On Tue, Mar 18, 2025 at 1:55 PM Tom Lane <t...@sss.pgh.pa.us> wrote: > 10s added to every check-world run doesn't sound "cheap" to me.
Really? If you're sensitive to the time the tests take to run, maybe use 'meson test' rather than 'make check-world'? I do that locally with MESON_TESTTHREADS=8 and the entire test suite finishes in under 3 minutes. A 10-second pgindent test will add something to that, presumably, because it'll take up one of those test threads for the time it's running, but if you assume perfect thread utilization before and after adding this test we're talking about the difference between 'meson test' finishing in 2:55 and 2:57, or something like that. That seems like pretty cheap insurance to me for what is, at least for me, a common error. (make check-world for me takes 12:57 -- and yes that could be faster with parallelism too, but then the output is jumbled. At any rate, if I'm waiting that long, I'm getting up and walking away from the computer and then the difference between being gone for 13 minutes and 13 minutes and 10 seconds isn't really material either, at least not to me.) > However, both pg_bsd_indent and the typedefs list are in-tree > nowadays, so I don't see any reason that this couldn't be a > totally self-contained check. Well, +1 from me. Every year when I go through the commit log from the previous year for contribution statistics, it always shocks me how fix-up commits we have. Now, some of that is unavoidable -- people are always going to miss things in review and discover them after commit. But it should be possible to reduce purely mechanical errors like failing to rerun pgindent after the very last cosmetic change, or whatever. I'd like to spend more time on substantive issues and less on things that could have been caught automatically. Tons and tons of commits have two, three, or even more fix-up commits and that all adds up to a lot of committer time that could be better spent. -- Robert Haas EDB: http://www.enterprisedb.com
