feat(native): add thread stackwalk mode option

[tustanivsky]

The crash daemon walks and symbolicates every thread of the crashed process. On applications with a high thread count this dominates the time needed to collect a crash.

sentry_options_set_thread_stackwalk_mode allows limiting client-side stackwalking to the crashed thread. Threads that did not crash are still reported with their id and name, only their stacktrace is omitted, so the thread list stays intact. On macOS the mode is also honored in the signal handler, which otherwise writes a stack copy per thread to disk.

Defaults to SENTRY_THREAD_STACKWALK_MODE_ALL, so existing behavior is unchanged.

Testing

Unreal Engine 5.8 editor crash on Windows, native backend with crash_reporting_mode=NATIVE_WITH_MINIDUMP, 134 threads and 1189 modules:

Phase	CRASHED_ONLY	ALL
Walk non-crashed threads	0 ms (0 threads)	2013 ms (133)
build_native_event total	548 ms	2557 ms
Crash processing, end to end	~2 s	~5 s
Envelope size	2.5 MB	2.7 MB

[github-actions]

	Messages
📖	Do not forget to update Sentry-docs with your feature once the pull request gets approved.

Generated by 🚫 dangerJS against 39ea8ae

[codecov]

Codecov Report

❌ Patch coverage is 76.19048% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.35%. Comparing base (dec0e5e) to head (39ea8ae).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2026      +/-   ##
==========================================
+ Coverage   74.24%   74.35%   +0.11%     
==========================================
  Files         104      104              
  Lines       26584    26603      +19     
  Branches     4838     4841       +3     
==========================================
+ Hits        19736    19781      +45     
+ Misses       5499     5472      -27     
- Partials     1349     1350       +1     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jpnurmi]

This makes me wonder what the use case is for client-side stackwalking outside the external crash reporter users. Given how expensive it is, what if we actually had a NONE mode and possibly even defaulted to that unless an external crash reporter is used?

I don't have the history of CRASHPAD_ENABLE_STACKTRACE for comparison/reference and especially why it's off by default, but I suspect it might be because of the DbgHelp and libunwind-ptrace dependencies. A NONE option would nicely leave the door open for turning those into optional runtime-loaded dependencies. 🤔

[tustanivsky]

what if we actually had a NONE mode and possibly even defaulted to that

Currently, we can switch the crash reporting mode to MINIDUMP to disable client-side stackwalking entirely so adding NONE feels a bit redundant as that would mean we could configure the same behavior via two different options.

what the use case is for client-side stackwalking outside the external crash reporter users

It allows us to still get a symbolicated stack trace for captured crashes on the event page even when debug symbols haven't been uploaded. For instance, this is a common scenario with the Unreal Editor where we deliberately skip the uploads for this target:

[jpnurmi]

what if we actually had a NONE mode and possibly even defaulted to that

Currently, we can switch the crash reporting mode to MINIDUMP to disable client-side stackwalking entirely so adding NONE feels a bit redundant as that would mean we could configure the same behavior via two different options.

Seems confusing and unintuitive because with Crashpad, client-side stackwalking specifically applies to minidumps. If Native's client-side stackwalking was later implemented for minidumps, would NONE then make sense? 🤯

what the use case is for client-side stackwalking outside the external crash reporter users

It allows us to still get a symbolicated stack trace for captured crashes on the event page even when debug symbols haven't been uploaded. For instance, this is a common scenario with the Unreal Editor where we deliberately skip the uploads for this target:

Ah, good to know, thanks! 🙏

[tustanivsky]

If Native's client-side stackwalking was later implemented for minidumps, would NONE then make sense?

Yes, if stackwalking becomes supported in Native's minidump mode (which should make its behavior similar to Crashpad's CRASHPAD_ENABLE_STACKTRACE configuration), having a way to opt out of it there could be useful 🤔

I'm still not sure whether NONE should be the default (if/once it's available) as it seems to defeat the whole purpose of the native crash reporting mode.

[jpnurmi]

I'm still not sure whether NONE should be the default (if/once it's available) as it seems to defeat the whole purpose of the native crash reporting mode.

My bad, I went too far with the NONE-as-default proposal. I didn't consider the scenario where the server wouldn't have symbols available. So an opt-out might make sense, but I don't think anymore that we necessarily need to add it to this PR at all. How about we just make the sentry_thread_stackwalk_mode_t values start from 1 so that we reserve the spot for a potential SENTRY_THREAD_STACKWALK_MODE_NONE=0 for later?