fix(tracing): finish active trace on crash by jpnurmi · Pull Request #1667 · getsentry/sentry-native

jpnurmi · 2026-04-23T17:01:16Z

Before this PR, an in-flight trace was silently dropped on crash. And even with a manual sentry_transaction_finish right before the crash, the crash event still floated at the trace root rather than nesting under the active span — sentry_transaction_finish clears scope->span / scope->transaction_object, so by the time the crash handler captures its event, scope-apply falls through to the stale propagation context.

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 ^{(manual finish)}	After ^{(auto-finish)}

For comparison:

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

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.}

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The prior crash-path trace finish only closed `scope->span` (the deepest active span) and its root transaction. Intermediate spans between them — e.g. Qt event dispatch handlers nested above a crashing slot — were never finished, never added to `tx.spans`, and ended up orphaning the crash event at the trace root in Sentry's UI. Track every span on its root transaction under `children_mutex` at `sentry__span_new`, deregister on normal finish, and drain the list in leaf-first order inside `sentry__trace_finish`. Matches sentry-cocoa's `SentryTracer` `_children` + `finishForCrash` and sentry-java's `SentryTracer.forceFinish`. Preserve `scope->span` / `scope->transaction_object` across the drain so the subsequent crash event inherits the full trace context from scope (mirrors cocoa's `finishTracer:shouldCleanUp:NO`). Without this, the crash event fell through to the stale propagation context and Sentry could not nest it under the active span chain. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Trim WHAT-narration comments in sentry__trace_finish and its tests, keeping the sentry-cocoa alignment anchor and the non-obvious "detach to skip remove-scan" rationale. Warn on the tracking-list allocation failure so a silent crash-finalize gap is audible. Use the typedef name for the forward declaration in sentry_sampling_context.h for consistency with the rest of the codebase. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Split sentry__trace_finish into save_active_trace / restore_active_trace (as a pair around the scope mutation) and finish_children for the atomic children swap + finish loop, so the top-level function reads as a short narrative. No behavior change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

sentry_transaction_finish_ts consumes its argument, so the ref that save_active_trace took for active_tx is released by the finish call. restore_active_trace was decref'ing it again. Harmless in crash context where the process exits, but a real refcount underflow otherwise. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

sentry_span_finish_ts consumes its argument (decref at sentry_core.c:1607 on success, :1611 on fail), so the explicit sentry__span_decref after the finish call released the children-list ref a second time. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The transaction's `children` list held an incref'd reference on each live span, while each span holds a ref on its transaction via `span->transaction`. Any code path that decref'd a span without going through `sentry_span_finish_ts` (e.g. the `span_data` unit test using the low-level `sentry__span_new` / `sentry__span_decref` API) left both sides stuck at refcount 1, leaking both. Make `tx->children` weak: `sentry__span_new` no longer increfs on add, and `sentry__span_decref` unlists the span from the children list on its final drop. `sentry__transaction_remove_child` correspondingly no longer decrefs. The drain path in `sentry__trace_finish` continues to work — the spans on the swapped-out list are alive via their other refs (scope / saved_span / user var), and `sentry_span_finish_ts` consumes one of those refs as the caller's, exactly as in the non-crash flow. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

After the weak-children refactor, `finish_children` handed each span to `sentry_span_finish_ts` without a caller ref. `finish_ts` consumes one ref on its argument, so when the drained span had only the user's own ref outstanding, finish_ts would drop it to zero and free the span — leaving the user's pointer dangling. Incref inside the drain loop so finish_ts consumes "our" ref, leaving user refs intact. Also update the `trace_finish` unit test to release the refs it held on `tx`/`child`/`grand` (without the children-list strong ref, unfinished spans now properly leak if the user forgets to decref). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

SENTRY_WITH_SCOPE_MUT calls sentry__scope_flush_unlock on exit, which invokes the backend's flush_scope_func — file I/O in both crashpad and native backends. sentry__trace_finish runs from signal-handler context in the inproc/breakpad/native crash handlers, so the flush is unsafe there. Use the NO_FLUSH variant the codebase provides for this exact situation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Call restore_active_trace before the early return so any refs taken by save_active_trace are released even if active_tx is NULL. No-op under the current invariant (saved_span->transaction is always non-NULL), but defensive if the invariant ever changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

This reverts commit b0c1aca.

