Reuse AOT DI scripts when compile fingerprint matches by koriym · Pull Request #484 · bearsunday/BEAR.Package

koriym · 2026-07-10T17:30:48Z

Summary

Fixes #483: PackageInjector::factory() no longer always recompiles prod DI scripts.

Replaces the earlier source-fingerprint approach (this PR's first revisions) with a compile marker: a boolean "a compile has run here" file, not a freshness check.

After a successful compile, write {scriptDir}/.compile-marker.
On later factory() calls, if the marker is present, construct CompiledInjector directly (no write, no recompile).
If the marker is missing, compile on demand and emit E_USER_NOTICE ("Not precompiled; ...") so a forgotten bin/compile.php is loud.
A marker with broken scripts is a deploy error, not a recoverable one: the boot is left to throw instead of falling back to a runtime recompile (which would also die under a read-only FS).
Compiler::compile() writes the marker after the final DI scripts (AOT / __invoke path).

The marker is written after a successful compile, so a crashed compile leaves no marker and the next boot recovers via the on-demand path.

Why not a fingerprint

A source fingerprint (src/ mtimes + composer.lock + .env*) can never prove the scripts still match the source: mtime is not content, appDir in the hash makes image-path relocation a guaranteed mismatch, and max() mtime misses rolled-back files. The fingerprint was the framework trying to be smarter than it can be. The marker changes the question from "do these scripts still match the source?" (unprovable) to "did a compile run here?" (trivially, perfectly knowable). Freshness becomes the deploy's responsibility — the same requirement the existing serialized-injector cache already imposes.

Foundation (image reuse / read-only FS)

Reusing AOT output safely requires the output to be complete and portable, which is being fixed in the Ray libraries:

fix: resolve #[ScriptDir] to __DIR__ instead of baking the compile-time absolute path ray-di/Ray.Compiler#136 — compiled scripts resolve #[ScriptDir]/InjectorInterface to __DIR__ instead of baking the build-time absolute path (path portability). Merged.
fix: re-emit weaved class file when declared in-process but missing on disk ray-di/Ray.Aop#261 — Compiler::compile() re-emits the weaved class file when it is declared in-process but missing on disk, so AOT output includes AOP-woven class definitions (completeness). Without this, reusing AOT output throws Class not found for intercepted resources (incl. @Embed).

This PR should land alongside the Ray.Aop release that includes #261 (bump ray/aop), so reused AOT output is complete.

Test plan

composer test, phpstan, psalm, phpcs
Marker present → script inodes unchanged (reuse, no recompile)
Marker missing → on-demand compile + E_USER_NOTICE
Source change with marker present → still reuses (freshness is not checked)
Marker present but scripts corrupted → throws (does not silently rebuild)

Skip unconditional Ray\Compiler recompile in PackageInjector::factory() when {scriptDir}/.compile-fingerprint still matches the source tree. Fixes ahead-of-time compile being discarded on first request (bearsunday#483).

coderabbitai · 2026-07-10T17:30:56Z

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 98fd1679-3a78-4c3c-bb65-f4bb37302241

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

codecov · 2026-07-10T17:32:13Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (7483252) to head (1d5a808).

Additional details and impacted files

@@             Coverage Diff             @@
##                 1.x      #484   +/-   ##
===========================================
  Coverage     100.00%   100.00%           
- Complexity       252       257    +5     
===========================================
  Files             51        52    +1     
  Lines            753       770   +17     
===========================================
+ Hits             753       770   +17

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

NaokiTsuchiya

The stamp won't match in the case #483 was about: {tmpDir}/di shipped in an image.

The tests only cover reuse within the same tree. Compiling in one tree and booting from another — the whole point of AOT — is untested, and breaks for two independent reasons:

Absolute path in the hash. appDir is hash input, so a different runtime path is a guaranteed mismatch (see line comment).
mtime, not content. maxMtime() maxes filemtime() instead of hashing. Any copy that doesn't preserve mtimes — git clone, CI artifacts, rsync without -t — misses. (max() also can't see a file rolled back.)
Fixing the path alone still leaves the mtime problem.

On mismatch, prodInjector() falls through to compile(), which under readOnlyRootFilesystem dies with FileNotWritable: _bindings.log. Impact 2 from #483 isn't just unfixed — the app doesn't boot.

