CI: share iOS port build across iOS workflows via reusable workflow by liannacasper · Pull Request #4837 · codenameone/CodenameOne

liannacasper · 2026-04-30T14:33:21Z

Summary

The biggest remaining iOS performance gap from the original audit: the iOS port is built from scratch in every iOS workflow on every PR (5 builds across scripts-ios.yml, scripts-ios-native.yml, and ios-packaging.yml's 3-row matrix). This PR makes that build run at most once per workflow run, and shares the result across workflows when source hasn't changed.

How

A new reusable workflow .github/workflows/_build-ios-port.yml (workflow_call) runs setup-workspace + build-ios-port. Each of the 3 iOS workflows now has 2 jobs:

build-port (uses _build-ios-port.yml)  →  test job (needs: build-port)

The test job just restores the populated caches and runs build-ios-app.sh + tests. It uses actions/cache/restore@v4 with fail-on-cache-miss: true so a missing cache is a clear error, not silent regression.

The cache that does the heavy lifting

A new cn1-built cache stores the built CN1 + iOS port artifacts:

~/.m2/repository/com/codenameone
Themes/
Ports/iOSPort/nativeSources/

Keyed on a composite hash of:

All Java / Objective-C / theme source under CodenameOne/src, Ports/iOSPort, vm/JavaAPI, vm/ByteCodeTranslator, Themes/
Every pom.xml
setup-workspace.sh, build-ios-port.sh, build-native-themes.sh, and the reusable workflow itself

No restore-keys — exact match only. A stale artifact set never satisfies a different source state.

The cn1-built cache is restored after the broader ~/.m2/repository cache so it overwrites any stale com/codenameone/ subtree pulled in by the POM-keyed m2 cache.

On cache hit, build-port skips both setup-workspace and build-ios-port and the job finishes in ~1–2 min instead of ~30 min. The 3 test jobs then restore the cache and run.

Cross-workflow benefit (the M6 part)

The cn1-built-${runner.os}-${source-hash} cache key is identical across all 3 iOS workflows. When scripts-ios.yml runs first on a PR and populates the cache, ios-packaging.yml and scripts-ios-native.yml later in the same PR cache-hit and skip their entire iOS port rebuild.

A PR that doesn't touch CN1 / iOS port source at all will hit the cache from master's last successful run.

Expected savings

Rough timings:

Scenario	Before	After
First run, source changed	~33 min × 3 matrix in parallel = ~37 min wall, ~99 min CPU	~30 min build-port → ~10 min × 3 in parallel = ~40 min wall, ~60 min CPU
Subsequent run, source unchanged	~37 min wall, ~99 min CPU	~2 min build-port → ~10 min × 3 in parallel = ~12 min wall, ~32 min CPU
Cross-workflow on same PR	3 × ~33 min = ~99 min macOS minutes	First workflow: ~33 min, others hit cache: ~12 min each → ~57 min total

So the warm-cache path (the common case for typical PRs) goes from ~37 min wall-clock to ~12 min, and three iOS workflows on the same PR collectively go from ~3 × 33 min to ~33 + 12 + 12 min.

Bonus: parparvm-tests.yml self-trigger bug

parparvm-tests.yml's paths: filter didn't include itself, so my changes to that workflow in #4836 never actually triggered the workflow's own validation. Fixed by adding .github/workflows/parparvm-tests.yml to its paths: list.

Risks / things to watch

fail-on-cache-miss: true on the test jobs. If the cn1-built cache somehow doesn't survive the build-port → test-job handoff (e.g., GitHub cache eviction in the seconds between jobs), the test jobs fail loudly. This is intentional — silent fall-through could mask a problem.
Hash completeness. The source hash covers Java, Objective-C, theme, properties, XML, CSS files plus all POMs and the build scripts. If a build output changes based on a file we didn't include, we'd serve a stale artifact. The composite is conservative; happy to tighten or loosen if you spot a gap.
First-run wall-clock is slightly slower (~40 min vs ~37 min) because build-port is serial before the matrix. The CPU savings (~40 min less per first run) offset that, and every subsequent run is dramatically faster.
Within-PR cancellation. The reusable workflow has its own concurrency: group (mac-ios-port-${{ github.workflow }}-${{ github.ref_name }}) so a new push cancels the in-flight build-port too.

