Re: 3.3.1 GCJ Cross compiler configuration issues

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Peter" == Peter Blemel <pblemel@hotmail.com> writes:

Peter> o The libjava/configure.host script sets up the build based on the
Peter> host configuration (as you'd expect, given it's name).   The "problem"
Peter> is that it's really configuring the library for the target, not the
Peter> host.
Peter> Given that, the case statements  case "${host}" ... need to be case
Peter> "${target}".

This comes up from time to time.  libgcj really is correct here, but
confusing.  The "host" is the platform on which the code will run.  If
you think about it, a target library must be configured with its
"host" set to the top-level's notion of "target".

Peter> o It appears as tho the only 'active' port to powerpc is Darwin, and
Peter> the various configuraton case statements all assume that powerpc*-* ==
Peter> powerpc*-*-darwin.   I'd like to refine that so my
Peter> powerpc-motorola-elf build can have it's own configuration cases.

Sounds good, unobjectionable really.

Peter> o configure.in setc GCC_UNWIND_INCLUDE to the GCC source tree by
Peter> default but then goes on to (basically) clear the definition for cross
Peter> compiler builds.  The problem is that exception.cc is still included
Peter> in the cross compiler build, and it requires unwind-pe.h from the GCC
Peter> include directory.   The 'undef' statement in fact has a comment
Peter> questioning the need to clear the definition.   If the intent was to
Peter> omit exception.cc then the change lies there.   Otherwise, this
Peter> statement needs to be removed from the configuration.

We definitely need exception.cc in the build.  Without it, exceptions
won't work.  I haven't looked at the issues around this in a long
time, but I suggest removing the statement and trying it out.

Tom