CallNewArray sites need the original feedback cell at crankshaft time.

CallNewArray sites need the original feedback cell at crankshaft time.

This CL addresses a TODO in the hydrogen-based array constructor code,
to pass through the actual type feedback cell, rather than the contents
of the cell.

BUG=
R=danno@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15004

[mvstanton, 2013-06-05 16:24:50]

Hi Danno, this addresses a TODO that we discussed some time back. And Andreas,
you might be interested in these permutations of oracle/type info.
Thanks,
--Michael

[rossberg, 2013-06-06 09:24:19]

https://codereview.chromium.org/16408005/diff/1/src/ast.cc
File src/ast.cc (right):

https://codereview.chromium.org/16408005/diff/1/src/ast.cc#newcode663
src/ast.cc:663: if (value->IsSmi()) {
I'm not sure how this can ever be true, given that allocation_info_cell_ has
type JSGlobalPropertyCell (which is a HeapObject). Either the logic must be
wrong or the type. ;)

[rossberg, 2013-06-06 10:36:49]

https://codereview.chromium.org/16408005/diff/1/src/ast.cc
File src/ast.cc (right):

https://codereview.chromium.org/16408005/diff/1/src/ast.cc#newcode663
src/ast.cc:663: if (value->IsSmi()) {
On 2013/06/06 09:24:19, rossberg wrote:
> I'm not sure how this can ever be true, given that allocation_info_cell_ has
> type JSGlobalPropertyCell (which is a HeapObject). Either the logic must be
> wrong or the type. ;)

Never mind, I was overlooking the value method... 8-}

[danno, 2013-06-07 10:58:37]

lgtm

[mvstanton, 2013-06-07 13:21:29]

Committed patchset #1 manually as r15004 (presubmit successful).