[cfe-commits] tgmath.h

[Chris Lattner]

On Feb 18, 2009, at 9:48 AM, Douglas Gregor wrote:

>
> On Feb 18, 2009, at 9:30 AM, Howard Hinnant wrote:
>
>> On Feb 18, 2009, at 12:26 PM, Douglas Gregor wrote:
>>
>>> Committed, with a little twist:
>>>
>>> typedef void _Argument_type_is_not_arithmetic;
>>> static _Argument_type_is_not_arithmetic __tg_promote(...) \
>>>  __attribute__((__unavailable__,__overloadable__));
>>
>> Nice.  The '\' confused me at first, but seems harmless enough.
>
> I guess we could give the attribute a named macro, e.g.,
>
> static _Argument_type_is_not_arithmetic __tg_promote(...)
> __ATTRACT_BAD_TGMATH_ARG
>
> ?
>
>>> We're recovering badly from failures when calling __tg_promote. I'm
>>> working on some parser/sema tweaks to improve the situation, and
>>> will send an updated error message when it's done.
>
> Here's what we get now when we try to take the sine of a pointer to
> short:
>
> tgmath_test.c:4:3: error: call to function '__tg_promote' that has
> been intentionally made unavailable
>   sin(sp);
>   ^~~~~~~
> ./tgmath.h:293:27: note: instantiated from:
> #define sin(__x) __tg_sin(__tg_promote1((__x))(__x))
>                           ^
> ./tgmath.h:56:50: note: instantiated from:
> #define __tg_promote1(__x)           (__typeof__(__tg_promote(__x)))
>                                                  ^
> In file included from tgmath_test.c:1:
> ./tgmath.h:41:41: note: unavailable function is declared here
> static _Argument_type_is_not_arithmetic __tg_promote(...) \
>                                         ^
> 2 diagnostics generated.

Nice!!

-Chris