From 305443112571026f2852565ddfc4941012093545 Mon Sep 17 00:00:00 2001 From: gsxdsm Date: Wed, 3 Jun 2026 18:55:50 -0700 Subject: [PATCH] =?UTF-8?q?docs(test):=20U8=20gate=20decision=20=E2=80=94?= =?UTF-8?q?=20defer=20vitest=204.x=20upgrade=20to=20follow-up=20PR=20(evid?= =?UTF-8?q?ence=20recorded)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- docs/test-speed-baseline-2026-06-03.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/docs/test-speed-baseline-2026-06-03.md b/docs/test-speed-baseline-2026-06-03.md index 5d01a431a8..a39e69c811 100644 --- a/docs/test-speed-baseline-2026-06-03.md +++ b/docs/test-speed-baseline-2026-06-03.md @@ -204,3 +204,7 @@ change is to `routes-planning-tracking.test.ts`, which is not a top-time file (i sibling `routes-planning.test.ts` is a different file) and whose post-fix duration is unchanged. A later full refresh via `node scripts/ci-test-shard.mjs --write-timings` remains the canonical mechanism. + +## U8 gate decision (2026-06-03): Vitest 4.x upgrade DEFERRED + +Gate criterion: proceed only if cold-start/transform cost is a top contributor blocking the 30s/5min targets. Evidence: cold-start probe ~1.3-1.8s/process (~35-40s aggregate across ~25 invocations, ~10% of full-suite wall-clock); heavy real-SQLite/real-git integration tests dominate and are irreducible-by-design (U7). Transform/collect is the largest *remaining addressable* cost (dashboard lanes show collect ~= test time), so the upgrade is worthwhile as a dedicated follow-up PR — but it is not the top blocker, and a major-version bump across 28 configs + an experimental cache flag does not belong on this already-large branch. Pre-paid: deprecation delta confirmed zero (U5 audit). Re-open with: normalize caret ranges across ALL packages (incl. desktop/mobile/droid-cli/pi-*), pre-bump peer-dep audit, fsModuleCache with kill switch + stale-transform invalidation test, inventory EQUALITY before/after.