From ec8ec7368f558f47a79e1ca65213e6568643cb30 Mon Sep 17 00:00:00 2001 From: HackingGate Date: Thu, 29 Oct 2026 11:14:46 +0900 Subject: [PATCH] The hooks job leaves the engine's tests and clippy to the engine job On ubuntu-slim, capped at 15 minutes a job, the pre-commit leg rebuilt the 32 test binaries the engine job had already built for the same commit and was cancelled partway through engine-tests: twice on one pull request, and 2 of the last 7 runs on main. Skipping engine-tests and engine-clippy there, as coverage and kani already are, removes about six and a half minutes from the pre-commit and prek legs. --- .github/workflows/test.yml | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml index e64f726..8f8c8a1 100644 --- a/.github/workflows/test.yml +++ b/.github/workflows/test.yml @@ -179,10 +179,14 @@ jobs: # survives; running it once per runner here would build the tree # instrumented three more times to answer the same question. The Kani # proofs are skipped for the same reason, and have a job of their own below. + # So are the engine's tests and clippy, which the `engine` job runs on the + # same commit: ubuntu-slim cancels a job at its time cap, and building the + # test binaries a second time on its one CPU is what pushed the pre-commit + # leg past it. Their hook entries still run at every local pre-push. # What this job asks is whether each runner can DRIVE the configuration, # and the hooks left in it answer that. env: - SKIP: coverage,kani + SKIP: coverage,kani,engine-tests,engine-clippy # The private-name guards ask the forge whether a name is a private # repository, and they ask through `gh`. With no token `gh` answers # nothing -- and until an unaskable forge became its own state, that was