KeyError 'refdoc' when missing references

[Anntoin]

I'm seeing the same issue as noted in abakan-zz/ablog#75

Seems to be related to this:

def missing_reference(app, env, node, contnode):

    target = node['reftarget']
    return _missing_reference(app, target, node['refdoc'],
contnode, node['refexplicit'])

https://github.com/sunpy/ablog/blob/master/ablog/post.py#L458

If I change node['refdoc'] to something like node.source it works - I'm not sure if this is correct behaviour however.

Haven't been able to reproduce a minimal example yet, it's just happening in my main env, any pointers appriciated. Using sphinx 1.7.6, ablog 0.9.2

[nabobalis]

Could you provide me a link to your environment if its possible? I will try to see if I can replicate it with the code I have but I can't see any issues with the (small) testing we have enabled on the repo. I plan to add some more in a PR.

[Anntoin]

Unfortunately I can't share my env (work), but have done a little more digging.

What I'm seeing is that revise_pending_xrefs isn't being called when I run a build, so refdoc isn't being set on any nodes. In fact it looks like the doctree-resolved event isn't being emitted at all, so process_postlist isn't called which should call revise_pending_xrefs.

As doctree-resolved is a core sphinx event I'm sure there's something funky up with the version of sphinx I'm using, but it'd be good if the extension handled this gracefully too.

Not sure what the best behaviour would be but should revise_pending_xrefs be handled as part of missing_reference (or a different event?) or just fall back if refdoc isn't found, e.g. to node.source.

[Anntoin]

Actually, looks like the problem is due to event ordering. If I change node['refdoc'] to node.source that allows the build to proceed, so afterwards doctree-resolved is emitted and process_postlist is called.

[nabobalis]

So I should just make that change and that should fix whatever this bug is?

[Anntoin]

Just updating https://github.com/sunpy/ablog/blob/master/ablog/post.py#L459 to
return _missing_reference(app, target, node.get('refdoc'), fixes the immediate issue for me. There's more going on with missing_references and postlists that probably needs looking into at some point - but I'll raise an issue if I spot anything.

[nabobalis]

I've made the change in PR #28, but I also changed the other dictionary index just in case. They probably should be changed in other places. But I will do a release 0.9.3 for now to make this change.

[Anntoin]

Awesome, thanks 👍

[nabobalis]

I just pushed the update to PyPi. Conda will take a while to update.

[Anntoin]

Just to confirm, 0.9.3 is working for me. Getting warnings for missing references but builds succeed. Thanks for the quick turn-around