Make accessors for oddball objects return Oddball* instead of Object*.

Make accessors for oddball objects return Oddball* instead of Object*.

Fix a use of the hole value and the undefined value before initialization when
initializing V8. Before we just read a NULL value from them.
Committed: http://code.google.com/p/v8/source/detail?r=9557

[fschneider, 2011-10-04 10:48:51]

http://codereview.chromium.org/8130002/diff/1/src/heap.cc
File src/heap.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/heap.cc#newcode1813
src/heap.cc:1813: set_undefined_value(Oddball::cast(obj));
heap->undefined_value() is used by CreateApiObjects, which does
AllocateJSObjectFromMap which calls AllocateFixedArray.

http://codereview.chromium.org/8130002/diff/1/src/isolate.cc
File src/isolate.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/isolate.cc#newcode1772
src/isolate.cc:1772: InitializeThreadLocal();
Calls clear_pending_exception() which uses heap->the_hole_value().

The oddballs are only initialized after deserialization.

[Vyacheslav Egorov (Chromium), 2011-10-04 11:00:51]

LGTM

http://codereview.chromium.org/8130002/diff/1/src/debug.cc
File src/debug.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/debug.cc#newcode2161
src/debug.cc:2161: reinterpret_cast<Object**>(uncaught ?
factory->true_value().location() :
line is too long, also consider making Factory::ToBoolean analogue of
Heap::ToBoolean

http://codereview.chromium.org/8130002/diff/1/src/debug.cc#newcode2188
src/debug.cc:2188: reinterpret_cast<Object**>(before ?
factory->true_value().location() :
line is too long, also consider making Factory::ToBoolean analogue of
Heap::ToBoolean

http://codereview.chromium.org/8130002/diff/1/src/heap.cc
File src/heap.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/heap.cc#newcode2210
src/heap.cc:2210: undefined_value()->Initialize("undefined", nan_value(),
Oddball::kUndefined);
long line

http://codereview.chromium.org/8130002/diff/1/src/type-info.cc
File src/type-info.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/type-info.cc#newcode76
src/type-info.cc:76: :
Handle<Object>::cast(Isolate::Current()->factory()->undefined_value());
you have isolate_ now

http://codereview.chromium.org/8130002/diff/1/src/v8utils.h
File src/v8utils.h (right):

http://codereview.chromium.org/8130002/diff/1/src/v8utils.h#newcode146
src/v8utils.h:146: static inline void MemsetPointer(T** dest, U* value, int
counter) {
well. this one is spooky. 

at least at assignability check to T to U (it will not be checked if we are
generating inline assembly).

(see Handle constructor in handles.h on how to check assignability).

[fschneider, 2011-10-04 11:28:16]

http://codereview.chromium.org/8130002/diff/1/src/debug.cc
File src/debug.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/debug.cc#newcode2161
src/debug.cc:2161: reinterpret_cast<Object**>(uncaught ?
factory->true_value().location() :
On 2011/10/04 11:00:51, Vyacheslav Egorov wrote:
> line is too long, also consider making Factory::ToBoolean analogue of
> Heap::ToBoolean

Done.

http://codereview.chromium.org/8130002/diff/1/src/debug.cc#newcode2188
src/debug.cc:2188: reinterpret_cast<Object**>(before ?
factory->true_value().location() :
On 2011/10/04 11:00:51, Vyacheslav Egorov wrote:
> line is too long, also consider making Factory::ToBoolean analogue of
> Heap::ToBoolean

Done.

http://codereview.chromium.org/8130002/diff/1/src/heap.cc
File src/heap.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/heap.cc#newcode2210
src/heap.cc:2210: undefined_value()->Initialize("undefined", nan_value(),
Oddball::kUndefined);
On 2011/10/04 11:00:51, Vyacheslav Egorov wrote:
> long line

Done.

http://codereview.chromium.org/8130002/diff/1/src/type-info.cc
File src/type-info.cc (right):

http://codereview.chromium.org/8130002/diff/1/src/type-info.cc#newcode76
src/type-info.cc:76: :
Handle<Object>::cast(Isolate::Current()->factory()->undefined_value());
On 2011/10/04 11:00:51, Vyacheslav Egorov wrote:
> you have isolate_ now

Done. Of course, I totally forgot about that :)

http://codereview.chromium.org/8130002/diff/1/src/v8utils.h
File src/v8utils.h (right):

http://codereview.chromium.org/8130002/diff/1/src/v8utils.h#newcode146
src/v8utils.h:146: static inline void MemsetPointer(T** dest, U* value, int
counter) {
On 2011/10/04 11:00:51, Vyacheslav Egorov wrote:
> well. this one is spooky. 
> 
> at least at assignability check to T to U (it will not be checked if we are
> generating inline assembly).
> 
> (see Handle constructor in handles.h on how to check assignability).

Done.

[fschneider, 2011-10-04 12:50:11]

Adding Vitaly: 

r8537 moved the call to InitializeThreadLocal up. In order to avoid using
uninitialized (zeroed) hole_value from the roots array, I split
InitializeThreadLocal into two.
Is there potentially other things that depend on this call before the
deserialization is done?

[Vitaly Repeshko, 2011-10-05 04:41:10]

On 2011/10/04 12:50:11, fschneider wrote:
> Adding Vitaly: 
> 
> r8537 moved the call to InitializeThreadLocal up. In order to avoid using
> uninitialized (zeroed) hole_value from the roots array, I split
> InitializeThreadLocal into two.
> Is there potentially other things that depend on this call before the
> deserialization is done?

Potentially, yes :) To be safe you could add debug-only checks that only
initialized fields are accessed.