test: fix overly restrictive cufile skipping logic

[cpcloud]

This PR enables testing cufile on a larger variety of filesystems where the root filesystem (/) might have a different mount point than the mount point of the checkout directory, by looking at all of /proc/mounts instead of only the first matching mount point.

[copy-pr-bot]

Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[cpcloud]

/ok to test

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-1271/
https://nvidia.github.io/cuda-python/pr-preview/pr-1271/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-1271/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-1271/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

[cpcloud]

/ok to test

[cpcloud]

/ok to test

[cpcloud]

Turns out these tests are never run on main because they are skipped. See https://github.com/NVIDIA/cuda-python/actions/runs/19510167362/job/55859660002 and search for cuda.bindings, you'll see this:

[cpcloud]

The logic I have here is I believe correct, but I'm not sure if these tests are just broken, or this is a latent bug that we've been living with for a while.

[rwgk]

Turns out these tests are never run on main because they are skipped.

Tangential:

Yeah, I ran into that twice with the ctk-next testing. SWQA was the first to run in environments where they are not skipped, but I didn't think of checking first if they actually ran in the cuda-python CI and started debugging painstakingly assuming they did.

It'd be awesome to have some sort of automatic flagging, after all jobs finish, to report which tests

• are skipped everywhere
• are always skipped on linux-64, linux-aarch64, win-64

[leofang]

1. For cuFILE related changes, let us always rope in the GDS team for consultation (cc @sourabgupta3).
2. Currently lots of things cannot be run in the CI ([FEA]: Improve CI coverage of cuFile #1143), not even in cuFILE's compat mode IIUC.

[cpcloud]

It's really an issue of incorrect skipping logic, not whether a thing runs in CI. This should have been running, because the runners have a compatible file system.

This wasn't running, but it was unintentionally not running.

[cpcloud]

Given all the failures happen on 13.0.2, and the 12.9.x tests all pass, that seems like a good place to start digging.

[cpcloud]

Ah, well those are not running the cuda.bindings tests, so that wouldn't explain the failure.

[cpcloud]

/ok to test

[cpcloud]

/ok to test

[leofang]

Let's give Sourab some time. I pinged him offline and he's looking into it.

[leofang]

Let's give Sourab some time. I pinged him offline and he's looking into it.

Ping'd you both in an internal thread.