Review of remaining xm-host headers

[Zack Weinberg]

There are 13 remaining xm-host headers.  Some of them are genuinely
necessary, others could be removed with more or less machine
independent work.

xm-interix.h:
	This is the only one left in the top level of the config
	directory.  It sets DO_GLOBAL_CTORS_BODY and
	DO_GLOBAL_DTORS_BODY.  This logic belongs in a target header,
	but I'm not sure which one would be appropriate.  There's no
	generic interix.h.  (We could create one...)  Interix ports
	exist for alpha and i386.
alpha/xm-alpha-interix.h:
	Forces HOST_WIDE_INT and HOST_WIDEST_INT to be long
	long... presumably long is only 32 bits on this host.
	(Why?)  Could go away with better WIDE(ST)_INT probing
	(several places have to do this) or a real multiprecision
	library.
i386/xm-i386-interix.h:
	Sets HOST_WIDEST_INT to long long or __int64 depending on
	bootstrap compiler.  Logic belongs in hwint.h.

alpha/xm-vms.h:
vax/xm-vms.h:
	Lots of stuff, which I am not qualified to evaluate.  I
	suspect a lot of it is unnecessary with modern VMS.

i386/xm-cygwin.h:
i386/xm-mingw32.h:
	Sets up pathname handling for DOSish environment.  This area
	needs a cleanup but there's no real junk in this file.
i386/xm-djgpp.h:
	Same, but has to do more work due to less veneer over DOS.
i386/xm-vsta.h:
	Sets PATH_SEPARATOR.  I don't know anything about VSTA.

i386/xm-osf1elf.h:
	Sets HZ.  Logic belongs in autoconf.

mips/xm-iris5.h:
	Undefines HAVE_INTTYPES_H with a comment about configure not
	properly detecting a clash between inttypes.h and
	sys/types.h.  Could go away with a better configure probe.
rs6000/xm-darwin.h:
	Forces HAVE_DESIGNATED_INITIALIZERS off, apparently there's a
	system compiler bug.  Wait for bug to be fixed?

rs6000/xm-beos.h:
	Sets STANDARD_INCLUDE_DIR and SYSTEM_INCLUDE_DIR.  Why does
	rs6000 beos do this and not i386 beos?

zw