make html && make latexpdf fails with: *** No rule to make target 'all-pdf'. Stop.

[marxin]

Describe the bug

It's caused by my b1390c4 where now the following happens:

$ make clean ; make html  SPHINXBUILD=~/bin/venv/bin/sphinx-build && make latexpdf SPHINXBUILD=~/bin/venv/bin/sphinx-build
...
Running LaTeX files through pdflatex...
make LATEXMKOPTS="-interaction=nonstopmode -f" -C _build/latex all-pdf
make[1]: Entering directory '/home/marxin/Programming/texi2rst-generated/sphinx/gdc/_build/latex'
make[1]: *** No rule to make target 'all-pdf'.  Stop.
make[1]: Leaving directory '/home/marxin/Programming/texi2rst-generated/sphinx/gdc/_build/latex'
make: *** [Makefile:58: latexpdf] Error 2

Where _build/latex/ is an empty directory. Is there anything I forgot to change in the aforementioned commit or should I revert it?
@AA-Turner

How to Reproduce

See steps above.

Expected behavior

No response

Your project

Sphinx

Screenshots

No response

OS

Linux

Python version

3.10

Sphinx version

master

Sphinx extensions

No response

Extra tools

No response

Additional context

No response

[jfbu]

You can see same phenomenon emitting three successive commands in your shell:

• make clean
• make html
• make man
  Then _build/man is actually empty and man build did not happen.

I did not investigate why but if last step is replaced by make text, the _build/text/index.txt is created. But make texinfo behaves like make latex or make man: nothing in the corresponding build directories appears.

I am not familiar enough with builders/__init__.py, but indeed this commit should be reverted or problem fixed prior to release, naturally. I marked it as regression but it has not yet made it to a release... only to stamp it as a "must-fix" issue ;-)

[jfbu]

I did not investigate why but if last step is replaced by make text

It appears to_build in build_update() from builders/__init__.py is then a generator not a string and list(to_build) produces the list ['index'] and not ['__all__'] so the b1390c4 logic does not inhibit the build process then.

[marxin]

All right, so it's likely related to the fact the some builders don't inform about outdated docs and they rather return a string which means effectively a rebuild:

sphinx/sphinx/builders/manpage.py

Lines 40 to 41 in c51a88d

	def get_outdated_docs(self) -> Union[str, List[str]]:
	return 'all manpages' # for now

sphinx/sphinx/builders/latex/__init__.py

Lines 127 to 128 in c51a88d

	def get_outdated_docs(self) -> Union[str, List[str]]:
	return 'all documents' # for now

That said, one can't effectively distinguish when a rebuild is needed or not. Thus I'm going to revert my commit.

On the other hand, HTML has a more fine-grained control over the outdated docs:

sphinx/sphinx/builders/html/__init__.py

Lines 402 to 415 in c51a88d

	def get_outdated_docs(self) -> Iterator[str]:
	try:
	with open(path.join(self.outdir, '.buildinfo'), encoding="utf-8") as fp:
	buildinfo = BuildInfo.load(fp)
	if self.build_info != buildinfo:
	logger.debug('[build target] did not match: build_info ')
	yield from self.env.found_docs
	return
	except ValueError as exc:
	logger.warning(__('Failed to read build info file: %r'), exc)
	except OSError:
	# ignore errors on reading
	pass

[AA-Turner]

I wonder if the right fix here is ensuring a consistent contract in get_outdated_docs? I'll merge the revert though for now.

A