Fix array concat for holey arrays and concat of double to object arrays.

- Initialize the result array with holes if we concat a double array into an object array, since it may cause a marking step while boxing a double.
- Ensure we go holey if we are concatting any holey array.

Committed: https://code.google.com/p/v8/source/detail?r=13038

[Toon Verwaest, 2012-11-22 15:18:56]

PTAL

[mvstanton, 2012-11-22 15:55:41]

One nit and one suggestion. LGTM.

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc
File src/builtins.cc (right):

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc#newcod...
src/builtins.cc:1045: AssertNoAllocation no_gc;
Here you have a scoping brace, for the AssertNoAllocation? Style-wise, I
wouldn't do it because it's not done at other sites.

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc#newcod...
src/builtins.cc:1210: has_double = has_double ||
IsFastDoubleElementsKind(arg_kind);
This works, but you might like something that I have in my current project:

ElementsKind GetUnifiedFastElementsKind(k1,k2);

It does the right thing for our elements transition tree.
For example DOUBLE and HOLEY_SMI unifies to HOLEY_DOUBLE.

You'd still need to keep the has_double boolean for the predicate below.

[Toon Verwaest, 2012-11-22 16:23:13]

Addressed comment

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc
File src/builtins.cc (right):

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc#newcod...
src/builtins.cc:1045: AssertNoAllocation no_gc;
On 2012/11/22 15:55:41, mvstanton wrote:
> Here you have a scoping brace, for the AssertNoAllocation? Style-wise, I
> wouldn't do it because it's not done at other sites.

Done.

https://chromiumcodereview.appspot.com/11413142/diff/1/src/builtins.cc#newcod...
src/builtins.cc:1210: has_double = has_double ||
IsFastDoubleElementsKind(arg_kind);
Seems like what I have is faster. You can replace it with your
GetUnifiedFastElement once it's in and fast.

On 2012/11/22 15:55:41, mvstanton wrote:
> This works, but you might like something that I have in my current project:
> 
> ElementsKind GetUnifiedFastElementsKind(k1,k2);
> 
> It does the right thing for our elements transition tree.
> For example DOUBLE and HOLEY_SMI unifies to HOLEY_DOUBLE.
> 
> You'd still need to keep the has_double boolean for the predicate below.