Built-in JSON support

- Extended lazy loading to general objects, not just functions.
- Added lazily loaded JSON object.

[Lasse Reichstein, 2009-04-23 15:28:49]

drive-by comments on json-delay.js

http://codereview.chromium.org/93066/diff/1/12
File src/json-delay.js (right):

http://codereview.chromium.org/93066/diff/1/12#newcode40
Line 40: // include them in the reg exps in all places where whitespace is
allowed.
Is this our comment? 
Since Irregexp does include \u2028 and \u2029 in \s, we shouldn't need to add it
(as in [\s\u2028\u2029] below).

http://codereview.chromium.org/93066/diff/1/12#newcode165
Line 165: indent = stepback;
indent is a local variable here, so setting it just before returning won't do
anything. This line can be removed.
The intended behavior is achieved anyway, since the calling function's indent
hasn't been modified.

http://codereview.chromium.org/93066/diff/1/12#newcode179
Line 179: if (ObjectHasOwnProperty.call(replacer, i)) {
The specification says that the properties of replacer must be iterated in
ascending numeric order.
(Actually, it doesn't, but it does explicitly say that the list of elements
should be in ascending order, and then that something must be done for each
element in the list, which suggests that it should be done in the same order).

http://codereview.chromium.org/93066/diff/1/12#newcode185
Line 185: member += " ";
Should be on previous line (or indented and put in squiggly brackets).

http://codereview.chromium.org/93066/diff/1/12#newcode216
Line 216: indent = stepback;
Again unnecessary.

[Mads Ager (chromium), 2009-04-23 19:09:25]

LGTM after addressing these comments.

We should fix the xcode project and make sure to have a v8.gyp change ready for
when this is pushed to chromium.

http://codereview.chromium.org/93066/diff/1/3
File src/arm/ic-arm.cc (left):

http://codereview.chromium.org/93066/diff/1/3#oldcode280
Line 280: // Check that the value is a JSFunction.
We need this JSFunction check.  We invoke r1 below as a function, so we need to
check that it is a function and otherwise go to the runtime system.

That is the reason for not using GenerateCheckNonFunctionOrLoaded here.

http://codereview.chromium.org/93066/diff/1/3
File src/arm/ic-arm.cc (right):

http://codereview.chromium.org/93066/diff/1/3#newcode140
Line 140: // Check if the function has been loaded.
function -> object

http://codereview.chromium.org/93066/diff/1/8
File src/handles.cc (right):

http://codereview.chromium.org/93066/diff/1/8#newcode685
Line 685: // If lazy loading failed, restore the unloaded state of fun.
fun -> obj

http://codereview.chromium.org/93066/diff/1/9
File src/handles.h (right):

http://codereview.chromium.org/93066/diff/1/9#newcode304
Line 304: void SetupLazy(Handle<JSObject> fun,
fun -> object?

http://codereview.chromium.org/93066/diff/1/9#newcode308
Line 308: void LoadLazy(Handle<JSObject> fun, bool* pending_exception);
fun -> object?

http://codereview.chromium.org/93066/diff/1/11
File src/ia32/ic-ia32.cc (left):

http://codereview.chromium.org/93066/diff/1/11#oldcode491
Line 491: // Check that the value is a JavaScript function.
You are removing a JSFunction check here - you shouldn't.  The object will be
invoked below, so it needs to be a function.

http://codereview.chromium.org/93066/diff/1/12
File src/json-delay.js (right):

http://codereview.chromium.org/93066/diff/1/12#newcode198
Line 198: member += " ";
Indentation.

http://codereview.chromium.org/93066/diff/1/18
File src/objects.h (right):

http://codereview.chromium.org/93066/diff/1/18#newcode2586
Line 2586: // But positions for but field 2
Bit positions for bit field 2?

[arv (Not doing code reviews), 2009-04-23 21:07:58]

This is an exciting change.

http://codereview.chromium.org/93066/diff/1/12
File src/json-delay.js (right):

http://codereview.chromium.org/93066/diff/1/12#newcode57
Line 57: return global.eval('(' + s + ')');
I hope this is just a temporary solution. This solution is necessary in pure JS
but the whole point of moving JSON into the language was to get away from
relying on regualr expressions and eval.

[Christian Plesner Hansen, 2009-04-24 08:11:21]

No comment means "fixed".

http://codereview.chromium.org/93066/diff/1/3
File src/arm/ic-arm.cc (left):

http://codereview.chromium.org/93066/diff/1/3#oldcode280
Line 280: // Check that the value is a JSFunction.
Oh, good point.

http://codereview.chromium.org/93066/diff/1/12
File src/json-delay.js (right):

http://codereview.chromium.org/93066/diff/1/12#newcode40
Line 40: // include them in the reg exps in all places where whitespace is
allowed.
Good point.

http://codereview.chromium.org/93066/diff/1/12#newcode57
Line 57: return global.eval('(' + s + ')');
That's not the reason I'm adding it but more to have JSON available without
having to send the code over the wire.  I didn't have any plans to change this.

I'll submit the code as it is and then I'll take a look at how difficult it
would be to move the critical parts into the vm.  It might just be a matter of
adding a special flavor of global eval that has an extra check that the AST is
benign before executing it that bails out if it isn't.

http://codereview.chromium.org/93066/diff/1/12#newcode165
Line 165: indent = stepback;
Good point.  I was following the spec a bit too closely.

http://codereview.chromium.org/93066/diff/1/12#newcode179
Line 179: if (ObjectHasOwnProperty.call(replacer, i)) {
I got ascending and descending mixed up.  Even though what I have is called
"right/left confusion" it actually means that I confuse any pair of directions,
as this demonstrates.  Also good and evil.

http://codereview.chromium.org/93066/diff/1/12#newcode185
Line 185: member += " ";
Gah.  But you're right.

http://codereview.chromium.org/93066/diff/1/18
File src/objects.h (right):

http://codereview.chromium.org/93066/diff/1/18#newcode2586
Line 2586: // But positions for but field 2
Whoops.

[Christian Plesner Hansen, 2009-04-24 13:41:11]

http://codereview.chromium.org/93066/diff/1/12
File src/json-delay.js (right):

http://codereview.chromium.org/93066/diff/1/12#newcode57
Line 57: return global.eval('(' + s + ')');
Update: it turned out to be straightforward to move parsing and validation into
native code so I've done that.  No more regexps and eval.