structs are represented in LLVM. This means whether code has UB or not depends on surprising aspects, such as whether a struct has 2 or 3 (non-zero-sized) fields. Now that both hashbrown and scopeguard have released fixes to be compatible with field retagging, I think it is time to enable full field retagging by default.
`@saethlin` do you have an idea of how much fallout enabling full field retagging by default will cause? Do you have objections to enabling it by default?
Fixes https://github.com/rust-lang/miri/issues/2528
rustc_target: drop duplicate code
Drop duplicate helper methods on `Layout`, which are already implemented on `LayoutS`. Note that `Layout` has a `Deref` implementation to `LayoutS`, so all accessors are automatically redirected.
The methods are identical and have been copied to `rustc_abi` in:
commit 390a637e29
Author: hamidreza kalbasi <hamidrezakalbasi@protonmail.com>
Date: Mon Nov 7 00:36:11 2022 +0330
move things from rustc_target::abi to rustc_abi
This commit left behind the original implementation. Drop it now.
(originally moved by ``@HKalbasi)``
error/E0691: include alignment in error message
Include the computed alignment of the violating field when rejecting transparent types with non-trivially aligned ZSTs.
ZST member fields in transparent types must have an alignment of 1 (to ensure it does not raise the layout requirements of the transparent field). The current error message looks like this:
```text
LL | struct Foobar(u32, [u32; 0]);
| ^^^^^^^^ has alignment larger than 1
```
This patch changes the report to include the alignment of the violating field:
```text
LL | struct Foobar(u32, [u32; 0]);
| ^^^^^^^^ has alignment of 4, which is larger than 1
```
In case of unknown alignments, it will yield:
```text
LL | struct Foobar(u32, [u32; 0]);
| ^^^^^^^^ may have alignment larger than 1
```
This allows developers to get a better grasp why a specific field is rejected. Knowing the alignment of the violating field makes it easier to judge where that alignment-requirement originates, and thus hopefully provide better hints on how to mitigate the problem.
This idea was proposed in 2022 in #98071 as part of a bigger change. This commit simply extracts this error-message change, to decouple it from the other diagnostic improvements.
(Originally proposed by `@compiler-errors` in #98071)
Prototype: Add unstable `-Z reference-niches` option
MCP: rust-lang/compiler-team#641
Relevant RFC: rust-lang/rfcs#3204
This prototype adds a new `-Z reference-niches` option, controlling the range of valid bit-patterns for reference types (`&T` and `&mut T`), thereby enabling new enum niching opportunities. Like `-Z randomize-layout`, this setting is crate-local; as such, references to built-in types (primitives, tuples, ...) are not affected.
The possible settings are (here, `MAX` denotes the all-1 bit-pattern):
| `-Z reference-niches=` | Valid range |
|:---:|:---:|
| `null` (the default) | `1..=MAX` |
| `size` | `1..=(MAX- size)` |
| `align` | `align..=MAX.align_down_to(align)` |
| `size,align` | `align..=(MAX-size).align_down_to(align)` |
------
This is very WIP, and I'm not sure the approach I've taken here is the best one, but stage 1 tests pass locally; I believe this is in a good enough state to unleash this upon unsuspecting 3rd-party code, and see what breaks.
style-guide: Some cleanups from the fmt-rfcs repo history
There were a few commits that seem to have gone missing between the fmt-rfcs repo and the style guide. Re-apply those commits to the version of the style-guide in `rust-lang/rust`.
All of these are cleanups that don't change meaning.
Still more complexity, but this allows computing exact `NaiveLayout`s
for null-optimized enums, and thus allows calls like
`transmute::<Option<&T>, &U>()` to work in generic contexts.
avoid clone path prefix when lowering to hir
Found this while trying to parallelize `lower_to_hir`.
When lowering to hir, `Nested` paths in `ast` will be split and the prefix segments will be cloned. This could be omited, since the only consequence is that the prefix segments in `Path`s in hir will have the same `HirId`s, and it seems harmless.
This simplifies the process of lowering to hir and avoids re-modification of `ResolverAstLowering`.
r? `@Aaron1011`
cc #99292
Substitute types before checking inlining compatibility.
