Build and test the documentation in Azure Pipelines #3854
Conversation
|
|
soupault
force-pushed
the
azure_docs
branch
2 times, most recently
from
dba768c
to
d673e8b
Compare
|
Alright! The documentation testing seems to work! I had to implement several workarounds:
Otherwise, the PR is ready for review. |
soupault
changed the title
| @@ -93,17 +93,17 @@ class BRIEF(DescriptorExtractor): | |||
| >>> extractor.extract(square2, keypoints2) | |||
| >>> descriptors2 = extractor.descriptors | |||
| >>> matches = match_descriptors(descriptors1, descriptors2) | |||
| >>> matches | |||
| >>> matches # doctest: +SKIP | |||
There was a problem hiding this comment.
why these tests are skipped?
| # Extract all possible features to be able to select the most salient. | ||
| feature_coord, feature_type = \ | ||
|
|
||
| if __name__ == '__main__': |
There was a problem hiding this comment.
i'm a big -1 on this. This really confuses new users, who become confused since they don't knkow what __main__ means. why do we need to do this?
There was a problem hiding this comment.
@hmaarrfk I'm still trying to figure out a better solution. The issue is described here - #3854 (comment), specifically, in the StackOverflow post.
In short, on Windows multiprocessing imports the full contents of the file when spawning extra processes. This leads to starting an infinite number of processes.
So, basically, the multiprocessing code has to be wrapped somehow.
| @@ -64,18 +64,18 @@ class BRIEF(DescriptorExtractor): | |||
| >>> import numpy as np | |||
| >>> square1 = np.zeros((8, 8), dtype=np.int32) | |||
| >>> square1[2:6, 2:6] = 1 | |||
| >>> square1 | |||
| >>> square1 # doctest: +SKIP | |||
There was a problem hiding this comment.
I think you should get linux setup, and run tehse doctests on linux. Linux will be able to handle the doctests and all.
There was a problem hiding this comment.
@Borda I think conftest.py file is not being discovered by pytest on Windows. In that file we force numpy to use legacy printoptions. Therefore, the builds expect the newer notation (that is numerically equivalent, but visually different).
There was a problem hiding this comment.
@hmaarrfk on Azure you mean? I'm not sure if anyone from the core team is willing to switch from Travis to Azure at the moment. Of course, it shouldn't be an issue to have a Linux instance running, but I really want to have Windows build with the full coverage.
Unfortunately, googling "pytest, windows, conftest" doesn't yield any info. I'm going to check if forcing conftest could be done with the env.variables...
There was a problem hiding this comment.
We could also just test the doctests on numpy 1.14 and higher
or just bunp the minimum reqs all the way up to 1.14 ;)
There was a problem hiding this comment.
@soupault, it might be worthwhile too post an issue on pytest's github page. I couldn't figure this out either, it could very well be a bug on windows.
I understand that you want full test coverage on windows, but I would argue that having the doctests be readable is more important. I've often seen my comments copied verbatim in other's code. I wouldn't watn to pollute their dropbox with # doctest: +SKIP ;)
|
The work is continued in #3873. |
Description
Implement documentation building and testing in Azure Pipelines, similarly to https://github.com/scikit-image/scikit-image/blob/master/tools/travis/script.sh.
For reviewers
later.
__init__.py.doc/release/release_dev.rst.@meeseeksdev backport to v0.14.x