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
Regression with setuptools 27.2.0 on appveyor with Windows + Python 3.5 + numpy 1.11.1 #790
Comments
So I think you're reporting a regression with 27.1.0 from 27.0.0, and here are the differences. Basically, the monkeypatching logic was consolidated in a new monkey module. There were known defects (specifically addressing the patching of MSVC) with v27.1.0 which were fixed in v27.1.1 and v27.1.2. Can you test v27.1.2? |
I've updated the docs so you can more readily see the history of releases. |
27.1.2 is working. |
Then i guess the problem come from But i guess, if Then maybe there is no other choices than something like that to really patch |
Mentioning b6f2fee to get the link. |
Well, the issue is certainly with the quoting of args, which looks suspiciously similar to #739. Since you seem to be getting inconsistent results, I'm not sure what's to be done next. Is it possible that the modules are getting patched multiple times? |
I have no idea. |
Then, the proposed API looks to work with our appveyor context (windows python 2.7, 3.4, 3.5).
What do you think about? I also inverted the order or original and replacement |
I appreciate you working out a technique that works. That's helpful. I'd like to work out what it is about your implementation that fixes the numpy build. I'd like to either have a test in setuptools that captures the expectation that numpy has about the monkeypatching or somehow add an integration test that will capture the actual numpy expectation. Based on your investigation, can you imagine a unit test that would capture the expectation that your changes meet? |
I checked that this patch allows me to build |
Sorry, i was very busy this days. The problem is that part of Right now the setuptools code is nice, but using
I guess the problem is somewhere here. There is no way to use Anyway, i dont really know why the behaviour on my Windows computer and on appveyor is not the same. I dont really reach deeply on that, cause it is laborious. |
Thanks. That helps characterize the cause. I'll put together a patch and try to somehow capture the numpy expectation. |
Hi. I try the last version on my project. It is working. Thanks a lot. |
Hi,
We have a regression on Windows + Python 3.5 + numpy 1.11.1 on our project when compiling with appveyor.
The issue is relative to unquoted path at linking step:
LINK : fatal error LNK1181: cannot open input file 'Files.obj'
See https://ci.appveyor.com/project/ESRF/pyfai/build/0.13.170/job/vrotml5gkymirvad
I try to reproduce the problem with an own Windows machine without success. Then i only have the issue on appveyor + Python 3.5 + numpy 1.11.1
Here is few tests executed on appveyor according to setuptools versions
While i can't reproduce the problem on an own computer it is very difficult to know what's append. But do you have an idea?
BTW, there is a very mysterious message:
No module named 'numpy.distutils._msvccompiler' in numpy.distutils; trying from distutils
. Maybe there is a problem on relative/absolute import or something like that? But this message is also displayed on setuptools 27.0.0, the version which works.The text was updated successfully, but these errors were encountered: