Use static type information from IDL to streamline the wrapping and unwrappin...

Use static type information from IDL to streamline the wrapping and unwrapping process
between DOM nodes and JS objects.

The basic idea is that, IDL files provide static type information. Certain types, such as subtypes of Node,
only need 'NODE' has type for wrapping and unwrapping. So, intead of going through a gigatic switch statement,
IDL compiler generates fast path for know types, the CL only does it for Node and subtypes.

I have seen it improves DOM-peerable example by 5% when running standalone, and 35% when running with
whole Dromaeo tests.

I missed the another point of this CB. It removed expensive MaybeDOMWrapper checks in production code. It contributes a lot to the overhead.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=2490

[Mike Belshe, 2008-09-23 00:36:13]

looks good; would be nice to see the generated code, but I think I get it.

http://codereview.chromium.org/3195/diff/1/5
File webkit/port/bindings/scripts/CodeGenerator.pm (right):

http://codereview.chromium.org/3195/diff/1/5#newcode284
Line 284: print "Scanning interface " . $interface . " in " . $directory . "\n"
if $verbose;
Do we want this print statement?  Or is this for debugging.

http://codereview.chromium.org/3195/diff/1/9
File webkit/port/bindings/v8/v8_proxy.cpp (right):

http://codereview.chromium.org/3195/diff/1/9#newcode1916
Line 1916: NODE_WRAPPER_TYPES(MAKE_CASE)
I wonder if we shouldn't #ifndef NDEBUG these to avoid having a large switch
statement here

[Mads Ager (chromium), 2008-09-23 06:45:08]

LGTM.  Great simplifications.

http://codereview.chromium.org/3195/diff/1/3
File webkit/port/bindings/scripts/IDLParser.pm (right):

http://codereview.chromium.org/3195/diff/1/3#newcode60
Line 60: sub ParseInheritance
Add a one line comment stating what the return value will be?

http://codereview.chromium.org/3195/diff/1/6
File webkit/port/bindings/v8/v8_proxy.h (right):

http://codereview.chromium.org/3195/diff/1/6#newcode444
Line 444: // use NODE as cptr_type. JS wrapper has store cptr_type and impl as
JS wrapper has store -> JS wrappers store

[Feng Qian, 2008-09-23 21:46:04]

http://codereview.chromium.org/3195/diff/1/5
File webkit/port/bindings/scripts/CodeGenerator.pm (right):

http://codereview.chromium.org/3195/diff/1/5#newcode284
Line 284: print "Scanning interface " . $interface . " in " . $directory . "\n"
if $verbose;
On 2008/09/23 00:36:13, mbelshe wrote:
> Do we want this print statement?  Or is this for debugging.

For debugging purpose, the condition variable $verbose is off by default.

http://codereview.chromium.org/3195/diff/1/9
File webkit/port/bindings/v8/v8_proxy.cpp (right):

http://codereview.chromium.org/3195/diff/1/9#newcode1916
Line 1916: NODE_WRAPPER_TYPES(MAKE_CASE)
Here I'd like to let it return NULL so the render will crash right after this
call. This shouldn't happen although.

I looked at the generated assembly, a jump table is used for the big switch
statement. So the runtime overhead isn't high.

On 2008/09/23 00:36:13, mbelshe wrote:
> I wonder if we shouldn't #ifndef NDEBUG these to avoid having a large switch
> statement here
> 
>

http://codereview.chromium.org/3195/diff/1/6
File webkit/port/bindings/v8/v8_proxy.h (right):

http://codereview.chromium.org/3195/diff/1/6#newcode444
Line 444: // use NODE as cptr_type. JS wrapper has store cptr_type and impl as
On 2008/09/23 06:45:09, Mads Ager wrote:
> JS wrapper has store -> JS wrappers store

Done.