Re: CVS-19981209: Patch for "unsigned HOST_WIDE_INT" on Solaris2.6

[Jeffrey A Law <law at cygnus dot com>]

  In message <9812180442.AA13939@vlsi1.ultra.nyu.edu>you write:
  > You seem to be taking the stance that *some* interface changes are
  > acceptable and others not, which is in itself a potentially reasonable
  > position,
Yes.

  > but where I'm confused is that you seem to be saying that
  > the smaller and safer change is the one that is *not* acceptable.
I am suggesting we make the smaller and safer change *AND* avoid making
large incompatible changes throughout the compiler.

Paul's suggestion to start using unsigned_HOST_WIDE_INT would have been
a large incompatible change throughout the compiler.

If you go back to one of the original message you'll find that one of the
core problems with using intmax_t on irix to define HOST_WIDE_INT prevents
us from using "unsigned HOST_WIDE_INT".  Paul suggestwe we could use
"unsigned_HOST_WIDE_INT", which I objected to because it would be large
incomaptible change all over the compiler.

  > It is totally unreasonable to assume that all front ends will have
  > release schedules synchronized with either EGCS or GCC2; indeed it's
  > likely there are front ends that neither of us even know about.
I disagree strongly, at least for popular front ends.

The only major outliers in the egcs world are GNAT and Pascal.  I strongly 
suspect we'll have Pascal integrated in the near future -- which leaves GNAT
out in the cold.  I'd love ot see GNAT integrated and hopefully it will happen
someday.

egcs has shown that it can and does work.  C, C++, ObjC, Fortran, Java, and
Chill are all integrated in egcs and development on those langauges happens in
egcs.  When changes are made to the language independent parts of the compiler,
all the front ends are updated accordingly.

jeff