optimizely / swift-sdk / 28870957209

94%

Build Type

push

github

Committed by web-flow

Commit Message

[FSSDK-12729] fix: prevent EXC_BAD_ACCESS from premature URLSession deallocation (#649)

* fix: move URLSession.finishTasksAndInvalidate() into async callbacks

The defer block fires at synchronous scope exit — before the async
URLSession callback runs — deallocating the session while the request
is still in-flight. This causes EXC_BAD_ACCESS, especially when
Firebase Performance Monitoring swizzles URLSession delegate methods.

Move session.finishTasksAndInvalidate() from the outer defer into the
completion callback body (or inner defer for ODP modules) so the
session stays alive until the response is fully processed.

Affected modules:
- DefaultDatafileHandler.downloadDatafile()
- DefaultEventDispatcher.sendEvent()
- OdpEventApiManager.sendOdpEvents()
- OdpSegmentApiManager.fetchSegments()

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* test: fix race in MockUrlSession session counter

MockUrlSession used lock.async for both incrementing (init) and
decrementing (finishTasksAndInvalidate) validSessions. With the
URLSession fix moving invalidation into the async callback, the
decrement could still be pending when tearDown asserts the counter
is zero.

Change lock.async to lock.sync so counter updates complete before
the caller proceeds, eliminating the race between finishTasksAndInvalidate
and the test assertion in tearDown.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: address review feedback from Copilot

- DefaultDatafileHandler: call completionHandler with failure result
  when getRequest returns nil, preventing caller deadlock on
  DispatchGroup.wait()
- MockUrlSession: make lock static so all instances share a single
  serial queue for the static validSessions counter

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Coverage Stats

1 of 1 new or added line in 1 file covered. (100.0%)

1 existing line in 1 file now uncovered.

1131 of 1205 relevant lines covered (93.86%)

8933.59 hits per line