References outside function

[hoxbro]

I have seen a similar problem as mentioned here. Problem is for docstrings but only for docstrings outside of functions. I have attached an example, and by running the command sphinx-build -b singlehtml -d _build/doctrees . _build/singlehtml l i get the following for output:

If I delete :members: from manual.rst the docstring outside the function works as expected.

Am i doing anything wrong? I'm running the newest version (0.9.1).

[sebp]

I observed the same using sphinx==1.7.9 and numpydoc==0.9.1, the reference is rendered correctly for a function, but is messed up when used in a class.

[MatthewFlamm]

I have the same issue with version 0.9.1 for a reference in a class as well. I reverted to version 0.8.0 and all works as expected.

I confirmed that the only package changed when reverting to version 0.8.0 was numpydoc.

[larsoner]

Are either of you comfortable enough with git to do a git blame to isolate the commit that causes this issue?

[MatthewFlamm]

I thought I was, but your comment makes me question that. I thought that git blame requires knowledge of what file or line is suspected to cause the issue. Did you mean git bisect?

[larsoner]

Argh yes, sorry. git bisect is what I meant

[larsoner]

(I don't mean to scare you off -- it can be difficult for people totally unfamiliar with git but if you've already used git before you'll probably be fine.)

[MatthewFlamm]

I'll give it a try. Although I first need to figure out how to get my environment to work with a development version of numpydoc, as I currently have it installed from pypi. Is it as simple as cloning the repo and then pip install -e numpydoc?

[larsoner]

Usually I use pip install -e . after cloning and cd to the clone dir.

[larsoner]

(But looking at it I guess this would be functionally equivalent!)

[sebp]

According to my investigation, commit c8513c5 is to blame.

[MatthewFlamm]

Beat me to it. This is the same commit that breaks functionality for me.

[hoxbro]

I seem to have fixed this issue:

[larsoner]

@hoxbro do you want to open a PR to fix it?

[dizcza]

I still have this problem https://pytorch-mighty.readthedocs.io/en/latest/reference/loss.html

[hoxbro]

I think your problem is related to #130, which is also commented in the documentation https://numpydoc.readthedocs.io/en/latest/format.html.

[dizcza]

@hoxbro
The workaround posted in #130, unfortunately, does not help. Now it parses the references in an overview section as [Rfe594912252b-R1] instead of [Rfe594912252b-1].

[hoxbro]

What I was trying to say is that there have been some problems with reference in tables for a long time.

If I were you I will make a new issue with a minimal working example.