[Arguments] Introduce a new backing store for non-strict arguments objects.

[Arguments] Introduce a new backing store for non-strict arguments objects.

The elements of a non-strict arguments object go through a level of
indirection.  I've lifted the term "parameter map" from the ES5 spec.  The
parameter map maps property indexes to context slot indexes for the indexed
properties that alias the parameters.  It contains the context and a
separate backing store for indexed properties that do not alias the
parameters.

For non-strict arguments allocation, the arguments allocation stub goes
immediately to the runtime (temporarily).  Inline caching is disabled for
most non-strict arguments objects.

R=ager@chromium.org

Committed: http://code.google.com/p/v8/source/detail?r=7334

[Kevin Millikin (Chromium), 2011-03-23 16:07:20]

http://codereview.chromium.org/6725030/diff/1/src/accessors.cc
File src/accessors.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/accessors.cc#newcode736
src/accessors.cc:736: UNIMPLEMENTED();
I have placed UNIMPLEMENTED() in places that will need to change or be inspected
to make sure they still work.

http://codereview.chromium.org/6725030/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/arm/code-stubs-arm.cc#newco...
src/arm/code-stubs-arm.cc:4843: if (type_ == NEW_NON_STRICT) {
This is never true, but only temporarily.

http://codereview.chromium.org/6725030/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/objects.cc#newcode2704
src/objects.cc:2704: UNIMPLEMENTED();
I'm pretty sure some of these UNIMPLEMENTEDs are actually UNREACHABLE (like this
one), but I haven't inspected them all to verify.

http://codereview.chromium.org/6725030/diff/1/src/objects.cc#newcode2708
src/objects.cc:2708: case EXTERNAL_SHORT_ELEMENTS:
I changed all these switches over the elements kind to be exhaustive, with no
default, so it's a compilation error to forget to handle one.

http://codereview.chromium.org/6725030/diff/1/src/objects.cc#newcode7588
src/objects.cc:7588: UNIMPLEMENTED();
This is the point where I really want to use the existing code to work on the
indirect actual backing store holding the unaliased indexed properties.

The existing code is not well factored for that.  It either needs functions that
take the elements backing store as an extra argument or else needs to be taught
that some objects have an extra level of indirection to get to the real backing
store.

I'll decide on a case-by-case basis how to refactor.

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc
File src/runtime.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc#newcode7196
src/runtime.cc:7196: if (argument_count > 0) {
I'm aware this is monolithic.

There will eventually be a function in handles.cc (unless we think of a better
place) to do all this because there are other places where we allocate an
arguments object explicitly via Factor::NewArgumentsObject and then fill it in.

[Mads Ager (chromium), 2011-03-24 08:04:04]

LGTM

http://codereview.chromium.org/6725030/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/arm/code-stubs-arm.cc#newco...
src/arm/code-stubs-arm.cc:4873: __ mov(r1, Operand(r1, LSR, kSmiTagSize));  //
Untag the length for the loop.
Long line.

http://codereview.chromium.org/6725030/diff/1/src/objects.cc
File src/objects.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/objects.cc#newcode2708
src/objects.cc:2708: case EXTERNAL_SHORT_ELEMENTS:
On 2011/03/23 16:07:20, Kevin Millikin wrote:
> I changed all these switches over the elements kind to be exhaustive, with no
> default, so it's a compilation error to forget to handle one.

Thanks. Good idea.

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc
File src/runtime.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc#newcode7212
src/runtime.cc:7212: // Store the context and the arguments array at the end of
the
end -> beginning?

elements array -> parameter map?

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc#newcode7244
src/runtime.cc:7244: // elements array.
elements array -> parameter map

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc#newcode7248
src/runtime.cc:7248: // The context index goes in the elements array with a hole
in the
elements array -> parameter map

[Kevin Millikin (Chromium), 2011-03-24 08:42:51]

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc
File src/runtime.cc (right):

http://codereview.chromium.org/6725030/diff/1/src/runtime.cc#newcode7212
src/runtime.cc:7212: // Store the context and the arguments array at the end of
the
On 2011/03/24 08:04:04, Mads Ager wrote:
> end -> beginning?
> 
> elements array -> parameter map?

Thanks.  Those were some stale comments.