Repository navigation
feat(native): Provide crash metadata to on_crash - #2174
Merged
Merged
Conversation
|
![sentry[bot]](https://avatars.githubusercontent.com/in/12637?s=60&v=4){kind=small}
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## v1 #2174 +/- ##
==========================================
- Coverage 75.65% 75.00% -0.65%
==========================================
Files 103 103
Lines 28707 28902 +195
Branches 5294 5338 +44
==========================================
- Hits 21718 21679 -39
- Misses 5624 5854 +230
- Partials 1365 1369 +4 🚀 New features to boost your workflow:
|
mujacica
approved these changes
Oct 8, 2026
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit bcf18cf. Configure here.
jpnurmi
force-pushed
the
jpnurmi/feat/native-crash-stack
branch
from
October 8, 2026 18:46
bcf18cf to
59a96d8
Compare
jpnurmi
added this pull request to stack #2181
October 8, 2026 18:47
jpnurmi
force-pushed
the
jpnurmi/feat/native-crash-stack
branch
from
October 9, 2026 07:49
59a96d8 to
fda25c6
Compare
Provide exception metadata, registers, and a basic crashing-thread stack for crash identification in native `on_crash`, matching `inproc`. Use the same payload for `before_send` when `on_crash` is unset. - Capture up to 128 fault-context IPs and unwind source in shared memory. Retain the faulting IP when available. - Linux: SDK `libunwind` with signal-safe logging. - macOS: SDK frame-pointer walker, guarded Mach reads, PAC stripping, `__DARWIN_OPAQUE_ARM_THREAD_STATE64` accessors, and `instruction_addr_adjustment=none`. - Android/Linux without `libunwind`: bounded frame-pointer capture via `process_vm_readv`; avoids `libunwindstack`'s C++ allocation. - Windows: MSVC x64 `RtlLookupFunctionEntry`/`RtlVirtualUnwind` walking without DbgHelp; SP progress permits recursive frames. Frame-pointer capture on x86/ARM64 and other x64 builds. - ARM32 frame-pointer fallbacks retain only the faulting IP. The daemon reuses partial captures and enriches existing frames. Preserve exception/register/frame edits and removals, and honor callback rejection.
jpnurmi
force-pushed
the
jpnurmi/feat/native-crash-stack
branch
from
October 15, 2026 14:00
800ead5 to
69b83ca
Compare
jpnurmi
removed this pull request from stack #2181
October 15, 2026 14:06
jpnurmi
added this pull request to stack #2209
October 15, 2026 14:07
jpnurmi
added a commit
that referenced
this pull request
Oct 16, 2026
The recently introduced crashed-thread stack unwinding in #2174 switched the native signal handler to the SDK unwinder. This brought its 4 KiB `/proc/self/maps` read buffer onto the handler stack as an unintended side effect. Crash handlers can run on alternate signal stacks as small as 16 KiB, as noted in #2029, which added stack monitoring for #997. The inproc backend preserves existing alternate stacks, so the SDK's 64 KiB default is not a universal budget. The signal frame, callbacks, and handler work must fit; overflowing can prevent the original crash from being reported. Reduce the buffer to 512 bytes, preserving support for oversized lines and address prefixes split across reads without adding allocations in the signal handler. With GCC 15 on Linux x86-64, both 512-byte and 256-byte buffers reduce the native handler peak from 6,344 to 3,424 bytes and the inproc peak from 7,592 to 4,672 bytes. The 2,920-byte reduction is about 18% of a 16 KiB stack budget. Warm timings for two maps lookups (code and stack addresses), using GCC 15 `-O3` on a Ryzen AI 9 HX 370, were: 4096 B 512 B 256 B GNOME Shell 0.406 ms 0.490 ms 0.583 ms Edge 0.776 ms 0.972 ms 1.288 ms These are medians of nine rotating batches, excluding stack unwinding and the rest of crash handling. Choose 512 bytes because it preserves the measured handler stack reduction with less read overhead than 256.
jpnurmi
added a commit
that referenced
this pull request
Oct 16, 2026
#2174 brought the SDK unwinder's 4 KiB `/proc/self/maps` buffer onto the native signal-handler stack. Use 512 bytes; preserve oversized lines and address prefixes split across reads without new allocations. Inproc preserves existing alternate stacks, which may have only 16 KiB for the signal frame, callbacks, and handler, despite the 64 KiB default. Overflow can lose the report (#997, #2029). GCC 15 `-O3`, Linux x86-64, Ryzen AI 9 HX 370: 4096 B 512 B 256 B Native handler peak (B) 6,344 3,424 3,424 Inproc handler peak (B) 7,592 4,672 4,672 GNOME maps lookups (ms) 0.406 0.490 0.583 Edge maps lookups (ms) 0.776 0.972 1.288 512 saves 2,920 bytes (~18% of 16 KiB), matching 256's handler peaks with less read overhead. Lookup times are warm code/stack lookups: medians of nine rotating batches, excluding unwinding and other crash handling.
jpnurmi
added a commit
that referenced
this pull request
Oct 16, 2026
#2174 brought the SDK unwinder's 4 KiB `/proc/self/maps` buffer onto the native signal-handler stack. Use 512 bytes; preserve oversized lines and address prefixes split across reads without new allocations. GCC 15 `-O3`, Linux x86-64, Ryzen AI 9 HX 370: | Measurement | 4096 B | 512 B | | --- | ---: | ---: | | Native handler peak (B) | 6,344 | 3,424 | | Maps lookups, 190 KiB / 1,900 mappings (ms) | 0.406 | 0.490 | | Maps lookups, 308 KiB / 3,053 mappings (ms) | 0.776 | 0.972 | This saves 2,920 bytes of handler stack at the cost of more reads. Lookup timings are warm code/stack lookup pairs: medians of nine rotating batches, excluding unwinding and other crash handling.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Reuse and extend the existing SDK unwinders already used by
inprocto provide exception metadata, registers, and a basic crashing-thread stack for crash identification in nativeon_crash, matchinginproc.libunwindwith signal-safe logging.__DARWIN_OPAQUE_ARM_THREAD_STATE64accessors, andinstruction_addr_adjustment=none.libunwind: bounded frame-pointer capture viaprocess_vm_readv; avoidslibunwindstack's C++ allocation.RtlLookupFunctionEntry/RtlVirtualUnwindwalking without DbgHelp; SP progress permits recursive frames. Frame-pointer capture on x86/ARM64 and other x64 builds.The daemon reuses partial captures and enriches existing frames. Preserve exception/register/frame edits and removals, and honor callback rejection.
Ref: #2102