Linux documentation build memory hog

[marxin]

Describe the bug

As mentioned here:
https://lore.kernel.org/all/cb916eae-fdf6-504f-8f38-7928e0fa8344@gmail.com/

the Linux kernel documentation builds https://docs.kernel.org/doc-guide/sphinx.html?highlight=sphinx#sphinx-build gets slower and consumer much more memory.

I bisected that to 080517c.
@AA-Turner: Can you please take a look?

How to Reproduce

Clone Linux repo and apply:
https://lore.kernel.org/all/87wn629ggg.fsf@meer.lwn.net/

make htmldocs

Environment Information

Aforementioned Sphinx version.

Sphinx extensions

No response

Additional context

No response

[AA-Turner]

@AA-Turner: Can you please take a look?

I'm taking a break due to other work until Easter; apologies for this. I'd advise pinning to an earlier version if a root cause can't be found.

A

[marxin]

Do you think it would make sense to bisect docutils in order to find what caused that?

[marxin]

So the issue is caused by the following removed hunk:

sphinx/sphinx/util/nodes.py

Lines 630 to 643 in 4660b62

	def _new_copy(self: Element) -> Element:
	"""monkey-patch Element.copy to copy the rawsource and line
	for docutils-0.16 or older versions.
	refs: https://sourceforge.net/p/docutils/patches/165/
	"""
	newnode = self.__class__(self.rawsource, **self.attributes)
	if isinstance(self, nodes.Element):
	newnode.source = self.source
	newnode.line = self.line
	return newnode
	nodes.Element.copy = _new_copy # type: ignore

which now relies on docutils copy function which also copies _document` property:
https://github.com/docutils/docutils/blob/bc9bf5fc2c7f905507a49e8e37c12368c2f4a0cc/docutils/docutils/nodes.py#L1056-L1061

If I add copying of _document to the _new_copy function, I get the very same memory hog.

[AA-Turner]

I have released v6.1.3 which should fix this.

A

[marxin]

Thanks for the reversion. Should we keep this issue open for a proper fix in docutils project? Or should I close it?

[jfbu]

I will close the issue for time being but we can always reopen at later date.

The fix happened at dcb4429 and was merged into 6.1.3.

The original removal of the monkeypatch of DocUtils happened at 080517c from #10470 which was part of the 6.0.0 release. (as @marxin mentioned in original report, thanks!)