Re: [PATCH] IBM z/OS + EBCDIC support

[Paul Eggert]

A few non-ctype-related comments:

Omit parens around arguments of 'defined', e.g., say "defined __MVS__" 
not "defined (__MVS__)".

I agree with Eric that we should rename "__string" rather than fiddle 
with #undefing it.  It's just a placeholder name.  I suggest renaming it 
to "__str".  We can backport this to glibc eventually.

In strtod.c, don't bother with "if (errno == ERANGE) errno = 0;". Just 
do "errno = 0;".

Also in strtod.c, don't assume isnand exists.  That is, replace 
"!isnand(num)" with "num == num".

Update serial numbers in changed .m4 files.

In m4/fclose.m4, gl_FUNC_FCLOSE should AC_REQUIRE([AC_CANONICAL_HOST]).  
Also, it should test $host_os rather than $host.

In m4/strstr.m4, the __MVS__ failure should be at compile-time, with 
#error, rather than at run-time.  That's better if cross-compiling.

In comments, prefer imperatives, e.g., "Instead, temporarily redefine 
..." rather than "Instead, we temporarily redefine ...". This is 
standard GNU style and is shorter.

The get_rusage_as code has duplications.  Simpler would be:

  uintptr_t
  get_rusage_as (void)
  {
    /* On Mac OS X, AIX, Cygwin, and z/OS, get_rusage_as_via_setrlimit
       exists but does not work.  */
  #if (! ((defined __APPLE__ && defined __MACH__)                         \
      || defined _AIX || defined __CYGWIN__ || defined __MVS__)       \
       && HAVE_SETRLIMIT && defined RLIMIT_AS && HAVE_SYS_MMAN_H && 
HAVE_MPROTECT)
    /* Prefer get_rusage_as_via_setrlimit() if it succeeds,
       because the caller may want to use the result with setrlimit().  */
    uintptr_t result = get_rusage_as_via_setrlimit ();
    if (result != 0)
      return result;
  #endif
    return get_rusage_as_via_iterator ();
  }