Abort incremental marking before GCs when test assumes that an object is collected.

Abort incremental marking before GCs when test assumes that an object is collected.

Also port the related bleeding edge r8892 to gc branch.
It avoid using shared static variables in threaded tests.

TEST=Threaded4

Committed: http://code.google.com/p/v8/source/detail?r=8896

[Vyacheslav Egorov (Chromium), 2011-08-11 11:37:24]

LGTM

http://codereview.chromium.org/7617005/diff/1/test/cctest/test-api.cc
File test/cctest/test-api.cc (right):

http://codereview.chromium.org/7617005/diff/1/test/cctest/test-api.cc#newcode402
test/cctest/test-api.cc:402: if (HEAP->incremental_marking()->IsMarking()) {
Maybe we should always abort incremental marking when we call
CollectAllAvailableGarbage.

It has this 'All' in the name after all.

We also have a kMakeHeapIterableMask GC flag that cancels incremental marking
internally.

I think we can have another flag that cancels marking but does not do precise
sweeping.

[Lasse Reichstein, 2011-08-11 12:20:13]

http://codereview.chromium.org/7617005/diff/1/test/cctest/test-api.cc
File test/cctest/test-api.cc (right):

http://codereview.chromium.org/7617005/diff/1/test/cctest/test-api.cc#newcode402
test/cctest/test-api.cc:402: if (HEAP->incremental_marking()->IsMarking()) {
Actaully CollectAllAvailableGarbage does use the kMakeHeapIterableMask flag, so
this call is safe.
On the other hand CollectAllGarbage doesn't do it (and it has "All" in the name
too (which is odd naming - the unqualified CollectAllGarbage collects less than
the qualified CollectAllAvailableGarbage, and "Available" seems odd to use about
garbage which, by definition, isn't reachable)).

I've changed the code to use CollectAllAvailableGarbage instead of
CollectAllGarbage instead of inserting these manual aborts.