Enable code flushing even when objects were marked incrementally,

Enable code flushing even when objects were marked incrementally,
since the incremental visitor is setting the gc_age fields correctly.
This means the ResetSharedFunctionInfoCountersDuringIncrementalMarking
test passes even if it hits a marking stack overflow.
Committed: https://code.google.com/p/v8/source/detail?r=11481

[Erik Corry, 2012-04-30 07:59:45]

Updated with the removal of an incorrect assert.

This currently passes all tests.  If it is not correct then we should add tests
that show that.

[ulan, 2012-04-30 09:00:54]

On 2012/04/30 07:59:45, Erik Corry wrote:
> Updated with the removal of an incorrect assert.
> 
> This currently passes all tests.  If it is not correct then we should add
tests
> that show that.

I am not confident here. May I ask to redirect to Slava?

[Vyacheslav Egorov (Chromium), 2012-04-30 11:10:41]

Description of the change list is misleading. 

ic_age has nothing to do with _code_ flushing, it affects only _ic_ flushing.

It seems you don't need to enable code flushing (it would not be able to flush a
lot (if anything) because everything is marked and no candidate collected) but
instead you should move    

if (shared->ic_age() != heap->global_ic_age()) {
  shared->ResetForNewContext(heap->global_ic_age());
}

from 

VisitSharedFunctionInfoAndFlushCodeGeneric

to the beginning of 

VisitSharedFunctionInfoAndFlushCode

to clear those shared-function-infos that are marked after incremental marking
is complete.

https://chromiumcodereview.appspot.com/10258001/diff/2002/test/cctest/test-he...
File test/cctest/test-heap.cc (right):

https://chromiumcodereview.appspot.com/10258001/diff/2002/test/cctest/test-he...
test/cctest/test-heap.cc:1660: HEAP->CollectAllGarbage(Heap::kNoGCFlags,
It seems that if !IsStopped() and !should_hurry() we will postpone a full gc at
this point. 

Maybe force should_hurry?