In article <ebcb0c28-7e32-7a71-7912-4bec6ee63...@gmail.com>, Rin Okuyama <rokuyama...@gmail.com> wrote: >Hi, > >On 2019/02/13 6:07, Paul Goyette wrote: >> On Tue, 12 Feb 2019, Rin Okuyama wrote: >> >>> Hi, >>> >>> As Martin pointed out, it is useful for debugging to turn on >>> DIAGNOSTIC for modules (for non-release branches). >>> >>> Now, all modules for amd64 are successfully built with DIAGNOSTIC. >>> >>> I'd like to commit the patch below, if there's no objection. >> >> This would be very helpful. >> >> I would also wonder if we could increase the WARNS?= level from 3 to 5 >(to match the current WARNS?= level used for kernel builds). Has anyone >tried to see how many modules would fail with WARNS?=5 ?? > >Thank you for your comment. > >Well, I examined that (both for GCC7 & clang). Among ~ 360 modules, >- 2 (lua and zfs) need WARNS=0 >- 1 (solaris) needs WARNS=1 >- 136 need WARNS=3 (mostly due to sign-compare) >- 4 need WARNS=4 >- Others can be compiled with WARNS=5 > >I propose this patch: >http://www.netbsd.org/~rin/modules_bump_warns_20190213.patch > >- Bump default value of WARNS for modules from 3 to 5 >- Explicitly set WARNS for modules that fail with WARNS=5 >- Then, expect someone in charge will fix them ;-) > >Thoughts?
Go for it, we can fix the ones that don't come from 3rd party sources opportunistically. christos