See #60154
We should add the build and fix the existing errors
Comment From: iabhi4
take
Comment From: iabhi4
I did some deep testing across numpy versions and here’s what I’m seeing: The existing xfail condition for np.float32(1.1) was correct in spirit, this behavior starts changing already in numpy 1.25, not just 1.26.
If we only add np_version_gte1p26 to xfail, tests still fail under numpy 1.25.x (confirmed locally with 1.25.2).
The previous condition using NPY_PROMOTION_STATE was guarding this for <1.24 or promotion changes but now it looks like we should generalize to:
or np_version_gte1p25
That makes the test robust across current and future versions, Should I proceed with this change and send a PR?
Also regarding the CI build, I saw the earlier discussion in #60154 , would you like me to add that CI job (actions-310-numpy-126.yaml) in this PR itself, or would you prefer it as a followup PR?
Comment From: datapythonista
I think the idea is to test for users who can't upgrade to numpy 2, which are expected to use the latest numpy 1 version.
Probably better to add the CI job in the first PR, so it's seen what fails.
CC @jorisvandenbossche
Comment From: jorisvandenbossche
would you like me to add that CI job (
actions-310-numpy-126.yaml) in this PR itself, or would you prefer it as a followup PR?
It's probably easiest to do both in a single PR, because we need to add the extra CI build to verify if the fixes are working, but we will also need those fixes to get a green CI to be able to merge it.
Comment From: debajyoticss
Hi! I’m new to open source and would love to work on this issue. Please assign it to me 🙂