Re: Compiler warning in costsize.c

On Wed, Apr 5, 2017 at 2:54 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Michael Paquier <michael(dot)paquier(at)gmail(dot)com> writes:
>> In builds where USE_ASSERT_CHECKING is not enabled, costsize.c can
>> generate warnings. Here is for example with MSVC:
>> src/backend/optimizer/path/costsize.c(4520): warning C4101: 'rte' : unreferenced local variable [C:\Users\ioltas\git\postgres\postgres.vcxproj]
>> src/backend/optimizer/path/costsize.c(4640): warning C4101: 'rte' : unreferenced local variable [C:\Users\ioltas\git\postgres\postgres.vcxproj]
>
>> a9c074ba7 has done an effort, but a bit more is needed as the attached.
>
> That doesn't look right at all:
>
> +#ifdef USE_ASSERT_CHECKING
> RangeTblEntry *rte PG_USED_FOR_ASSERTS_ONLY;
> +#endif
>
> The entire point of the PG_USED_FOR_ASSERTS_ONLY annotation is to suppress
> this type of warning without a need for an explicit #ifdef like that one.
>
> We should either fix PG_USED_FOR_ASSERTS_ONLY to work on MSVC as well,
> or decide that we don't care about suppressing such warnings on MSVC,
> or give up on PG_USED_FOR_ASSERTS_ONLY and remove it everywhere in
> favor of plain #ifdefs.

Visual does not have any equivalent of __attribute__((unused))... And
visual does not have an equivalent after looking around. A trick that
I can think of would be to change PG_USED_FOR_ASSERTS_ONLY to be
roughly a macro like that:

#if defined(__GNUC__)
#define PG_ASSERT_ONLY(x) ASSERT_ONLY_ ## x __attribute__((unused))
#else
#define PG_ASSERT_ONLY(x) ((void) x)
#endif

But that's ugly...

> (I'm personally not that much in love with PG_USED_FOR_ASSERTS_ONLY,
> because it tends to confuse pgindent.)

I would be incline to just do that, any other solution I can think of
is uglier than that.
--
Michael