Re: TR1 Math tarball

[Paolo Carlini <pcarlini at suse dot de>]

Paolo Carlini wrote:

> Anyway, I plan to post a few additional comments later today...

As promised, in totally sparse order (if you want more, just ask ;)

1- What is __floating_point_constant buying us?
2- There are spurious ^M around.
3- Please wrap everything < 79 columns.
4- MI_P & co are *not* ok for long double, only for double. I would 
suggest the solution adopted elsewhere in the library, bring in the long 
double version from somewhere (e.g., the glibc math.h header or a 
software for arbitrary precision computations)
5- int(__nu + 0.5) isn't really ok for rounding, see the thread around:

   http://gcc.gnu.org/ml/gcc-patches/2006-10/msg00971.html

6- Please double check that everything is 64-bit safe: ints, at variance 
with longs, are *always* 32 bits (ok, sorry about the pedantic point, 
but apparently way too many people don't remember this basic point and 
*awful* things happen)

Please try to make progress on those issues, but the std::tr1::cmath one 
is the most important blocker, in my opinion, because bootstrap can fail 
(pre-compiled headers are produced also for tr1 facilities).

Thanks again a lot for your work,
Paolo.