pip download does not conclude downloaded requirements with new resolver

[McSinyx]

For instance,

$ pip download pip
Collecting pip
  Using cached pip-20.2.1-py2.py3-none-any.whl (1.5 MB)
  Saved ./pip-20.2.1-py2.py3-none-any.whl
Successfully downloaded pip
$ rm pip*
$ pip download pip
Collecting pip
  Using cached pip-20.2.1-py2.py3-none-any.whl (1.5 MB)
  Saved ./pip-20.2.1-py2.py3-none-any.whl

This traces back to

pip/src/pip/_internal/commands/download.py

Lines 137 to 139 in ee4371c

	downloaded = ' '.join([req.name # type: ignore
	for req in requirement_set.requirements.values()
	if req.successfully_downloaded])

and that the new resolver never set InstallRequirement.successfully_downloaded. With GH-8685 merged, we can use successfully_downloaded in place of not needs_more_preparation and solve the issue:

pip/src/pip/_internal/operations/prepare.py

Lines 481 to 498 in ee4371c

	def prepare_linked_requirement(self, req, parallel_builds=False):
	# type: (InstallRequirement, bool) -> Distribution
	"""Prepare a requirement to be obtained from req.link."""
	assert req.link
	link = req.link
	self._log_preparing_link(req)
	wheel_dist = self._fetch_metadata(link)
	if wheel_dist is not None:
	req.needs_more_preparation = True
	return wheel_dist
	return self._prepare_linked_requirement(req, parallel_builds)
	def prepare_linked_requirement_more(self, req, parallel_builds=False):
	# type: (InstallRequirement, bool) -> None
	"""Prepare a linked requirement more, if needed."""
	if not req.needs_more_preparation:
	return
	self._prepare_linked_requirement(req, parallel_builds)

This happens to be the part I'm working on at the moment, so if it is not urgent, I will submit a patch in around 12 hours.

[pradyunsg]

It was "fun" realizing this while we were walking through the codebase. I'm genuinely surprised that no one reported it prior to us stumbling upon this. :P

[McSinyx]

That makes me question if anybody is expecting it at all. uranusjr's point in #8709 (comment) got me thinking we should do something about it so that the output would be helpful to/needed by the users. I'll file an issue asking for more opinions, especially from the UI/UX team in case this was mention during the earlier survey. Edit: filed as GH-8754.

[uranusjr]

Yeah, I get the impressions that the current representation (from the legacy resolver) is so bad that everyone just ignores them, and therefore never notices they are missing from the 2020 resolver.