Use a separate marker value to allocate the arguments object on deoptimzation...

Use a separate marker value to allocate the arguments object on deoptimzation.

Before we used the hole value for this purpose, but this does not work once we
start using the hole value for other purposes in the optimizing compiler.


Committed: http://code.google.com/p/v8/source/detail?r=6210

[fschneider, 2011-01-06 13:48:01]

New version.
Added a new oddball object arguments_marker.
Changed all the places that used the hole to indicate the arguments object to
use the new marker instead.

[Kevin Millikin (Chromium), 2011-01-06 14:00:47]

You will have to update the classic code generators for ARM and c64, because
they use the same code in accessors.cc.

Other than that, LGTM.

http://codereview.chromium.org/6116001/diff/8001/src/heap.h
File src/heap.h (right):

http://codereview.chromium.org/6116001/diff/8001/src/heap.h#newcode119
src/heap.h:119: V(Object, arguments_marker, ArgumentsMarker)
I guess this list is unordered, but it makes sense to put all the oddballs
together.

http://codereview.chromium.org/6116001/diff/8001/src/objects-debug.cc
File src/objects-debug.cc (right):

http://codereview.chromium.org/6116001/diff/8001/src/objects-debug.cc#newcode371
src/objects-debug.cc:371: ASSERT(value == 0 || value == 1 || value == -1 ||
This seems a bit silly, but I guess it could catch a corrupted oddball?  Maybe
add a comment that we use distinct negative integers for ToNumber on 'hidden'
oddballs, and then rewrite the assert as:
const int kLeastHiddenOddballNumber = -4;
ASSERT(value <= 1);
ASSERT(value >= kLeastHiddenOddballNumber);

http://codereview.chromium.org/6116001/diff/8001/src/objects.h
File src/objects.h (right):

http://codereview.chromium.org/6116001/diff/8001/src/objects.h#newcode596
src/objects.h:596: INLINE(bool IsArgumentsMarker());
I wonder why we INLINE exactly these two?

[fschneider, 2011-01-06 16:22:49]

http://codereview.chromium.org/6116001/diff/8001/src/heap.h
File src/heap.h (right):

http://codereview.chromium.org/6116001/diff/8001/src/heap.h#newcode119
src/heap.h:119: V(Object, arguments_marker, ArgumentsMarker)
On 2011/01/06 14:00:47, Kevin Millikin wrote:
> I guess this list is unordered, but it makes sense to put all the oddballs
> together.

Done.

http://codereview.chromium.org/6116001/diff/8001/src/objects-debug.cc
File src/objects-debug.cc (right):

http://codereview.chromium.org/6116001/diff/8001/src/objects-debug.cc#newcode371
src/objects-debug.cc:371: ASSERT(value == 0 || value == 1 || value == -1 ||
On 2011/01/06 14:00:47, Kevin Millikin wrote:
> This seems a bit silly, but I guess it could catch a corrupted oddball?  Maybe
> add a comment that we use distinct negative integers for ToNumber on 'hidden'
> oddballs, and then rewrite the assert as:
> const int kLeastHiddenOddballNumber = -4;
> ASSERT(value <= 1);
> ASSERT(value >= kLeastHiddenOddballNumber);

Done.

http://codereview.chromium.org/6116001/diff/8001/src/objects.h
File src/objects.h (right):

http://codereview.chromium.org/6116001/diff/8001/src/objects.h#newcode596
src/objects.h:596: INLINE(bool IsArgumentsMarker());
On 2011/01/06 14:00:47, Kevin Millikin wrote:
> I wonder why we INLINE exactly these two?

Good question. I don't know. Maybe we should use INLINE for the others too.