Hi Nathan and Mark,
On Fri, Jul 10, 2020 at 11:30 PM Nathan Chancellor
<natechancel...@gmail.com> wrote:
>
> Sure, you could hide it behind an ifdef for either CONFIG_OF or MODULE
> (since you could build this as a module with CONFIG_OF disabled).
>
> I just figured this would be something frowned upon but if that is how
> you would prefer it to be fixed, then I have no objections to it.
>
That's how things used to work in the past: we had to #ifdef CONFIG_OF
all the driver of_XXX code, so it could be built on !CONFIG_OF.
I remember problems with this approach: it generated a lot of
visual noise, it was fragile in both directions: including too much,
and excluding too much. Last but not least, it required us ARM people
to test build each patch on !CONFIG_OF, which many conveniently forgot
to do :)
My "vote" would be to fix the warning using compiler magic. For
example:
#if !CONFIG_OF
// #define of_match_ptr(x) NULL
#define of_match_ptr(x) ((x) == NULL ? NULL : NULL)
#endif
That seems to eliminate the warning on my gcc version 7.5.0, but of
course as compilers get more clever, this could eventually
generate a warning (if it doesn't already on clang).
So perhaps use compiler attributes instead?
Stand-alone testable code snippet below.
======================================================
// gcc -Wall unused.c
// results in: match_table is NULL
// gcc -Wall -DCONFIG_OF unused.c
// results in: match_table[0] = 5
#include <stdio.h>
#ifdef CONFIG_OF
#define of_match_ptr(x) x
#else
#define of_match_ptr(x) ((x) == NULL ? NULL : NULL)
#endif
struct of_device_id {
int id;
};
static const struct of_device_id some_ids[] = {
{ .id = 5, },
{ /* sentinel */ }
};
struct driver {
const struct of_device_id *of_match_table;
};
static const struct driver my_driver = {
.of_match_table = of_match_ptr(some_ids),
};
int main()
{
if (my_driver.of_match_table)
printf("match_table[0] = %d\n", my_driver.of_match_table[0].id);
else
printf("match_table is NULL\n");
}
