Lower HTML memory usage by lowering the number of unnecessary spans

[bolshoytoster]

Is your feature request related to a problem? Please describe.
I'm using the angr API reference, which is built with sphinx. Currently the linked page uses ~500MB of memory, mostly due to the sheer number of elements (yes, I know that page is much bigger than the average sphinx page at ~14MB of HTML).

I've noticed that a lot of this is from the many <span class="pre">blah</span>s that are everywhere (they make up ~2MB of the raw HTML). A comment tells me that the spans are there to prevent code from line-wrapping, which is fair, but they're unneccessary if the code is just a single word (in fact, furo, the theme that angr uses, ignores these spans, so they're completely useless).

Describe the solution you'd like
Sphinx shouldn't output these spans for a single word (i.e. len(self.words_and_spaces.findall(encoded)) == 1), in this case, the pre class should be added to the parent element instead.

It's probably hard to do this well with this basically being a string builder, but it could be something like replacing

sphinx/sphinx/writers/html5.py

Lines 826 to 835 in c4929d0

	for token in self.words_and_spaces.findall(encoded):
	if token.strip():
	# protect literal text from line wrapping
	self.body.append('<span class="pre">%s</span>' % token)
	elif token in {' ', '\n'}:
	# allow breaks at whitespace
	self.body.append(token)
	else:
	# protect runs of multiple spaces; the last one can wrap
	self.body.append('&#160;' * (len(token) - 1) + ' ')

with

tokens = self.words_and_spaces.findall(encoded)
# Very fragile, will probably break
if len(tokens) == 1:
    self.body[-1] = f'{self.body[-1][:-2]} pre">'
for token in tokens:
    if token.strip():
        if len(tokens) == 1:
            self.body.append(token)
        else:
            # protect literal text from line wrapping
            self.body.append('<span class="pre">%s</span>' % token)
    elif token in {' ', '\n'}:
        # allow breaks at whitespace
        self.body.append(token)
    else:
        # protect runs of multiple spaces; the last one can wrap
        self.body.append('&#160;' * (len(token) - 1) + ' ')

Describe alternatives you've considered
Just dealing with it, and closing the tab when I'm not using it

[AA-Turner]

Currently the linked page uses ~500MB of memory

What's the best way of measuring this?

[bolshoytoster]

@AA-Turner I'm measuring it in firefox's memory profiler, and it also appears in firefox's process manager (shift+esc).