[Ecls-list] Big patch

[Matthew Mondor]

On Sun, 17 Oct 2010 15:22:04 +0200
Juan Jose Garcia-Ripoll <juanjose.garciaripoll at googlemail.com> wrote:

> I already commented here the problem with POSIX locales and ECL's routine
> for reading and printing numbers.
> 
> Well, the fact is that I had to go through a huge set of changes to add ECL
> some standalone routines for those tasks. These routines are hand-written C
> equivalents of SBCL's FLONUM-TO-STRING and are now reused in various parts
> of ECL, including the core routines for printing and FORMAT.

I tested today's ECL CVS HEAD;

GCC 4.1.3 without glibc (but NetBSD-5 using its own libc) do not yet
support C99 long double math functions.  An initial build failed with
undefined reference errors to the following l-suffixed long double math
functions:
- atanl ceill coshl cosl expl floorl frexpl ldexpl logl sinhl sinl sqrtl
  tanhl tanl

However, building again with the --disable-longdouble configure option
worked fine.  If it's easy enough (and reliable) to detect if these
functions are available via autoconf, it might be a good idea to
automatically toggle the default boolean of longdouble depending on
environment eventually.

Otherwise, when I work again on a pkgsrc stub for a future ECL release,
I could also create an option (disabled by default), which when disabled
issues --disable-longdouble, and for some OS/versions known to support
those into their libc to enable it issueing --enable-longdouble
instead, which is not hard to do.

Other than some of my own software I didn't run test suites, but I've
observed no other problems yet despite the huge changes.

Thanks again,
-- 
Matt