AssertionError with LocalGitClient on prior Git cloned repo into MemoryRepo?

[dAnjou]

Hi,

after reporting #1157 I finally got around to having another try, and unfortunately I'm getting a different error now:

Traceback (most recent call last):
  File "/builds/dAnjou/fs-code/src/codefs/_core.py", line 98, in _delegate
    self.mount(mount_point, self.fetch_archive(ref))
                            ^^^^^^^^^^^^^^^^^^^^^^^
  File "/builds/dAnjou/fs-code/src/codefs/gitfs/__init__.py", line [87](https://gitlab.com/dAnjou/fs-code/-/jobs/4301134873#L87), in __call__
    result = self._client.fetch(self._path, self._repo)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/dulwich/client.py", line 1455, in fetch
    refs = r.fetch(
           ^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/dulwich/repo.py", line 417, in fetch
    target.object_store.add_pack_data(count, pack_data, progress)
  File "/usr/local/lib/python3.11/site-packages/dulwich/object_store.py", line 1056, in add_pack_data
    self.add_object(unpacked_object.sha_file())
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/dulwich/pack.py", line 223, in sha_file
    assert self.obj_type_num is not None and self.obj_chunks is not None
                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError

I noticed the error in one of my CI jobs: https://gitlab.com/dAnjou/fs-code/-/jobs/4301134873, but it can also be reproduced like this:

import subprocess
from dulwich.client import get_transport_and_path
from dulwich.repo import MemoryRepo

# needs pytest for tmp_path fixture

# fails
def test_clone_repo(tmp_path: Path) -> None:
    subprocess.run(["git", "--version"])  # my output: git version 2.40.1
    subprocess.run(["git", "clone", "https://gitlab.com/dAnjou/fs-code.git", str(tmp_path)])
    client, path = get_transport_and_path(str(tmp_path))
    repo = MemoryRepo()
    assert client.fetch(path, repo)

# passes
def test_init_repo(tmp_path: Path) -> None:
    subprocess.run(["git", "--version"])  # my output: git version 2.40.1
    subprocess.run(["git", "init", str(tmp_path)])
    (tmp_path / "file").write_text("lorem ipsum")
    subprocess.run(["git", "add", "file"], cwd=tmp_path)
    subprocess.run(["git", "commit", "-m", "foobar"], cwd=tmp_path)
    client, path = get_transport_and_path(str(tmp_path))
    repo = MemoryRepo()
    assert client.fetch(path, repo)

UPDATE 1: Noticed that it does not happen with a Urllib3HttpGitClient, but it does with a LocalGitClient.

UPDATE 2: I also have a passing test where I create a local repo with Dulwich, the failure seems to happen with a Git cloned repo.

UPDATE 3: Better code sample for reproduction, needs pytest for tmp_path fixture.

UPDATE 4: Yup, needs to be a Git cloned repo, a Git initialized repo does not show the failure.