Proposal: valid XHTML for basic theme output.

[jayaddison]

Is your feature request related to a problem? Please describe.
It seems that most of the HTML output generated when building Sphinx's own documentation using the basic theme (not currently what we have configured) is XHTML-compliant - that is, it's also valid XML.

In fact, of 146 HTML files, 144 are valid XML. The only two exceptions currently are: search.html and domains.html.

• On search.html, there is a defer attribute that is not valid XML.
• In domains.html, there are unclosed <span> tags that are not valid XML.

I ran the following utility script to determine that, after using sphinx-build -b html doc _build to self-build the documentation -- with the html_theme set to basic.

from pathlib import Path
from xml.etree.ElementTree import ElementTree

element_tree = ElementTree()
valid = 0
total = 0

build_dir = Path('_build')
for path in sorted(build_dir.rglob('**/*.html')):
    print(path)
    try:
        element_tree.parse(path)
        valid += 1
    except Exception as e:
        print(str(e))
    total += 1

print(f'Valid X(HT)ML : {valid}')
print(f'Total files   : {total}')

Trimmed output:

$ python check.py
...
_build/usage/restructuredtext/domains.html
mismatched tag: line 61, column 2
_build/usage/restructuredtext/field-lists.html
_build/usage/restructuredtext/index.html
_build/usage/restructuredtext/roles.html
_build/usage/theming.html
Valid X(HT)ML : 144
Total files   : 146

Describe the solution you'd like

• Output valid XHTML for all pages when using the basic theme -- in other words, fix the output for search.html and domains.html.
• Add some test coverage -- perhaps to the html5 builder tests -- to check some simple cases for valid XML.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
I think this would enable us to drop permissive HTML parsers such as html5lib and lxml from our sphinx[test] dependencies.

• Discovered during review-of / experimentation-with [tests] deprecate html5lib in favor of lxml.html #12157.

[chrisjsewell]

Sounds reasonable to me 👍

I guess,from a"theoretical" perspective, I'm interested to know does anyone think there is ever a valid reason to have non-compliant XHTML? Or would we loosely ascribe this to "bugs" in these templates?