Patch for 'long long' in preprocessor.

[Jeffrey A Law]

  In message < 199902100201.NAA00368@geoffk.wattle.id.au >you write:
  > 
  > I have not tested this patch, because on my system it is not
  > needed---although it does at least compile.
  > 
  > A similar thing should be done for cppexp.c, but (at least for the
  > version in egcs 1.1.1) it also needs to have parse_number changed to
  > be the same as in cexp.c.
  > 
  > This causes glibc 2.1 to build improperly if it is built under an egcs
  > compiled under glibc 2.0.
No.  We want HOST_WIDEST_INT, not HOST_WIDE_INT for this stuff in cpp.

ie, HOST_WIDE_INT should mean the same thing regardless of whether or not we 
are in the cpp sources or elsewhere in the compiler.

cpp needs to perform some operations in the widest possible mode, thus we
should create a type for this -- HOST_WIDEST_INT seems the most natural
choice.  Then we fix cpp to use it.

There may also be places in gcc itself that could benefit from having 
HOST_WIDEST_INT available.


jeff