Re: gcj 3.2.1

[Tom Tromey]

>>>>> "Brian" == Brian Jones <address@hidden> writes:

Brian> I've noticed I cannot compile CVS with gcj 3.2.1, is any other
Brian> version capable?

I'd heard that 3.2 *could* compile CVS Classpath and that 3.3 (and
trunk, currently aka 3.4) could not.  This belief forms the basis for
PR java/9695.

In a perverse way it would be good news if 3.2 couldn't do the job.
That could mean we don't have a gcj regression, hence one less
required bug to fix before release.  (A somewhat silly approach to
bugs, but gcj time is limited these days.)

In the past people have checked in workarounds in Classpath to avoid
gcj bugs.  At the time I think I thought this was a bad idea.
Nowadays, well, there's a precedent... :-)

Seriously, we could try to work around the problem.  Or we could just
report gcj bugs and compile Classpath with a different free compiler
(jikes, or the Eclipse compiler, or kopi, or ... are there others?).
I'm not too bothered if the workarounds don't go in; for libgcj we'll
maintain a temporary divergence if need be.

How does gcj 3.2 fail on CVS Classpath?  Maybe that can help guide us.

Tom