Mail Archives: djgpp/1999/02/26/01:40:12
> Uhhh... if you're referring to page 660 (section 22.3)
> it most certainly does not make any such guarantee with
> regards to -long double- datatypes.
S22.3, page 661: "double atan (double) ..... In addition, <cmath> and
<math.h> supply these functions for float and long double arguments."
You're wrong, I'm right. Long double versions of atan and friends ARE
demanded by the standard. I would like to see them very soon so that I may
use them (the long double versions anyways) when I want that bit of extra
precision.......
> at least on my box (sparc solaris), libm is vendor provided.
Not on an IBM PC clone, where it comes with the DJGPP/Cygwin/whatever
development environment you install. I installed the DJGPP/EGCS development
environment and got a broken libm.
> gnu/include/g++/cmath -does- make reference to the math
> functions that use long doubles for arguments, but I don't
> think any library on my box actually supports them.
Then those libraries are broken and not conformant. I suggest you complain
to the vendor. You may also want to ask the bookstore where you got your
copy of Stroustrup for an exchange or a refund; I sure would have if my
copy had turned out to have pp. 661-662 torn out or missing.
Anyways, since these are function overloads in C++, I would expect the long
double and float versions to be in libstdc++, not libm. And libstdc++ comes
with gcc/egcs rather than being vendor supplied, on all platforms. If the
egcs libstdc++ is lacking long double overloads for atan and the like, then
this is a bug in egcs.
--
.*. "Clouds are not spheres, mountains are not cones, coastlines are not
-() < circles, and bark is not smooth, nor does lightning travel in a
`*' straight line." -------------------------------------------------
-- B. Mandelbrot |http://surf.to/pgd.net
_____________________ ____|________ Paul Derbyshire pderbysh AT usa DOT net
Programmer & Humanist|ICQ: 10423848|
- Raw text -