On Sat, Apr 22, 2023 at 1:42 PM Andrew Dunstan <and...@dunslane.net> wrote: > > > On 2023-04-22 Sa 04:50, Michael Paquier wrote: > > On Fri, Apr 21, 2023 at 09:58:17AM +0200, Jelte Fennema wrote: > > For 2 the upstream thread listed two approaches: > a. Install a pre-receive git hook on the git server that rejects > pushes to master that are not indented > b. Add a test suite that checks if the code is correctly indented, so > the build farm would complain about it. (Suggested by Peter E) > > I think both a and b would work to achieve 2. But as Peter E said, b > indeed sounds like less of a divergence of the status quo. So my vote > would be for b. > > FWIW, I think that there is value for both of them. Anyway, isn't 'a' > exactly the same as 'b' in design? Both require a build of > pg_bsd_indent, meaning that 'a' would also need to run an equivalent > of the regression test suite, but it would be actually costly > especially if pg_bsd_indent itself is patched. I think that getting > more noisy on this matter with 'b' would be enough, but as an extra > PG_TEST_EXTRA for committers to set. > > Such a test suite would need a dependency to the 'git' command itself, > which is not something that could be safely run in a release tarball, > in any case. > > > Perhaps we should start with a buildfarm module, which would run pg_indent > --show-diff. That would only need to run on one animal, so a failure wouldn't > send the whole buildfarm red. This would be pretty easy to do.
Just to be clear, you guys are aware we already have a git repo that's supposed to track "head + pg_indent" at https://git.postgresql.org/gitweb/?p=postgresql-pgindent.git;a=shortlog;h=refs/heads/master-pgindent right? I see it is currently not working and this has not been noticed by anyone, so I guess it kind of indicates nobody is using it today. The reason appears to be that it uses pg_bsd_indent that's in our apt repos and that's 2.1.1 and not 2.1.2 at this point. But if this is a service that would actually be useful, this could certainly be ficked pretty easy. But bottom line is that if pgindent is as predictable as it should be, it might be easier to use that one central place that already does it rather than have to build a buildfarm module? //Magnus
