[gclist] Incremental collection and time limit in Boehm

[David Chase]

At 10:36 AM 1/16/2002 -0800, wink wrote:
>Related to my previous thread with a GC on an embedded system.
>
>It appears that using the incremental mode and setting GC_time_limit
>appropriately my also also be a solution to minimizing the the hit GC
>will take at any one time.
>
>With regards to the Boehm Collector:
>
>Is incremental collection and using GC_time_limit reliable?

Don't know for sure.

>Is it widely used?
>
>Is there a Boehm Collector specific email list?

This is about as good as it will get.

I think you should not be quite so worried about modifying
the B-W collector.  The code that you will need to fiddle is
well-isolated (last I checked) from the nuts and bolts of
garbage collection.  You should be able to modify it to make
it not stop "your threads".  In fact, this might even make
sense as a reasonable interface to export in general, if you
care to put that much work into it.  (A thread could, for
instance, call something like "gc_leave_this_thread_alone()").
Conversely, you could require that it only fiddle with
explicitly registered threads.  In either case, the only
gotcha that I see comes if you drop in BW gc_malloc as a
replacement for malloc, and some library code that your
RT code calls happens to invoke malloc.  That would be bad,
but presumably that is typical of many things in the RT
world.

About 11 years ago (yes, it has
been around that long) I modified it to run on Mach, and most
of that work involved iterating over the (POSIX) threads and
stopping all but the current thread, then fishing the stack
bounds out for each thread.  Cross-my-heart, this was not
tough code, and that is from my 11-years-ago, little-experience-
with-x86 point-of-view.

David Chase