Re: run pgindent on a regular basis / scripted manner

Hi,

On 2023-10-16 20:45:00 -0400, Tom Lane wrote:
> Peter Geoghegan <pg@bowt.ie> writes:
> 2. We could raise awareness of this issue by adding indent verification
> to CI testing.  I hesitate to suggest that, though, for a couple of
> reasons:
>    2a. It seems fairly expensive, though I might be misjudging.

Compared to other things it's not that expensive. On my workstation, which is
slower on a per-core basis than CI, a whole tree pgindent --silent-diff takes
6.8s.  For That's doing things serially, it shouldn't be that hard to parallelize
the per-file processing.

For comparison, the current compiler warnings task takes 6-15min, depending on
the state of the ccache "database". Even when ccache is primed, running
cpluspluscheck or headerscheck is ~30s each. Adding a few more seconds for an
indentation check wouldn't be a problem.


>    2b. It's often pretty handy to submit patches that aren't fully
>    indent-clean; I have such a patch in flight right now at [1].
>
> 2b could be ameliorated by making the indent check be a separate
> test process that doesn't obscure the results of other testing.

The compiler warnings task already executes a number of tests even if prior
tests have failed (to be able to find compiler warnings in different compilers
at once). Adding pgindent cleanliness to that would be fairly simple.


I still think that one of the more important things we ought to do is to make
it trivial to check if code is correctly indented and reindent it for the
user.  I've posted a preliminary patch to add a 'indent-tree' target a few
months back, at
https://postgr.es/m/20230527184201.2zdorrijg2inqt6v%40alap3.anarazel.de

I've updated that patch, now it has
- indent-tree, reindents the entire tree
- indent-head, which pgindent --commit=HEAD
- indent-check, fails if the tree isn't correctly indented
- indent-diff, like indent-check, but also shows the diff

If we tought pgindent to emit the list of files it processes to a dependency
file, we could make it cheap to call indent-check repeatedly, by teaching
meson/ninja to not reinvoke it if the input files haven't changed.  Personally
that'd make it more bearable to script indentation checks to happen
frequently.


I'll look into writing a command to update typedefs.list with all the local
changes.

Greetings,

Andres Freund