Progression not functional once a single error / fatal line was encountered

[mxrch]

To run the script properly, just create a creds.py file with a token variable, and a .tmp folder

import creds

from git import Repo, RemoteProgress
from pathlib import Path

import uuid


username = "gitpython-developers"
repo_name = "GitPython"
uuid = uuid.uuid4().hex[:20]

# ---------------

class Progress(RemoteProgress):
 def update(self, op_code, cur_count, max_count=None, message=''):
 print(f"{op_code} {cur_count} {max_count} {message}")

repo_url = f'https://{creds.token}:x-oauth-basic@github.com/{username}/{repo_name}'
repo_path = Path('.tmp') / Path(uuid)
print(f'Cloning the commits history of {repo_name}...')
err = None
try:
 Repo.clone_from("https://github.com/gitpython-developers/GitPython", repo_path, progress=Progress(), multi_options=["--filter=tree:0"])
 #Repo.clone_from(repo_url, repo_path, progress=Progress(), multi_options=["--filter=tree:0"])
except Exception as error:
 err = error
 print(err)

If you comment the 1st Repo.clone_from and uncomment the 2nd, the progress will not work anymore, even though the progress is showing up using the git binary.

[Byron]

Thanks a lot for putting in the time to make this issue reproducible. With the latest GitPython from main and tested with my own token, the following code produced progress messages just like the one above.

import creds

from git import Repo, RemoteProgress
from pathlib import Path

import uuid


username = "gitpython-developers"
repo_name = "GitPython"
uuid = uuid.uuid4().hex[:20]

# ---------------

class Progress(RemoteProgress):
 def update(self, op_code, cur_count, max_count=None, message=''):
 print(f"{op_code} {cur_count} {max_count} {message}")

repo_url = f'https://{creds.token}:x-oauth-basic@github.com/{username}/{repo_name}'
repo_path = Path('.tmp') / Path(uuid)
print(f'Cloning the commits history of {repo_name}...')
err = None
try:
 #Repo.clone_from("https://github.com/gitpython-developers/GitPython", repo_path, progress=Progress(), multi_options=["--filter=tree:0"])
 Repo.clone_from(repo_url, repo_path, progress=Progress(), multi_options=["--filter=tree:0"])
except Exception as error:
 err = error
 print(err)

Thus I was unable to reproduce the issue, running on python 3.8 on MacOS. This is something I would expect as there is no special case for the URL at all.

GitPython/git/repo/base.py

Lines 1012 to 1016 in 96f8f17

	proc = git.clone(multi, Git.polish_url(url), clone_path, with_extended_output=True, as_process=True,
	v=True, universal_newlines=True, **add_progress(kwargs, git, progress))
	if progress:
	handle_process_output(proc, None, to_progress_instance(progress).new_message_handler(),
	finalize_process, decode_streams=False)

Maybe there is something else that influences whether or not progress is sent that isn't present in my particular setup?

[mxrch]

Thanks for your answer.
I updated Git SCM and GitPython to the latest version, but I can't manage to get it working. Maybe it's because of Windows ? But it would be no sense since we get the same progress with the two different URLs using the Git CLI.
Do you know a GitPython contributor on Windows that can test the same thing ? @Byron

[mxrch]

Maybe it's the way you get the stream that doesn't work the same way on Windows.
I'll try to play with GitPython and put some print() everywhere.

[mxrch]

Okay it works well on my Linux VM.

So the problem should be Windows

[Byron]

Thanks for sorting this out, very helpful! It's definitely something the next major release of GitPython will fix.
In the mean time, you could call git directly, using subprocess, and have its stderr/progress written directly to the terminal if that's an option. Parsing it yourself is what GitPython is trying to do, which apparently is an issue on windows at least the way it's implemented currently.
If you come across a fix for this it would be great to read about it here.
Thank you.

[mxrch]

I'll let you know, thank you, for the moment I'm trying to understand all your pumps / streams things aha 😄
By the way, were you able to confirm this case doesn't work for another Windows user ? Just to be sure that it's not because of my environment

[mxrch]

I added a print in this function, before the two handlers :

GitPython/git/cmd.py

Lines 78 to 89 in 96f8f17

	def pump_stream(cmdline, name, stream, is_decode, handler):
	try:
	for line in stream:
	if handler:
	if is_decode:
	line = line.decode(defenc)
	handler(line)
	except Exception as ex:
	log.error("Pumping %r of cmd(%s) failed due to: %r", name, remove_password_if_present(cmdline), ex)
	raise CommandError(['<%s-pump>' % name] + remove_password_if_present(cmdline), ex) from ex
	finally:
	stream.close()

And the stream is well received in the two case, here is the x-oauth-basic case :

And with a normal link :

So the issue should be in the output processing.

[Byron]

I can't confirm it nor deny it. My spotty memory tells me this isn't the first time I hear about an issue like it though, if that helps.
Awesome, I have a feeling you will soon have a PR ready which fixes this issue once and for all.

[mxrch]

I finally figured out the issue !

GitPython printed this line at the end so I didn't think it was the issue, but it was : "fatal: credential-cache unavailable; no unix socket support" and the stream got it at the beginning.
And since in your parsing function here :

GitPython/git/util.py

Lines 426 to 446 in 96f8f17

	def _parse_progress_line(self, line: AnyStr) -> None:
	"""Parse progress information from the given line as retrieved by git-push
	or git-fetch.
	- Lines that do not contain progress info are stored in :attr:`other_lines`.
	- Lines that seem to contain an error (i.e. start with error: or fatal:) are stored
	in :attr:`error_lines`."""
	# handle
	# Counting objects: 4, done.
	# Compressing objects: 50% (1/2)
	# Compressing objects: 100% (2/2)
	# Compressing objects: 100% (2/2), done.
	if isinstance(line, bytes): # mypy argues about ternary assignment
	line_str = line.decode('utf-8')
	else:
	line_str = line
	self._cur_line = line_str
	if self.error_lines or self._cur_line.startswith(('error:', 'fatal:')):
	self.error_lines.append(self._cur_line)
	return

if there is only one line that starts with "error:" or "fatal:", it put all the next output in the "self.error_lines" variable even if they are not errors, and so that condition still True until the end.
My solution was to fix the error itself by doing it :

git config --global credential.helper wincred

But what do you think about editing the parsing function to let the progress going on even if there was an error at the beginning ?