Compute flags for target in configure, not Makefile

[Alexandre Oliva]

On Jul 29, 2000, Chris Faylor <cgf@cygnus.com> wrote:

> And, now I'm generically annoyed since cygwin builds, are, in fact
> broken.

Actually, it is Canadian crosses that are broken in general.  I had
tested building with --target=i686-cygwin, but not a Canadian cross,
but D.J. told me I broke them.

> Your use of -nostdinc is ill-advised since it eliminates the
> searching of gcc-lib include directories.

Yep.  It shouldn't be used on Canadian crosses.

> I also see that you're using gcc specific options if the gcc source
> directory exists, which, IMO, is not a very good method for determining
> when it is ok to use gcc options.

If gcc isn't being built, those options won't make any difference, and
it is impossible to tell at configure time whether someone is going to
remove the gcc build sub-tree after configure, so I'm inclined to keep
them instead of performing the test in the Makefile, as we used to do.
Do you see any reason to do otherwise?

> Please fix this ASAP.

Yes, sir :-)

I'm working on that.  I'll post a patch in a couple of minutes, but
since it may take a long while for me to set up a working cygwin
cross-build environment to be able to test the patch for a Canadian
cross, if there's anybody with a working cygwin environment on-line
willing to test the patch, let me know, and I'll post the patch.
Meanwhile, I'll proceed with my own builds to test it.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me