BUG: Change dtype comparison in _view_is_safe to be independent of name fields. #13261
Conversation
|
Haven't looked at this in detail, but I think it might relate to #8514 |
|
@lukasbrauncom do you wish to move forward with this? It needs tests that fail before this PR and now pass, and those tests should address the problem of padding. |
|
@eric-wieser Is this impacted by your recent fixups? |
| return ndtype.type | ||
| return newdtype | ||
|
|
||
| if oldtype.names or oldtype.names: |
There was a problem hiding this comment.
Typo, one of these should say newtype.
Should also use .names is not None here and above.
|
@eric-wieser Want to fix this up? |
|
We discussed this briefly, and the backport candidate label seemed unnecessary. I am going to close this due to the fact that it stalled for a longer time and would at least need new tests. I also expect that the machinery/fixes from gh-19346 could be interesting to generalize this more, although they are only available in C and not quite aligned with this, I think checking |
I have changed the _view_is_safe in _internal.py such that the comparison of dtypes is independent of the name field in structured arrays. This fixes issue #13213