$ the-wire · showcase
Polonius constraint passes get faster, RawVec drops Unique
By RepoJournal · Filed · About Rust · Composed from the cited sources · methodology
A quiet day across rust-lang: perf work lands in the compiler and Miri's shim layer, plus a rustfmt toolchain bump and routine crates.io dependency churn.
A couple polonius constraints perf improvements rust-lang/rust
jackh726's pair of polonius constraint changes cut time spent in traversal-heavy cases, independent of the lazy liveness work already in flight. The PR author notes the two changes conflict textually but are conceptually independent.
Auto merge of #162804 - maxdexh:nuke-unique-in-vec, r=hanna-kruppe rust-lang/rust
maxdexh's removal of `Unique` from `RawVec` merged via bors after review by hanna-kruppe, following the t-libs discussion on whether `Unique` can be nuked. It touches allocation code that is sensitive to build times and debug mode, so the PR flagged for a perf run.
Run more tests natively rust-lang/miri
Running more of Miri's test suite natively surfaced a segfault in the pthread_set/getname_np tests: bad `pthread_t` values do not always return an error as previously assumed. The PR text states plainly that "previously we thought they'd return an error when given a bad `pthread_t`, turns out they don't always do that."
windows: reject some FILE_FLAG_OPEN_REPARSE_POINT uses we do not support rust-lang/miri
Miri's Windows shims previously ignored `FILE_FLAG_OPEN_REPARSE_POINT` in most paths; the flag uses it does not support are now rejected rather than silently accepted.
chore: bump rustfmt toolchain to nightly-2026-09-21 rust-lang/rustfmt
rustfmt's subtree push moved its toolchain from 1.100.0-nightly (bff8e12ff 2026-08-26) to 1.100.0-nightly (bba531001 2026-09-20). The rest of the day was routine: crates.io-index crate bumps such as `csusage` and `bevy_cef_render_process`, and Renovate dependency updates on crates.io including @percy/cli 1.32.10 and @pierre/diffs 1.4.3.
