sourcelink.html template from sphinx basic theme is used instead of theme one

[OriolAbril]

I don't have time right now to try and get a MWE but figured it might be useful to report already. Feel free to ignore until it is reproduceable though.

Some doc builds end up using sourcelink.html from sphinx itself instead of the sourcelink.html defined by the theme. Consequently, this ends up in right sidebars that look like the following:

instead of the expected version using the theme provided template:

Versions used:

sphinx 8.1.3
pydata-sphinx-theme 0.16.1

[Charlie-XIAO]

This is seen in scikit-learn as well with sphinx==7.4.7 and pydata-sphinx-theme==0.16.1 (from conda-forge).

I downloaded the .conda package from conda-forge and it seems that sourcelink.html became sourcelink.h which might be the reason. Note that the distribution on PyPI looks fine.

Screenshot of the unzipped .conda package

Maybe @trallard?

[nyoma-diamond]

+1

Downgrading to 0.16.0 fixes the issue. Seems to be a bug exclusive to 0.16.1

[trallard]

Hey y'all, thanks for flagging this up.

We will look into this as soon as we can.

[ukukas]

can confirm that pydata-sphinx-theme=0.16.1 and sphinx=8.1.3 from conda-forge were used in #2129

[drammock]

@ukukas does it work if you install pydata-sphinx-theme==0.16.1 from PyPI instead of conda-forge? That would confirm what @Charlie-XIAO saw, that only the conda-forge distribution is broken.

[trallard]

It seems to match @drammock - sourcelinks work as expected in our docs (see #2129) so it would seem the issue is with the conda-forge distribution

[nyoma-diamond]

I observed this behavior using pip/PyPl, so not clear that conda-forge is specifically the issue (unless my python virtualenv decided to do something weird and alias pip to conda-forge somehow without my knowledge)

[trallard]

Hmmm 🤔 ok will check if there is something odd going on when we bundle the theme.
I don't think we made changes to this between the last releases but will check anyway.

[ukukas]

Just did some more testing and this is what I found. The issue was present in version 0.16.1 when installing either via conda or installing the wheel via pip. When installing version 0.16.1 from source, the issue was no longer present.

[trallard]

Yeah this smells like an issue with our packaging workflow. I will look into it.
Thanks for the data points folks.

[trallard]

This seems to be a rather odd issue with a flaky build. I checked through our build workflow and found that only the release for 0.16.1 has such bug:

• 0.16.1rc0: sourcelink.html is correctly generated (see Package contents in https://github.com/pydata/pydata-sphinx-theme/actions/runs/12257024106)
• 0.16.1: sourcelink.h is generated as pointed out above (see https://github.com/pydata/pydata-sphinx-theme/actions/runs/12371373877), so since this is automatically done, then the conda-forge inherited this odd build
• Subsequent builds have all been ok (see, for example, https://github.com/pydata/pydata-sphinx-theme/actions/runs/13361619320), so I am unsure how this happened for that one workflow.

I might add a further check to ensure we do not have weird extension files through our build. But the one way to get rid of this bug would be to make a new release soon. cc @drammock

Perhaps need to update #2077 and merge before that WDYT @drammock?

[drammock]

I might add a further check to ensure we do not have weird extension files through our build. But the one way to get rid of this bug would be to make a new release soon. cc @drammock

Perhaps need to update #2077 and merge before that WDYT @drammock?

good detective work. We're due for a release I think so no harm in pushing that out soonish to get this fixed. re: #2077 is there anything else you want me to do over there? I think the last request for me was the org-secret -> environment-secret thing, which is done.

[trallard]

re: #2077 is there anything else you want me to do over there? I think the last request for me was the org-secret -> environment-secret thing, which is done.

Nope, I just need to make some minor fixes to the workflows due to changes in #2122 and we should be good to go.