Re: gcj and jikes

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Andrew" == Andrew Haley <aph at redhat dot com> writes:

Tom> The gcj bytecode generator has some advantages over the one in
Tom> jikes.

Andrew> I can't say that I believe this to be very important.

Yeah.  I'd prefer to do an excellent job if this were to be done.  But
objectively the payoff from better bytecode generation is pretty
small.

Andrew> The problem that bothers me is divergence.  For example, I
Andrew> want to do something with Generic Java ASAP, and I suppose I'd
Andrew> have to keep both Jikes and gcj up to date.

It would be a situation analogous to the other stuff we import: the GC
and big chunks of the class libraries.  We could maintain divergences
when necessary, etc.

There are some worries.  For instance, what's the value we provide to
the jikes project, to help convince them not to change things around
to break our model?  There may already be areas of jikes we'd have to
change; for instance I didn't look too hard at how, say `foo.class' is
implemented (if the call to the synthetic method is generated early,
that is bad for us, since we'd rather have a direct reference in most
cases).

[ licensing ]
Andrew> Ah.  Oh.  Darn.

Not impossible, but yeah.

Tom