Skip to content

ci(toolchain-sync): gate nightly bump on clippy, drop redundant cargo check #324

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
githubrobbi merged 1 commit into from
Jun 1, 2026
Merged

ci(toolchain-sync): gate nightly bump on clippy, drop redundant cargo check #324

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
93 changes: 68 additions & 25 deletions just/shared.just
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,9 +103,24 @@ toolchain-ensure:
# short-circuits straight to step-back because it is not transient.
# Post-install health probe (`rustc +CANDIDATE --version`) catches the
# rare "install reported success but toolchain is corrupt" case.
# 3. If install succeeds, run `cargo check --workspace --all-targets` with
# `+TOOLCHAIN` override (so rust-toolchain.toml stays untouched during the
# probe). If check fails — step back one day and retry.
# 3. Run clippy ONCE per feature set with the SAME strict flag stack as the
# PR merge gate (`--workspace --all-targets --locked --no-deps -- -D
# warnings`) under the `+TOOLCHAIN` override (so rust-toolchain.toml stays
# untouched during the probe). Clippy is a SUPERSET of `cargo check` — it
# runs the full compile front-end and THEN the lint passes — so this one
# gate proves the workspace both COMPILES and is lint-clean; a separate
# `cargo check` is deliberately NOT run because it would only recompile
# the same code for no extra signal (clippy can't reuse the check driver's
# artifacts, so running both compiled the workspace twice). Probe BOTH
# the `--all-features` and `--no-default-features` sets the gate uses,
# because cfg/feature-gated code (e.g. the non-windows MftReader stubs) can
# trip a lint in one set but not the other. A nightly can compile yet
# still fail the gate by tripping a new/renamed lint or leaving a
# now-stale `#[expect(...)]` unfulfilled — exactly how the v0.5.106 bump
# to nightly-2026-05-31 compiled here but failed the merge gate (the
# `unused_async` → `unused_async_trait_impl` lint split, and
# `duration_suboptimal_units` no longer firing). If clippy fails — step
# back one day, so sync only pins a nightly the merge gate will accept.
# 4. Verify every component listed in rust-toolchain.toml installs cleanly on
# the candidate. If a component conflict fires (e.g. the rustup rename
# `llvm-tools-preview` → `llvm-tools` tripping a stale bookkeeping entry),
Expand All @@ -122,8 +137,12 @@ toolchain-ensure:
# cache wipe before stepping back — handles the common `.partial` rename
# failure mode and short network blips without losing a day on the pin),
# • corrupt toolchain installs (`rustc --version` health probe),
# • upstream nightly regressions (ethnum, rustc ICE, etc.) that break cargo
# check on specific dates,
# • upstream nightly regressions (ethnum, rustc ICE, etc.) that fail to
# COMPILE on specific dates — caught by the Phase 2 clippy gate, which
# compiles the workspace as part of linting (no separate `cargo check`),
# • upstream nightly *lint* changes — new lints, lint renames, or lints that
# stop firing and strand an `#[expect(...)]` — that compile clean but fail
# the merge gate's `clippy -D warnings` (also caught by the Phase 2 gate),
# • rustup component-rename bookkeeping conflicts that would otherwise fire
# on the *next* rustup invocation (breaking `cargo fmt` / `ship-fresh`
# after a successful sync).
Expand Down Expand Up @@ -311,28 +330,52 @@ toolchain-sync:
continue
fi

# ── Phase 2: cargo check --workspace --all-targets ─────
# ── Phase 2: clippy gate (same strict flags as the PR merge gate) ──
# `+TOOLCHAIN` override bypasses rust-toolchain.toml, so a failed probe
# never mutates the pin. --all-targets covers tests/benches/examples,
# which catch regressions that lib-only check misses (e.g. ethnum-via-
# polars-parquet only surfaces under test targets). CARGO_TARGET_DIR
# is scoped per-candidate so cross-toolchain artifact hash mismatches
# ("please recompile that crate using this compiler") don't produce
# false-negative probe failures.
printf "\033[0;34m Running cargo +%s check --workspace --all-targets...\033[0m\n" "$CANDIDATE"
if CARGO_TARGET_DIR="$PROBE_ROOT/$CANDIDATE" \
cargo "+$CANDIDATE" check --workspace --all-targets --quiet \
>/tmp/toolchain-sync-check.log 2>&1; then
printf "\033[0;32m ✅ cargo check ok\033[0m\n"
else
printf "\033[0;33m ⚠️ cargo check failed under %s — stepping back one day.\033[0m\n" "$CANDIDATE"
printf "\033[0;33m (last 3 lines from /tmp/toolchain-sync-check.log:)\033[0m\n"
tail -n 3 /tmp/toolchain-sync-check.log | sed 's/^/ /'

