[armedbear-devel] Supporting object finalizers (to support trivial-garbage)

[Erik Huelsmann]

In r13250, I committed a different approach which uses a collection
thread (started as late as the first call to EXT:FINALIZE).

Bye,

Erik.

On Sun, Mar 13, 2011 at 12:54 AM, Ville Voutilainen
<ville.voutilainen at gmail.com> wrote:
> On 13 March 2011 01:33, Erik Huelsmann <ehuels at gmail.com> wrote:
>> One of the higher systems on the list of Quicklisp downloads is the
>> trivial-garbage system (http://www.cliki.net/trivial-garbage).
>> One of the things this package does is to provide 'normalized' access
>> to object finalization functionality. I thought I'd quickly add that
>> support (and did so in r13248). However, it seems our bootstrap
>> compilation has now slowed down to a crawl.
>> So, I'm ready to revert the commit and reopen the ticket I had opened
>> to track progress. However, are you seeing the same? Do you have any
>> good alternatives? I'd love some discussion on this subject:
>> supporting Quicklisp downloaded packages looks like a good way to
>> promote ourselves.
>
> It's not just a slowdown problem:
>
>     [java] Exception in thread "interpreter"
> java.lang.OutOfMemoryError: GC overhead limit exceeded
>     [java]     at java.lang.ref.Finalizer.register(Finalizer.java:72)
>     [java]     at java.lang.Object.<init>(Object.java:20)
>     [java]     at org.armedbear.lisp.LispObject.<init>(LispObject.java:39)
>
> The build fails for me, with the culprit apparently being the finalizer stuff.
> It would perhaps be an idea worthy of investigating to not burden
> every LispObject with that support, but rather try and do it as an opt-in
> somehow.
>