Re: build failure i686-pc-linux-gnu (today's CVS)

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Gerald" == Gerald Pfeifer <pfeifer@dbai.tuwien.ac.at> writes:

>> Thanks for your report.  This is a known problem.  The experts are
>> arguing over who should fix it... but it will be fixed.

Gerald> The problem is, that more often or not, Java fails to
Gerald> bootstrap, at least according to my highly subjective
Gerald> experience on non-GNU/Linux platforms (Solaris,
Gerald> FreeBSD). Usually the Java folks are quite responsive, but
Gerald> still these bootstrap failures are annoying.

I was just responding to the bug report so that the reporter would get
some feedback.  I wasn't making criticism in this case.

But to respond: I, and the other Java hackers, try to make sure things
bootstrap properly.  However sometimes a bug gets checked in that is
out of our ability to fix.  This bug is one such.

Why does this happen?  Because not everyone bootstraps with Java, even
on those platforms where it is known to work.  Then, once it is broken
by such a change, that brokenness is used as an excuse not to build it
ever again.

If I had more free time I'd go back over the logs to see why Java
bootstraps have failed for the last N times.  My gut feeling is that
80% or so are unrelated to changes made by Java hackers.

Gerald> I'd really like to see Java enabled by default for GCC 3.1,
Gerald> but we must reach a bit more of stability for that.

I think it should be enabled by default on known good platforms in GCC
3.0.  The list of known good platforms isn't very long, and a few
weeks ago we bootstrapped fine on all the ones (I think) we care
about.  Unfortunately even on those platforms you can't build it right
now due to broken changes to gcc.

Furthermore once it works on our target platforms I think there should
be a rule prohibiting a checkin that breaks Java, just as breaking C
or C++ is disallowed.

Tom