Re: [PATCH] staging: wfx: avoid compiler warning on empty array

[Arnd Bergmann]

On Thu, Apr 30, 2020 at 10:42 AM Jerome Pouiller
<Jerome.Pouiller@xxxxxxxxxx> wrote:
> On Wednesday 29 April 2020 22:34:56 CEST Arnd Bergmann wrote:
> > On Wed, Apr 29, 2020 at 6:04 PM Jerome Pouiller
> > <Jerome.Pouiller@xxxxxxxxxx> wrote:
> > > On Wednesday 29 April 2020 16:21:09 CEST Arnd Bergmann wrote:
> > > >
> > > > -static const struct of_device_id wfx_sdio_of_match[];
> > > > +static const struct of_device_id wfx_sdio_of_match[] = {
> > > > +       { .compatible = "silabs,wfx-sdio" },
> > > > +       { .compatible = "silabs,wf200" },
> > > > +       { },
> > > > +};
> > > > +MODULE_DEVICE_TABLE(of, wfx_sdio_of_match);
> > >
> > > I suggest to keep the '#ifdef CONFIG_OF' around this definition. If
> > > CONFIG_OF is undefined, of_match_ptr() and of_match_node() will be NULL
> > > and it should compile.
> >
> > I would generally always go for fewer #ifdef instead of more when the result
> > is the same. Are you worried about wasting 600 bytes of object code size for
> > the array on systems that need this driver but not CONFIG_OF, or something
> > else?
>
> I am not very concerned about the size of the object. However, I think
> that all the modules should apply the same policy regarding the device
> tables. With a few greps, I found 3954 struct of_device_id. About 500 are
> inside #ifdef and about 1000 use of_match_ptr().
>
> Should we consider that the structs of_device_id have to be defined even
> if CONFIG_OF is not defined? And In this case, should we drop
> of_match_ptr()?
>
> Or in contrary, when kernel is compiled without CONFIG_OF, no modules
> should contains OF entries in its device table?

I think the drivers that use an #ifdef here just do so for historic reasons.
In the linux-2.6 days, this caused build failures, but just leaving them
defined has worked for a long time.

       Arnd