Re: linux-6.12-rc1/drivers/iio/imu/bmi323/bmi323_core.c:133: Array contents defined but not used ?

[Nathan Chancellor <nathan@kernel.org>]

On Mon, Sep 30, 2024 at 03:58:06PM +0200, Javier Carrasco wrote:
> On 30/09/2024 15:49, David Binderman wrote:
> > Hello there,
> > 
> > I just tried to build linux-6.12-rc1 with clang. It said:
> > 
> > drivers/iio/imu/bmi323/bmi323_core.c:133:27: warning: variable 'bmi323_ext_reg_savestate' is not needed and will not be emitted [-Wunneeded-internal-declaration]
> > 
> > A grep for the identifier shows the following strange results::
> > 
> > inux-6.12-rc1 $ grep bmi323_ext_reg_savestate drivers/iio/imu/bmi323/bmi323_core.c
> > static const unsigned int bmi323_ext_reg_savestate[] = {
> > 	unsigned int ext_reg_settings[ARRAY_SIZE(bmi323_ext_reg_savestate)];
> > 	for (unsigned int i = 0; i < ARRAY_SIZE(bmi323_ext_reg_savestate); i++) {
> > 	for (unsigned int i = 0; i < ARRAY_SIZE(bmi323_ext_reg_savestate); i++) {
> > linux-6.12-rc1 $ 
> > 
> > I see no mention of bmi323_ext_reg_savestate[ i]. Is there a possible
> > cut'n'paste error in one of the two for loops ?
> > 
> > Regards
> > 
> > David Binderman
> 
> 
> I think that is a bug in clang:
> 
> https://bugs.llvm.org/show_bug.cgi?id=33068
> 
> That happens because clang sees that bmi323_ext_reg_savestate is not
> used but to gets its size, and that means for it that the variable is
> not needed. That does not happen for example with
> bmi323_ext_reg_savestate (right above bmi323_ext_reg_savestate) because
> that one is used beyond ARRAY_SIZE.
> 
> Safe to ignore?

As later comments in this thread point out, this is a bug in the code:

https://git.kernel.org/jic23/iio/c/506a1ac4c4464a61e4336e135841067dbc040aaa

That bug report is pretty misguided, as that is exactly the type of code
that warning tries to catch: the programmer has a static variable that
is only used in sizeof(), a compile time evaluation, so they won't get a
-Wunused-variable since it is used in that expression, but did they
intend to use it elsewhere? If they didn't, they can just slap
'__maybe_unused' / '__attribute__((unused))' on it or use
sizeof(<type>), which removes all ambiguity. Otherwise, the code should
be fixed, like in this case.

Cheers,
Nathan