Skip to content

0.1.6

Choose a tag to compare

View all tags
@ichinya ichinya released this 17 Oct 05:16
· 2 commits to docs/anonymous-case-01-benchmark since this release
0.1.6
d6438a7
This commit was signed with the committer’s verified signature.
ichinya Ichi
SSH Key Fingerprint: kD8o1XWkHGSHQ/SWo4jmiArJ59QznppixPIjxSH868U
Verified
Learn about vigilant mode.

Changes since 0.1.5

  • Skip impossible argument shapes before expensive physical-source parsing in the possible-callback-tuple, guarded-local-model-tuple and model-property argument filters. The shortcut can only retain a diagnostic; suppression still requires the existing physical-source and native-metadata proofs. Parenthesized, commented, dynamic and unknown syntax keeps the full proof path.
  • Add handler CPU profiling, reproducible benchmark tooling, parser-based shape controls and public analyzer-depth witnesses.
  • Strengthen benchmark integrity: rediscover the complete source path/hash set after each run, verify full diagnostic multisets, retain malformed-output failures and reject incomplete resume state. Stop elapsed timing immediately after the analyzer exits.

Measured result

On the documented anonymous Laravel workload, three counterbalanced Linux runs reduced median elapsed time from 308.84 s to 236.68 s (23.37%), and median CPU from 539.73 s to 411.24 s (23.81%). All 13,196 complete diagnostic records remained equal, including duplicates. This is local, workload-specific evidence on a shared workstation, not a universal speed claim. The optimized run still took 4.89 times the Larastan level-10 run; their check sets differ.

CPU profile, repeated measurements and limitations

Validation

  • Parser-based syntax oracle: 79 controls.
  • Four focused native suites, including genuine diagnostic removals, complete residual diagnostics, source/SDK mutation and one/three-worker controls. Three suites used a test-only 600,000 ms IPC request deadline after an unmodified local control exceeded 120,000 ms; assertions and production source were unchanged.
  • Fresh native/candidate depth follow-up: 132 marked observations across 33 defect/control pairs, matching the archived matrices. The PHPStan levels 0-10 matrix is an archived reference, not a newly rerun matrix.
  • Cache controls: 10 syntax assertions, 28 query assertions and 62 consumer controls.
  • Benchmark protocol: 13 tests, including five Linux-runner tests with synthetic analyzers.
  • PHP syntax checks and strict Composer manifest validation.

Full composer check and hosted analyzer CI were not run. Failed pilots and the rejected cache-compression experiment are documented and excluded from accepted timing samples; production syntax-cache behavior is unchanged.

Update

composer require --dev ichinya/laramago:^0.1.6

Full comparison from 0.1.5

Assets 2
Loading