Makie started segfaulting quite regularly :(

[SimonDanisch]

Making a tracking issue for segfaults on julia 1.8.3+ to 1.9beta2, which have been happening quite frequently lately!

CairoMakie benchmark CI: https://github.com/MakieOrg/Makie.jl/actions/runs/3937490334/jobs/6735073575

wsl2 + linux + 1.9beta2 (this one always reproduces):

Tyler.jl + Linux segfaults on 1.8 while building docs (likely flaky and maybe HTTP.jl related): https://github.com/MakieOrg/Tyler.jl/actions/runs/3937793009/jobs/6735719538#step:7:87

Then there's #2585
I've also heard from other users an increase in segfaults!

[SimonDanisch]

Intuitively I'd say this isn't our fault, since, at least in Makie itself, we shouldn't do anything unsafe.
But there's a chance I'm overlooking something, or some of our dependencies do something unsafe.
There's also a chance, that julia 1.8+ has become less stable, or that what we do in our precompilation is slightly unsafe.
The stacktrace in the CairoMakie benchmark on Julia@1.8.5 does point to serialization:

signal (11): Segmentation fault
in expression starting at none:1
jl_deserialize_struct at /cache/build/default-amdci4-2/julialang/julia-release-1-dot-8/src/dump.c:2081 [inlined]
jl_deserialize_value_any at /cache/build/default-amdci4-2/julialang/julia-release-1-dot-8/src/dump.c:2186 [inlined]
jl_deserialize_value at /cache/build/default-amdci4-2/julialang/julia-release-1-dot-8/src/dump.c:2319

@timholy don't want to use up your time here, but maybe you can give a few pointers about what could be going on here.
I'll try to see today, if I can clean up our precompilation, to make sure we don't have any dangling globals that are not serializable (although it still shouldn't segfault 🤷 )

[timholy]

Have you cleaned your .julia/compiled/v1.9 directory and rebuilt? That's fixed all other such segfaults I've heard about. But that suggests we're not checking that something needs rebuilding. It's quite possible that the heroic @KristofferC has already fixed it:

• also check build ids when checking for staleness of precompile files JuliaLang/Pkg.jl#3308
• add a suffix to a new cache files in case of failure of renaming it to an exisiting cache file already in use JuliaLang/julia#48137 (unlikely to be this error but fixes other Windows build errors)

It looks like beta3 is just about to be released, please try that when it comes out. I believe it may have both of these fixes in it.

[timholy]

For me GLMakie compiles just fine on WSL.

[timholy]

But yes this seems serious. It took me a bit to realize you're seeing this on 1.8 also. Not good. Can you try reverting back to SnoopPrecompile 1.0.1? There's no good reason that 1.0.3 should be able to cause segfaults, but there may be bad reasons (i.e., Julia bugs).

[SimonDanisch]

Thanks, I'll try downgrading!
For the segfaults I run into locally (e.g. wgl2), I did indeed rm -rf ./julia/compiled/v1.9 and it's still segfaulting... But, I'm less concerned about a segfault with a pre-release Julia version, and there's a chance that I somehow messed up my WSL setup.

[SimonDanisch]

Uhhh, I do have a pretty good lead for at least the GLMakie precompilation segfaults:

Seems like cleaning up open windows in GLFW when exiting julia can segfault - which could interact badly with precompiling GLMakie!

[SimonDanisch]

Hmpf, I expected this to be easier and GLFW just doing something unsafe with the open window, but it doesn't seem so and the stacktrace points to the nvidia driver & pthreads:

#0  0x00007fffc9f3545a in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#1  0x00007fffc9f341be in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#2  0x00007fffc9f34136 in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#3  0x00007ffff7f9e609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#4  0x00007ffff7ec3133 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
(gdb) bt 100
#0  0x00007fffc9f3545a in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#1  0x00007fffc9f341be in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#2  0x00007fffc9f34136 in ?? ()
   from /usr/lib/wsl/drivers/nvlti.inf_amd64_ac791a13a77aa333/libnvwgf2umx.so
#3  0x00007ffff7f9e609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#4  0x00007ffff7ec3133 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

[SimonDanisch]

Ah wow, so this is completely unrelated to the other segfaults and now also segfaults in older Julia versions, and since I just recently tried out GLMakie in WSL2 and it worked perfectly, this must be due to a driver update / windows update -.- Why is everyone suddenly conspiring to segfault (GL)Makie ??

[SimonDanisch]

Haven't seen segfaults in the last weeks besides the wsl2 one, so closing this issue for now