Reapply revisions 1432, 1433, 1469 and 1472 while fixing issue 279.

Reapply revisions 1432, 1433, 1469 and 1472 while fixing issue 279.

Committed: http://code.google.com/p/v8/source/detail?r=1571

[Erik Corry, 2009-03-16 14:07:26]

LGTM

I'd like to see the test include a huge nested literal so we can test GC during
clone of boilerplate.

The clone operation is recursive.  Should we be checking for stack overflow?

Are there tests of making sure that nested arrays and objects have the correct
context when created?  I'm a bit hazy on this part of the VM to be honest, but
it seems there is an issue here, but no tests.

Otherwise LGTM if it lints, passes tests and works on ARM.

http://codereview.chromium.org/48006/diff/1/7
File src/codegen-arm.cc (right):

http://codereview.chromium.org/48006/diff/1/7#newcode2637
Line 2637: // If the entry is undefined we call the runtime system to computed
spolling (not yours)

http://codereview.chromium.org/48006/diff/1/5
File src/runtime.cc (right):

http://codereview.chromium.org/48006/diff/1/5#newcode73
Line 73: // Cast the given object to an int and store it in a variable with
Should say smi instead of int in name and comments.

http://codereview.chromium.org/48006/diff/1/5#newcode117
Line 117: if (attributes != NONE) continue;
I think this line warrants some explanation.

http://codereview.chromium.org/48006/diff/1/5#newcode120
Line 120: if (value->IsFailure()) return result;
You just asserted this couldn't happen...

http://codereview.chromium.org/48006/diff/1/5#newcode140
Line 140: if (value->IsFailure()) return result;
ditto

http://codereview.chromium.org/48006/diff/1/5#newcode154
Line 154: HandleScope scope;
I wonder why we need this.  DeepCopyBoilerplate isn't handle-based so it can't
call handle code.

http://codereview.chromium.org/48006/diff/1/2
File test/mjsunit/regress/regress-279.js (right):

http://codereview.chromium.org/48006/diff/1/2#newcode62
Line 62: assertEquals(0, b[0].length, "Array in array");
We should test that /re/ regexp literals in arrays and objects perform as we
expect and that.

var n = new Object();

function make() { return [ n ]; }

var a = make();
a[0].foo = "bar";
assertEquals("bar", n.foo) and similar cases.

[olehougaard, 2009-03-19 13:17:18]

The change resulted in a performance regression for code that didn't use nested
objects or arrays, so I changed it to call different runtime functions dependent
on the depth of the object (or array).

Turns out that a boilerplate object can have slow properties if it gets its map
from the map cache, so I've dealt with that as a special case.

Please review again.

[Erik Corry, 2009-03-20 09:09:36]

LGTM

http://codereview.chromium.org/48006/diff/1027/1033
File src/codegen-arm.cc (right):

http://codereview.chromium.org/48006/diff/1027/1033#newcode1
Line 1: // Copyright 2006-2008 the V8 project authors. All rights reserved.
2009

http://codereview.chromium.org/48006/diff/1027/1031
File src/runtime.cc (right):

http://codereview.chromium.org/48006/diff/1027/1031#newcode112
Line 112: if (copy->HasFastProperties()) {
I think we can use Kasper's idea here to avoid the DescriptorReader and just get
the properties by their index from 0 to length - 1.

http://codereview.chromium.org/48006/diff/1027/1031#newcode129
Line 129: } else {
It seems wrong that a boilerplate object should be non-fast case.  Lets submit
this as it is then see if we can work out where the slow case maps are coming
from.  Ideally boilerplate objects should be all-in-object.

http://codereview.chromium.org/48006/diff/2054/2067
File src/objects.h (right):

http://codereview.chromium.org/48006/diff/2054/2067#newcode1401
Line 1401: // Access to set in object properties.
÷set

[olehougaard, 2009-03-23 07:26:45]

http://codereview.chromium.org/48006/diff/1027/1033
File src/codegen-arm.cc (right):

http://codereview.chromium.org/48006/diff/1027/1033#newcode1
Line 1: // Copyright 2006-2008 the V8 project authors. All rights reserved.
On 2009/03/20 09:09:36, Erik Corry wrote:
> 2009

Fixed a couple of places.

http://codereview.chromium.org/48006/diff/1027/1031
File src/runtime.cc (right):

http://codereview.chromium.org/48006/diff/1027/1031#newcode112
Line 112: if (copy->HasFastProperties()) {
On 2009/03/20 09:09:36, Erik Corry wrote:
> I think we can use Kasper's idea here to avoid the DescriptorReader and just
get
> the properties by their index from 0 to length - 1.

Let's do that in a later cl.

http://codereview.chromium.org/48006/diff/1027/1031#newcode129
Line 129: } else {
On 2009/03/20 09:09:36, Erik Corry wrote:
> It seems wrong that a boilerplate object should be non-fast case.  Lets submit
> this as it is then see if we can work out where the slow case maps are coming
> from.  Ideally boilerplate objects should be all-in-object.

Ok

http://codereview.chromium.org/48006/diff/2054/2067
File src/objects.h (right):

http://codereview.chromium.org/48006/diff/2054/2067#newcode1401
Line 1401: // Access to set in object properties.
On 2009/03/20 09:09:36, Erik Corry wrote:
> ÷set

Fixed