On Sun, Oct 11, 2020 at 3:04 PM Masahiro Yamada <masahi...@kernel.org> wrote:
> On Tue, Oct 6, 2020 at 6:08 AM Andrew Lunn <and...@lunn.ch> wrote:
>
>
> I am not a big fan of KBUILD_CFLAGS_W1_<timestamp>
> since it is ugly.
>
> I'd like to start with adding individual flags
> like drivers/gpu/drm/i915/Makefile, and see
> how difficult it would be to maintain it.
I don't really like that either, to be honest, mostly because that is
somewhat incompatible with my plan to move all the warning flags
out the command line and into _Pragma() lines in header files.
> One drawback of your approach is that
> you cannot set KBUILD_CFLAGS_W1_20200930
> until you eliminate all the warnings in the
> sub-directory in interest.
> (i.e. all or nothing approach)
>
> At best, you can only work out from 'old -> new' order
> because KBUILD_CFLAGS_W1_20200326 is a suer-set of
> KBUILD_CFLAGS_W1_20190907, which is a suer-set of
> KBUILD_CFLAGS_W1_20190617 ...
>
>
>
> If you add flags individually, you can start with
> low-hanging fruits, or ones with higher priority
> as Arnd mentions about -Wmissing-{declaration,prototypes}.
>
>
> For example, you might be able to set
> 'subdir-ccflags-y += -Wmissing-declarations'
> to drivers/net/Makefile, while
> 'subdir-ccflags-y += -Wunused-but-set-variable'
> stays in drivers/net/ethernet/Makefile.
I agree with the goal though. In particular it may be helpful to pick
a set of warning flags to always be enabled without also setting
-Wextra, which has different meanings depending on compiler
version, or between gcc and clang.
I wonder how different they really are, and we can probably find
out from https://github.com/Barro/compiler-warnings, but I have
not checked myself.
Arnd
