$ the-wire · showcase
cc-rs stops forwarding profile flags, miri gains utimensat
By RepoJournal · Filed · About Rust · Composed from the cited sources · methodology
12 people shipped this
malezjaa
@malezjaa
1 cited
Trevor Gross
@tgross35
1 cited
cvijdea-bd
@cvijdea-bd
1 cited
bhaskargurram-ai
@bhaskargurram-ai
1 cited
aliheidary1381
@aliheidary1381
1 cited
nihalpasham
@nihalpasham
1 cited
Hamdan-Khan
@Hamdan-Khan
1 cited
programskillforverification
@programskillforverification
1 cited
bors
@bors
NobodyXu
@NobodyXu
jhpratt
@jhpratt
RalfJung
@RalfJung
cc-rs stopped passing rustc's profiler flags to the C compiler after the LLVM-version mismatch broke mixing rustc and clang instrumentation, while miri closed out its filesystem gap for fs::set_times.
fix: stop forwarding -Cprofile-generate/-Cprofile-use to the C compiler rust-lang/cc-rs
Since that change, -Cprofile-generate=<dir> and -Cprofile-use=<path> from CARGO_ENCODED_RUSTFLAGS were forwarded to clang as -fprofile-generate / -fprofile-use on the assumption that both use LLVM; that only holds when the C compiler's LLVM matches the one bundled with rustc, since the instrprof runtime and raw/indexed profile formats are tied to the LLVM version. The example that broke it: Rus...
implement utimensat to support fs::set_times and fs::set_times_nofollow rust-lang/miri
A new utimensat shim makes fs::set_times and fs::set_times_nofollow work under miri on Linux, FreeBSD, Solaris and illumos; futimens keeps its old behavior, with the times parsing factored out and shared. Unsupported cases (a path relative to a dirfd other than AT_FDCWD, a NULL path, flags beyond AT_SYMLINK_NOFOLLOW) raise "unsupported operation" rather than silently doing the wrong thing.
Skip runtime fma detection when enabled at compile time on x86_64 rust-lang/compiler-builtins
On x86, fma/fmaf previously always went through the cached function pointer from select_once!, even when fma or fma4 was enabled at compile time, which blocked inlining. With -C target-cpu=x86-64-v3 the call compiled to a jmp into libm followed by an indirect jmp; it now compiles to a single vfmadd213ss, and builds without those features are unchanged.
fix: do not apply target crt-static when target is overridden rust-lang/cc-rs
Cargo sets CARGO_CFG_* variables strictly for the build's primary target, so a build script compiling a host-side code generator via cc::Build::new().target(&env::var("HOST")) was still picking up the primary target's crt-static setting because add_default_flags inspected the ambient CARGO_CFG_TARGET_FEATURE. Overridden-target builds no longer get that flag applied.
Add exp2m1 and exp2m1f rust-lang/compiler-builtins
exp2m1 and exp2m1f were ported by hand from CORE-MATH as groundwork for rust-lang/rust#155489, with tests and literal arrays done with LLM help; malezjaa is asking tgross35 for review. The earlier commit adding exp2m1 from CORE-math does the same for the f64 path.
Expose regular argument attributes in rustc_public rust-lang/rust
ArgAttributes::regular() and ArgAttributeFlags expose the argument flags rustc already computes to rustc_public consumers, preserving all ten including the overlapping capture flags, and the docs now spell out what pointee_size guarantees at function entry and return and how no_free affects LLVM dereferenceability attributes.
fix misleading diagnostics and improve diagnostics of return type mismatch rust-lang/rust
A return-position check stops the compiler from suggesting you add a return and semicolon to an expression already in return position, and a de-duplicated suggest_missing_break_or_return_expr ends the duplicate E277 emission that came from a failed normalization.
test: Remove `#![feature(optimize_attribute)]` rust-lang/compiler-builtins
compiler-builtins dropped #![feature(optimize_attribute)] from its tests now that the attribute is stable in recent nightlies, and the rust-lang/rust rollup carried the test-float-parse table sharing, trait-span diagnostics, and debuginfo changes alongside it.
Action items