Suggestions

Hash src/ contents (hash_file, sort() first — iterator order is filesystem-dependent) instead of mtimes.
Make fileStamp() paths relative to appDir.
Add a test: compile in one dir, copy src/ and di/ elsewhere clobbering mtimes, boot there. Minimal repro of baking into an image.
Longer term, the opt-in from the end of #483 — declaring these scripts were compiled from this tree — is cheaper and safer for containers. With compilation verified in CI the runtime check can go entirely. Content hash as the safe default, opt-in for immutable images, covers both.

NaokiTsuchiya · 2026-07-11T06:59:33Z

+            return '0';
+        }
+
+        return $path . ':' . (string) filemtime($path);


The absolute path is in the hash input, so appDir is part of the stamp. Compile in /build/app, run from /var/www/html, and it never matches — identical files or not. Should be relative to appDir.

Replace CompileStamp (mtime + composer.lock + .env* fingerprint) with CompileMarker, a boolean "a compile has run" file. The fingerprint could never prove the scripts still matched the source: mtime is not content, appDir was hash input (image path -> guaranteed mismatch), and the fallback compile() died under readOnlyRootFilesystem (bearsunday#483 impact 2 unfixed). The marker only records that a compile ran; freshness is the deploy's responsibility (ship the marker with the scripts, or recompile). factory() boots straight from CompiledInjector when the marker is present and lets a broken-scripts boot throw instead of silently recompiling -- a marker without working scripts is a deploy error, and a runtime recompile would also die on a read-only FS. When the marker is missing the cold path compiles on demand and emits E_USER_NOTICE so a forgotten bin/compile.php is loud. The marker is written after a successful compile (both the on-demand path in factory() and the AOT Compiler::compile() path), so a crashed compile leaves no marker and the next boot recovers via the on-demand path. deleteFiles() now removes dotfiles too (preserving the .placefolder dir sentinel) so a stale marker does not survive a test clean and short-circuit the next boot. Fixes bearsunday#483 (replaces the fingerprint approach in bearsunday#484).

@see

"Run bin/compile.php for production." assumed the 1.21+ skeleton compile entry, which older projects do not have (they use bear.compile or Compiler::fromInjector). Drop the command name from the E_USER_NOTICE and point to the production manual via @see instead, on both prodInjector() and CompileMarker.

@see

The notice now carries a direct link to the Compilation Recommended section, matching the existing cache-failure notice that embeds a URL. A developer who forgets to pre-compile gets the pointer at runtime, not only in the source @see.

Reuse AOT DI scripts when compile fingerprint matches

9f6bcd5

Skip unconditional Ray\Compiler recompile in PackageInjector::factory() when {scriptDir}/.compile-fingerprint still matches the source tree. Fixes ahead-of-time compile being discarded on first request (bearsunday#483).

koriym mentioned this pull request Jul 10, 2026

Ahead-of-time compiled DI scripts are discarded and rebuilt on the first request #483

Open

koriym added 4 commits July 11, 2026 02:37

cs-fix and cover CompileFingerprint / AOT reuse paths

5a009d9

Rename CompileFingerprint to CompileStamp

7d7ec13

Remove accidentally committed AGENTS.md

ed5d724

Fix Psalm ArgumentTypeCoercion on prodInjector scriptDir

5f8b130

koriym mentioned this pull request Jul 10, 2026

Migrate from bear.compile to Compiler::fromInjector #482

Open

3 tasks

NaokiTsuchiya reviewed Jul 11, 2026

View reviewed changes

koriym added 3 commits July 11, 2026 20:02

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Reuse AOT DI scripts when compile fingerprint matches#484

Reuse AOT DI scripts when compile fingerprint matches#484
koriym wants to merge 8 commits into
bearsunday:1.xfrom
koriym:fix/reuse-aot-di-scripts

koriym commented Jul 10, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Jul 10, 2026 •

edited

Loading

Review skipped

Uh oh!

codecov Bot commented Jul 10, 2026 •

edited

Loading

Uh oh!

NaokiTsuchiya left a comment

Uh oh!

NaokiTsuchiya Jul 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

koriym commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why not a fingerprint

Foundation (image reuse / read-only FS)

Test plan

Uh oh!

coderabbitai Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

codecov Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

NaokiTsuchiya left a comment

Choose a reason for hiding this comment

Uh oh!

NaokiTsuchiya Jul 11, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

koriym commented Jul 10, 2026 •

edited

Loading

coderabbitai Bot commented Jul 10, 2026 •

edited

Loading

codecov Bot commented Jul 10, 2026 •

edited

Loading