Critical: Repeated Bun bridge crashes (SIGILL segfault) since v3.5.5-beta

[Zutyosh]

Summary

Since upgrading to version 3.5.5-beta, the Empryo desktop application experiences repeated crashes during normal usage. At least 7 distinct incidents occurred on a single day of use (2026-07-29), with 3 confirmed Bun bridge process crashes showing identical SIGILL segmentation fault signatures.

Environment

Component	Value
OS	CachyOS (Arch-based)
Kernel	Linux 7.1.5-1-cachyos
Installation Method	Arch Linux package
Application Build	Empryo Desktop 3.5.5-beta
Electron	43.2.0
Bridge Runtime	Bun v1.3.14 (0d9b296a)
System RAM	33.57 GB
Peak RSS at Crash	≤ 0.63 GB (no memory pressure)

Impact

• Application becomes completely unusable without restart
• Sessions interrupted during normal navigation (tab/workspace switching)
• No user action triggers the crash predictably
• Work/files remain untouched but workflow is disrupted

Incident Timeline (2026-07-29)

#	Time	Type	Status
1	~12:03 → 13:55	Complete app restart (no proper shutdown sequence)	Probable
2	14:21 → 14:23	Renderer reload + "picker was stuck invisible" warning	Probable
3	15:03:53	Bridge segfault (Bun), signal=SIGILL	✅ Confirmed
4	15:33:17	Bridge segfault (Bun), signal=SIGILL	✅ Confirmed
5	15:39 → 15:52	Renderer reload (~12 min after opening)	Probable
6	~18:15	Incomplete shutdown sequence	To verify
7	19:11:03	Bridge segfault (Bun), signal=SIGILL	✅ Confirmed

Confirmed Crash Details

All 3 confirmed Bun crashes return the exact same hash, indicating a deterministic/reproducible bug rather than random memory corruption:============================================================
Bun v1.3.14 (0d9b296a) Linux x64
Linux Kernel v7.1.5 | glibc v2.44
CPU: sse42 popcnt avx avx2
panic(main thread): Segmentation fault at address 0x0
oh no: Bun has crashed. This indicates a bug in Bun, not your code.

Crash Report

Bridge Lifetime Before Crash

• Crash 1 (15:03:53): ~40 minutes after bridge startup (cwd: Git/PNW-SS)
• Crash 2 (15:33:17): ~29 minutes after bridge startup (cwd: Git/PNW-SS)
• Crash 3 (19:11:03): ~7 minutes after bridge startup (cwd: Git/Cyberia)

Key observation: Crashes occur across different workspaces, suggesting this is not project-specific but a systemic runtime issue.

Reproducibility

No specific user action reliably triggers the crash. It occurs during normal navigation (tab switching, workspace changes) after variable durations (7-40 minutes) of bridge runtime.

User Interface Evidence

Screenshot shows the "Engine disconnected" dialog that appears when the bridge process terminates unexpectedly:

• Error: bridge exited code=null signal=SIGILL
• User must manually reconnect to resume work

Steps Taken

• Full application reinstall attempted
• Memory pressure ruled out (plenty of available RAM at crash time)
• Debug bundle generated via app menu (available upon request)
• Full session logs captured covering multiple days with all 3 complete crash traces

Expected Behavior

Bridge process should remain stable throughout normal application usage without unexpected terminations.

Additional Notes

• This issue appeared specifically after upgrading to 3.5.5-beta
• Previous versions did not exhibit this behavior
• SIGILL (Illegal Instruction) suggests potential CPU instruction compatibility issue or memory corruption in the Bun runtime

[proxysoul]

Investigated this. The crash matches a known Bun FFI bug in JSCallback({ threadsafe: true }), which is used by the FFF file watcher.

Related upstream reports:

• bun:ffi JSCallback invoked from different thread crashing after a while oven-sh/bun#24529
• Segfault when native code repeatedly invokes JSCallback({ threadsafe: true }) oven-sh/bun#28113
• bun:ffi: threadsafe JSCallback called from a non-Worker OS thread crashes with heap corruption (SIGSEGV/SIGTRAP) oven-sh/bun#33840

Upstream fixes:

• ffi: avoid copying the threadsafe callback wrapper on the calling thread oven-sh/bun#31332
• Fix four crash/correctness bugs: node:vm link(), Worker name, FFI threadsafe callbacks, sliced Bun.file oven-sh/bun#34140

The fixes are merged, but Bun 1.3.14 does not include them. Empryo embeds Bun 1.3.14 inside the desktop bridge, so upgrading the user's local Bun will not help.

I am weighing two options:

1. Rebuild the bridge with a pinned Bun 1.4 canary containing the fixes.
2. Wait for Bun 1.4 stable and use a temporary watcher fallback.

I will update this after verifying the patched runtime on Linux x64.

[proxysoul]

Update: a hotfix is implemented.

The crash is a Bun 1.3.14 runtime bug in threadsafe FFI callbacks, triggered by the native file watcher of our search engine (FFF). We reproduced it in a Linux container within seconds on Bun 1.3.14 and confirmed the same workload survives on Bun 1.4, which contains the upstream fixes (oven-sh/bun#31332, oven-sh/bun#34140).

The hotfix: Empryo no longer subscribes to the FFF native watcher when running on Bun older than 1.4 and falls back to the OS file watcher instead. File search, grep and pickers are unaffected. This removes the crash entirely. The only cost is slightly less granular sidebar refresh for edits made outside Empryo on Linux.

This is temporary. Once Bun 1.4 stable ships we will move the bridge runtime to it and re-enable the native watcher.

The fix will be part of release 3.5.6-beta, coming soon.

[Zutyosh]

Thank you, I appreciate all of your hard work; it's truly incredible!

[proxysoul]

I am in contact with FFF's author, he is looking into an upstream fix! in the meantime a new patch release is cooking! should have the workaround "skip" of FFF's file watcher, fallback is slower but it'll do for now 🫡 Time to sleep! work tomorrow 😂 (release will be published soon)