regression: bash crashes intermittently, sometimes on startup

[un_spacyar]

After updating to hrev58021 (32 bit version), when the booting process is about to end (all the boot icons are on), instead of load the Desktop, it lands in the debugger, with the following message:

Debugger for team 107 on interface local (shredder)created and initialized sucessfully! [Thread stopped: 107 "sh"]

If I boot in safe mode, the issue is not reproducible.

[waddlesplash]

Please type "save-report", then reboot back into safe mode and upload the saved crash report here.

[waddlesplash]

(And hit enter after typing "save-report" and wait for Debugger to finish saving the report of course.)

[un_spacyar]

Thank you. After typing 'save-report' and wait for a few seconds, the Desktop appeared (no need to restart). However, I do not see any crash report in the Desktop.

[waddlesplash]

Check the syslog at least, it may have a backtrace (or backtraces.)

[diver]

IIRC 'save-report' save to /boot/home

[madmax]

Happened to me twice in a row today. hrev58021 32bits, QEMU KVM.

[pulkomandy]

I saw it as well. I tried to type "bt" to investigate, but nothing happened, so I have no idea where it crashed

[waddlesplash]

Well, there's a backtrace in the attached syslog if nothing else.

It looks like Bash was last recompiled back in January. The crash appears to be in something related to stack unwinding. Perhaps we just need to rebuild with the fixed GCC 13?

[waddlesplash]

Ah, but bash is still built with the legacy GCC, so that's not likely to be the issue.

[waddlesplash]

This appears to be reproducible when rebuilding the whole HaikuPorts repository: delete the "repository" folder in a HaikuPorts checkout, then run HaikuPorter to regenerate it from all the recipes, and (at least for me) there will be a few Bash crashes along the way.

I tried re-running bash on just some of the recipes that crashed, but it didn't reproduce then. So whatever this is appears to be somehow intermittent.

[waddlesplash]

The crash appears to be much more reproducible with the debug allocator. I got repeated crashes in atomic_test_and_set inside heap_allocate_from_bin with HaikuPorter when using it.

[pulkomandy]

Because there is a function called "unwind" in the stacktrace doesn't mean it is stack unwinding. This one comes from bash, and as far as I can see it's pretty boring C code managing some kind of linked list: ​https://github.com/gitGNU/gnu_bash/blob/master/unwind_prot.c

[waddlesplash]

Fixed in hrev58035 +beta5. This also seems to fix the Git crashes as well.