- Move the type parameter from `encode` and `decode` methods to
the trait.
- Remove `UseSpecialized(En|De)codable` traits.
- Remove blanket impls for references.
- Add `RefDecodable` trait to allow deserializing to arena-allocated
references safely.
- Remove ability to (de)serialize HIR.
- Create proc-macros `(Ty)?(En|De)codable` to help implement these new
traits.
LLVM IR coverage encoding aligns closer to Clang's
I found some areas for improvement while attempting to debug the
SegFault issue when running rust programs compiled using MSVC, with
coverage instrumentation.
I discovered that LLVM's coverage writer was generating incomplete
function name variable names (that's not a typo: the name of the
variable that holds a function name).
The existing implementation used one-up numbers to distinguish
variables, and correcting the names did not fix the MSVC coverage bug,
but the fix in this PR makes the names and resulting LLVM IR easier to
follow and more consistent with Clang's implementation.
I also changed the way the `-Zinstrument-coverage` option is supported in
symbol_export.rs. The original implementation was incorrect, and the
corrected version matches the handling for `-Zprofile-generate`, as it
turns out.
(An argument could be made that maybe `-Zinstrument-coverage` should
automatically enable `-Cprofile-generate`. In fact, if
`-Cprofile-generate` is analagous to Clang's `-fprofile-generate`, as
some documentation implies, Clang always requires this flag for its
implementation of source-based code coverage. This would require a
little more validation, and if implemented, would probably require
updating some of the user-facing messages related to
`-Cprofile-generate` to not be so specific to the PGO use case.)
None of these changes fixed the MSVC coverage problems, but they should
still be welcome improvements.
Lastly, I added some additional FIXME comments in instrument_coverage.rs
describing issues I found with the generated LLVM IR that would be
resolved if the coverage instrumentation is injected with a `Statement`
instead of as a new `BasicBlock`. I describe seven advantages of this
change, but it requires some discussion before making a change like
this.
r? @tmandry
Constified str::from_utf8_unchecked
This would be useful for const code to use an array to construct a string using guaranteed utf8 inputs, and then create a `&str` from it.
As a relative beginner, it took a while for me to figure out I could just steal the references to avoid partially moving the child and thus retain ability to call functions on it (and store it in structs etc).
Stop BTreeMap casts from reborrowing
Down in btree/node.rs, the interface and use of `cast_unchecked` look a bit shady. It's really just there for inverting `forget_type` which does not borrow. By borrowing we can't write the same `cast_unchecked` in the same way at the Handle level.
No change in undefined behaviour or performance.
I found some areas for improvement while attempting to debug the
SegFault issue when running rust programs compiled using MSVC, with
coverage instrumentation.
I discovered that LLVM's coverage writer was generating incomplete
function name variable names (that's not a typo: the name of the
variable that holds a function name).
The existing implementation used one-up numbers to distinguish
variables, and correcting the names did not fix the MSVC coverage bug,
but the fix in this PR makes the names and resulting LLVM IR easier to
follow and more consistent with Clang's implementation.
I also changed the way the `-Zinstrument-coverage` option is supported
in symbol_export.rs. The original implementation was incorrect, and the
corrected version matches the handling for `-Zprofile-generate`, as it
turns out.
(An argument could be made that maybe `-Zinstrument-coverage` should
automatically enable `-Cprofile-generate`. In fact, if
`-Cprofile-generate` is analagous to Clang's `-fprofile-generate`, as
some documentation implies, Clang always requires this flag for its
implementation of source-based code coverage. This would require a
little more validation, and if implemented, would probably require
updating some of the user-facing messages related to
`-Cprofile-generate` to not be so specific to the PGO use case.)
None of these changes fixed the MSVC coverage problems, but they should
still be welcome improvements.
Lastly, I added some additional FIXME comments in instrument_coverage.rs
describing issues I found with the generated LLVM IR that would be
resolved if the coverage instrumentation is injected with a `Statement`
instead of as a new `BasicBlock`. I describe seven advantages of this
change, but it requires some discussion before making a change like
this.
Clean up some mir transform passes
I noticed a few places where there were intermediates being created
in MIR optimization passes, so I removed them.
I also changed some `Some(..)` into just `..` and wrap `Some(..)` at the function end, doing early returns for `None`.
I was generally looking for some easy optimizations in theses passes, and hopefully these should improve the runtime of these passes by a tinnnyyyyy bit.
r? @oli-obk
Run standard library unit tests without optimizations in `nopt` CI jobs
This was discussed in #73288 as a way to catch similar issues in the future. This builds an unoptimized standard library with the bootstrap compiler and runs the unit tests. This takes about 2 minutes on my laptop.
I confirmed that this method works locally, although there may be a better way of implementing it. It would be better to use the stage 2 compiler instead of the bootstrap one.