Test plan

All 5 modified/new workflows parse (verified locally with yaml.safe_load).
First PR run (cold cache): all 3 iOS workflows succeed; build-port does the full build, test jobs find the cn1-built cache and skip ahead to build-ios-app.sh.
Second push on same PR (warm cache): build-port logs "cn1-built cache HIT — setup-workspace and build-ios-port were skipped." Test jobs run as before.
Verify cross-workflow share: if scripts-ios.yml runs first then ios-packaging.yml runs later on the same PR, ios-packaging.yml's build-port logs a cache hit.
Confirm parparvm-tests.yml runs on this PR (since it modifies .github/workflows/parparvm-tests.yml, which is now in the paths filter).

Out of scope (kept as future work)

M3/M4: Dedupe the Maven + Ant double-build inside pr.yml.
L6: Drop redundant push: triggers on test-only workflows.
Apply the same reusable-workflow pattern to scripts-android.yml + scripts-javase.yml.

🤖 Generated with Claude Code

github-actions · 2026-04-30T14:46:37Z

Developer Guide build artifacts are available for download from this workflow run:

Developer Guide quality checks:

AsciiDoc linter: No issues found (report)
Vale: Vale failed (exit code 2) (report)
Image references: No unused images detected (report)

github-actions · 2026-04-30T14:58:34Z

✅ Continuous Quality Report

Test & Coverage

✅ Tests: 2431 total, 0 failed, 0 skipped
📊 Line coverage: 53.34% [HTML preview] [Download]
- Lowest covered classes
  - com.codename1.ui.Display$EdtException – 0.00%
  - com.codename1.ui.plaf.CSSBorder$LinearGradient – 0.00%
  - com.codename1.util.EasyThread$InQueueRunnable – 0.00%
  - com.codename1.components.FloatingActionButton$ReleaseActionListener – 0.00%
  - com.codename1.io.Oauth2$ShowAuthenticationActionListener – 0.00%
  - com.codename1.io.Oauth2$RefreshTokenActionListener – 0.00%
  - com.codename1.util.EasyThread$RunAndWaitRunnable – 0.00%
  - com.codename1.components.ToastBar$FlushAnimationCallback – 0.00%
  - com.codename1.ui.plaf.CSSBorder$ColorStop – 0.00%
  - com.codename1.ui.ElevationComparator – 0.00%

Static Analysis

SpotBugs [Report archive]
- ✅ ByteCodeTranslator: 0 findings (no issues)
- ✅ android: 0 findings (no issues)
- ✅ codenameone-maven-plugin: 0 findings (no issues)
- ✅ core-unittests: 0 findings (no issues)
- ✅ ios: 0 findings (no issues)
✅ PMD: 0 findings (no issues) [Report archive]
✅ Checkstyle: 0 findings (no issues) [Report archive]

Generated automatically by the PR CI workflow.

github-actions · 2026-04-30T15:04:19Z

Cloudflare Preview

URL: https://pr-4837-website-preview.codenameone.pages.dev
Branch: pr-4837-website-preview

