search: support searching for (sub)titles

[marxin]

trafficstars

Collect all titles from all pages and utilize a full match (case insensitive) in Search page.

Fixes: #10689

[AA-Turner]

Does this work for partial matches?

A

[marxin]

Does this work for partial matches?

Yes, I've just added partial match support.

[AA-Turner]

Please add tests and a CHANGES entry.

A

[marxin]

I've just added CHANGES, but for a working test, I would need an HTML inspection of rendered search.html page. Do we have such tests?

[marxin]

As an improvement, it should be possible to emit a link directly to the title in a page. But I can't find a function that would give me for a nodes.title Node an anchor:

{'rawsource': 'Demo documentation', 'children': [<#text: 'Demo documentation'>], 'attributes': {'ids': [], 'classes': [], 'names': [], 'dupnames': [], 'backrefs': []}, 'tagname': 'title', 'parent': <section "demo documentation": <title...><compound...><paragraph...><substitution_defin ...>, '_document': <document: <substitution_definition "gol"...><section "demo documen ...>, 'source': '/home/marxin/Programming/texi2rst-generated/sphinx/demo/index.rst', 'line': 2}

Can you please help me?

[jbms]

I think you may need to look at the parent section element to find the ids.

[jbms]

Note: In https://github.com/jbms/sphinx-immaterial I have implemented something similar entirely client side, but there are a few differences:

• The sub-sections are parsed from the HTML document itself, while extracting snippets.
• If the search text is found within the page, the result link is to the nearest containing section.

This is the source code of my implementation, for reference: https://github.com/jbms/sphinx-immaterial/blob/main/src/assets/javascripts/sphinx_search.ts

In general handling this when building the index is probably better, though given that the HTML must be parsed anyway to handle the snippets I'm not sure.

[marxin]

In general handling this when building the index is probably better, though given that the HTML must be parsed anyway to handle the snippets I'm not sure.

Yes, I do prefer the server side implementation and I'm still curious about the title links as mentioned in my previous comment. One should be able to get a link to them.

[AA-Turner]

One should be able to get a link to them.

Something like node.parent["names"] should work? The anchor link is on the docutils.nodes.section node as I recall.

A

[marxin]

Something like node.parent["names"] should work? The anchor link is on the docutils.nodes.section node as I recall.

Yep, that almost works:

diff --git a/sphinx/search/__init__.py b/sphinx/search/__init__.py
index bbb28c0b9..7916d26f0 100644
--- a/sphinx/search/__init__.py
+++ b/sphinx/search/__init__.py
@@ -216,6 +216,8 @@ class WordCollector(nodes.NodeVisitor):
         elif isinstance(node, nodes.title):
             title = node.astext()
             self.found_titles.append(title)
+            print('node:', node)
+            print('node.parent[names]:', node.parent['names'])
             self.found_title_words.extend(self.lang.split(title))
         elif isinstance(node, Element) and self.is_meta_keywords(node):
             keywords = node['content']

emits something like:

node: <title>Comparison of GCC docs in Texinfo and Sphinx</title>
node.parent[names]: ['comparison of gcc docs in texinfo and sphinx']
node: <title>HTML output</title>
node.parent[names]: ['html output']
node: <title>Formatting</title>
node.parent[names]: ['formatting']

So the last missing piece is probably an escaping that will emit e.g. comparison-of-gcc-docs-in-texinfo-and-sphinx?

[AA-Turner]

Ahh, can you try ["ids"]?

A

[marxin]

Ahh, can you try ["ids"]?

Works for me, added that.

[marxin]

Can you please @AA-Turner review the pull request now?

[AA-Turner]

Something seems to be wrong:

https://sphinx--10717.org.readthedocs.build/en/10717/search.html?q=More+topics+to+be+covered

https://www.sphinx-doc.org/en/master/search.html?q=More+topics+to+be+covered

The PR only shows 5 results, and doesn't highlight the title, whereas the current master shows the title, albeit as the third result.

A

[marxin]

The PR only shows 5 results, and doesn't highlight the title, whereas the current master shows the title, albeit as the third result.

Yeah, it's a fancy feature of Read the Docs, it must be a plug-in that is used for Sphinx docs.

Please compare it with another pull request: https://sphinx--10807.org.readthedocs.build/en/10807/search.html?q=More+topics+to+be+covered

[AA-Turner]

Rebased.

A

[AA-Turner]

If running an incremental build, searchindex.js is only updated after loading it, so we need to bump the environment version.

A

[AA-Turner]

Thanks @marxin!

A

[marxin]

Thanks for merging that!