Notably, there are currently four `libstd` unit tests that fail in debug mode on `i686-unkown-linux-gnu` (a tier one target):
```
failures:
f32::tests::test_float_bits_conv
f32::tests::test_total_cmp
f64::tests::test_float_bits_conv
f64::tests::test_total_cmp
```
These are the tests that prompted #73288 as well as the ones added in #72568, which is currently broken due to #73328.
Emit E0288 (lifetime bound for trait object cannot be deduced) only on
bare trait objects. When the trait object is in the form of
`&dyn Trait`, E0106 (missing lifetime specifier) will have been emitted,
making the former redundant.
This also explicitly checks that the types are `bool`. `try_eval_bool` also appears to just
succeed for `u8`, so this ensures that it actually is a bool before casting.
Disable zlib in LLVM on aarch64-apple-darwin
For some reason, building rustc on the Apple Silicon DTK fails with some undefined symbols from zlib, which I guess is similar to some issues that appeared on *-apple-ios and *-apple-tvos.
Expand function pointer docs
Be more explicit in the ABI section, and add a section on how to obtain a function pointer, which can be somewhat confusing.
Cc https://github.com/rust-lang/rust/issues/75239
Change registered "program name" for -Cllvm-args usage messages
While debugging a codegen issue, I tried adding LLVM options with
the rustc -Cllvm-args option, and was confused by the error and usage
messaging.
The LLVM "program name" argument is set to "rustc", and command line
error messages make it look like invalid arguments are "rustc"
arguments, not LLVM.
I changed this argument so error messages and the "-help" usage feedback
is easier to understand and react to. (Clang does something similar.)
r? @wesleywiser
fix LocalInfo doc comment
The doc comment makes it sound like this is diagnostics-only, but that is not true -- even [unsafety checking uses this information](ded20c98be/src/librustc_mir/transform/check_unsafety.rs (L206)), so it is crucial for soundness, not just "nice to have".
Cc @oli-obk
self-profile: Cache more query key strings when doing self-profiling.
This PR adds optimized `SpecIntoSelfProfilingString` implementations for two common query key types (`LocalDefId` and `WithOptConstParam`). This makes raw self-profiling data on disk 8-9% smaller for my two test cases (`regex` and `cargo`).
The on-disk format is not affected, so no tooling updates need to happen.
I also tried adding an impl for `Ty<'tcx>` (which should reduce size quite a bit) but the compiler did not allow me to add a specialized impl parameterized with `'tcx`. I don't know if there is an actual problem with that or if the implementation of specialization just doesn't support it yet.
cc @wesleywiser @Mark-Simulacrum
Hard way to respect BTreeMap's minimum node length
Resolves#74834 the hard way (though not the hardest imaginable).
Benchmarks (which are all biased/realistic, inserting keys in ascending order) say:
```
benchcmp r0 r1 --threshold 10
name r0 ns/iter r1 ns/iter diff ns/iter diff % speedup
btree::map::clone_slim_100_and_clear 2,183 2,723 540 24.74% x 0.80
btree::map::clone_slim_100_and_drain_all 3,652 4,173 521 14.27% x 0.88
btree::map::clone_slim_100_and_drain_half 3,320 3,940 620 18.67% x 0.84
btree::map::clone_slim_100_and_into_iter 2,154 2,717 563 26.14% x 0.79
btree::map::clone_slim_100_and_pop_all 3,372 3,870 498 14.77% x 0.87
btree::map::clone_slim_100_and_remove_all 5,111 5,647 536 10.49% x 0.91
btree::map::clone_slim_100_and_remove_half 3,259 3,821 562 17.24% x 0.85
btree::map::iter_0 1,733 1,509 -224 -12.93% x 1.15
btree::map::iter_100 2,714 3,739 1,025 37.77% x 0.73
btree::map::iter_10k 3,728 4,269 541 14.51% x 0.87
btree::map::range_unbounded_unbounded 28,426 36,631 8,205 28.86% x 0.78
btree::map::range_unbounded_vs_iter 28,808 34,056 5,248 18.22% x 0.85
```
This difference is not caused by the `debug_assert`-related code in the function `splitpoint`, it's the same without.
First iteration of simplify match branches
This is a simple MIR pass that attempts to convert
```
bb0: {
StorageLive(_2);
_3 = discriminant(_1);
switchInt(move _3) -> [0isize: bb2, otherwise: bb1];
}
bb1: {
_2 = const false;
goto -> bb3;
}
bb2: {
_2 = const true;
goto -> bb3;
}
```
into
```
bb0: {
StorageLive(_2);
_3 = discriminant(_1);
_2 = _3 == 0;
goto -> bb3;
}
```
There are still missing components(like checking if the assignments are bools).
Was hoping that this could get some review though.
Handles #75141
r? @oli-obk
We already have builders which built standard library *test*s without
optimizations, but we previously did not have builders which built the standard
library itself without optimizations and then tested that.
This adds those builds for i686 and x86_64 linux.