glib bad system call

[synkathairo]

Executing glib-2.86.3-r0.trigger
* Bad system call
ERROR: lib/apk/exec/glib-2.86.3-r0.trigger: exited with error 159

using stable app store version of ish (1.3.2 build 494) with latest v3.23 https://dl-cdn.alpinelinux.org/alpine/v3.23/ and installing glib or upgrading using apk upgrade -U && apk fix

[strike20023]

The main difficulty is that iSH does not yet fully emulate the Linux syscalls, memory layout, TLS, and dynamic linking that glibc relies on. Musl is simpler, so iSH can support it more reliably at this stage.

[emkey1]

There are also a lot of issues with /proc, centered around various race conditions due to a lack of proper locking.

[strike20023]

Thanks for pointing this out — I really appreciate it.

I’ve been experimenting a bit with trying to get an arm64 version of iSH working (mostly using tools like Codex), and I’m still very much learning as I go. Issues like this around /proc and synchronization are honestly quite challenging for me to fully understand yet, so this kind of feedback is really helpful.

My rough intuition is that moving to an arm64 guest might make it possible to support more things (e.g., better compatibility with glibc-based programs, or improving some networking-related behavior like ifconfig), but I’m not very experienced with systems-level details, so I could definitely be missing important constraints.

I’m just a CS student learning by experimenting, so any pointers or explanations would be greatly appreciated.

[emkey1]

You might take a look at my fork of iSH, it's called iSH-AOK...

https://github.com/emkey1/ish-AOK

It started out a few years ago with me doing all the programming and has more recently been revived with the help of Codex. It runs Debian12 and Alpine 3.23.x reasonably well and I'm thinking of tackling 64 bit next.

My biggest advice in regards to Codex is the lean on it to create regression tests for EVERYTHING and be sure they get run regularly, but always before new releases. Another project of mine is something I call PSCAL, which was 99.9% AI written over the past year plus. It's tens of thousands of lines at this point and I never would have gotten this far without a combination of constant advances in the state of the art and LOTS of regression testing.

https://github.com/emkey1/pscal

[strike20023]

Thanks for sharing — really appreciate it.

I’m mostly just experimenting a bit on my side and still trying to understand how all these pieces fit together, so this is very helpful.

I’ll take a look at your fork. Thanks again!