Re: Running gump

[Dalibor Topic <robilad at kaffe dot org>]

Tom Tromey wrote:
> Anthony> I'll figure out how to host this properly (maybe
> Anthony> gcc.gnu.org?).
>
> gcc.gnu.org is basically dedicated to hosting, I think running any
> kind of test setup there would be forbidden.
>
> Dalibor> Cool, I wanted to setup a gump with gcj, jamvm and others already on
> Dalibor> kaffe.brainfood.com, and you've beaten me to it;)
>
> Dalibor> We could host it on kaffe.brainfood.com, as we run the kaffe
> Dalibor> tinderboxes there already, and we've got a xen instance for playing
> Dalibor> there.
>
> How powerful is this machine?  I turned off my nightly builds a while
> ago; I would like to restart them but preferably not hosted on my main
> development machine.  These builds were nontrivial: gcc including
> tests (mauve and jacks) , classpath (with gcj and jikes), kawa,
> inetlib, jessie, crypto, rhug, and japi comparisons every night.

It's a xen instance with 512M of ram. Two kaffe builds in parallel with 
jikes can crunch it a bit, as 2X200+M can come close to its capacity, 
but it's pretty fast in general, so we could host more stuff on it. If 
you're interested, drop me or man-di an e-mail, and one of us can give 
you an account.

Regarding the gump: I've looked into it again. The bad results of gcj on 
the brainfood (and green's own) box compared to kaffe on brutus seem to 
come from many gump builds relying on pre-installed non-free jars, like 
jaxp, which are installed on brutus, so that the most basic things, like 
xml-resolver fail because their prerequisites are not met.

I'm not sure how to proceed here: should we fix gump not to rely on 
them, or install the yucky non-free stuff? Former sounds more promising, 
to me. Anthony, do you want to investigate what needs to be done in that 
direction?

cheers,
dalibor topic