bug#12381: Assume C89 or later for math functions.

[Eli Zaretskii]

> Date: Fri, 07 Sep 2012 13:22:38 -0700
> From: Paul Eggert <eggert@cs.ucla.edu>
> CC: Eli Zaretskii <eliz@gnu.org>
> 
> In doing signal-handling cleanup I ran afoul of a lot of crufty old
> floating-point code that hasn't been needed in Emacs for many years.
> Since it's getting in the way of improving Emacs I plan to install
> the following patch to clean out the cruft.  I don't think this
> affects the Windows port but am CC'ing to Eli just in case.

You missed msdos/*.inp scripts.  sed2v2.inp does define
FLOAT_CHECK_DOMAIN for the MSDOS port (although I'm not sure the DOS
port needs this nowadays, see below).

> +2012-09-07  Paul Eggert  <eggert@cs.ucla.edu>
> +
> +     Assume C89 or later for math functions.
> +     This simplifies the code, and makes it a bit smaller and faster,
> +     and (most important) makes it easier to clean up signal handling
> +     since we can stop worring about floating-point exceptions in
> +     library code.  That was a problem before C89, but the problem
> +     went away many years ago on all practical Emacs targets.
> +     * data.c, image.c, lread.c, print.c:
> +     Don't include <math.h>; no longer needed.
> +     * data.c, floatfns.c (IEEE_FLOATING_POINT): Don't worry that it
> +     might be autoconfigured, as that never happens.
> +     * data.c (fmod):
> +     * doprnt.c (DBL_MAX_10_EXP):
> +     * print.c (DBL_DIG):
> +     Remove.  C89 or later always defines these.
> +     * floatfns.c (HAVE_MATHERR, FLOAT_CHECK_ERRNO, FLOAT_CHECK_DOMAIN)
> +     (in_float, float_error_arg, float_error_arg2, float_error_fn_name)
> +     (arith_error, domain_error, domain_error2):
> +     Remove all this pre-C89 cruft.  Do not include <errno.h> as that's
> +     no longer needed -- we simply return what C returns.  All uses removed.
> +     (IN_FLOAT, IN_FLOAT2): Remove.  All uses replaced with
> +     the wrapped code.
> +     (FLOAT_TO_INT, FLOAT_TO_INT2, range_error, range_error2):
> +     Remove.  All uses expanded, as these macros are no longer used
> +     more than once and are now more trouble than they're worth.
> +     (Ftan): Use tan, not sin / cos.
> +     (Flogb): Assume C89 frexp.
> +     (fmod_float): Assume C89 fmod.
> +     (matherr) [HAVE_MATHERR]: Remove; no longer needed.
> +     (init_floatfns): Remove.  All uses removed.

I think we should discuss this first.  Issues that I consider
non-trivial are:

 . Are we sure all supported systems implement the math functions as
   assumed by this patch?  I'd suggest reviewing the results of
   running the configure script on *BSD systems, on AIX, and on old
   Solaris, before making the decision.

 . Do we really want to remove a possibility of signaling an error on
   EDOM and ERANGE?  Right now, this is a configure-time option, but
   the proposed patch removes it forever.  (For that matter, is it TRT
   for Emacs to silently continue with HUGE_VAL etc., disregarding
   errors in math functions?)

 . The pieces related to 'matherr' might still be needed on platforms
   which provide that function.  I don't remember the details, but
   AFAIR the default version of that function may print an error
   message, which I think we need to prevent.

 . Last, but not least: what exactly does the patch assume about the
   math library functions?  Those assumptions are not spelled out in
   the log entries; without them, it's hard to reason about the
   possible implications of the change.  E.g., the log entry says "we
   can stop worrying about floating-point exceptions in library code",
   but AFAIK you _can_ build a C89 program that will generate SIGFPE
   in library functions.  Maybe I don't understand what exactly is
   meant by "floating-point exceptions".