Repository navigation
numpy 2.0: a possible API inconsistency in numpy.ndarray.astype (copy semantics) ? #25925
Description
As I'm adapting astropy to the new copy semantics from #25168 and #25922, I observe that numpy.ndarray.astype wasn't updated to follow these changes. As a result, it still supports the old semantics where copy=False means "copy only if you must" but never raises an error. Is that intended ?
numpy.astype, introduced in #25079, is also affected by this question.
Activity
rgommers commentedon Mar 4, 2024
@neutrinoceros thanks for the question. Yes, this was intentional, since there wasn't too much appetite for changing the behavior of
copy=Falsegiven thatastypeexplicitly casts to a different dtype - and that usually cannot be done without a copy. It's noted in NEP 56: https://numpy.org/neps/nep-0056-array-api-main-namespace.html#copy-keyword-semantics.I checked that
astypedoes accept all ofTrue/False/None.Falsejust doesn't mean "never copy". Does that seem reasonable?neutrinoceros commentedon Mar 4, 2024
Absolutely ! Just wanted to check with you before the first release candidate if I could :)
Let's just close this then. Good luck with the release !