Can/should pip download directly to target directory

[McSinyx]

Environment

• pip version: ea10d0e
• Python version: 3.8.5, but I think it is irrelevant
• OS: Debian, but I think it is irrelevant

Description

Currently pip download (and pip wheel also I believe) download distributions to a temporary directory. I think this is to unpack sdists (and historically wheels as well), and probably it also has something to do with caching (I'm not really familiar with the caching mechanism and would love to know the answer).

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

Lines 134 to 136 in ea10d0e

	# let's download to a tmp dir
	from_path, content_type = _download_http_url(
	link, downloader, temp_dir.path, hashes

Later these will be copied to the download_dir:

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

Lines 545 to 560 in ea10d0e

	if download_dir:
	if link.is_existing_dir():
	logger.info('Link is a directory, ignoring download_dir')
	elif local_file:
	download_location = os.path.join(
	download_dir, link.filename
	)
	if not os.path.exists(download_location):
	shutil.copy(local_file.path, download_location)
	download_path = display_path(download_location)
	logger.info('Saved %s', download_path)
	if self._download_should_save:
	# Make a .zip of the source_dir we already created.
	if link.is_vcs:
	req.archive(self.download_dir)

I'm wondering is there any reason not to download, at least wheels, directly to the user-specified directory and avoid copying.

This was initially raised as a TODO in #8737 (comment) but was suggested by @chrahunt to be filed in a separate issue instead.

[uranusjr]

IIRC download_dir is explicitly set to a non-temporary location in pip download and pip wheel contexts.

[McSinyx]

I'm closing this since as shown in GH-8827, we wouldn't want to download directly to target directory.