Removed transitions from the accessor pair descriptors.

Removed transitions from the accessor pair descriptors.

AccessorPair related transitions are now also stored as single map links, simplifying the code that handles transitions. AccessorPairs can now be shared between descriptor arrays, since they can only be mutated after another transition anyway; during which the pair is copied before writing.

Committed: https://code.google.com/p/v8/source/detail?r=12097

[Toon Verwaest, 2012-07-16 13:13:41]

PTAL. The main parts of the change are JSObject::DefineFastAccessor,
NewCallbackTransition and TryAccessorTransition. Most of the other code are
syntactical drive-by cleanups and simplifying the code since transitions can now
only be maps (mostly ClearNonLiveTransitions).

[Sven Panne, 2012-07-16 13:49:43]

LGTM with some minor nits...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4368
src/objects.cc:4368: MaybeObject* maybe_accessors = CreateAccessorPairFor(name);
Using a local scope makes it very clear that maybe_accessors is only used for
testing/conversion immediately afterwards. It is a common v8 idiom (a.k.a. Maybe
monad ;-), so I would prefer to leave the scope as it is.

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4442
src/objects.cc:4442: MaybeObject* maybe_ok =
NormalizeProperties(CLEAR_INOBJECT_PROPERTIES, 0);
See above.

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4449
src/objects.cc:4449: MaybeObject* maybe_new_map = map()->CopyDropDescriptors();
See above.

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4461
src/objects.cc:4461: maybe_ok = SetNormalizedProperty(name, structure, details);
See above.

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4615
src/objects.cc:4615: MaybeObject* maybe_accessors;
Scope again...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode4962
src/objects.cc:4962: // Attention, tricky index manipulation ahead: Two
consecutive indices are
Does this comment still hold?

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode5760
src/objects.cc:5760: MaybeObject* maybe_result = descriptor->KeyToSymbol();
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode5772
src/objects.cc:5772: MaybeObject* maybe_result = descriptor->KeyToSymbol();
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode5781
src/objects.cc:5781: MaybeObject* maybe_descriptors = Allocate(new_size,
MAY_BE_SHARED);
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode5814
src/objects.cc:5814: MaybeObject* maybe_result = Allocate(number_of_descriptors,
shared_mode);
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode5924
src/objects.cc:5924: MaybeObject* maybe_copy = heap->AllocateAccessorPair();
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode12564
src/objects.cc:12564: MaybeObject* maybe_key =
heap->LookupSymbol(String::cast(k));
Scope...

http://codereview.chromium.org/10784014/diff/1/src/objects.cc#newcode12607
src/objects.cc:12607: MaybeObject* maybe_new_map =
obj->map()->CopyReplaceDescriptors(descriptors);
Scope...

http://codereview.chromium.org/10784014/diff/1/src/property.h
File src/property.h (right):

http://codereview.chromium.org/10784014/diff/1/src/property.h#newcode363
src/property.h:363: return GetValue();
Can we put an ASSERT here about the lookup_type_?