> Btw, can anyone advise me on getting an actual account on a natively
> 64-bit machine somewhere? I don't really like the model of checking in
> broken code, waiting for the tinderbox to get around to testing it,
> blindly fixing things, and repeating. Especially when the tinderbox
> machines frequently die for random reasons unrelated to the code.
Actually, you don't need a native 64-bit system. A big-endian (e.g.
SPARC) system with 'long long' is at least useful. It might be that the
Compaq "Test Drive" machines could be useful. I know little about them,
however.
I agree the tinderbox can be clunky to use, and it isn't always useful.
For example, right now, the PPC machine is still running Parrot 0.0.6
(because of the cvs repository move some time ago), and the Solaris 64-bit
machines are (apparently) core-dumping in the pointer-alignment test of
Configure.pl.
But that doesn't mean the tinderbox is useless. For quite a while, the
Solaris/JIT boxes were failing simply because of a missing test_prep
target (effectively). Once I fixed that, they suddenly started passing.
For quite a while, the systems that needed 'ranlib' were failing until I
added 'ranlib'. Similarly, the Solaris 32+64 machine is currently failing
due to a core pmc int/INTVAL mixup. I've committed my patch for that, and
you should see that machine turning green again next time it runs. When I
hacked in a crude alignment workaround, the Tru64 machines all went green.
When it was backed out accidentally, they all went orange again. So it
can be useful.
I fear we have a viscous cycle. Few people pay much attention to the
tinderboxes because so many of them are broken. You can't tell if your
change broke anything because so many of them are already broken anyway.
But few people seem bother to try to fix the tinderboxes because few
people pay attention to them. By trying to fix them, I'm hoping we can
break that cycle.
I certainly agree that blindly trying to patch something on x86 that only
breaks on 64-bit SPARC, with a several-hour-turnaround, is really really
hard, and probably a very poor use of time. However, if we can keep the
tinderbox mostly green, it can at least be useful in pointing out whether
a problem exists.
--
Andy Dougherty [EMAIL PROTECTED]