Addresses https://github.com/rust-lang/rust/issues/112332 and https://github.com/rust-lang/rust/issues/113781
I don't have a minimal test, but I this seems to remove the ICE locally.
This whole pre-inlining validation mirrors the "real" MIR validation pass to verify that inlined MIR will still pass validation.
The debuginfo loop is added because MIR validation check projections in debuginfo.
Likewise, MIR validation only checks `is_subtype`, so there is no reason for a stronger check.
The types were not being substituted in `check_equal`, so we were not bailing out of inlining if the substituted MIR callee body would not pass validation.
Include the computed alignment of the violating field when rejecting
transparent types with non-trivially aligned ZSTs.
ZST member fields in transparent types must have an alignment of 1 (to
ensure it does not raise the layout requirements of the transparent
field). The current error message looks like this:
LL | struct Foobar(u32, [u32; 0]);
| ^^^^^^^^ has alignment larger than 1
This patch changes the report to include the alignment of the violating
field:
LL | struct Foobar(u32, [u32; 0]);
| ^^^^^^^^ has alignment of 4, which is larger than 1
In case of unknown alignments, it will yield:
LL | struct Foobar<T>(u32, [T; 0]);
| ^^^^^^ may have alignment larger than 1
This allows developers to get a better grasp why a specific field is
rejected. Knowing the alignment of the violating field makes it easier
to judge where that alignment-requirement originates, and thus hopefully
provide better hints on how to mitigate the problem.
This idea was proposed in 2022 in #98071 as part of a bigger change.
This commit simply extracts this error-message change, to decouple it
from the other diagnostic improvements.
Drop duplicate helper methods on `Layout`, which are already implemented
on `LayoutS`. Note that `Layout` has a `Deref` implementation to
`LayoutS`, so all accessors are automatically redirected.
The methods are identical and have been copied to `rustc_abi` in:
commit 390a637e29
Author: hamidreza kalbasi <hamidrezakalbasi@protonmail.com>
Date: Mon Nov 7 00:36:11 2022 +0330
move things from rustc_target::abi to rustc_abi
This commit left behind the original implementation. Drop it now.
Signed-off-by: David Rheinsberg <david@readahead.eu>
Use SHA256 source file checksums by default when targeting MSVC
Currently, when targeting Windows (more specifically, the MSVC toolchain), Rust will use SHA1 source file checksums by default. SHA1 has been superseded by SHA256, and Microsoft recommends migrating to SHA256.
As of Visual Studio 2022, MSVC defaults to SHA256. This change aligns Rust and MSVC.
LLVM can already use SHA256 checksums, so this does not require any change to LLVM.
MSVC docs on source file checksums: https://learn.microsoft.com/en-us/cpp/build/reference/zh?view=msvc-170
Minor improvements to Windows TLS dtors
This does a few things:
* Moves keyless dtors into the same module as the `on_tls_callback` function because of dylib mess. We keep the `inline(never)` hints as a precaution (see also the issue they link to).
* Introduces the `HAS_DTORS` atomic as an optimization hint. This allows removing (most) of the TLS dtor code if no dtors are ever run. Otherwise it's always included because of a `#[used]`.
* Only run either keyed dtors or keyless dtors but not both. They should be mutually exclusive as keyed dtors are a fallback. I've also added an `assert` to make sure this is true.
Support `--print KIND=PATH` command line syntax
As is already done for `--emit KIND=PATH` and `-L KIND=PATH`.
In the discussion of #110785, it was pointed out that `--print KIND=PATH` is nicer than trying to apply the single global `-o` path to `--print`'s output, because in general there can be multiple print requests within a single rustc invocation, and anyway `-o` would already be used for a different meaning in the case of `link-args` and `native-static-libs`.
I am interested in using `--print cfg=PATH` in Buck2. Currently Buck2 works around the lack of support for `--print KIND=PATH` by [indirecting through a Python wrapper script](d43cf3a51a/prelude/rust/tools/get_rustc_cfg.py) to redirect rustc's stdout into the location dictated by the build system.
