CI job tox -e py time:

• first run 6min 51s
• second run 1min 11s

The job failed once with some urllib error, so that's a bit worrying.

The job failed once with some urllib error, so that's a bit worrying.

What exactly was it?

what exactly was it?

See here
https://github.com/scipp/essreflectometry/actions/runs/13651506808/job/38163329797

Hmm, that just looks like a network interuption. It should hopefully just be a one off.

It should hopefully just be a one off.

Just a coincidence ;)

If it happens again we can look at it more.

What are we optimising here for btw? Maybe I am missing the exact context.
I ran tox with and without this change. After clearing local cache both times I don't see any specific speedups.
With the new URL:
1st run: 44 passed in 370.73s
2nd run: 44 passed in 10.79s
With the old URL:
1st run: 44 passed in 372.07s
2nd run: 44 passed in 10.89s

What are we optimising here for btw? Maybe I am missing the exact context.

We're optimizing for the total time to run the tests when the pooch cache is empty. So from what you wrote it seems to me like we're on the same page.

But the results you show are unexpected, it looks like the local pooch cache is not cleared, are you sure it is?

(I think you already did this, but just to clarify, the pooch cache need to be cleared in between every run.)

(I think you already did this, but just to clarify, the pooch cache need to be cleared in between every run.)

I did not!! The change makes sense now!

I'm not 100% sure this is stable yet, we had the network error earlier for example.
But if we try it out for a while we will find out, so I'll merge it anyway and if it breaks we can change the url back again.

For the record the caching code is here: https://github.com/scipp/fileserver-cache