On Tue, Oct 17, 2023 at 10:43 PM Jelte Fennema <postgres@jeltef.nl> wrote:
>
> On Tue, 17 Oct 2023 at 18:53, Maciek Sakrejda <m.sakrejda@gmail.com> wrote:
> > Git push does have an --atomic flag to treat the entire push as a single operation.
>
> I decided to play around a bit with server hooks. Attached is a git
> "update" hook that rejects pushes to the master branch when the new
> HEAD of master does not pass pgindent. It tries to do the minimal
> amount of work necessary. Together with the --atomic flag of git push
> I think this would work quite well.
>
> Note: It does require that pg_bsd_indent is in PATH. While not perfect
> seems like it would be acceptable in practice to me. Its version is
> not updated very frequently. So manually updating it on the git server
> when we do does not seem like a huge issue to me.
If it doesn't know how to rebuild it, aren't we going to be stuck in a
catch-22 if we need to change it in certain ways? Since an old version
of pg_bsd_indent would reject the patch that might include updating
it. (And when it does, one should expect the push to take quite a long
time, but given the infrequency I agree that part is probably not an
issue)
And unless we're only enforcing it on master, we'd also need to make
provisions for different versions of it on different branches, I
think?
Other than that, I agree it's fairly simple. It does nede a lot more
sandboxing than what's in there now, but that's not too hard of a
problem to solve, *if* this is what we want.
(And of course needs to be integrated with the existing script since
AFAIK you can't chain git hooks unless you do it manually - but that's
mostliy mechanical)
--
Magnus Hagander
Me: https://www.hagander.net/
Work: https://www.redpill-linpro.com/