Re: Support for long long type for C99 and C++11 compliant compilers

[Corinna Vinschen <vinschen at redhat dot com>]

On Oct  9 16:20, Thomas Uhle wrote:
> On Thu, 2014-10-09 at 14:34 +0200, Corinna Vinschen wrote:
> 
> > Hi Thomas,
> > 
> > On Sep  9 20:05, Thomas Uhle wrote:
> > > Hello,
> > > 
> > > the attached patch fixes some issues with long long type support for
> > > compilers that have C99 or C++11 compliance. For instance, llabs(),
> > > lldiv() and atoll() have been missing if g++ was invoked with option
> > > '-std=c++11'.
> > 
> > Your patch is missing a ChangeLog entry.  Also, why do you replace "long
> > long" with _LONG_LONG_TYPE?  Isn't _LONG_LONG_TYPE always "long long"?
> > Also, while you're doing this in math.h, you're sticking to "long long"
> > for other declarations like for atoll.
> > 
> > What's the reasoning here?
> 
> 
> Hi Corinna,
> 
> thanks for your reply!  The macro _LONG_LONG_TYPE has been used rather
> inconsistently through math.h, for llroundl() it is _LONG_LONG_TYPE, for
> llroundf() it is "long long" for instance. I just tried to get its usage
> aligned for all of these functions. As far as I can see, _LONG_LONG_TYPE
> is only used in math.h, nowhere else. That is why I did not touch
> anything else. And _LONG_LONG_TYPE is defined in _ansi.h to be "long
> long" if _HAVE_STDC is defined and to be just "long" if _HAVE_STDC is
> not defined. I do not know the rationale, but it seems that it has been
> there since the time of the following entry in the ChangeLog:
> 
> | 2007-12-19  Dave Korn  <dave.korn@artimi.com>
> |             Jeff Johnston  <jjohnstn@redhhat.com>
> | 
> |         * libc/include/_ansi.h: Add _LONG_LONG definition.
> |         * libc/include/math.h (llrint, llrintf, rintl, lrintl, llrintl): Add
> |         prototypes.
> 
> I could perfectly live without _LONG_LONG_TYPE, though. But I do not
> know whether replacing _LONG_LONG_TYPE by "long long" does break things
> on any other platform. That is why ...

That makes sense.  I'm not sure either, why _LONG_LONG_TYPE is used in
math.h and then only for the aforementioned set of functions (and even
then inconsistently, as you observed).

Jeff, I see the addition of using _LONG_LONG_TYPE is from you.  I don't
quite grok why this is done for the functions added by Dave Korn, but
not for other long long functions added later on.  It looks like an
oversight, or the importance of !_HAVE_STDC got negligible.

AFAICS, using _LONG_LONG_TYPE instead of "long long" for other functions
now might break the ABI dependend on _HAVE_STDC.  I'm not sure there's
still any compiler out there not having long long, so *if* we go in any
ABI-breaking direction, I'd prefer to get rid of _LONG_LONG_TYPE and
define the affected functions always as "long long".


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

Attachment: pgp2HmApM5vZU.pgp
Description: PGP signature