Bugzilla Bug 116259

I had to drop the ~ppc keyword because you still haven't fixed bug 112207.

nichoj: Would probably be good to get cacao ~amd64 keywording too.

Fixed amd64.

cacao-0.94 was released so I suggest testing that instead.

cacao 0.94 compiled with jikes 1.22 and gnu-classpath 0.20 is failing some
regression tests on ppc, e.g.

getfield (J): FAILED: 8213163617277379787 != 1234567890123

Interesting numbers:
8213163617277379787 = 0x71fb04cb71fb04cb
1234567890123 =            0x11f71fb04cb

The top word of the longword is a repeat of the bottom word instead of having
the correct value. Note that this test passes when run under jamvm 1.4.1.

Can we do another bump of cacao to 0.95? This problem appears to be fixed in
the next upstram release:

http://www.cacaojvm.org/

"CACAO 0.95 released.

This is a bugfix release. Most important release features:
...
powerpc bugfix: (M_COPY): Handle longs and doubles properly."

(In reply to comment #5)
> Can we do another bump of cacao to 0.95? This problem appears to be fixed in
> the next upstram release:
> 

Done.

Ugh, problem didn't go away in 0.95 as I expected. 

Filed a bug in the cacao bugzilla
http://b2.complang.tuwien.ac.at/cgi-bin/bugzilla/show_bug.cgi?id=21

cacao 0.96 fixed the problem -- makes it through test phase on ppc. It requires
gnu-classpath 0.91. Can we do version bumps of both cacao and classpath? For
powerpc, these seem to be the best candidates for eventually stabling. jamvm
1.4.2 also works better with gnu-classpath 0.91 and could be stabilized when
classpath is.

(In reply to comment #8)
> cacao 0.96 fixed the problem -- makes it through test phase on ppc. It requires
> gnu-classpath 0.91. Can we do version bumps of both cacao and classpath? For
> powerpc, these seem to be the best candidates for eventually stabling. jamvm
> 1.4.2 also works better with gnu-classpath 0.91 and could be stabilized when
> classpath is.
> 

Sure. The only problem with gnu-classpath is that 0.91 breaks binary
compatibility and I need to figure out how to handle the situation so that the
vms and classpath are compatible. One solution is to slot gnu-classpath.

any news here from the java team?

(In reply to comment #10)
> any news here from the java team?
> 

nichoj just committed a slotted version of gnu-classpath to the tree and a
jamvm using it. I tried to do the same to cacao but it does not work with the
current implementation of jikes we have and compiling cacao with sun javac
somewhat defeats the point. ecj on the other hand needs a working vm. Of course
we could use the jamvm --> cacao route here and be done with it.

Okay, add us back when you know what you want with regards to keywords.