Re: run pgindent on a regular basis / scripted manner

On Tue, 17 Oct 2023 at 17:47, Peter Geoghegan <pg@bowt.ie> wrote:
> Once you figure all that out, you're still obligated to hand-polish
> typedefs.list to be consistent with whatever Bruce's machine's copy of
> objdump does (or is it Tom's?). You need to sort the entries so they
> kinda look like they originated from the same source as existing
> entries, since my Debian machine seems to produce somewhat different
> results to RHEL, for whatever reason. It's hard to imagine a worse use
> of committer time.
>
> I think that something like this new policy could work if the
> underlying tooling was very easy to use and gave perfectly consistent
> results on everybody's development machine. Obviously, that just isn't
> the case.

To make koel pass you don't need to worry about hand-polishing
typedefs.list. koel uses the typedefs.list that's committed into the
repo, just like when you run pgindent yourself. If you forget to
update the typedefs.list with new types, then worst case the pgindent
output will look weird. But it will look weird both on your own
machine and on koel. So afaik the current tooling should give
perfectly consistent results on everybody's development machine. If
you have an example of where it doesn't then we should fix that
problem.

On Tue, 17 Oct 2023 at 17:47, Peter Geoghegan <pg@bowt.ie> wrote:
>
> On Tue, Oct 17, 2023 at 8:24 AM Robert Haas <robertmhaas@gmail.com> wrote:
> > I also just discovered that my pre-commit hook doesn't work if I pull
> > commits into master by cherry-picking. I had thought that I could have
> > my hook just check my commits to master and not all of my local dev
> > branches where I really don't want to mess with this when I'm just
> > banging out a rough draft of something. But now I see that I'm going
> > to need to work harder on this if I actually want it to catch all the
> > ways I might screw this up.
>
> Once you figure all that out, you're still obligated to hand-polish
> typedefs.list to be consistent with whatever Bruce's machine's copy of
> objdump does (or is it Tom's?). You need to sort the entries so they
> kinda look like they originated from the same source as existing
> entries, since my Debian machine seems to produce somewhat different
> results to RHEL, for whatever reason. It's hard to imagine a worse use
> of committer time.
>
> I think that something like this new policy could work if the
> underlying tooling was very easy to use and gave perfectly consistent
> results on everybody's development machine. Obviously, that just isn't
> the case.
>
> --
> Peter Geoghegan