PATCH: Add tgmath.h

[Matt Austern]

On Aug 15, 2004, at 5:21 PM, Mark Mitchell wrote:

> Jakub Jelinek wrote:
>
>> On Sun, Aug 15, 2004 at 10:07:21AM +0000, Joseph S. Myers wrote:
>>
>>>> It seems like that is more of a bug in the current implementation, 
>>>> and easily
>>>> fixed.
>>>>
>>> It is true that there is a conformance bug in <math.h> and 
>>> <complex.h> - not in <tgmath.h> - and I would deal with the ABI 
>>> issue a different way myself (arrange for calls to the long double 
>>> versions to be redirected to calls to the double versions, though 
>>> doing this for code that doesn't include the headers would itself 
>>> involve compiler trickery), but in the
> It might be a reasonable compromise to provide a fixinclude that would 
> (a) create tgmath.h if not provided at all by the system, or (b) fix 
> the existing implementation if broken, for some declaration of 
> "broken".  So, for example, on GNU/Linux, this fixinclude might do 
> nothing at all if the glibc version is declared sufficiently 
> non-broken.

That makes sense---at least if we agree that a reasonable first step is 
to do a very crude probe: provide our own tgmath on system that don't 
have one at all, and assume that the system-provided tgmath is good 
enough if we can find one.  Yes, that's too crude and it'll need 
incremental refinement.  But I claim it's good enough as a starting 
point.

			--Matt