-
-
Notifications
You must be signed in to change notification settings - Fork 9.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BUG: Unusual test failures (whitespace-related?) on ppc32 with numpy-1.25.1 #24239
Comments
Using meson does very much not seem advisable at all, assuming it works, you still should get a slower than wanted build. The main failure I see seem to me thing like:
Can you check if the first and/or second fail? I suppose it would have to be related with the PPC32 platform somehow, but not sure where to look. |
wrt meson: yeah, that's fine, we've already filed a bunch of bugs for things. Having Meson makes life a lot easier for us because of the hacks needed for distutils to get cross working. We've already discussed it w/ Ralf (not saying he loves it, but he's aware).
b'3\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00'
b'1\x00\x00\x00\x001\x00\x00\x00\x001\x00\x00\x00\x001\x00\x00\x00\x001\x00\x00\x00\x00' |
Hummm, well, the second is nonense, its filling in 1 not 3. Although it should end up doing much the same code path internally, unless string to string copies are broken 🤔. But first let's see, does this part work? (i.e. the first step, converting the NumPy 3 to the right string). That seems more likely to be wrong then:
|
ppc32-testing ~ # cat /tmp/foo2.py
import numpy as np
s = np.int_(3)
print(repr(s))
print(s)
a = np.array(3)[()]
print(s)
ppc32-testing ~ # python /tmp/foo2.py
1
1
1 Huh! |
OK. that was helpful, thanks. @mattip another oversight in gh-23746 from everyone glancing at the code, although I am very surprised that this could fly under the radar long enough?! I suppose it hits seriously only on a big-endian machine that also has a "weird" size for C @thesamesam if you want to look at that, the casts in that PR are clearly wrong mixing bit-sized names with C names. |
Many thanks for your help! Let me give that a whack. |
numpy#23746 introduced a fast path for scalar int conversions, but the map between Python types and C types was subtly wrong. This fixes tests on at least ppc32 (big-endian). Many thanks to Sebastian Berg for debugging this with me and pointing out what needed to be fixed. Bug: numpy#24239 Fixes: 81caed6
numpy#23746 introduced a fast path for scalar int conversions, but the map between Python types and C types was subtly wrong. This fixes tests on at least ppc32 (big-endian). Many thanks to Sebastian Berg for debugging this with me and pointing out what needed to be fixed. Closes numpy#24239. Fixes: 81caed6
@thesamesam could you also post the last remaining (clear) error you are seeing, or maybe just create a new issue? |
numpy#23746 introduced a fast path for scalar int conversions, but the map between Python types and C types was subtly wrong. This fixes tests on at least ppc32 (big-endian). Many thanks to Sebastian Berg for debugging this with me and pointing out what needed to be fixed. Closes numpy#24239. Fixes: 81caed6
numpy#23746 introduced a fast path for scalar int conversions, but the map between Python types and C types was subtly wrong. This fixes tests on at least ppc32 (big-endian). Many thanks to Sebastian Berg for debugging this with me and pointing out what needed to be fixed. Closes numpy#24239. Fixes: 81caed6
Describe the issue:
I get several more test failures with numpy-1.25.1 than I do with numpy-1.24.4 on ppc32:
It's worth noting that we've switched prematurely to building with Meson in Gentoo for 1.25.x, although I hope that's not relevant for this bug.
What's especially interesting is that a bunch of the failures seem to be whitespace related in TestArray2String.
There are some other errors in the log as well which look endianness related. Some of these aren't new to us (see https://bugs.gentoo.org/707116, #11831, #12638).
Reproduce the code example:
Error message:
Full log (xz-compressed): https://bugs.gentoo.org/attachment.cgi?id=866019.
The text was updated successfully, but these errors were encountered: