morluto/rea

Fork 0

mirror of https://github.com/morluto/rea.git synced 2026-11-05 06:45:22 +00:00

[PR #2131] [MERGED] ci: narrow native-host lane selection and reuse build caches #5050

New issue

Closed

opened 2026-10-31 01:00:48 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 01:00:48 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/2131
Author: @kaoru0822-kitauji
Created: 10/12/2026
Status: ✅ Merged
Merged: 10/12/2026
Merged by: @kaoru0822-kitauji

Base: main ← Head: ci/narrow-lane-selection

📝 Commits (2)

e395e8c ci: narrow native-host lane selection and reuse build caches
d2e75e2 ci: match package script entrypoints against scope globs

📊 Changes

5 files changed (+413 additions, -45 deletions)

View changed files

📝 .github/workflows/ci.yml (+11 -4)
📝 docs/testing.md (+44 -22)
📝 scripts/ci/plan.mjs (+43 -15)
📝 scripts/ci/scopes.mjs (+104 -2)
📝 tests/boundary/filesystem/ciChangeScope.test.ts (+211 -2)

📄 Description

Why

Replaying scripts/ci/plan.mjs over the last 120 first-parent commits on main showed that 93/120 (77.5%) selected the full baseline. That means the 4-host package matrix, the Windows curated lane, the Apple macOS lane, NativeAOT and all 8 real-provider lanes. The main cause was the Unclassified input fallback: almost every src/ file without an owner landed there.

Changes

Owned scopes replace the full fallback for runtime sources.
- New apple and package scopes in scripts/ci/scopes.mjs. They own the Apple artifact lane and the package matrix plus Windows curated lane (Windows native, Inspector, setup/config, packaging entrypoints).
- The test files those lanes invoke explicitly now select their lane.
- An unowned src/** or bridge/** file selects the Linux suite plus the Linux x64 installed-package/Inspector baseline. .agents/skills/** does the same and also selects docs.
- Shared boundaries, dependencies, unknown top-level files and the planner itself still select full.
Manual-only verifier entrypoints are code-only. MANUAL_VERIFIERS lists real Ghidra/Hopper/IDA/Electron, client, JEB/apktool/ADB, packaged-Ghidra-Windows and WASM verifiers. No automatic lane runs them. A new test fails if an automatic workflow (ci.yml, workflow_call or pull_request workflows) ever reaches one through npm run or node scripts/....
macOS x64 runs on PRs only with ci:full. Main pushes, schedule and dispatch keep it.
More doc-only paths: CHANGELOG.md, SECURITY.md, .github/ISSUE_TEMPLATE/**, .github/pull_request_template.md, native/windows/README.md.
Windows curated lane: drops the duplicate npm run typecheck, which the Linux static job already runs, and restores rea-build-cache before build:cached.
Apple artifact lane restores rea-build-cache before its build:cached verifiers.
Linux test shards go from 4 to 6. The timing-balanced sequencer already handles any count.

Main pushes keep the full baseline. CI required and all job names are unchanged.

Replay (same 120 commits, PR event)

	full baseline	package (4-host + Windows)	apple	runtime	code
before	93 (77.5%)	93	93	101	115
after	33 (27.5%)	43	35	100	115

Every file owned by a provider scope (336 occurrences), the Apple scope (14) or the package scope (59) still selected its lane.
No lane that was selected without the full fallback was lost.
The one commit that lost runtime had been full only because it touched a manual verifier.

Coverage tradeoffs

Change 1: a change to unowned shared runtime code (src/application, src/domain, src/artifacts, …) can fail only on arm64, macOS, Windows, the Apple lane, or a provider lane that does not own the file. Such a failure is no longer caught on the PR; the main-push full baseline reports it after merge. Ownership globs do not follow imports. Use ci:full when a change may have effects on other hosts or providers.
Change 4: PRs no longer verify the darwin x64 installed package and Inspector unless labelled ci:full. Main pushes still do.
Cost increase: editing a test file that the Windows or Apple lane lists now selects that lane. For Windows tests this includes the package matrix, because package implies runtime.

Validation

npm run test:focused -- tests/boundary/filesystem/ciChangeScope.test.ts tests/process-global/curatedWindowsLane.test.ts: 80 passed
npm run check:fast, npm run docs:check: pass
actionlint 1.7.7 (-shellcheck= -pyflakes=): pass
No real-provider verification is involved; this changes CI selection only.

