-
-
Notifications
You must be signed in to change notification settings - Fork 18.4k
CI: Numpy 1.22 caused CI to fail #45182
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
Comments
In the other PR/issue, this:
was said to be the only "real" issue (not typing related). I think this is a duplicate of numpy/numpy#20699 (comment) probably, I will look into it (I guess there may be non-nativ byte-order involved, although maybe not necessarily always) |
fyi: |
not sure why but versions of numpy below 1.22 are causing whitesource security vulnerabilities triggers to go off per https://www.whitesourcesoftware.com/vulnerability-database/CVE-2021-41496 . It would seem that 1.21.5 would solve it as the fix (numpy/numpy@16f7824) shows it's present in 1.21.5 |
@padrepitufo we have been getting nonsense CVEs for NumPy, this is one of them. I am pretty disappointed by these CVEs appearing out of no-where and being assigned ridiculous severity levels. |
I‘ll prepare a pr adressing the typing issues |
For the NumPy issue, not sure if this will be fixed by numpy/numpy#20735 |
@rhshadrach there was just a new numpy release that includes a fix for some reductions (reducelikes) failing with a weird type resolution error. I would hope that fixes the issue here. Please make sure to ping me if pandas still has any problems with the 1.22.1 release! |
Thanks @seberg - I am still seeing this with 1.22.1. Here is the test that's failing: pandas/pandas/tests/frame/test_ufunc.py Lines 277 to 286 in 96f2f2a
and the error message I'm seeing is
on
|
I was previously running with numba 0.53.0; updating to 0.55.0 I'm no longer seeing a failure. |
Numba was broken for a long time for NumPy 1.21+ (they rely on the Python calling convention not changing, but Python optimized things by changing the calling convention, and I used that in NumPy 💥). So I expect if you were on old numba, it is just a numba bug. If you use a dynamic numba vectorization (the types are not specified ahead of time), then there is still a problem on 1.22.1. I am planning to fix it soon, but do not feel like rushing it considering that Numba support for NumPy 1.21.x support took 3 months or so. |
moving to 1.4.1 and removing blocker label |
Not sure what the right course of action here is - would we want to bump the minimum version of Numba? It would need to be bumped to a recent version, which seems undesirable. We could always just document this somewhere - a warning in the whatsnew as well as in the Numba section of the docs. |
removing milestone. |
Looks like these have been addressed so closing |
The test
is failing.
Also we have some typing problems
The text was updated successfully, but these errors were encountered: