Skip to content

ci: warm OCaml env cache on base branches - #223

Merged
tiensonqin merged 2 commits into
feat/ml-backendfrom
ci/warm-opam-cache
Oct 12, 2026
Merged

tiensonqin merged 2 commits into
feat/ml-backendfrom
ci/warm-opam-cache

Conversation

@tiensonqin

Copy link
Copy Markdown
Collaborator

Windows native-tests went from ~3 min to ~20-25 min since the week-42 cache epoch rolled: setup-lui-ocaml missed and rebuilt cygwin+opam+5.5.1 from scratch (setup step 900-1400 s).

Why every PR missed: PR runs can only read caches saved on their own ref or the base branch. The only Windows caches are refs/pull/*/merge (not shareable); CI push runs on feat/ml-backend are cancel-in-progress and get cancelled by the next merge before the cold setup saves.

Fix: a separate, never-cancelled workflow that builds/saves the env on main/feat/ml-backend when deps or the setup action change, weekly right after the epoch rolls, and on demand. After it runs once, PR Windows jobs restore it (exact or restore-keys hit) and drop back to ~3-4 min.

(The real failures exposed by the cold env are fixed separately in #222.)

@tiensonqin
tiensonqin merged commit 42b5eac into feat/ml-backend Oct 12, 2026
8 checks passed
@tiensonqin
tiensonqin deleted the ci/warm-opam-cache branch October 12, 2026 23:43
tiensonqin added a commit that referenced this pull request Oct 13, 2026
Picks up #224 (retained scene fragments), #217 (dead-clip ghost
walk), #213, #209, #223. Resolution:

- walk keeps both features: the ghost pass takes clip so the spatial
  keep test can prune its popup search, and both body and ghost
  enumerate children through the shared candidates helper.
- paint stays ?spatial — the retained fragment pass is the
  production path and owns its own cull, so the host's spatial
  provider is dropped; the index still serves hit-testing via
  Assembly.handle and Lui_paint.paint callers.
- Lui_host keeps only the invalidation counter the hit-index key
  needs.
- Restores Lui_paint.resolve_in, lost from #224's squash merge
  (same fix as 21b58d7 on perf/ml-p2-3-scene-batch-cache).

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant