[Cfp-interest] TS draft comments

[Joel C. Salomon]

On 01/16/2012 12:56 PM, Jim Thomas wrote:
> On Mon, 16 Jan 2012 11:42:55 -0500 Joel C. Salomon wrote:
>> Is there a way to say that a freestanding implementation can conform if
>> it has certain bits normally included in a hosted implementation?
>> (E.g., freestanding + <fenv.h> + <math.h> + ....)
>>
>> I understand that the C committee won't want to create new subsets of
>> the Standard, but this project is creating an extension to the Standard,
>> so why not define the extension for freestanding implementations as well
>> as hosted ones?  (There's nothing stopping an extended freestanding
>> implementation from including the CFP binding as an extension without
>> such a definition, but then they can't claim, "we support IEC 60559:2011
>> per WG14 TS 00000".)
> 
> <stdio.h> support would also be required. Would that be acceptable for freestanding implementations?

Ooh, yuck.  Without those string-to-binary functions an implementation
is not 754-2008 compliant (Section 5.12), but if we require <stdio.h> we
may as well require a hosted implementation.

How about, a freestanding implementation conforms to this TS if it
provides (the relevant parts of) <fenv.h> & <math.h>, but it does not
conform to 754 unless it also provides, in some implementation-defined
manner, the functions mandated in 754-2008 Section 5.12.  Recommended
practice:  if the implementation provides some subset of <stdio.h>
functionality, it should follow the guidelines this TS provides.

This is getting unwieldy very quickly.

--Joel