-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Dlopening MKL on Windows gives "Access is denied" error #38993
Comments
Might be the same underlying issue as #38411 |
Does it happen on 1.5.2? |
|
I suspect this is not an issue in Julia but in the JLL package (not 100% sure though): JuliaPackaging/BinaryBuilder.jl#982 might fix it in |
Also running into this with CUDA.jl on Windows 10 using Julia 1.6:
CUDA.jl doesn't use a JLL for its artifacts, and is using LazyArtifacts already, so that PR doesn't seem like it would help.
|
On a suggestion by @KristofferC, reinstalling on 1.5 fixes the issue there. So this is still 1.6 only; adding it back to the milestone.
Although the 1.5: 1.6: |
So the difference is maybe how the artifact is unpacked? |
Unsurprisingly bisected to a Pkg.jl/Tar.jl bump, #38678, which includes this change: JuliaIO/Tar.jl@1b63f2a. Reverting this hunk allows me to Anyway, this looks like something BinaryBuilder should fix (both MKL_jll and CUDA_jll's libraries are |
Hmmm, yeah, I think the proper fix here is to just have BB generate the artifacts with the proper permissions; this seems like things are finally working as intended. It is unfortunate that MKL and CUDA both have incorrectly-chmod'ded files, perhaps we need to add an audit pass to BB to auto-chmod stuff in |
We don't run autofix on MKL for Windows at the moment: https://github.com/JuliaPackaging/Yggdrasil/blob/e17db19b1fba946c256e875639c23db2446f282f/M/MKL/build_tarballs.jl#L55-L57. I'm looking forward to finding what bugs we discover with that. |
Fixing the CUDA artifacts, JuliaPackaging/Yggdrasil@aabb21c, has resolved this. I think we can close this then (but MKL_jll still needs fixing). |
Got same error when using CUDA in Julia 1.7 (and 1.6.0-DEV). However, CUDA works well in Julia 1.5.2. |
There hasn't been a release of CUDA.jl with the new artifacts yet. Use the master branch (or include JuliaGPU/CUDA.jl@749d7e0) if you need to use Julia 1.6. |
A file in an artifact that should be executable isn't: sglyon/PlotlyBase.jl#40, with Julia 1.6. |
As I explained on discourse, the issue is that the creator of the tarball put a file with the wrong file permissions. It isn't a Julia issue. |
I see, somehow I gathered that the tarball was unpacked incorrectly. So, you're saying that the file was put into the ball without the proper permissions? Got it. |
@giordano - I think this diagnosis is incorrect. I can reproduce the problem shown by @maleadt without any artifacts at all. It appears to be introduced by We encountered this here: JuliaGraphics/Gtk.jl#605 The issue can be reproduced simply:
|
Hmm...
produces "drw-rw-rw-" even though the current user and the user's group both should have execute permissions and Everyone should have no permissions set. There seems to be some issues mapping Windows ACL to Linux mode values.
And then here's what chmod does; how it treats the current user/owner and Everyone seems odd:
|
I guess this is an unsolved problem in |
Just
using MKL_jll
on Windows should reproduce the error. This is from tests inFFTW.jl
:This is a regression from v1.5.
The text was updated successfully, but these errors were encountered: