Repository navigation
Conversation
3d592ab to
047ba0e
Compare
pllim
left a comment
pllim
left a comment
There was a problem hiding this comment.
Less hacky solutions welcome!
| >>> a = [1.0, np.ma.masked] | ||
| >>> b = [np.ma.masked, 'val'] | ||
| >>> Table([a, b], names=('a', 'b')) | ||
| >>> b = [np.ma.masked, 'value1'] |
There was a problem hiding this comment.
I have to expand this to 6-char so the table print out is the same whether it reports back str or object.
There was a problem hiding this comment.
I'd rather not corrupt the user documentation like this, namely showing the user a table output that looks different from what they will see (i.e. the ... for the column type).
If need be you can use >>> print(Table([a, b], names=('a', 'b'))) to avoid showing the line with types.
taldcroft
left a comment
taldcroft
left a comment
There was a problem hiding this comment.
Overall I'm not sure about the approach. At least if I understand, eventually numpy-dev will be released and then this problem will occur with the latest numpy release. Then the NUMPY_DEV tag won't be enough.
Or is the assertion that the problem is somehow related to the packaging of numpy-dev in contrast to normal releases?
| >>> a = [1.0, np.ma.masked] | ||
| >>> b = [np.ma.masked, 'val'] | ||
| >>> Table([a, b], names=('a', 'b')) | ||
| >>> b = [np.ma.masked, 'value1'] |
There was a problem hiding this comment.
I'd rather not corrupt the user documentation like this, namely showing the user a table output that looks different from what they will see (i.e. the ... for the column type).
If need be you can use >>> print(Table([a, b], names=('a', 'b'))) to avoid showing the line with types.
|
Yeah, I am not sure about this fix either. I think if possible, we need to understand why the |
|
Like I said in the issue, when I have numpy installed from the nightly wheel in my conda env, the tests actually passes if I move the test code outside of |
|
I do hope to get to this. My sense is that the size change is a red herring. Or perhaps a cython version incompatibility? |
|
See numpy/numpy#18425. |
|
Wow, thanks for tracking it down, @taldcroft ! I am closing this without merge then. |
Description
This pull request is to address dtype "S" and "U" turning into "object" in numpy-dev job. Probably a non-issue due to the moving target of numpy-dev and how Cython didn't compile against it, and we can just deal with it as long as we don't see this behavior in stable numpy?
Fix #11291
NOTE TO REVIEWERS: Make sure tests really pass now in the numpy-dev job!