Refactor collecting-PR script for release note

[nateanl]

refer to pytorch/vision#4626

[mthrok]

Looking at the code it is offending the bandit, it looks like the subprocess.Popen(command, ...) is causing the issue.

Before disabling the bandit, can you try changing the following commands into list of string?
I think then, we can remove shell=True.

audio/scripts/retrieve_prs.py

Line 35 in 184466a

cmd = f"git log -n 1 --pretty=format:%s {commit_hash}"

audio/scripts/retrieve_prs.py

Line 105 in 184466a

cmd = f"git merge-base {base_version} {new_version}"

audio/scripts/retrieve_prs.py

Line 111 in 184466a

cmd = f"git log --reverse --oneline {merge_base}..{new_version}"

[mthrok]

The git command should work just fine without shell=True, like the following

audio/setup.py

Lines 84 to 86 in 184466a

	sha = _run_cmd(['git', 'rev-parse', 'HEAD'])
	branch = _run_cmd(['git', 'rev-parse', '--abbrev-ref', 'HEAD'])
	tag = _run_cmd(['git', 'describe', '--tags', '--exact-match', '@'])

[carolineechen]

thanks for looking into and working on this!

[mthrok]

I think this is different from what it used to be.

Suggested change

	cmd = ['git', 'log', '-n', '1', f'--pretty=format:%s {commit_hash}']
	cmd = ['git', 'log', '-n', '1', '--pretty=format:%s', f'{commit_hash}']

[mthrok]

@nateanl Can you make a small PR for the part only related to shell=True so that we can merge that one first quickly? That will suppress the bandit failure on other PRs.

[carolineechen]

Suggested change

	"""Collect the PRs between the current and previous releases and
	"""Collect the PRs between two specified tags or commits and

[mthrok]

Suggested change

	def _parse_args(args):
	def _parse_args(args=None):

[mthrok]

Note: This would fail if the commit message is not encoded with utf-8. (i.e. if user configures their git with non-uff commit message https://www.git-tower.com/help/guides/faq-and-tips/faq/encoding/windows)

The script needs to handle such case, while logging the failure and keep processing the rest of commits.

[mthrok]

1. We need a way to detect the correct encoding from the result of check_output.
2. The for loop in the main function has to handle error properly and notify the user.

[nateanl]

Will follow-up the case in the future PR.