Regarding the copy_phase of the GC; I posted the following question:
>1)The first part of the "copy_phase" promotes survivors in gen 0 to
> gen 1 (assuming 2 generations) by calling "copy_object_simple". This
> function copies the objects to gen 1 and UPDATES (?) the references.
> If so, why is it necessary the second part of "copy_phase" (calls to
> "get_copied_object" and "GCHeap::Relocate" to perform the
> relocation fix up.
D. Stutz kindly replied to me that:
>The lines that you marked in your email are, indeed, updates. But
>remember that there are many places where other refs to the same object
>may be hanging out - the finalization queue, the elder generation, and
>the handle table, for example. The value of these refs need also to be
>updated - in the case of other calls on copy_object_simple, it is done
>immediately using GetRelocated. Otherwise, take a look at
>get_copied_object (and Relocate, which uses it) to see the rest of the
>relocation pass in action at the end of the copy_phase.
>Hope this helps!
>-- David
However I din't have time to analize the answer until now.
Here are the lines for the first pass of gc_heap::copy_phase
copy_through_cards_for_segments (copy_object_simple_const);
copy_through_cards_for_large_objects (copy_object_simple_const);
CNameSpace::GcScanRoots(GCHeap::Promote, ...
CNameSpace::GcScanHandles(GCHeap::Promote, ...
finalize_queue->GcScanRoots(GCHeap::Promote, heap_number, 0);
However the first pass (code above) already goes thru these places:
finalization queue, elder generation and handle table.
So, I still don't see why it is necessary a second pass (calls to
get_copied_object and GCHeap::Relocate) after the scavenge phase.
I'm assuming that this fase doesn't move any gen0 roots (all gen0 roots
were moved in the first pass)
I know that I should trace the code to see why is it necessary but I'm
lacking the time for it now.
If some one could help, I would be very thankful.
Best regards
P. Felix
