fix(tracing): finish active trace on crash

[jpnurmi]

Before, an in-flight trace was silently dropped on crash:

sentry-native/src/backends/sentry_backend_inproc.c

Lines 1147 to 1148 in 710def7

	// TODO(tracing): Revisit when investigating transaction flushing
	// during hard crashes.

Even with a (hypothetical) manual finish in the app right before a crash, the crash event still "floated" at the trace root rather than nested under the active span.

This PR auto-finishes the active span/transaction on crash with status aborted, and preserves scope->span / scope->transaction_object across the finish, so the trace reaches Sentry and the crash nests under it.

❌ Before (not finished, trace lost)	❌ Before (manual in-app force-finish)	✅ After (auto-finish)

For comparison:

• sentry-cocoa: sentry_finishAndSaveTransaction → finishForCrash
• sentry-java: UncaughtExceptionHint → forceFinish(ABORTED)

Android example

private fun crashWithNestedTrace() {
  val transaction =
    Sentry.startTransaction(
      "sentry-android.finish-on-crash",
      "demo.crash",
      TransactionOptions().apply { isBindToScope = true },
    )
  transaction.setTag("trace_finish_demo", "android")
  transaction.setData("expected", "sentry-android force-finishes this transaction on crash")

  fun openSpan(parent: ISpan, operation: String, description: String, delayMs: Long): ISpan {
    val span = parent.startChild(operation, description)
    SystemClock.sleep(delayMs)
    return span
  }

  val screen = openSpan(transaction, "ui.screen", "ErrorsScreen", 150)
  val click = openSpan(screen, "ui.action", "Crash with Nested Trace", 150)
  val work = openSpan(click, "demo.work", "leave nested spans unfinished", 150)
  val crash = openSpan(work, "demo.crash", "throw uncaught exception", 150)
  crash.setData("left_open", true)

  throw RuntimeException("Uncaught Exception with unfinished nested Sentry spans.")
}

https://sentry-sdks.sentry.io/issues/trace/80b71be4838c4130bd4954c2fe66ce13/?groupId=7506217367&node=span-8f25cb757ac94603&pageEnd=2026-05-26T17%3A50%3A07.509&pageStart=2026-05-25T17%3A50%3A07.509&project=5428559&query=is%3Aunresolved&referrer=issue-stream&source=issue_details&timestamp=1779774604.853

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

There are 4 total unresolved issues (including 2 from previous reviews).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6a2e71b. Configure here.}

[JoshuaMoelans]

Nice, LGTM! Small open Q on the trace-closing path now calling some SENTRY_WITH_SCOPE_MUT scope-flushes.

Also, maybe we want to add a note to our before_send_transaction that it too is potentially ran in a signal handler (similar to the ones we have for before_send/on_crash)

 * This function may be invoked inside a signal handler and must be safe for
 * that purpose, see https://man7.org/linux/man-pages/man7/signal-safety.7.html.
 * On Windows, it may be called from inside a `UnhandledExceptionFilter`, see
 * the documentation on SEH (structured exception handling) for more information
 * https://docs.microsoft.com/en-us/windows/win32/debug/structured-exception-handling

[JoshuaMoelans]

q: is it a problem that, during a crash, in here we do SENTRY_WITH_SCOPE_MUT (scope) {... which calls our backend scope flush function?

[jpnurmi]

Doesn't sound optimal to flush during crashes, but both out-of-process backends that have flush functions are guarded against flushing during a crash:

sentry-native/src/backends/sentry_backend_crashpad.cpp

Lines 282 to 286 in 710def7

	if (!data->event_path || data->crashed.load(std::memory_order_relaxed)
	|| !data->scope_flush.compare_exchange_strong(
	expected, true, std::memory_order_acquire)) {
	return;
	}

sentry-native/src/backends/sentry_backend_native.c

Lines 776 to 778 in 710def7

	if (sentry__atomic_fetch(&state->crashed)) {
	return;
	}