Packit-service: Unexpected username and e-mail in a changelog entry generated by propose_downstream job

[nforro]

See for example here.

rebase-helper <[email protected]> are fallback values used by rebase-helper in case it's not possible to obtain the proper values from git configuration (i.e. git config --get user.name and git config --get user.email fail).

[jpopelka]

hmm, the image looks OK to me:

$ docker run -ti --rm usercont/packit-service-worker:prod git config -l
user.name=Packit Service
[email protected]

[TomasTomecek]

...but sandcastle doesn't have any

$ docker run -ti --rm usercont/sandcastle:prod git config --get user.name
$ echo $?
1

[jpopelka]

I thought we use sandcastle only to build SRPM.

[lachmanfrantisek]

...but sandcastle doesn't have any

Can we move this issue to sandcastle then?

[TomasTomecek]

sandcastle is being used in multiple places and it seems bumpspec is executed in sandcastle

this definitely needs some investigation

anyway, @nforro this is great for you, your project is getting the credit :))

[lachmanfrantisek]

ping sync2jira

[lbarcziova]

Some more investigation:

• updating of the changelog is not happening in sandcastle

• the commits in which the changelog entries have these default values have the names in Signed off section set correctly, but not the email (should be [email protected]):
  Signed-off-by: Packit Service <[e-mail]>
Signed-off-by: Packit Service <[e-mail]> (haven't found record with worker-0)
  -> the ~/.gitconfig is present (user.name in commit message correct), values are overwritten for some local git repo?

• happened in that time in other projects too - https://src.fedoraproject.org/rpms/python-ogr/blob/f32/f/python-ogr.spec#_58, https://src.fedoraproject.org/rpms/packit/blob/f32/f/packit.spec#_84

[TomasTomecek]

what does rpmdev-packager print?

$ rpmdev-packager                                                                                                                                
Tomas Tomecek <[email protected]>

[lbarcziova]

In workers:

$ rpmdev-packager
id: cannot find name for user ID ...
Packit Service <[email protected]>

:thinking:
Anyway, when updating changelog, the name and email should be obtained from git config command.

[TomasTomecek]

It seems that we are using rebase-helper to update the changelog:

    def update_changelog_in_spec(self, changelog_entry):
        if hasattr(self, "update_changelog"):
            # new rebase-helper
            self.update_changelog(changelog_entry)

So could the issue be in rebase-helper itself? We should investigate how it obtains the username and email.

[nforro]

So could the issue be in rebase-helper itself? We should investigate how it obtains the username and email.

As mentioned in the first post, rebase-helper just calls git config --get user.name and git config --get user.email (using GitPython, but that shouldn't matter).

[lbarcziova]

It seems that we are using rebase-helper to update the changelog

Yes, I was checking the method calls, that's how I came to the rebase-helper code which I linked in the comment above.

So could the issue be in rebase-helper itself? We should investigate how it obtains the username and email.

I don't think so, as it was said, it just uses GitPython methods.

<[e-mail]> but these emails in Signed-off are weird as well.

[lbarcziova]

I was checking some sentry issues from times when this was happening and found some relevant logs, the git config --get commands failed in worker in /src dir:

subprocess      git config --get user.name
subprocess.cwd: /src

rebasehelper.helpers.git_helper
Failed to get configured git user name, using 'rebase-helper'



md5-44031a8f4288195da8e48c14f6906e1f



subprocess      git config --get user.email
subprocess.cwd: /src



md5-44031a8f4288195da8e48c14f6906e1f



rebasehelper.helpers.git_helper
Failed to get configured git user email, using '[email protected]'

Tried to reproduce, but commands return correct values now, and I haven't found any occurrences of this error since 21th September, so I propose to close this issue and possibly revisit it if it happens in future.

[TomasTomecek]

@lbarcziova oh my, what a cluster-f... mess; thanks for such a thorough investigation

since we are now confident that the way %changelog user is being populed is via git config --get, would it make sense to have a test case which would run in the worker deployed in p-s and check that we are getting the value from the git-config-get we are expecting?

[lachmanfrantisek]

@lbarcziova this is solved(=not occurring anymore and a test added) by your PR, isn't it?

btw, Thanks for your deep investigation!

[lbarcziova]

@lbarcziova this is solved(=not occurring anymore and a test added) by your PR, isn't it?

Yes, closing.