GLFW crashes entire julia session after re-attaching in tmux

[lassepe]

Migrating from MakieOrg/Makie.jl#3092

I frequently use GLMakie from within a remote tmux ssh session via x-forwarding. That generally works quite well; however, it does not survive re-attaching. Indeed, the issue can be reproduced with a plain GLFW window:

Steps to reproduce:

1. Connect to a ssh remote, e.g., ssh <user@hostname>
2. run tmux on the remote
3. Start a Julia REPL on the remote, run using GLFW; GLFW.Window()
4. Detach from the tmux session using CTRL+b, d
5. Disconnect from ssh client using exit
6. Re-connect to ssh client ssh <user@hostname>
7. Re-attach to tmux session tmux -u attach-session
8. Open another window GLFW.Window().

Step 8 here fails reliably with

XIO:  fatal IO error 2 (No such file or directory) on X server "localhost:10.0"
      after 212 requests (212 known processed) with 1111 events remaining.

Note: this error is slightly different from the one originally reported in MakieOrg/Makie.jl#3092 (here we now have "No such file or directory").

I verified that the DISPLAY variable is set correctly upon re-attaching. I suspect, however, that maybe some socket is moved/re-created upon re-attaching which invalidates some GLFW internal state? In fact, if I start a fresh Julia session and load GLFW after re-attaching, things work as expected.

Thoughts/Questions

1. Ideally, GLFW should be robust to this kind of re-attaching.
2. Even if GLFW itself cannot recover, it would be desirable to not kill the REPL upon this error.
3. As a temporary workaround, is there some kind of "reset"-function that I can call to reset the internal state of GLFW upon re-attaching?

[SimonDanisch]

You should open this with the C GLFW repository, since here it's basically just me reading the issue^^

[lassepe]

Thanks, I have filed an issue in the C GLFW repository. Irrespective of the upstream fix, would it be possible to handle this issue more gracefully on the Julia end (somehow catch the error instead of closing the REPL?)