🤖 Generated with Claude Code

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/2131 **Author:** [@kaoru0822-kitauji](https://github.com/kaoru0822-kitauji) **Created:** 10/12/2026 **Status:** ✅ Merged **Merged:** 10/12/2026 **Merged by:** [@kaoru0822-kitauji](https://github.com/kaoru0822-kitauji) **Base:** `main` ← **Head:** `ci/narrow-lane-selection` --- ### 📝 Commits (2) - [`e395e8c`](https://github.com/morluto/rea/commit/e395e8c797b8323b72cc7e2ada178ec1ed54c0f6) ci: narrow native-host lane selection and reuse build caches - [`d2e75e2`](https://github.com/morluto/rea/commit/d2e75e22115621cc0a2a89aa24904635fc93134f) ci: match package script entrypoints against scope globs ### 📊 Changes **5 files changed** (+413 additions, -45 deletions) <details> <summary>View changed files</summary> 📝 `.github/workflows/ci.yml` (+11 -4) 📝 `docs/testing.md` (+44 -22) 📝 `scripts/ci/plan.mjs` (+43 -15) 📝 `scripts/ci/scopes.mjs` (+104 -2) 📝 `tests/boundary/filesystem/ciChangeScope.test.ts` (+211 -2) </details> ### 📄 Description ## Why Replaying `scripts/ci/plan.mjs` over the last 120 first-parent commits on main showed that **93/120 (77.5%)** selected the full baseline. That means the 4-host package matrix, the Windows curated lane, the Apple macOS lane, NativeAOT and all 8 real-provider lanes. The main cause was the `Unclassified input` fallback: almost every `src/` file without an owner landed there. ## Changes 1. **Owned scopes replace the full fallback for runtime sources.** - New `apple` and `package` scopes in `scripts/ci/scopes.mjs`. They own the Apple artifact lane and the package matrix plus Windows curated lane (Windows native, Inspector, setup/config, packaging entrypoints). - The test files those lanes invoke explicitly now select their lane. - An unowned `src/**` or `bridge/**` file selects the Linux suite plus the Linux x64 installed-package/Inspector baseline. `.agents/skills/**` does the same and also selects docs. - Shared boundaries, dependencies, unknown top-level files and the planner itself still select full. 2. **Manual-only verifier entrypoints are code-only.** `MANUAL_VERIFIERS` lists real Ghidra/Hopper/IDA/Electron, client, JEB/apktool/ADB, packaged-Ghidra-Windows and WASM verifiers. No automatic lane runs them. A new test fails if an automatic workflow (ci.yml, `workflow_call` or `pull_request` workflows) ever reaches one through `npm run` or `node scripts/...`. 3. **macOS x64 runs on PRs only with `ci:full`.** Main pushes, schedule and dispatch keep it. 4. **More doc-only paths:** `CHANGELOG.md`, `SECURITY.md`, `.github/ISSUE_TEMPLATE/**`, `.github/pull_request_template.md`, `native/windows/README.md`. 5. **Windows curated lane:** drops the duplicate `npm run typecheck`, which the Linux `static` job already runs, and restores `rea-build-cache` before `build:cached`. 6. **Apple artifact lane** restores `rea-build-cache` before its `build:cached` verifiers. 7. **Linux test shards go from 4 to 6.** The timing-balanced sequencer already handles any count. Main pushes keep the full baseline. `CI required` and all job names are unchanged. ## Replay (same 120 commits, PR event) | | full baseline | package (4-host + Windows) | apple | runtime | code | |---|---|---|---|---|---| | before | 93 (77.5%) | 93 | 93 | 101 | 115 | | after | **33 (27.5%)** | 43 | 35 | 100 | 115 | - Every file owned by a provider scope (336 occurrences), the Apple scope (14) or the package scope (59) still selected its lane. - No lane that was selected without the full fallback was lost. - The one commit that lost `runtime` had been full only because it touched a manual verifier. ## Coverage tradeoffs - **Change 1:** a change to unowned shared runtime code (`src/application`, `src/domain`, `src/artifacts`, …) can fail only on arm64, macOS, Windows, the Apple lane, or a provider lane that does not own the file. Such a failure is no longer caught on the PR; the main-push full baseline reports it after merge. Ownership globs do not follow imports. Use `ci:full` when a change may have effects on other hosts or providers. - **Change 4:** PRs no longer verify the darwin x64 installed package and Inspector unless labelled `ci:full`. Main pushes still do. - **Cost increase:** editing a test file that the Windows or Apple lane lists now selects that lane. For Windows tests this includes the package matrix, because `package` implies `runtime`. ## Validation - `npm run test:focused -- tests/boundary/filesystem/ciChangeScope.test.ts tests/process-global/curatedWindowsLane.test.ts`: 80 passed - `npm run check:fast`, `npm run docs:check`: pass - `actionlint` 1.7.7 (`-shellcheck= -pyflakes=`): pass - No real-provider verification is involved; this changes CI selection only. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-31 01:00:48 +00:00

closed this issue
added the
pull-request
label

bsakaguchi referenced this issue

2026-11-01 21:42:54 +00:00

[PR #5065] [CLOSED] test(process): reduce macOS wall-clock capture test flakiness #6765

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

morluto/rea#5050

No description provided.

Rows
Columns