Skip to content
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

devsetup.sh fails with current Python and Yarn #507

Open
bikubi opened this issue May 23, 2024 · 1 comment
Open

devsetup.sh fails with current Python and Yarn #507

bikubi opened this issue May 23, 2024 · 1 comment

Comments

@bikubi
Copy link

bikubi commented May 23, 2024

So possibly deeper incompatibilities?
After updating my system (Arch Linux) and (re-)running devsetup.sh I ran into errors and found workarounds:

  • Python 3.12 is maybe missing pip? Related workarounds seemed to work for a minute.
    • workaround: downgraded to Python 3.10 via pyenv
  • Yarn 1.22.22 and/or node 22.2.0 fail random tasks (e.g. node_modules…exif…yarn install) with a weird "/bin/sh bad interpreter, text file busy" error, at a temporary wrapper script like /tmp/yarn--1716461794779-0.007020274052261355/node. This apparently usually means that an executable file is being edited while executed. Huh.
    • workaround: downgraded to node 16, yarn 1.21 via nvm. Haven't bisected which version is the true culprit here.
bikubi added a commit that referenced this issue May 24, 2024
- torch 2.0.1 wouldn't detect my locally installed cuda libs
  (various .so's "missing from system path")
- now cublas + cuda seem to be installed as deps?
- this might be a quirk of my setup (Arch Linux).
  fwiw, i could then uninstall my local cuda package...
- i poked this until it worked. idk if it makes sense.
@bikubi
Copy link
Author

bikubi commented May 24, 2024

Got it working. Several issues at play.

  • Python: still using 3.10 via pyenv
  • Python deps: had to downgrade torch to avoid errors re: various cublas/cuda so-files "missing from system path". See commit linked above. Maybe related to Pytorch 2.0.1 pypi wheel does not install dependent cuda libraries pytorch/pytorch#100974
  • Yarn issue: /usr/bin/env: 'node': Text file busy apparently points to a problem with new Arch kernel. Downgrading 6.9→6.8 per this comment on docker-node fixed this. Wow.
  • Yarn footgun: I re-created the whole setup in a new directory. Yarn kept symlinks to the old directory in, thru several rm -rf node_modules and such. Quick solution was to rm ~/.config/yarn/link/froide-*; yarn unlink would be proper? Symptoms: makerequest.js wasn't transpiled, keeping all import foo from 'foo-no-extension' modules intact. This resulted in (uncircumventable) CORS errors in Firefox and 404s in a lowered-security Chromium, resp.
  • Yarn/node versions weren't an issue, setup now works with 22/1.22, sans nvm.

I can't really tell if any of this is relevant to anybody else (in the future), so I'm not closing the issue just yet.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant