Use absolute path to set .ipynb file as dependency

[mgeier]

Getting the relative path seems to be broken, but Sphinx seems to be fine with absolute paths as well.

With a bit of luck, this fixes mgeier/sphinx-last-updated-by-git#26.

[mgeier]

I'm actually not sure whether absolute paths are "officially supported" by Sphinx.

Maybe it would be better to fix the relative paths?

Feel free to close this PR and replace it with one that fixes the relative paths.

[vidartf]

I'm not sure if I follow what the bug is. Would you mind giving a bit more detail?

[mgeier]

Sure. I didn't look too closely into this anyway, so it might be good to take a closer look:

When adding this to the nbsphinx-link source code:

def _env_updated(app, env):
    for docname in env.found_docs:
        print('dependencies for', docname)
        for dep in env.dependencies[docname]:
            print('   ', dep)

... and then adding the following to the setup() function:

    app.connect('env-updated', _env_updated)

... and running Sphinx on the nbsphinx-link docs, I get this output:

...
dependencies for introduction
    ../../notebooks/introduction.ipynb
    /home/[...]/nbsphinx-link/notebooks/smile.png
    ../../notebooks/images/smile.png
    /home/[...]/nbsphinx-link/notebooks/images/smile.png
    notebooks/introduction.ipynb
    /home/[...]/nbsphinx-link/notebooks/images/subsubdir/smile.png
    smile.png
    ../../notebooks/images/subsubdir/smile.png
    images/smile.png
    ../../notebooks/smile.png
    images/subsubdir/smile.png
...

The problem is the path notebooks/introduction.ipynb, which doesn't exist (relative to the source directory).

However, what I didn't know before, there seems to already be a correct path for the notebook: ../../notebooks/introduction.ipynb.

I don't know why there are two different paths for the same notebook, but I know that one of them is wrong, which I think is the cause for the issue I got: mgeier/sphinx-last-updated-by-git#26.

Some of the other dependencies seem to be duplicated as well, but that's probably a separate issue.

[mgeier]

Any news on this?

[mgeier]

Ping?

[vidartf]

I can really see that this will change anything: Currently the code uses path, which was set using

path = utils.relative_path(None, abs_path)

When None is the first argument, this function only normalizes the path to the format that docutils like. It should still be an absolute path...

[mgeier]

It should still be an absolute path...

I don't know about that, but my example clearly shows that there are (invalid) relative paths. Can you reproduce this?