# never mutates the pin. This is the SINGLE compile step: clippy is a
# superset of `cargo check` — it runs the full compile front-end and
# THEN the lint passes — so we deliberately do NOT also run `cargo
# check` (that would just recompile the same code for no extra signal;
# clippy can't reuse the check driver's artifacts, so keeping both
# compiled the workspace twice). This one gate therefore proves the
# workspace COMPILES *and* passes the PR gate's `-D warnings` lint.
# The PR gate (`pr-fast.yml`) runs clippy with `-D warnings`, so a
# nightly that compiles can still fail it by tripping a new/renamed lint
# or leaving a now-stale `#[expect(...)]` unfulfilled — exactly the
# v0.5.106 → nightly-2026-05-31 episode (`unused_async` →
# `unused_async_trait_impl` split; `duration_suboptimal_units` no longer
# firing — which a bare `cargo check` sailed straight past). Probe
# BOTH the `--all-features` and `--no-default-features` sets the gate
# uses, because cfg/feature-gated code (e.g. the non-windows MftReader
# stubs) can trip a lint in one set but not the other; `--all-targets`
# covers tests/benches/examples (e.g. the ethnum-via-polars-parquet
# path that only surfaces under test targets). CARGO_TARGET_DIR is
# scoped per-candidate so cross-toolchain artifact hash mismatches
# ("please recompile that crate using this compiler") can't produce
# false negatives, and the two feature-set runs share it. Windows-only
# lints stay the Nightly Canary's job; this probe runs on the
# maintainer's host. `clippy` ships in the `--profile default` install
# (Phase 1), so it is already present.
CLIPPY_OK=1
for CLIPPY_FEATURES in "--all-features" "--no-default-features"; do
printf "\033[0;34m Running cargo +%s clippy %s -- -D warnings...\033[0m\n" "$CANDIDATE" "$CLIPPY_FEATURES"
if CARGO_TARGET_DIR="$PROBE_ROOT/$CANDIDATE" \
cargo "+$CANDIDATE" clippy --workspace --all-targets "$CLIPPY_FEATURES" --locked --no-deps --quiet -- -D warnings \
>/tmp/toolchain-sync-clippy.log 2>&1; then
printf "\033[0;32m ✅ clippy ok (%s)\033[0m\n" "$CLIPPY_FEATURES"
else
printf "\033[0;33m ⚠️ clippy failed under %s (%s) — stepping back one day.\033[0m\n" "$CANDIDATE" "$CLIPPY_FEATURES"
printf "\033[0;33m (last 5 lines from /tmp/toolchain-sync-clippy.log:)\033[0m\n"
tail -n 5 /tmp/toolchain-sync-clippy.log | sed 's/^/ /'
CLIPPY_OK=0
break
fi
done
if [ "$CLIPPY_OK" != "1" ]; then
if [ "$CANDIDATE" = "$CURRENT" ]; then
printf "\n\033[0;31m❌ Current pin %s installs but no longer compiles the workspace.\033[0m\n" "$CURRENT"
printf "\033[0;31m Full log: /tmp/toolchain-sync-check.log\033[0m\n"
printf "\033[0;31m Manual intervention required — the pinned toolchain is broken.\033[0m\n"
printf "\n\033[0;31m❌ Current pin %s no longer compiles + passes clippy -D warnings.\033[0m\n" "$CURRENT"
printf "\033[0;31m Full log: /tmp/toolchain-sync-clippy.log\033[0m\n"
printf "\033[0;31m Manual intervention required — fix the lints; the merge gate will reject this pin.\033[0m\n"
exit 1
fi
CANDIDATE_DATE=$(date_minus_one "$CANDIDATE_DATE")
Expand Down
Loading