anton-akhmerov:
@cwg conda-forge has an annoying build problem on windows for tinyarray 1.2.4. (Annoying because the build logs tell no information at all. Do you know anyone who could test it?
@cwg conda-forge has an annoying build problem on windows for tinyarray 1.2.4. (Annoying because the build logs tell no information at all. Do you know anyone who could test it?
cwg:
And 1.2.3 works in the same context?
And 1.2.3 works in the same context?
anton-akhmerov:
Let me see.
Let me see.
cwg:
What about other CPython versions? Also, the failure with tinyarray 1.2.4 is only with CPython 3.10 or also with other versions?
What about other CPython versions? Also, the failure with tinyarray 1.2.4 is only with CPython 3.10 or also with other versions?
anton-akhmerov:
The failures are:
The failures are:
- 1.2.4 + windows + 3.10
- 1.2.3 + any OS + 3.10
cwg:
Isn't there really any information? I see some errors in the build log. It seems to start in line 457 of the log where it complains about
Isn't there really any information? I see some errors in the build log. It seems to start in line 457 of the log where it complains about
d:\bld\tinyarray_1640058215207\work\src\array.hh(108): error C3646: 'object_size': unknown override specifier
d:\bld\tinyarray_1640058215207\work\src\array.hh(130): note: see reference to class template instantiation 'Array<T>' being compiled
d:\bld\tinyarray_1640058215207\work\src\array.hh(108): error C2059: syntax error: '('
d:\bld\tinyarray_1640058215207\work\src\array.hh(108): error C2238: unexpected token(s) preceding ';'
anton-akhmerov:
Ah, true. I see it now as well
Ah, true. I see it now as well
cwg:
Which is strange. This part of tinyarray hasn't been changed since 2016. Also, it does compile with CPython 3.9, on the same system, right?
Which is strange. This part of tinyarray hasn't been changed since 2016. Also, it does compile with CPython 3.9, on the same system, right?
anton-akhmerov:
Yes
Yes
anton-akhmerov:
Here's the successful build log for 3.9 on windows
Here's the successful build log for 3.9 on windows
cwg:
I diffed the logs. There are several differences, but nothing that should matter. The first serious difference is that error C3646 which is so far a mystery to me.
I diffed the logs. There are several differences, but nothing that should matter. The first serious difference is that error C3646 which is so far a mystery to me.
anton-akhmerov:
The docs are confusing.
The docs are confusing.
cwg:
Might be related to this: jpype-project/jpype#1009
Might be related to this: jpype-project/jpype#1009
anton-akhmerov:
Indeed, looks very similar
Indeed, looks very similar
cwg:
ssize_t seems to be POSIX - but so far it has worked on Windows as well. Something must have changed with CPython 3.10, most likely they no longer include some header.
ssize_t seems to be POSIX - but so far it has worked on Windows as well. Something must have changed with CPython 3.10, most likely they no longer include some header.
cwg:
Is it possible to check the conda-forge build without doing a release?
Is it possible to check the conda-forge build without doing a release?
anton-akhmerov:
It's also possible to patch sources during the build
It's also possible to patch sources during the build
anton-akhmerov:
Before we go that way, I wanted to check also if this solution is acceptable.
Before we go that way, I wanted to check also if this solution is acceptable.
anton-akhmerov:
https://stackoverflow.com/a/20987501/2217463 seems to suggest that it's correct
https://stackoverflow.com/a/20987501/2217463 seems to suggest that it's correct
basnijholt:
I was kind of happy to see that MacOS ARM build of kwant passed without any work: conda-forge/kwant-feedstock#97
However, I just tested it out and it seems to fail badly: http://files.nijho.lt/macosarm-fails.txt
basnijholt:
Any ideas what might cause these failures?
Any ideas what might cause these failures?
basnijholt:
The differences are quite extreme, e.g.,:
The differences are quite extreme, e.g.,:
actual = array([[ 0.76903213, -0.2516609 , 0.40139562],
[-0.2516609 , 0.96278288, -0.35304597],
[ 0.40139562, -0.35304597, 1.9477003 ]])
desired = array([[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.]]), decimal = 7, err_msg = '', verbose = True
basnijholt:
I opened an issue conda-forge/kwant-feedstock#99 BTW.
I opened an issue conda-forge/kwant-feedstock#99 BTW.
basnijholt:
Also, the incompatibility with Matplotlib 3.5 and Kwant 1.4.2 is blocking the Python 3.10 builds.
See this line.
Is there a plan for a patch release that addresses this problem soon? I see that the latest commit to Kwant (stable or master) is 9 months ago 😅
Anton:
Bas fixed compatibility with latest scipy, see the merge request.
Bas fixed compatibility with latest scipy, see the merge request.
cwg:
The minimal process is just making a tag and pushing. But releasing carefully for multiple platforms and doing checks takes indeed some time. Do you see items in RELEASE.rst that you find superfluous?
The minimal process is just making a tag and pushing. But releasing carefully for multiple platforms and doing checks takes indeed some time. Do you see items in RELEASE.rst that you find superfluous?
Anton:
Bas already knows https://github.com/conda-forge/kwant-feedstock/pull/100#issuecomment-1083841378
Bas already knows https://github.com/conda-forge/kwant-feedstock/pull/100#issuecomment-1083841378
cwg:
I bisected sympy to find the commit that introduces the compatibility problem with Sympy. It's the commit that fixes the following issue: sympy/sympy#19501
I bisected sympy to find the commit that introduces the compatibility problem with Sympy. It's the commit that fixes the following issue: sympy/sympy#19501
cwg:
git bisect rules!
Anton:
Yeah, sympy has caused us quite a bit of headache.
Yeah, sympy has caused us quite a bit of headache.
basnijholt:
NumPy has dropped Python 3.6 support almost 2 years ago: https://numpy.org/neps/nep-0029-deprecation_policy.html#support-table
NumPy has dropped Python 3.6 support almost 2 years ago: https://numpy.org/neps/nep-0029-deprecation_policy.html#support-table
Anton:
Yeah, it's just about testing
Yeah, it's just about testing
Anton:
The pipeline fails
The pipeline fails
Anton:
Did it work now?
Did it work now?
Anton:
Ha! Awesome!
Ha! Awesome!