perf: optimize CI/CD pipeline with caching and parallel builds

[anandgupta42]

Summary

• Parallel release builds: Split 11 sequential binary compilations into 3 parallel matrix jobs (linux/darwin/win32), reducing build time from ~11min to ~4min
• Dependency caching: Added Bun cache (~/.bun/install/cache) and pip cache across all workflows
• Faster CI lint: Separated Python linting into its own job with dev-only deps (skips heavy warehouse deps like snowflake/bigquery)
• Parallel release publishing: GitHub Release no longer waits for npm publish — runs right after build completes
• Engine publishes in parallel: publish-engine runs concurrently with build (no artifact dependency needed)

Files changed

File	Change
.github/workflows/release.yml	Matrix build strategy, Bun caching, fixed dependency chain, concurrency control
.github/workflows/ci.yml	Bun caching, pip caching, separate lint job
.github/workflows/docs.yml	pip caching
.github/workflows/publish-engine.yml	pip caching
packages/altimate-code/script/build.ts	Added --targets flag for OS-based filtering

Expected impact

Workflow	Before	After (estimated)
Release (wall clock)	~18 min	~8-9 min
Release build job	~11 min (1 job)	~4 min (3 parallel)
CI	~1 min	~45s (faster on cache hits)

Release workflow dependency graph

Before:                          After:
build (11m)                      build-linux ─┐
  └→ publish-npm (7m)            build-darwin ─┼─ (4m, parallel)
       └→ github-release (1m)    build-win32 ─┘
  └→ publish-engine (7m)            ├→ publish-npm (7m)
                                    ├→ github-release (1m)  ← parallel!
                                    └→ publish-engine (3m)  ← parallel!

Test plan

• All 1415 existing tests pass (bun test)
• YAML syntax validated (parsed all 4 workflow files)
• Semantic validation: matrix config, artifact patterns, merge-multiple, dependency chain, caching
• --targets flag unit tested: linux=6, darwin=3, win32=2, total=11
• Backward compatibility: no --targets flag builds all 11 targets
• Verify CI runs on this PR (GitHub Actions)
• Verify release works on next tag push

🤖 Generated with Claude Code

[jontsai]

thought: github-release now depends only on [build] instead of [build, publish-npm]. This means the GitHub Release is created before npm publish completes. If npm publish fails, you'll have a release with binaries but no npm package. Intentional tradeoff for speed?

[jontsai]

praise: Clean approach — the --targets flag keeps the build script backwards-compatible while enabling the matrix split. No changes needed to the build logic itself.

[jontsai]

suggestion: Since this lint job only needs ruff, you could install just ruff directly instead of .[dev] (which may pull in pytest and other dev deps). Would make this job even faster:

- name: Install linter
  run: pip install ruff

Though .[dev] ensures version consistency with what devs use locally, so this is a minor tradeoff.

[jontsai]

PR #42 Review Tour Guide

File Order

#	File	+/-	Purpose
1	.github/workflows/release.yml	+36/-10	Matrix build strategy, parallel jobs, caching
2	.github/workflows/ci.yml	+30/-4	Separate lint job, Bun + pip caching
3	packages/altimate-code/script/build.ts	+6/-1	--targets flag for OS-based filtering
4	.github/workflows/docs.yml	+2/-0	pip caching
5	.github/workflows/publish-engine.yml	+1/-0	pip caching

---

Stop 1: Release Workflow — Matrix Build (release.yml)

Context: The 11 binary targets are split into 3 parallel matrix jobs by OS (linux/darwin/win32). Each uploads dist-{os} artifacts, and downstream jobs use merge-multiple: true to combine them.

Review:

• ✅ fail-fast: false — correct, one OS failure shouldnt block others
• ✅ concurrency with cancel-in-progress: false — prevents overlapping releases
• ✅ merge-multiple: true on download — clean artifact merging
• ⚠️ github-release no longer depends on publish-npm (posted inline comment)

Stop 2: CI Workflow — Lint Separation (ci.yml)

Context: Python linting moved from the test matrix into its own job, avoiding heavy warehouse deps (snowflake/bigquery SDKs).

Review:

• ✅ Good separation — lint doesnt need warehouse deps
• 💡 Could use just pip install ruff instead of .[dev] (posted inline suggestion)
• ✅ Bun caching with bun.lock hash key

Stop 3: Build Script — --targets Flag (build.ts)

Context: Simple flag parsing to filter allTargets by OS name.

Review:

• ✅ Backward compatible — no flag builds all 11 targets
• ✅ Clean integration with existing singleFlag logic

Summary Checklist

Area	Status	Notes
Matrix build	✅ Good	3-way parallel, proper artifact naming
Caching	✅ Good	Bun + pip across all workflows
Dependency chain	⚠️ Thought	github-release no longer waits for npm
Build script	✅ Good	Backward compatible --targets flag
publish-engine	✅ Good	Correctly decoupled from build artifacts

LGTM — ship it! 🚢

[suryaiyer95]

Multi-Model Code Review — 6 AI Models

Verdict: APPROVE (with suggestions)
Critical: 0 | Major: 1 | Minor: 4 | Nits: 2

Reviewed by Claude Opus 4.6, Kimi K2.5, Grok 4, MiniMax M2.5, GLM-5, Qwen 3.5 Plus — converged in 1 round.

---

Major Issues

1. GitHub Release created before npm publish completes — partial release risk

Design — .github/workflows/release.yml:144 (needs: [build], changed from needs: [build, publish-npm])

Previously the GitHub Release was only created after npm publish succeeded. Now it runs in parallel. If npm publish fails:

• Users see a GitHub Release with npm install instructions that don't work
• No automatic rollback — the release stays up
• Downstream consumers (Homebrew tap) reference the release but need npm packages

This saves ~7min but the failure mode creates user confusion.

Fix: Either restore needs: [build, publish-npm] for atomicity, add a post-publish verification step, or add a YAML comment documenting the tradeoff.

Flagged by: Claude, Kimi, GLM-5, Grok (Consensus)

---

Minor Issues

2. docs.yml pip cache uses mkdocs.yml as dependency path

Bug / Performance — .github/workflows/docs.yml:29-30

cache: "pip"
cache-dependency-path: docs/mkdocs.yml

mkdocs.yml is a config file, not a pip requirements file. The cache key won't change when mkdocs-material releases a new version, serving stale packages.

Fix: Create docs/requirements.txt with mkdocs-material and reference it. (Claude, GLM-5, MiniMax, Qwen, Grok, Kimi — Consensus)

3. publish-engine.yml missing cache-dependency-path

Performance — .github/workflows/publish-engine.yml:21

Pip cache enabled but without cache-dependency-path, so invalidation relies on auto-detection which may miss pyproject.toml.

Fix: Add cache-dependency-path: packages/altimate-engine/pyproject.toml. (GLM-5, MiniMax, Grok, Kimi — Consensus)

4. --targets flag has no validation for invalid OS values

Code Quality — packages/altimate-code/script/build.ts:123

--targets=windows (instead of win32) silently produces zero targets. The rm -rf dist runs first, then nothing gets built — no error, no warning.

Fix: Validate against ['linux', 'darwin', 'win32'] and exit with an error for invalid values. (Claude, GLM-5, Grok, Qwen, Kimi — Consensus)

5. publish-engine has no needs: — runs immediately on tag push

Design — .github/workflows/release.yml:116-117

Engine publishes to PyPI independently, without waiting for build. While the engine doesn't need build artifacts, a PyPI release could exist for a broken CLI version.

Fix: Add a YAML comment documenting this is intentional, or restore needs: build. (Claude, GLM-5, Qwen, Kimi — Consensus)

---

Nits

• Redundant concurrency group name at release.yml:9 — release-${{ github.workflow }} evaluates to release-Release. Use just release or ${{ github.workflow }}.
• Repeated Bun cache block (3 identical 6-line blocks) — could be extracted to a composite action.

---

Positive Observations

• Matrix strategy is correct: fail-fast: false ensures all platforms build. needs: [build] correctly waits for ALL matrix jobs.
• Artifact merge pattern: dist-${{ matrix.os }} upload + pattern: dist-* with merge-multiple: true is the proper GitHub Actions pattern.
• Lint job separation: Moving lint to its own job with only [dev] deps (no heavy warehouse packages) is a smart optimization.
• Bun caching: Well-configured with hashFiles('bun.lock') and fallback restore-keys.
• --targets flag: Clean implementation that integrates naturally with existing flags.
• Concurrency control: cancel-in-progress: false is correct — never cancel a running release.
• Cross-compilation: Building darwin/win32 on ubuntu-latest using Bun's native cross-compile is efficient and correct.

Missing Tests

• --targets with invalid OS name (e.g., --targets=windows)
• --targets with empty string
• --targets combined with --single (precedence behavior)

---

Finding Attribution

Issue	Origin	Type
1. GitHub Release before npm publish	Claude, Kimi, GLM-5, Grok	Consensus
2. docs.yml pip cache misconfigured	Claude, GLM-5, MiniMax, Qwen, Grok, Kimi	Consensus
3. publish-engine.yml missing cache path	GLM-5, MiniMax, Grok, Kimi	Consensus
4. --targets flag no validation	Claude, GLM-5, Grok, Qwen, Kimi	Consensus
5. publish-engine no needs dependency	Claude, GLM-5, Qwen, Kimi	Consensus
6. Redundant concurrency group name	Claude, Kimi	Consensus
7. Repeated Bun cache block	Claude	Unique

Reviewed by 6 models: Claude Opus 4.6, Kimi K2.5, Grok 4, MiniMax M2.5, GLM-5, Qwen 3.5 Plus. Convergence: 1 round.

Notable rejected findings:

• Cross-compilation on ubuntu-latest (Bun supports this natively)
• cancel-in-progress: false race condition (concurrency group queues runs correctly)
• Matrix needs doesn't check success (needs: [build] waits for ALL matrix jobs by default)
• Cache key should use matrix.os (shared cache is optimal since all jobs run on same runner)