2021-03-16 09:34:25

by Rasmus Villemoes

[permalink] [raw]
Subject: Re: [PATCH 01/13] tools: disable -Wno-type-limits

On 16/03/2021 02.54, Yury Norov wrote:
> GENMASK(h, l) may be passed with unsigned types. In such case, type-limits
> warning is generated for example in case of GENMASK(h, 0).
>
> Signed-off-by: Yury Norov <[email protected]>
> ---
> tools/scripts/Makefile.include | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/tools/scripts/Makefile.include b/tools/scripts/Makefile.include
> index 84dbf61a7eca..15e99905cb7d 100644
> --- a/tools/scripts/Makefile.include
> +++ b/tools/scripts/Makefile.include
> @@ -38,6 +38,7 @@ EXTRA_WARNINGS += -Wswitch-enum
> EXTRA_WARNINGS += -Wundef
> EXTRA_WARNINGS += -Wwrite-strings
> EXTRA_WARNINGS += -Wformat
> +EXTRA_WARNINGS += -Wno-type-limits
>

I don't like that kind of collateral damage. I seem to recall another
instance where a macro was instead rewritten to avoid triggering the
type-limits warning (with a comment explaining the uglyness). Something like

foo > bar is the same as
!(foo <= bar) which is the same as
!(foo == bar || foo < bar)

Dunno if that would work here, but if it did, it would have the bonus
that when somebody builds the kernel proper with Wtype-limits enabled
(maybe W=1 or W=2) there would be no false positives from GENMASK to
wade through.

Alternatively, we really should consider making use of _Pragma to
locally disable/re-enable certain warnings.

Rasmus


2021-03-16 18:46:07

by Andy Shevchenko

[permalink] [raw]
Subject: Re: [PATCH 01/13] tools: disable -Wno-type-limits

On Tue, Mar 16, 2021 at 09:17:24AM +0100, Rasmus Villemoes wrote:
> On 16/03/2021 02.54, Yury Norov wrote:
> > GENMASK(h, l) may be passed with unsigned types. In such case, type-limits
> > warning is generated for example in case of GENMASK(h, 0).

...

> I don't like that kind of collateral damage. I seem to recall another
> instance where a macro was instead rewritten to avoid triggering the
> type-limits warning (with a comment explaining the uglyness). Something like
>
> foo > bar is the same as
> !(foo <= bar) which is the same as
> !(foo == bar || foo < bar)
>
> Dunno if that would work here, but if it did, it would have the bonus
> that when somebody builds the kernel proper with Wtype-limits enabled
> (maybe W=1 or W=2) there would be no false positives from GENMASK to
> wade through.
>
> Alternatively, we really should consider making use of _Pragma to
> locally disable/re-enable certain warnings.

Rasmus, in the kernel the same was fixed as per 355a3587d4ca.
I don't know why tools should be different to that.

--
With Best Regards,
Andy Shevchenko