* fix(logs): Fix test failures when session tracking is enabledThis commit fixes 3 test failures in test_logs.c that occur whenauto-session tracking is enabled (the default):- basic_logging_functionality- formatted_log_messages- logs_disabled_by_defaultRoot causes and fixes:1. validate_logs_envelope counted all envelopes but only validated logs   - Session envelopes from sentry_close() were triggering assertions   - Fixed by filtering: only count/validate log envelopes, skip others2. formatted_log_messages didn't wait for batching thread to start   - Added sleep_ms(20) after sentry_init() to match other tests3. batching_stop flag wasn't reset between sentry_init() calls   - Once set to 1 during shutdown, subsequent startups would fail   - Fixed by resetting to 0 in sentry__logs_startup()These issues were discovered in console SDK testing where sessiontracking is enabled by default and tests run sequentially in a singleprocess.🤖 Generated with [Claude Code](https://claude.com/claude-code)Co-Authored-By: Claude <noreply@anthropic.com>* fix(logs): Eliminate thread startup race with enum state machineThis commit eliminates the thread startup race condition by:1. Replacing batching_stop with enum-based thread_state   - SENTRY_LOGS_THREAD_STOPPED (0): Not running   - SENTRY_LOGS_THREAD_RUNNING (1): Thread active and processing logs   - Improves code clarity and makes thread lifecycle explicit2. Thread signals RUNNING state after initialization   - Batching thread sets state to RUNNING after mutex setup   - Provides deterministic indication that thread is ready3. Adding test-only helper: sentry__logs_wait_for_thread_startup()   - Polls thread_state until RUNNING (max 1 second)   - Zero production overhead (only compiled with SENTRY_UNITTEST)   - Tests explicitly wait for thread readiness instead of arbitrary sleeps4. Updating shutdown to use atomic state transition   - Atomically transitions from RUNNING to STOPPED   - Detects double shutdown or never-started cases   - Returns early if thread wasn't runningBenefits:- Eliminates race where logs could be enqueued before thread starts- Tests are deterministic (no arbitrary timing assumptions)- Code is clearer with explicit state names- No production overhead (test helper is ifdef'd out)The one remaining sleep in basic_logging_functionality test is intentional- it tests batch timing behavior (wait for buffer flush).🤖 Generated with [Claude Code](https://claude.com/claude-code)Co-Authored-By: Claude <noreply@anthropic.com>* fix(logs): Fix thread lifecycle race condition causing memory leaksThis commit fixes a race condition in the logs batching thread lifecyclethat caused valgrind to report 336 byte memory leaks in CI tests.## ProblemWhen `sentry__logs_shutdown()` ran before the batching thread transitionedfrom initial state to RUNNING, the two-state model couldn't distinguishbetween "never started" and "starting but not yet running", causingshutdown to skip joining the thread.## Solution1. Added three-state lifecycle enum:   - STOPPED (0): Thread never started or shut down   - STARTING (1): Thread spawned but not yet initialized   - RUNNING (2): Thread active and processing logs2. Added `sentry__atomic_compare_swap()` primitive for atomic state   verification (cross-platform: Windows InterlockedCompareExchange,   POSIX __atomic_compare_exchange_n)3. Startup sets state to STARTING before spawning thread4. Thread uses CAS to verify STARTING → RUNNING transition   - If CAS fails (shutdown already set to STOPPED), exits cleanly   - Guarantees thread only runs if it successfully transitioned5. Shutdown always joins thread if old state != STOPPED## Benefits- Eliminates race condition where shutdown could miss a spawned thread- Thread explicitly verifies state transition with CAS- No memory leaks in any shutdown scenario- All 212 unit tests pass- All log integration tests passFixes test failures:- test_before_send_log- test_before_send_log_discard🤖 Generated with [Claude Code](https://claude.com/claude-code)Co-Authored-By: Claude <noreply@anthropic.com>* docs(logs): Address code review comments with clarifying documentationAddresses bot review feedback by adding documentation without changing behavior:1. **CAS memory ordering**: Added comment explaining sequential consistency   usage for thread state transitions and why it's appropriate for   synchronization2. **Condition variable cleanup**: Added note explaining that static storage   condition variables don't require explicit cleanup on POSIX and Windows3. **CAS function documentation**: Enhanced docstring to document memory   ordering guarantees and note that ABA problem isn't a concern for simple   integer state machines4. **Shutdown race handling**: Added comment explaining how the atomic CAS   in the thread prevents the race when shutdown occurs during STARTING stateAll changes are documentation/comments only - no behavior changes.🤖 Generated with [Claude Code](https://claude.com/claude-code)Co-Authored-By: Claude <noreply@anthropic.com>* Update src/sentry_logs.cCo-authored-by: JoshuaMoelans <60878493+JoshuaMoelans@users.noreply.github.com>---------Co-authored-by: Claude <noreply@anthropic.com>Co-authored-by: JoshuaMoelans <60878493+JoshuaMoelans@users.noreply.github.com>

…em ANSI code page. (#1389)* fix(win): make symbolication independent of the system ANSI code page.* allow `NULL` module paths and symbol names.* add an integration test that runs the example from a cyrillic directory and validates the package paths* resolve relative paths + clean up subdir* remove the assertion that a frame must have a function* only assert on the frame_package being a file if it exists......however, no longer assert that a frame_package exists.* isolate package assertions to new test* don't conflate checking any function/package with checking package file validity* also adapt the windows modulefinder to be independent system ACP.The szExePath generated for actual UTF-8 paths was already filled with mojibake :-) so LoadLibrary couldn't find any local modules.This is actually connected to the symbolication:* in the server, if a module was found, the backend would assign packages to frames based on the instruction address and the module address range* if the module couldn't be found, as was the case previously, it had to use the frame package providedSo, now we fixed both and they should overlap.* update the CHANGELOG* explicitly specify `PSAPI_VERSION` because we only want to link kernel32* check string_from_wstr return values* use a heap-allocated 32K buffer for module paths* use a heap-allocated 32K buffer for symbol paths* move allocation into wrapping if* format after webui edit