On Thu, May 18, 2023 at 09:01:19AM -0700, Junio C Hamano wrote: > Taylor Blau <me@xxxxxxxxxxxx> writes: > > >> My inclination is punt and simply do not to claim that we have done > >> a good due diligence to ensure with all permutations of ifdef we are > >> including necessary headers. > > > > I think that's the best course of action, too. I see that it's already > > on 'next', thanks. > > Yeah, I am actuall hoping that somebody clever, with time, comes up > with a systematic way to give us better coverage, but until then, I > think it is better to honestly record where we are to future > developers. I faced a similar issue with the -Wunused-parameter patches. Just when I thought I had everything annotated, I'd find some obscure Makefile knob that compiled new code (or even in one case disabled some code that used a variable!). I never came up with a good solution, but relying on CI helped, since it just builds more of the combinations. Obviously that didn't catch this case. We could try to hit more platforms / combinations of knobs in CI, but there are diminishing returns on the compute time. But at least in this case, the old "if it is important, somebody will build it and report the problem" line of thinking worked out. So maybe that is enough. I guess maybe that was more philosophical than helpful. ;) -Peff
