heap exhausted with matrix usage



>>>>> "Robert" == Robert Dodier <robert.dodier at gmail.com> writes:

    Robert> On 2012-10-23, Raymond Toy <toy.raymond at gmail.com> wrote:
    >> But cmucl with gencgc has a known issue where cmucl hangs on to
    >> garbage for way too long, but that's usually only with specialized
    >> arrays, which aren't being used in this test.  The last time I saw
    >> this issue, I could work around it by setting the number of
    >> generations to 2 instead of the default 6.

    Robert> I found a web page somewhere in the CMUCL docs which says that different
    Robert> garbage collectors are used on different platforms. I'm working on an
    Robert> x86 platform (Celeron specifically) so I gather a conservative GC is
    Robert> employed, at least by default. Is there a way to change it to the exact
    Robert> GC?

That page is a bit out of date.  All supported platforms now default to gencgc
now[1].  In any case, the two available gc's for x86 are both
conservative, so there's no option for a precise gc, unfortunately.

In any case, the example I gave was not caused by the conservative
gc.  The gc is only conservative when it scans the stack, which is
both the lisp stack and the C stack.  Everything else is precise.

Ray
[1] Although called gencgc, for sparc and ppc, it is actually precise.