changes for approval, part one

[Phil Edwards <pedwards at disaster dot jaj dot com>]

These are the general changes that I've made.  A lot of the diffs are
formatting changes, e.g., wrapping seventy lines of script in if/then
means indenting them slightly, producing a 150-odd line diff.

Changes for Solaris all depend on these changes, so I'll come back
to those.  All the stuff below has been tested to the best of my
one-and-a-half-platforms ability.

1)  If no threads library is given, we shouldn't be trying to do any
    threadsafe stuff, since we /can't/.  Specifically, libio can't have
    MTSAFE_IO turned on.  So a new macro is added, _GLIBCPP_USE_THREADS,
    and that wraps the definition of _IO_MTSAFE_IO in _G_config.h.  If
    any thread package is requested, _GLIBCPP_USE_THREADS is defined
    to 1 and _IO_MTSAFE_IO stays on.

    The "typedef _IO_lock_t __c_lock;" also needs to be wrapped.

       [Background:  __c_lock is the type of the fstream's _M_lock member,
       which doesn't seem to be used in a non-MT-safe environment.
       In any case, that class member /must/ be there because it gets
       passed around in libio; all we can really change is its type.]

    If threads are in use, then _IO_lock_t is a valid resolvable type.
    Otherwise, it needs to be something else, something simple.  Using
    "void" or "void*" breaks.  I picked int.  So _GLIBCPP_USE_THREADS
    conditionally compiles one typedef or another:

        #ifdef _GLIBCPP_USE_THREADS
            typedef _IO_lock_t	__c_lock;
        #else
            typedef int       	__c_lock;
        #endif

    This seems slightly ugly to me, but since _IO_MTSAFE_IO is off, all
    the libio code gets safely skipped.  The parameter itself is assigned
    to 0 when passed around, so 'int' seemed safest.  (libio.h uses void,
    which isn't even legal C, but I don't think that code branch has
    ever been compiled; it's just a stub.)

2)  The --{enable,disable}-c-mbchar from GCC's configure is added, on by
    default.  This lets us disable wide characters at the 'configure' level.
    Doing so turns off _GLIBCPP_NEED_MBSTATE_T and _GLIBCPP_USE_WCHAR_T.

    If multibyte characters are left on, then "need_wlibio" is set to yes as
    per bkoz's "need to add checks to enable this" comment in acinclude.m4.
    That way the related libio files will be built.

3)  GLIBCPP_ENABLE_RELIBGCC is gone.  (Now that libsupc++ is here.)

4)  More comments and comment-cleanup in acinclude.m4 and configure.in.


Like I said, the number of lines added is small, but the number of lines
indented in acinclude.m4 makes for an ugly-looking diff.  Does this seem
good to check in, or do you want to try to read the diff?


phil