Eric Marsden <[EMAIL PROTECTED]> writes:

> >>>>> "ew" == Edi Weitz <[EMAIL PROTECTED]> writes:
> 
>   ew> * (test 1000 t)
>   ew> !!! CMUCL has run out of dynamic heap space. You can control heap size
>   ew> !!! with the -dynamic-space-size commandline option.
>   ew> *A2 gc_alloc_new_region failed, nbytes=8.
> 
> I can reproduce this on Linux/x86, on Solaris/sparcv9 with gencgc, and
> on SBCL. It looks like the gencgc GC trigger isn't working properly.
> 
> Maybe if we wait a while the SBCL folks will find the bug ;-)

I guess what's happening is that LOAD allocates lots of compiler
structures, for the thousands of nested PROGNs, MAYBE-GC gets called
to collect garbage in newspace but can't free anything because all
that data is live, and *GC-TRIGGER* constantly increases, so that
effectively very few GCs happen afterwards, if any.

Placing an upper bound on *GC-TRIGGER* would improve the situation, I
guess.



Reply via email to