patch for libstdc++/libio configuration problem

[Jeffrey A Law]

  In message < 199905180750.AAA09552@beauty.cygnus.com >you write:
  > >What are the precise host, target & build values?
  > 
  > If I configure my tree with
  >     host = cygwin, target = mips-elf, build = solaris2.5
  > Then the target libraries end up with
  >     host = mips-elf, target = mips-elf, build = solaris2.5
Thanks.

  > Because the hostness of the compiler used to build the libraries
  > is the same as the target I originally configured for (gcc emits
  > mips-elf so the host for that compiler is mips-elf).
Really?  Yup, I went in and looked at the toplevel Makefile.in, and sure
enough we set host_canonical to the value of target_canonical.  Wow.  You
learn something every day.

  > Since there is no way to know what the EXEEXT was for the host, the
  > only thing to really do is to test for the case of xgcc and xgcc.exe.
Yup.  Agreed and understood.

  > Or get rid of the test entirely.  It's is useful to some folks to be
  > able to build the target libraries with an already installed compiler.
If we really need that feature, then we need a way to override the test.
Presumably this is for building the target libraries in a canadian cross
build?

If so, then we are still OK -- we will check that a compiler exists for each
language.  True, we are checking the host-x-target compiler which we can not
execute.  But it will exist, which is all we need for the test to do what we
want.  We still end up building the libraries with the build-x-target compiler.

We do not want to remove the test since it is the most dependable way to 
determine what front-ends were build, and thus what target libraries need
to be built.

Thanks for the explanation.

jeff