dart:js - speed up constructor calls with few arguments.

dart:js - speed up constructor calls with few arguments.

This eliminates the 5x perf hit reported in http://dartbug.com/23514

R=sigmund@google.com

Committed: https://github.com/dart-lang/sdk/commit/4b2d2379cfdf79de11d014dc27f8d606ef8f627b

[sra1, 2015-05-23 01:04:23]

sra@google.com changed reviewers:
+ jakemac@google.com

[sra1, 2015-05-23 01:28:00]

Patchset #1 (id:1) has been deleted

[Siggi Cherem (dart-lang), 2015-05-26 16:35:17]

sigmund@google.com changed reviewers:
+ sigmund@google.com

[Siggi Cherem (dart-lang), 2015-05-26 16:35:18]

lgtm, just some questions below

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
File sdk/lib/js/dart2js/js_dart2js.dart (left):

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:157: var jsObj = JS('JavaScriptObject', 'new
#()', factoryFunction);
I'm not familiar with this annotation here - any reason to remove it at this
point? Or is it equivalent to not provide anything in the first arg?

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
File sdk/lib/js/dart2js/js_dart2js.dart (right):

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:158: return _wrapToDart(JS('', 'new #(#, #)',
constr, arg0, arg1));
any reason why not write these 3 lines as:
return _wrapToDArt(JS('',
    'new #(#[0], #[1])',
    constr, arguments, arguments));

or use a named template 'new #ctor(#argumets[0], #arguments[1])' ?

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:164: return _wrapToDart(JS('', 'new #(#, #,
#)', constr, arg0, arg1, arg2));
wrap

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:553: if (o is JsObject) {
just to make sure I'm not missing something, these changes in this function are
just style, right?

[sra1, 2015-05-26 17:58:59]

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
File sdk/lib/js/dart2js/js_dart2js.dart (left):

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:157: var jsObj = JS('JavaScriptObject', 'new
#()', factoryFunction);
On 2015/05/26 16:35:18, Siggi Cherem (dart-lang) wrote:
> I'm not familiar with this annotation here - any reason to remove it at this
> point? Or is it equivalent to not provide anything in the first arg?

The constructor could be a constructor for something understood by Dart, e.g
Blob or typed array.

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
File sdk/lib/js/dart2js/js_dart2js.dart (right):

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:158: return _wrapToDart(JS('', 'new #(#, #)',
constr, arg0, arg1));
On 2015/05/26 16:35:18, Siggi Cherem (dart-lang) wrote:
> any reason why not write these 3 lines as:
> return _wrapToDArt(JS('',
>     'new #(#[0], #[1])',
>     constr, arguments, arguments));

Not really, but the expression gets big and any automatic layout will either
make it ugly or make it take a lot of lines.

> 
> or use a named template 'new #ctor(#argumets[0], #arguments[1])' ?

I like this but named template parameters are not available to the JS 'function'
since the arguments are decomposed/spread at compile time.

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:164: return _wrapToDart(JS('', 'new #(#, #,
#)', constr, arg0, arg1, arg2));
On 2015/05/26 16:35:18, Siggi Cherem (dart-lang) wrote:
> wrap

Done.

https://codereview.chromium.org/1152673003/diff/20001/sdk/lib/js/dart2js/js_d...
sdk/lib/js/dart2js/js_dart2js.dart:553: if (o is JsObject) {
On 2015/05/26 16:35:18, Siggi Cherem (dart-lang) wrote:
> just to make sure I'm not missing something, these changes in this function
are
> just style, right?

Yes.  Initially I had just moved || to the end of the line.

[sra1, 2015-05-26 22:02:41]

Committed patchset #2 (id:30002) manually as
4b2d2379cfdf79de11d014dc27f8d606ef8f627b (presubmit successful).