github-actions · 2026-04-30T15:35:03Z

✅ ByteCodeTranslator Quality Report

Test & Coverage

✅ Tests: 644 total, 0 failed, 2 skipped

Benchmark Results

Execution Time: 10622 ms
Hotspots (Top 20 sampled methods):
- 23.99% java.lang.String.indexOf (445 samples)
- 19.14% java.util.ArrayList.indexOf (355 samples)
- 18.27% com.codename1.tools.translator.Parser.isMethodUsed (339 samples)
- 4.69% com.codename1.tools.translator.BytecodeMethod.addToConstantPool (87 samples)
- 4.53% java.lang.Object.hashCode (84 samples)
- 2.86% java.lang.System.identityHashCode (53 samples)
- 1.73% com.codename1.tools.translator.ByteCodeClass.updateAllDependencies (32 samples)
- 1.67% com.codename1.tools.translator.BytecodeMethod.optimize (31 samples)
- 1.56% com.codename1.tools.translator.ByteCodeClass.markDependent (29 samples)
- 1.35% com.codename1.tools.translator.Parser.generateClassAndMethodIndexHeader (25 samples)
- 1.29% com.codename1.tools.translator.ByteCodeClass.calcUsedByNative (24 samples)
- 1.29% com.codename1.tools.translator.BytecodeMethod.appendMethodSignatureSuffixFromDesc (24 samples)
- 1.02% java.lang.StringBuilder.append (19 samples)
- 0.65% com.codename1.tools.translator.Parser.cullMethods (12 samples)
- 0.65% com.codename1.tools.translator.BytecodeMethod.isMethodUsedByNative (12 samples)
- 0.59% com.codename1.tools.translator.BytecodeMethod.appendCMethodPrefix (11 samples)
- 0.59% sun.nio.ch.FileDispatcherImpl.write0 (11 samples)
- 0.54% com.codename1.tools.translator.Parser.writeOutput (10 samples)
- 0.49% com.codename1.tools.translator.ByteCodeField.equals (9 samples)
- 0.49% com.codename1.tools.translator.BytecodeMethod.addInstruction (9 samples)
⚠️ Coverage report not generated.

Static Analysis

✅ SpotBugs: no findings (report was not generated by the build).
⚠️ PMD report not generated.
⚠️ Checkstyle report not generated.

Generated automatically by the PR CI workflow.

shai-almog · 2026-04-30T16:14:26Z

Compared 85 screenshots: 85 matched.
✅ Native iOS screenshot tests passed.

Benchmark Results

VM Translation Time: 0 seconds
Compilation Time: 181 seconds

Build and Run Timing

Metric	Duration
Simulator Boot	64000 ms
Simulator Boot (Run)	1000 ms
App Install	25000 ms
App Launch	7000 ms
Test Execution	287000 ms

Detailed Performance Metrics

Metric	Duration
Base64 payload size	8192 bytes
Base64 benchmark iterations	6000
Base64 native encode	1540.000 ms
Base64 CN1 encode	1535.000 ms
Base64 encode ratio (CN1/native)	0.997x (0.3% faster)
Base64 native decode	1099.000 ms
Base64 CN1 decode	1673.000 ms
Base64 decode ratio (CN1/native)	1.522x (52.2% slower)
Base64 SIMD encode	578.000 ms
Base64 encode ratio (SIMD/native)	0.375x (62.5% faster)
Base64 encode ratio (SIMD/CN1)	0.377x (62.3% faster)
Base64 SIMD decode	489.000 ms
Base64 decode ratio (SIMD/native)	0.445x (55.5% faster)
Base64 decode ratio (SIMD/CN1)	0.292x (70.8% faster)
Image encode benchmark iterations	100
Image createMask (SIMD off)	64.000 ms
Image createMask (SIMD on)	15.000 ms
Image createMask ratio (SIMD on/off)	0.234x (76.6% faster)
Image applyMask (SIMD off)	141.000 ms
Image applyMask (SIMD on)	87.000 ms
Image applyMask ratio (SIMD on/off)	0.617x (38.3% faster)
Image modifyAlpha (SIMD off)	162.000 ms
Image modifyAlpha (SIMD on)	79.000 ms
Image modifyAlpha ratio (SIMD on/off)	0.488x (51.2% faster)
Image modifyAlpha removeColor (SIMD off)	173.000 ms
Image modifyAlpha removeColor (SIMD on)	105.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off)	0.607x (39.3% faster)
Image PNG encode (SIMD off)	1229.000 ms
Image PNG encode (SIMD on)	1074.000 ms
Image PNG encode ratio (SIMD on/off)	0.874x (12.6% faster)
Image JPEG encode	654.000 ms

