[Patchew-devel] [PATCH] importer: fix handling of missing commits

[Fam Zheng]

> On Mar 18, 2019, at 16:46, Paolo Bonzini <pbonzini at redhat.com> wrote:
> 
> On 17/03/19 03:56, Fam Zheng wrote:
>>> -        output = subprocess.check_output("""git log --format=%%b %s..%s |
>>> +        output = subprocess.check_output("""git show --format=%%b %s |
>>>                 awk 'BEGIN{IGNORECASE=1} /^message-id:/{print}'
>>>                 """% \
>>> -                (old_head, new_head), shell=True, cwd=clone).decode()
>>> +                ' '.join(new_commits), shell=True, cwd=clone).decode()
>> How does this make sure all message-ids in each commits between old_head and
>> new_head are collected? That list is used by the server to mark multiple series
>> merged.
> 
> It doesn't.  On the other hand this only happens if someone force-pushes
> to master, and the importer is redeployed in such a way that the
> old_head is not part of the importer's clone.  Given the circumstances,
> IMHO this is neither a big issue, nor a fully fixable issue.
> 
> On the other hand it can happen, and it is the cause of the broken
> next.patchew.org git imports (as I found out after dropping all the
> template crap from the logs...) so it's important to do _something_
> about it.

OK, I see the logic now. Thanks.

Reviewed-by: Fam Zheng <famz at redhat.com>