standards/175811: libstdc++ needs complex support in order use C99

Pedro Giffuni pfg at
Thu May 30 19:44:00 UTC 2013

( I stripped a bit the CC list )

On 30.05.2013 12:13, Steve Kargl wrote:
> On Thu, May 30, 2013 at 10:41:24AM -0500, Pedro Giffuni wrote:
>> I may be wrong but with long double support people that
>> need erfcl() and tgamma() can get them from boost.
>> The problem is therefore not implementing everything but
>> getting enough to turn on the features supported by
>> libstdc++ and boost.
> Of course, you're wrong. :-) :-) <-- Note smileys.

And I knew I could be likely wrong from the start ;).

> C99 defines many long double functions.  Anyone wanting
> to use C and libm, and not C++ and boost, will need
> quality implementations of these functions.  Of course,
> the lack of any actual C99 compiler tends to dampen
> this argument.
> What I find appalling is reading "people are tired
> of the situation with libm, so I'm  going to commit
> some atrocious hack".   The proper response should be
> "so I'm going to help implement and test the missing
> functionality".  It's unfortunate that only a few
> individuals are working to fix libm, but such is
> life.

I guess I was trying to hint that Boost is a good
place to look at to get ideas for the implementations
for such stuff. Stephen knows this well though since
he actually fixed some complex functions in boost :).

The implementations of erfc and tgamma in
OpenOffice are based on the Boost code with
the important difference that boost does the
automatic type promotion when they can.

FWIW, I was about to change OpenOffice to use
boost but then I noticed that the type promotion
doesn't work on FreeBSD (due to the lack of long
double math) so in general there was
not much gain in changing the status quo.


More information about the freebsd-numerics mailing list