ENH: Add lazy copy for sort_values #50643
Conversation
Co-authored-by: Matthew Roeschke <[email protected]>
| view = obj[:] | ||
| obj.sort_values(inplace=True, **kwargs) | ||
|
|
||
| assert np.shares_memory(obj.values, view.values) |
There was a problem hiding this comment.
Actually, this is failing with ArrayManager (failing CI)
There was a problem hiding this comment.
.values on a DataFrame is always a copy for ArrayManager, so the generic obj.values doesn't work here:
| assert np.shares_memory(obj.values, view.values) | |
| if isinstance(obj, DataFrame): | |
| assert np.shares_memory(get_array(obj, "a"), get_array(view, "a")) | |
| else: | |
| assert np.shares_memory(obj.values, view.values) |
There was a problem hiding this comment.
You will have to use the same pattern below, so can probably simplify this if we give the Series a name so can use get_array in all cases.
There was a problem hiding this comment.
While testing this, I just pushed the changes I was doing locally, hopefully you weren't yet working on it.
| using_copy_on_write | ||
| or using_array_manager | ||
| and isinstance(obj, DataFrame) | ||
| and pd.options.mode.copy_on_write |
There was a problem hiding this comment.
ArrayManager never uses Copy-on-Write (we should maybe error when setting those two options at the same time), so this check (and the or using_array_manager two lines above) shouldn't be needed. I suppose the reason it seemingly did give a copy is because of the .values giving a copy
|
Thx for fixing this, merged main to ensure that the dev builds get through |
doc/source/whatsnew/vX.X.X.rstfile if fixing a bug or adding a new feature.