Repository navigation
chore(ci): reuse CI artifacts in auto-release, cache dylint toolchain - #172
Merged
Merged
Conversation
…uted rustup Phase 2 of the setup-soldr audit (Refs #170): - Give the five native CI build wrappers explicit rust-targets so their wheels/binaries are identical to auto-release's (same plat-name tags, artifact names, and soldr cache keys). Also fixes the latent macOS x86 bug where the "native" build on Apple-Silicon macos-14 runners produced an arm64 binary under the x86 name. - auto-release: add a collect-artifacts job that waits for the CI build runs on the release SHA and reuses their artifacts; each of the six platform builds now runs only as a fallback when its CI artifact is missing, failed, or timed out. workflow_dispatch always builds. Eliminates ~80 min of duplicate release builds per version bump. - _lint.yml: cache the dylint nightly toolchain, cargo-dylint/dylint-link binaries, and dylint drivers (the 10+ min fixed cost of every lint job), and install the dylint nightly via `soldr rustup` instead of bare rustup so soldr-managed toolchain homes stay consistent. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This was referenced Jun 12, 2026
zackees
added a commit
that referenced
this pull request
Jun 12, 2026
…ative dylint cache (#174) The #172 change routed the dylint nightly install through `soldr rustup`, which installs into soldr's private toolchain home instead of the setup-soldr-exported RUSTUP_HOME, so the dylint driver build lost its rustc-dev components and failed with E0463 on all four lint platforms. Install the nightly with bare rustup (unconditional — it is not part of the dylint cache), and replace the hand-rolled actions/cache step with setup-soldr's native dylint-cache inputs, gating the cargo-dylint / dylint-link installs on its dylint-cache-hit output. Refs #170 Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
1 of 3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Phase 2 (structural fixes) of #170:
rust-targetas their auto-release counterparts, so CI wheels/binaries are release-identical (same plat-name tags, artifact names, soldr cache keys). Also fixes a latent bug:macos-14runners are Apple Silicon, so the previous "macOS x86" native build actually shipped an arm64 binary.collect-artifactsjob waits (≤35 min) for the six CI build runs on the release SHA and re-publishes their wheel/binary artifacts into the release run. Each platformbuild-*job is now a fallback that runs only when its CI artifact is missing, failed, or timed out;workflow_dispatchalways builds.create-tagacceptssuccess || skippedfrom each build, so a fully-reused release builds nothing. This removes ~80 min of duplicate platform builds per version bump._lint.yml: the nightly-2026-03-26 toolchain,cargo-dylint/dylint-linkbinaries, and~/.dylint_driversare cached viaactions/cachekeyed on the exact versions — the dominant 10+ min fixed cost of every lint job now pays only on cache miss. Per the maintainer's request, the remaining barerustupcall now goes throughsoldr rustup(therust-toolchain.tomltoolchain is already covered by setup-soldr'ssoldr toolchain ensure).Together with #171 this addresses the per-push compute waste quantified in #170 (~207 min baseline). The warm-then-fan-out restructure was deliberately deferred: with #171's unified cache keys, cross-push cache sharing already covers most of the duplicate dep compilation, and a
needs: warmgate would serialize the pipeline and increase wall-clock latency.Test plan
actionlintclean (validates theneedsgraph, expressions, and shellchecks the collect script)bash lintgreen, 18/18 unit testslinux-x86-build.yml→reused_linux_x86, etc.)_lint.yml+ build wrappers directly; first run warms the dylint cache, subsequent runs hit it)Refs #170