From skimming Cargo's usages of `--print`, it definitely seems like it would benefit from `--print KIND=PATH` too. Currently it is working around the lack of this by inserting `--crate-name=___ --print=crate-name` so that it can look for a line containing `___` as a delimiter between the 2 other `--print` informations it actually cares about. This is commented as a "HACK" and "abuse". 31eda6f7c3/src/cargo/core/compiler/build_context/target_info.rs (L242) (FYI `@weihanglo` as you dealt with this recently in https://github.com/rust-lang/cargo/pull/11633.)
Mentioning reviewers active in #110785: `@fee1-dead` `@jyn514` `@bjorn3`
Resurrect: rustc_llvm: Add a -Z `print-codegen-stats` option to expose LLVM statistics.
This resurrects PR https://github.com/rust-lang/rust/pull/104000, which has sat idle for a while. And I want to see the effect of stack-move optimizations on LLVM (like https://reviews.llvm.org/D153453) :).
I have applied the changes requested by `@oli-obk` and `@nagisa` https://github.com/rust-lang/rust/pull/104000#discussion_r1014625377 and https://github.com/rust-lang/rust/pull/104000#discussion_r1014642482 in the latest commits.
r? `@oli-obk`
-----
LLVM has a neat [statistics](https://llvm.org/docs/ProgrammersManual.html#the-statistic-class-stats-option) feature that tracks how often optimizations kick in. It's very handy for optimization work. Since we expose the LLVM pass timings, I thought it made sense to expose the LLVM statistics too.
-----
(Edit: fix broken link
(Edit2: fix segmentation fault and use malloc
If `rustc` is built with
```toml
[llvm]
assertions = true
```
Then you can see like
```
rustc +stage1 -Z print-codegen-stats -C opt-level=3 tmp.rs
===-------------------------------------------------------------------------===
... Statistics Collected ...
===-------------------------------------------------------------------------===
3 aa - Number of MayAlias results
193 aa - Number of MustAlias results
531 aa - Number of NoAlias results
...
```
And the current default build emits only
```
$ rustc +stage1 -Z print-codegen-stats -C opt-level=3 tmp.rs
===-------------------------------------------------------------------------===
... Statistics Collected ...
===-------------------------------------------------------------------------===
$
```
This might be better to emit the message to tell assertion flag necessity, but now I can't find how to do that...
style-guide: clean up "must"/"should"/"may"
Avoid using "should" or "may" for required parts of the default style.
The style guide inconsistently used language like "there should be a space" or
"it should be on its own line", or "may be written on a single line", for
things that are required components of the default Rust style. "should" and
especially "may" come across as optional. While the style guide overall now has
a statement at the top that the default style itself is a *recommendation*, the
*definition* of the default style should not be ambiguous about what's part of
the default style.
Rewrite language in the style guide to only use "should" and "may" and similar
for truly optional components of the style (e.g. things a tool cannot or should
not enforce in its default configuration).
In their place, either use "must", or rewrite in imperative style ("put a
space", "start it on the same line"). The latter also substantially reduces the
use of passive voice.
Looking for "should"s also flagged some recommendations the style guide made
for configurability of tools (e.g. a tool "should" have a given configuration
option). I've removed those recommendations, per discussion with the style
team; it's not the domain of the style guide to make such recommendations, only
to define the default Rust style.
In the process of making this change, I also fixed a typo, fixed a text structure
issue, fixed an example that didn't match the Rust style (missing a trailing
comma), and added an additional example for clarity. (Those changes would have
conflicted with this one.) Those changes appear in separate commits.
These are all purely editorial changes, and do not affect the semantic
definition of the Rust style.
std: remove an allocation in `Path::with_extension`
`Path::with_extension` used to reallocate (and copy) paths twice per call, now it does it once, by checking the size of the previous and new extensions it's possible to call `PathBuf::with_capacity` and pass the exact capacity required.
This also reduces the memory consumption of the path returned from `Path::with_extension` by using exact capacity instead of using amortized exponential growth.
remove the unstable `core::sync::atomic::ATOMIC_*_INIT` constants
Tracking issue: #99069
It would be weird to ever stabilise these as they are already deprecated.
THis significantly complicates `NaiveLayout` logic, but is necessary to
ensure that bounds like `NonNull<T>: PointerLike` hold in generic
contexts.
Also implement exact layout computation for structs.
