Didier 'OdyX' Raboud writes ("Bug#880658: dgit server rejects unattributed
commits"):
> Just did that, with that commit as HEAD:
> https://github.com/OdyX/test-broken-commits
> The push went through without problem.
Thanks for checking that.
> I don't get why it makes sense to block on git commits which are
> valid in git's eyes… dgit doesn't seem like the right place to be
> strict about these, frankly.
Well, the proximate reason I added this check is to prevent the spread
of corrupted commits generated by dgit due to #849041. (See [1].)
But in the general case, I think it is not sensible to have our repo
accept commits which are so broken that significant numbers of, or
important, servers will reject. That would apply to every repo,
really - I don't think the dgit repos server is special. It's just
that the dgit repos server needs to have a check because of #849041
and also provides a useful opportunity for checking.
Given the way git works and the pain of fixing things later, it is
better to discover these things early. And that, combined with the
lack of any coherent specification (either from git upstream, or
anyone else) means I wrote a defensive commit checker.
I'm sorry that this has got in your way, of course, and I will improve
the checker to be more subtle.
I don't find the argument "valid in git's eyes" at all convincing.
git is far from a good example of robust protocol design and
implementation. (Shall I tell you about the \0 feature signalling
hack in the git protocol, or about the strangeness one can experience
with extension headers in commit and tag objects?)
I hope that's enough of an explanation to convince you. If not I'm
happy to talk about it further but we should probably have a separate
bug for that conversation. Feel free to clone this one.
Ian.
[1] https://lists.debian.org/debian-devel-announce/2017/01/msg00001.html
--
Ian Jackson <[email protected]> These opinions are my own.
If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.
