Allocate global object as a normalized object

Allocate global object as a normalized object.

The global object is now allocated and initialized as a normalized object.

Committed: http://code.google.com/p/v8/source/detail?r=2582

[Søren Thygesen Gjesse, 2009-07-29 13:45:43]

I will run layout tests with this patch.

http://codereview.chromium.org/159587/diff/1001/11
File src/heap.cc (right):

http://codereview.chromium.org/159587/diff/1001/11#newcode2165
Line 2165: StringDictionary::Allocate(map->NumberOfDescribedProperties() * 2 +
4);
Shouldn't we allocate a larger dictionary here?

[Kasper Lund, 2009-07-29 14:00:51]

LGTM, but:

http://codereview.chromium.org/159587/diff/1001/11
File src/heap.cc (right):

http://codereview.chromium.org/159587/diff/1001/11#newcode2106
Line 2106: ASSERT(map->instance_type() != JS_FUNCTION_TYPE);
Maybe avoid using this for allocating global objects and extend the assert to
check that we don't?

http://codereview.chromium.org/159587/diff/1001/11#newcode2165
Line 2165: StringDictionary::Allocate(map->NumberOfDescribedProperties() * 2 +
4);
I think you're right: A bigger size would be good here. Maybe you should try to
count the number of resizes we do when running the bootstrapping code (maybe try
to use Date too to get that lazy loaded).

http://codereview.chromium.org/159587/diff/1001/11#newcode2171
Line 2171: Object* result = AllocateJSObjectFromMap(map, TENURED, dictionary);
Have you considered not using AlloacteJSObjectFromMap, but just call Allocate
and InitializeJSObjectFromMap directly from here? That way, you wouldn't have to
pass the initial_properties all the time...

[Søren Thygesen Gjesse, 2009-07-30 07:32:19]

http://codereview.chromium.org/159587/diff/1001/11
File src/heap.cc (right):

http://codereview.chromium.org/159587/diff/1001/11#newcode2106
Line 2106: ASSERT(map->instance_type() != JS_FUNCTION_TYPE);
On 2009/07/29 14:00:51, Kasper Lund wrote:
> Maybe avoid using this for allocating global objects and extend the assert to
> check that we don't?

Good point, JS_GLOBAL_OBJECT_TYPE and JS_BUILTINS_OBJECT_TYPE is now also
asserted to be not allocated this way. Also removed the initial_properties
parameter again as it has no use when the global objects are not allocated this
way.

http://codereview.chromium.org/159587/diff/1001/11#newcode2165
Line 2165: StringDictionary::Allocate(map->NumberOfDescribedProperties() * 2 +
4);
On 2009/07/29 14:00:51, Kasper Lund wrote:
> I think you're right: A bigger size would be good here. Maybe you should try
to
> count the number of resizes we do when running the bootstrapping code (maybe
try
> to use Date too to get that lazy loaded).

After measuring the sizes after bootstrapping is complete the initial size of
JS_GLOBAL_OBJECT_TYPE and JS_BUILTINS_OBJECT_TYPE is 64 and 512 respectively.

http://codereview.chromium.org/159587/diff/1001/11#newcode2171
Line 2171: Object* result = AllocateJSObjectFromMap(map, TENURED, dictionary);
On 2009/07/29 14:00:51, Kasper Lund wrote:
> Have you considered not using AlloacteJSObjectFromMap, but just call Allocate
> and InitializeJSObjectFromMap directly from here? That way, you wouldn't have
to
> pass the initial_properties all the time...

Done.