API: Allow SciPy to get away with assuming trapz is a Python function
#23034
Conversation
seberg
force-pushed
the
wrap-trapz
branch
2 times, most recently
from
7ba4883 to
3dcb7a5
Compare
numpy/lib/function_base.py
Outdated
| @functools.wraps(_trapz_orig) | ||
| def trapz(y, x=None, dx=1.0, axis=-1): | ||
| # Hack so that SciPy has some python code it can steal | ||
| return _trapz_orig(y, x=x, dx=dx, axis=axis) |
There was a problem hiding this comment.
Ah, this means we always pass defaults in __array_function__ (which is probably OK). We can also go back to *args, **kwargs. Or, I can just override all the attributes used by SciPy (__code__, and a couple of others).
There was a problem hiding this comment.
Finding @dhomeier comment at astropy:
E astropy.utils.exceptions.AstropyWarning: function 'trapz' is not known to astropy's Quantity. Will run it anyway, hoping it will treat ndarray subclasses correctly. Please raise an issue at https://github.com/astropy/astropy/issues.The issue seems to be that with the proposed fix np.trapz identifies as _trapz_orig, which is not in function_helpers.SUBCLASS_SAFE_FUNCTIONS.
Wrapping things up seems to confuse astropy a bit (the function is just an entry-point).
So I will fix this up to just steal all the attributes so that pretending that this is a Python function works (what SciPy does) to allow creating a new Python function from the non-python version.
There was a problem hiding this comment.
Thanks @seberg, this version (fbb915f) resolves astropy/astropy#14287 out of the box!
mattip
left a comment
mattip
left a comment
There was a problem hiding this comment.
LGTM. NIT: Should there be a test somewhere? Maybe not since this is a SciPy support shim that should go away in a few years.
Dunno, if its worthwhile, but seemed easier to add one than ponder whether or not :) |
seberg
force-pushed
the
wrap-trapz
branch
2 times, most recently
from
fbb915f to
eff8a9b
Compare
This wraps `trapz` into a proper python function and then copies all attributes expected on a Python function over from the "fake" version to the real one. This allows SciPy to pretend `trapz` is a Python function to create their own version.
|
thank you @seberg ! |
5 participants
This wraps
trapzinto a proper python function, I could also just copy over all the relevant attributes. In case we change things so thattrapzis a Python function anyway, the assert would fail.EDIT: Changed the approach to just copy over the stuff that SciPy needs to "clone" the function.