On 4/20/07, Patrick R. Michaud <[EMAIL PROTECTED]> wrote:
On Fri, Apr 20, 2007 at 10:18:41AM -0400, Andy Dougherty wrote:
> On Thu, 19 Apr 2007, Patrick R. Michaud via RT wrote:
> > This does prompt the question of removing -G from the tests,
> > but the last time I looked into doing this (about a month ago) Parrot
> > still had intermittent GC errors that would cause random
> > failures in the tests.
>
> [...] With -G, I get to test 395 in p5regex/p5rx.t.
> Without the -G, I memory usage still grows, and I only get to test 185
> before getting the PANIC.
Is it still the "out of memory" panic?
Just to add a "me too", I've also noticed that running PGE-based
routines with GC enabled causes them to consume memory much faster
than with GC disabled, which seems totally counterintuitive.
AFAIK, PGE isn't creating any global references (beyond its
static internal tables) that would cause objects to stick
around "forever". It certainly isn't part of PGE's design
for that to happen. So, maybe we need to do a bit of forensics
to find out where the memory consumption is occurring.
i have a feeling it has to do with strings and gc. it's likely that
the bug is not in pge, but since pge uses strings so heavily
(especially in a run of 900+ regex tests,) that this is where we see
the failures.
~jerry
