Fix test for pybind11 include on conda

[henryiii]

Checklist

• Used a fork of the feedstock to propose changes
• Bumped the build number (if the version is unchanged)
• Reset the build number to 0 (if the version changed)
• Re-rendered with the latest conda-smithy (Use the phrase @conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• Ensured the license file is being packaged.

This fixes the tests to make sure that the pybind11 include exists, rather than just assuming the base folder is the correct one. I expect this to fail, since currently the conda package does report the correct include directory.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[henryiii]

@conda-forge-admin, please rerender

[henryiii]

The problem is that this installs the header folder to ${PREFIX}/include/python3.8/* (etc) but reports ${PREFIX}/include for the include directory.

[ax3l]

Thank you for the report!
Wait, this is likely due to an adoption of the include logic and imho a bug. (But I cannot reproduce this issue, all tests are green here.) Please do not simply the test for simple file existence, we want to ensure the headers stay placed as we advertise them.

Do you see that issue only in local builds or also on CF? Cannot see an issue on master and the latest include logic update is not yet released.

Refs.:

• Install in LIBRARY_PREFIX #32
• Detect whether we are running in a Conda environment and adjust get_include() pybind/pybind11#1877
• Install headers using both headers and package_data pybind/pybind11#1995

[ax3l]

Please do NOT check only for file existence, that will always be true and is not what these two tests try to verify.
The modified test case checks that the files are installed where we intend them to be placed, according to conda-forge conventions. (We can additionally check if they exist.)

[henryiii]

Please do NOT check only for file existence, that will always be true

No, it was not true, this is why the PR failed until the patch was applied. The expected behavior is that pybind11.get_include() will return the include directory that contains the pybind11 folder with the pybind11 headers. It was not doing so. It was returning $PREFIX/include, which only contained pythonX.X.

It's a separate issue if that include directory is not the directory that is expected on conda that will also need fixing. I believe that this is being set by

pybind11-feedstock/recipe/build.sh

Line 7 in abbc6f7

-DUSE_PYTHON_INCLUDE_DIR=ON \

But only for unix (it's not in bld.bat), so I think on unix systems this is going into the nested folder. We can test this too, but we need to know whether it's supposed to be $PREFIX/include or $PREFIX/include/pythonX.X on conda.

[henryiii]

(I can test this out locally later and will update if we need to change the include location)

[ax3l]

Okay, I think the actual patch we need for CF is to avoid writing the nested (sub)folder in the first place.

[ax3l]

Got it, it's a unix-only problem, a leftover from #32.
Patch proposed in #44, thanks again for spotting this! Dang, how could we forgot to check as well for file existence...

[ax3l]

Thank you again for your spotting and reporting this!
I applied the patch for Unix in #44 and will close this PR so it does not get accidentally merged.

[henryiii]

44 and this PR are identical, by the way. Only change is I forgot to delete the unused patch. It's okay, though.

[ax3l]

Cheese, I did not see that you also adjusted build.sh in the last diff. Sorry for that! The additional patch that you applied is still in this PR though and should not be in (remconda.patch ).

Yes, this commit is the central one: f649a31