…inish

cursor Bot reviewed Apr 23, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

sentry Bot reviewed Apr 23, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

cursor Bot reviewed Apr 23, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

cursor Bot reviewed Apr 24, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

cursor Bot reviewed Apr 24, 2026

View reviewed changes

Comment thread src/sentry_tracing.c Outdated

sentry Bot reviewed Apr 24, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

jpnurmi mentioned this pull request Apr 24, 2026

fix(native): Skip scope flush during crash handling #1668

Merged

jpnurmi force-pushed the jpnurmi/fix/trace-finish branch from eaaeff4 to e78541b Compare April 24, 2026 09:16

sentry Bot reviewed Apr 24, 2026

View reviewed changes

Comment thread src/sentry_core.c

jpnurmi force-pushed the jpnurmi/fix/trace-finish branch from 1764758 to 570ed57 Compare April 24, 2026 15:35

jpnurmi changed the title ~~WIP: fix(tracing): finish active trace on crash~~ fix(tracing): finish active trace on crash Apr 25, 2026

jpnurmi force-pushed the jpnurmi/fix/trace-finish branch from 570ed57 to 264a69b Compare April 27, 2026 08:13

cursor Bot reviewed Apr 27, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

cursor Bot reviewed Apr 27, 2026

View reviewed changes

Comment thread src/sentry_tracing.c

jpnurmi changed the title ~~fix(tracing): finish active trace on crash~~ WIP: fix(tracing): finish active trace on crash Apr 27, 2026

sentry Bot reviewed Apr 27, 2026

View reviewed changes

Comment thread src/backends/sentry_backend_breakpad.cpp Outdated

cursor Bot reviewed Apr 27, 2026

View reviewed changes

Comment thread src/backends/sentry_backend_breakpad.cpp Outdated

Comment thread src/sentry_envelope.c Outdated

Comment thread src/backends/native/sentry_crash_daemon.c

cursor Bot reviewed Apr 27, 2026

View reviewed changes

Comment thread src/backends/sentry_backend_inproc.c

Comment thread src/backends/sentry_backend_native.c

jpnurmi and others added 12 commits May 20, 2026 16:16

fix(tracing): finish active trace on crash

fe461ad

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Revert "fix(tracing): Use NO_FLUSH scope macro in restore_active_trace"

257d2de

This reverts commit b0c1aca.

test: isolate native trace crash uploads

ea9dce5

jpnurmi added 7 commits May 20, 2026 16:35

fix: avoid live transport for crash trace finish

810bc78

Revert test changes

ab07daf

fix: dump crash trace transactions separately

39f28f3

Update CHANGELOG.md

3abfaff

ref: clean up unused function

8847ef8

fix: preserve transaction root trace when merging scope

e879be9

make format

336c007

jpnurmi force-pushed the jpnurmi/fix/trace-finish branch from 6a2e71b to 336c007 Compare May 21, 2026 05:15

Merge remote-tracking branch 'origin/master' into jpnurmi/fix/trace-f…

f5536fb

…inish

jpnurmi changed the title ~~WIP: fix(tracing): finish active trace on crash~~ fix(tracing): finish active trace on crash May 22, 2026

sentry Bot reviewed May 22, 2026

View reviewed changes

Comment thread src/backends/sentry_backend_inproc.c

decref transactions

30008f0

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(tracing): finish active trace on crash#1667

fix(tracing): finish active trace on crash#1667
jpnurmi wants to merge 21 commits into
masterfrom
jpnurmi/fix/trace-finish

jpnurmi commented Apr 23, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

jpnurmi commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

jpnurmi commented Apr 23, 2026 •

edited

Loading