ignore duplicated labels in autosectionlabel extension

[albertored]

When having duplicated section titles in the same file the autosectionlabel extension print a warning but adds the label anyway overriding the old one.

In this situation I think it would be better to simply not add labels for duplicated sections. In this way a user can define its own labels for that problematic sections avoiding inconsistencies.

This behaviour can be optionally enabled by a configuration option of the autosectionlabel extension.

If you want I can submit a PR.

[tk0miya]

Thank you for proposal. I understand your idea. But it is very difficult on the parallel build. So -0 for your idea.

[stsewd]

An idea I was having about this situation was to just create a label with a suffix, like repeated-label-2, repeated-label-3, etc. Not sure if that's possible with the parallel build that @tk0miya mention.

[albertored]

Yes you're right, I haven't thought about parallel builds.
Maybe we can simply check if a label is already defined for that section and in that case avoid adding the autogenerated one. In this way the user can bypass the autosectionlabel extension on certain sections by putting his own label.

[tk0miya]

@stsewd Do you mean generating the label "Hello Wolrd-label-2" from a section titled "Hello World", right? If so, it can't refer by :ref:`Hello World` , I think. In addition, we need to consider the way to make the sequence number unique in parallel mode.

[tk0miya]

@albertored I don't have a good idea to check the label is already defined or not in parallel mode. I'd not like to add an option only available only on non-parallel mode. I think the built document should be the same in both modes.

[albertored]

@tk0miya I don't the sphinx internal so maybe what I'm suggesting is not possible.

My assumption is that if a section has an explicit label added by the user then that label is already present in the app.env.domaindata['std']['labels'] dictionary at the time autosectionlabel parses that section. Please correct me if I'm wrong.

If this is the case we can do something like

def register_sections_as_label(app: Sphinx, document: Node) -> None:
    labels = app.env.domaindata['std']['labels']
    anonlabels = app.env.domaindata['std']['anonlabels']

    # ___ new code ___
    sectnames = [sectname for _, _, sectname in labels.values()]

    for node in document.traverse(nodes.section):
        # omitted code, here we get name, docname, labelid, sectname

       # ___ new code ___
        if name in sectnames:
            # here we ignore the section because the user has set an explicit label for it
            continue
 
        if name in labels:
            logger.warning(__('duplicate label %s, other instance in %s'),
                           name, app.env.doc2path(labels[name][0]),
                           location=node, type='autosectionlabel', subtype=docname)

        anonlabels[name] = docname, labelid
        labels[name] = docname, labelid, sectname

       # ___ new code ___
        sectnames.append(sectname)

[tk0miya]

My assumption is that if a section has an explicit label added by the user then that label is already present in the app.env.domaindata['std']['labels'] dictionary at the time autosectionlabel parses that section. Please correct me if I'm wrong.

Yes and No. Sphinx goes to process documents per file.

1. Parse file_a.rst
   1.1. Correct user-labels from the file.
   1.2. Call register_sections_as_label for the file.
2. Parse file_b.rst
   2.1. Correct user-labels from the file.
   2.2 Call register_sections_as_label for the file.
   ...

So you can check the existence of the label in the document, and you can't do it for labels in other documents. It causes duplicated warnings if file_a.rst contains a section named foobar and file_b.rst contains a user-label named foobar.

[fabiencelier]

Maybe we can simply check if a label is already defined for that section and in that case avoid adding the autogenerated one. In this way the user can bypass the autosectionlabel extension on certain sections by putting his own label.

+1 For that idea. That mode could be activated by a configuration to not change the current behavior. Adding a label only if there is no manual label makes sense as a general option and, I think, solve the initial issue

[jpihl]

I just read from a comment in #7728, that you can do it with suppress_warnings = ['autosectionlabel.*'].