getRoot() now retries up to 5 times with session verification to handle
race conditions when multiple paratest workers initialize simultaneously.
Previously, if account creation or session creation failed under load,
all subsequent test requests would fail with 401.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Stale in-memory project documents in ScheduleBase (and request-scoped
copies in api.php/general.php) were overwriting current DB state when
updateProjectAccess triggered. Because Database::updateDocument uses
array_merge with the passed document taking priority, cached projects
missing recent OAuth provider changes would silently disable them.
Now fetches a fresh project document from the DB before writing, so only
accessedAt is updated without clobbering other fields.
Replace raw string interpolation with escapeshellarg() for all
arguments passed to exec/shell_exec calls that build git commit,
gh pr create, gh api, and gh release commands. This prevents
shell injection from AI-generated changelog text or any other
dynamically constructed values.
Add a programmatic guard after parsing the AI response that rejects
major bumps or versions >= 1.0.0 for beta SDKs. When triggered, the
SDK is skipped with a warning instead of proceeding with an invalid
version.
Remove the manual commit message prompt. When AI is available and
produces a changelog, use it as the commit message. Otherwise fall
back to a descriptive message based on SDK name and version. A
manually provided --message flag still takes priority.
Replace the static supportedSDKS array with a getSupportedSDKs() method
that reads SDK keys from the sdks.php config file. This eliminates the
need to maintain the list in two places.
Beta SDKs (version < 1.0.0) should not be bumped to 1.0.0. The prompt
now checks the beta flag from sdk config and instructs the AI to use
minor bumps for both breaking changes and new features, and patch for
bug fixes only. Also adds a rule to wrap code identifiers in backticks
for better changelog rendering.
Realtime E2E tests were taking 24+ minutes due to intentional timeout waits. Many tests verify event filtering by expecting a TimeoutException, and each was waiting 45 seconds.
Changes:
- Reduce default WebSocket timeout from 45s to 2s in RealtimeBase
- Add optional timeout parameter to getWebsocket() methods
- Use longer timeouts (5-10s) for tests that legitimately wait for slow operations (function executions, test channel events)
- Use named parameters for improved readability
Performance impact:
- Before: 24:07 minutes (1,447 seconds)
- After: 1:09 minutes (69 seconds)
- Speedup: ~21x faster
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Remove async coroutine wrapper from event dispatch to simplify execution model and improve trace hierarchy. Listeners now execute synchronously in the caller's context, with dependency resolution inlined.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Introduce a generic event bus (Utopia\Bus) with typed events, listener
base class, Span instrumentation, and coroutine dispatch. Replace all
direct queueForExecutions and inline execution usage calls with
ExecutionCompleted event and dedicated listeners (Log, Usage).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
fogelito
Jake BarnbyandClaude Opus 4.6
Chirag AggarwalandGitHub
premtsd-codeandGitHub
Matej BačoandGitHub
Matej BačoandGitHub
Luke B. SilverandGitHub