Introduces a reusable workflow .github/workflows/_build-ios-port.yml (workflow_call) that runs scripts/setup-workspace.sh and scripts/build-ios-port.sh once per workflow run. The three iOS workflows (scripts-ios.yml, scripts-ios-native.yml, ios-packaging.yml) now have two jobs each: a build-port job that calls the reusable workflow, and the existing test job which "needs: build-port" and restores the populated caches. A new "cn1-built" cache stores the built CN1 + iOS port artifacts (~/.m2/repository/com/codenameone, Themes, Ports/iOSPort/nativeSources) keyed on a composite hash of: - CN1 / iOS port / VM Java/native source files - All pom.xml files - setup-workspace.sh / build-ios-port.sh / build-native-themes.sh / the reusable workflow itself There are no restore-keys on this cache (exact match only) so a stale artifact set never satisfies a different source state. On cache hit the build-port job skips both setup-workspace and build-ios-port and finishes in ~1-2 min instead of ~30 min. Cross-workflow benefit: the cn1-built cache is keyed only on the composite source hash, so the first iOS workflow that runs on a PR populates it and the other two skip their builds. On the same PR this typically saves ~2x ~30 min macOS minutes. Within ios-packaging.yml the matrix entries are now 3 small jobs that just restore caches and run build-ios-app + tests, replacing the prior 3 full duplicate pipelines. Also fixes a self-trigger bug in parparvm-tests.yml: changes to the workflow file itself didn't trigger a CI run because the paths filter omitted .github/workflows/parparvm-tests.yml. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…llow-up) #4837 (d119e5b) split the iOS port build into a reusable workflow whose output is consumed by build-ios via the cn1-built cache. build-ios-metal was still running setup-workspace.sh + build-ios-port.sh inline, costing ~12 minutes of redundant work per PR even though the source SHA matches what build-port already cached. Add the same cache-restore steps build-ios uses: - needs: build-port so the cache is populated before the Metal job starts - Compute CN1 source hash + setup-workspace hash (same formulas as the reusable workflow + build-ios) - Cache codenameone-tools, Maven repository, cn1-binaries, CocoaPods/gems - Restore the cn1-built bundle (~/.m2/repository/com/codenameone, Themes, Ports/iOSPort/nativeSources) with fail-on-cache-miss so a missed cache surfaces immediately rather than re-running the full build Drops Setup workspace + Build iOS port from the Metal job (the cache restore covers both). Timeout reduced from 60 to 45 minutes since the remaining work is the sample-app build + Metal UI test run, both of which fit under 30 minutes already. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

liannacasper force-pushed the ci-ios-port-share branch from 13c6c34 to c349fb0 Compare April 30, 2026 17:12

shai-almog merged commit d119e5b into master May 1, 2026
17 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

CI: share iOS port build across iOS workflows via reusable workflow#4837

CI: share iOS port build across iOS workflows via reusable workflow#4837
shai-almog merged 1 commit intomasterfrom
ci-ios-port-share

liannacasper commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026 •

edited

Loading

Uh oh!

shai-almog commented Apr 30, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

liannacasper commented Apr 30, 2026

Summary

How

The cache that does the heavy lifting

Cross-workflow benefit (the M6 part)

Expected savings

Bonus: parparvm-tests.yml self-trigger bug

Risks / things to watch

Test plan

Out of scope (kept as future work)

Uh oh!

github-actions Bot commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Continuous Quality Report

Test & Coverage

Static Analysis

Uh oh!

github-actions Bot commented Apr 30, 2026

Cloudflare Preview

Uh oh!

github-actions Bot commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ ByteCodeTranslator Quality Report

Test & Coverage

Benchmark Results

Static Analysis

Uh oh!

shai-almog commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Benchmark Results

Build and Run Timing

Detailed Performance Metrics

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Apr 30, 2026 •

edited

Loading

github-actions Bot commented Apr 30, 2026 •

edited

Loading

shai-almog commented Apr 30, 2026 •

edited

Loading