Conversation
Rust 1.99 broke CI on `main`: `Clippy`, `Documentation`, all three `Build`/`Test` matrices, the three storage-filesystem jobs and the WebRTC devnet job fail, because the workflow sets `RUSTFLAGS: -D warnings` and the new toolchain reports two new lint families here. The same tree passed every job on 1.98.1, and PR #244 is blocked by it rather than by anything in that PR. `Atomic::fetch_update` is deprecated, renamed `try_update` for consistency with the new infallible `update`. It is the same method under a new name, but only stable from 1.95, so the rename takes the MSRV with it. `assert_is_empty` is a new pedantic lint, picked up through the blanket `pedantic = "warn"`. Its point is that a bare `assert!` prints nothing useful on failure; clippy's suggestion is of the form `assert_eq!(dirs, [] as [String; 0])`, so these use a message carrying the value instead, which gives the same diagnostic and reads better in a test. - rename the two `fetch_update` calls in `web_rtc` - raise `rust-version` to 1.95 and the MSRV job to 1.95.0, and update the MSRV quoted in `README.md` and `docs/WEBRTC_DIRECT_TESTNET.md` - give the 37 flagged assertions a failure message, binding a local first where the subject was a method call so the value can be printed All assertion conditions are unchanged, so every test passes or fails exactly as before. Verified on 1.99.0: `cargo fmt --all -- --check`, `cargo clippy --all-targets --all-features --keep-going -- -D warnings` and `RUSTDOCFLAGS="-D warnings" cargo doc --all-features --no-deps`, all clean. The sweep used `--keep-going` deliberately: clippy aborts at the first failing target, so the lints hid behind one another across `src`, `tests/e2e` and the devnet tests. Closes V2-1399 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
dirvine
left a comment
dirvine
left a comment
There was a problem hiding this comment.
Approved after focused review of head 6f73aa3336b762302353efc67f059b8d48a8d6be. No blocking findings.
Reviewed all assertion changes: predicates remain unchanged and newly bound method results are evaluated once. Both WebRTC atomic call sites retain their closures and memory orderings. Rust 1.99.0 core source confirms fetch_update delegates directly to try_update, stable since 1.95.0. Manifest, MSRV CI and documented toolchain floor agree. Rust 1.95 MSRV and Clippy checks pass. Consumer-visible caveat: source builds now require Rust 1.95; preserve the stated release-versioning treatment.
Review scope: full PR diff and relevant source/API semantics, live CI and review-thread inspection, plus one lightweight GLM-5.2 independent cross-check (no full panel). No local crate build/test run; execution evidence is GitHub CI.
Still pending at approval: Test (ubuntu-latest), Test (macos-latest), Test (windows-latest), Storage on ext4, Storage on xfs, Storage on btrfs, Build (ubuntu-latest), Build (macos-latest), Build (windows-latest), WebRTC Direct five-node devnet. Code approval does not waive these checks or authorise merging before the normal CI gates are satisfied.
2 participants
Rust 1.99 (released 2026-09-28) broke CI on
main.Clippy,Documentation, all threeBuild/Testmatrices, the three storage-filesystem jobs and the WebRTC devnet job fail, becauseci.ymlsetsRUSTFLAGS: -D warningsand the new toolchain reports two new lint families in thistree. The same tree passed every job on 1.98.1 (the Clippy run on #237, merged 2026-09-28). This is
also what fails #244 — those failures are pre-existing on
main, not caused by that PR, and thisunblocks it.
Atomic::fetch_updateis deprecated, renamedtry_updatefor consistency with the new infallibleupdate. It is the same method under a new name, but only stable from 1.95, so the rename takesthe MSRV with it.
assert_is_emptyis a new pedantic lint, picked up through this crate's blanketpedantic = { level = "warn" }. Its point is that a bareassert!prints nothing useful onfailure; clippy's own suggestion is of the form
assert_eq!(dirs, [] as [String; 0]), so these usea message carrying the value instead — same diagnostic, better reading in a test.
fetch_updatecalls inweb_rtcrust-versionto 1.95 and the MSRV job to 1.95.0, and update the MSRV quoted inREADME.mdanddocs/WEBRTC_DIRECT_TESTNET.mdmethod call so the value can be printed
One of four PRs for V2-1399; the others are in
saorsa-transport(the same deprecation, same MSRVbump),
saorsa-coreandant-protocol.ant-clientandevmlibwere checked and need nothing.Linear issue
Closes V2-1399
Risk tier
Proposing T0: the only non-test source change is the
try_updaterename, which is the samemethod, signature and memory orderings. Every assertion condition is unchanged — only the failure
message is new. If the MSRV bump counts as consumer-facing rather than tooling, this is T1.
Compatibility
Semver impact
The lint fixes alone would be a patch, but raising
rust-versionis visible to consumers — anyoneon 1.91–1.94 can no longer build the crate — so it wants a minor bump.
Test evidence
T0: repo CI only. All runs on 1.99.0 (b940084d7, 2026-09-28), the toolchain that produced the
failures:
cargo fmt --all -- --check— cleancargo clippy --all-targets --all-features --keep-going -- -D warnings— cleanRUSTDOCFLAGS="-D warnings" cargo doc --all-features --no-deps— clean (theDocumentationjob)--keep-goingwas load-bearing for the sweep: clippy stops at the first failing target, so thelints hid behind one another. The 37 sites came out in three passes — 27 in the lib, then 8 across
tests/e2eandtests/webrtc_direct_devnet.rs, then 2 intests/poc_bootstrap_stall.rs— and the--keep-goingrun is what confirms none are left.Test suites were not run locally; CI runs them on this PR. The jobs to watch are the ones that were
failing on
main, plusRust 1.95 MSRV, which this PR re-points.New dependency
none
ADR
n/a
Mitigation / rollback
Revert the commit. The two halves are independent if only one turns out to be unwanted: the
deprecation can instead be silenced with
#[allow(deprecated)]at the two call sites, keeping the1.91 floor, and the lint can instead be turned off with
assert_is_empty = "allow"under[lints.clippy].🤖 Generated with Claude Code