Convert legacy 'native code' to JS-forms in dart:html

Convert legacy 'native code' to JS-forms in dart:html

This should enable minification to rename function arguments. Issue 5821 and Issue 5822 need to be fixed first.

BUG=dart:5808
Committed: https://code.google.com/p/dart/source/detail?r=13690

[sra1, 2012-10-12 03:02:27]

This + issue 5821 + issue 5822 is too big for M1, so we need to disable renaming
on --minify until after M1 is cut.

[erikcorry, 2012-10-12 07:13:03]

https://codereview.chromium.org/11094082/diff/5001/lib/html/dart2js/html_dart...
File lib/html/dart2js/html_dart2js.dart (left):

https://codereview.chromium.org/11094082/diff/5001/lib/html/dart2js/html_dart...
lib/html/dart2js/html_dart2js.dart:18538: var vendors = ['ms', 'moz', 'webkit',
'o'];
This way the 'vendors' variable does not get minified.

[Anton Muhin, 2012-10-12 11:40:38]

DBC

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
File lib/html/scripts/systemhtml.py (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
lib/html/scripts/systemhtml.py:856: '  $MODIFIERS$TYPE $(HTML_NAME)($PARAMS)
native "$NAME";\n')
just for my education, any reasons why it still uses native?  If it is switched
to JS helper, we can remove some cruft.

[floitsch, 2012-10-12 13:40:17]

LGTM.
I would appreciate if somebody else could look at the python code too, though.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
File lib/html/templates/html/dart2js/factoryprovider_CSSMatrix.darttemplate
(right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
lib/html/templates/html/dart2js/factoryprovider_CSSMatrix.darttemplate:6: static
CSSMatrix createCSSMatrix([String cssValue = '']) =>
by convention we put "=>" into the next line.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
File lib/html/templates/html/dart2js/factoryprovider_ShadowRoot.darttemplate
(right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
lib/html/templates/html/dart2js/factoryprovider_ShadowRoot.darttemplate:6:
static ShadowRoot createShadowRoot(Element host) =>
move "=>" to next line.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
File lib/html/templates/html/impl/impl_Element.darttemplate (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
lib/html/templates/html/impl/impl_Element.darttemplate:865: static Element
createElement_tag(String tag) =>
move '=>' to next line.

[blois, 2012-10-12 16:55:19]

lgtm

JS changes look good to me.

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
File lib/html/scripts/systemhtml.py (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
lib/html/scripts/systemhtml.py:791: ' JS("void", "#.$NAME = #", this, value);
}\n',
Nit- should this be expanded to multiple lines? Not sure what the style guide is
for this and how close we want to try to follow it.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
File lib/html/templates/html/impl/impl_Element.darttemplate (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
lib/html/templates/html/impl/impl_Element.darttemplate:746: if (JS('bool',
'!!#.insertAdjacentText', this)) {
indentation

[sra1, 2012-10-12 21:39:46]

I'll check in after M1 is cut.

https://codereview.chromium.org/11094082/diff/5001/lib/html/dart2js/html_dart...
File lib/html/dart2js/html_dart2js.dart (left):

https://codereview.chromium.org/11094082/diff/5001/lib/html/dart2js/html_dart...
lib/html/dart2js/html_dart2js.dart:18538: var vendors = ['ms', 'moz', 'webkit',
'o'];
On 2012/10/12 07:13:03, erikcorry wrote:
> This way the 'vendors' variable does not get minified.

Yes, but writing this code in Dart is extremely painful at the moment.
The Dart-compiled loops etc cause more code to be compiled in than would be
saved.
One way to improve the situation would be to parse the JS string into a JS AST.
Then it could be minified too.

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
File lib/html/scripts/systemhtml.py (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
lib/html/scripts/systemhtml.py:791: ' JS("void", "#.$NAME = #", this, value);
}\n',
On 2012/10/12 16:55:19, blois wrote:
> Nit- should this be expanded to multiple lines? Not sure what the style guide
is
> for this and how close we want to try to follow it.

Done.

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
lib/html/scripts/systemhtml.py:856: '  $MODIFIERS$TYPE $(HTML_NAME)($PARAMS)
native "$NAME";\n')
On 2012/10/12 11:40:38, Anton Muhin wrote:
> just for my education, any reasons why it still uses native?  If it is
switched
> to JS helper, we can remove some cruft.

Some form where there is a simple renaming will continued to be supported.  We
will transition to using metadata at some point instead of the keyword.

There are two thing here that can't be achieved with JS(...).

1. The compiler needs to know there is already a member called $NAME to avoid
problems in choosing field names and minification.

2. The compiler can't inline the code since it does not know that the blob of
JavaScript generated by JS has the correct properties.
In general, you can't inline here, even if you have resolved the foo method
precisely:

   x.foo()

You can only inline if you know that the body of foo will throw a
NullPointerException before any other  effect.  JS(...) generated code is opaque
to the compiler, so has no guarantee and will prevent inlining.

It is also faster to directly generate the inlined code than inline an
expression.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
File lib/html/templates/html/dart2js/factoryprovider_CSSMatrix.darttemplate
(right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
lib/html/templates/html/dart2js/factoryprovider_CSSMatrix.darttemplate:6: static
CSSMatrix createCSSMatrix([String cssValue = '']) =>
On 2012/10/12 13:40:17, floitsch wrote:
> by convention we put "=>" into the next line.

I disagree that this is the convention.
In pkg, it is 140:0 in favor of  =>\n

To be resolved by Issue 5853.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
File lib/html/templates/html/dart2js/factoryprovider_ShadowRoot.darttemplate
(right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/da...
lib/html/templates/html/dart2js/factoryprovider_ShadowRoot.darttemplate:6:
static ShadowRoot createShadowRoot(Element host) =>
On 2012/10/12 13:40:17, floitsch wrote:
> move "=>" to next line.

To be resolved outside of this CL.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
File lib/html/templates/html/impl/impl_Element.darttemplate (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
lib/html/templates/html/impl/impl_Element.darttemplate:746: if (JS('bool',
'!!#.insertAdjacentText', this)) {
On 2012/10/12 16:55:19, blois wrote:
> indentation

Done.

https://codereview.chromium.org/11094082/diff/5001/lib/html/templates/html/im...
lib/html/templates/html/impl/impl_Element.darttemplate:865: static Element
createElement_tag(String tag) =>
On 2012/10/12 13:40:17, floitsch wrote:
> move '=>' to next line.

To be resolved outside of this CL.

[Anton Muhin, 2012-10-15 10:27:34]

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
File lib/html/scripts/systemhtml.py (right):

https://codereview.chromium.org/11094082/diff/5001/lib/html/scripts/systemhtm...
lib/html/scripts/systemhtml.py:856: '  $MODIFIERS$TYPE $(HTML_NAME)($PARAMS)
native "$NAME";\n')
I see, thanks, Stephen.  May main concern is this form is not supported for
statics, so either we should fork it and resort to JS for statics or keep
statics unsupported.

On 2012/10/12 21:39:47, sra1 wrote:
> On 2012/10/12 11:40:38, Anton Muhin wrote:
> > just for my education, any reasons why it still uses native?  If it is
> switched
> > to JS helper, we can remove some cruft.
> 
> Some form where there is a simple renaming will continued to be supported.  We
> will transition to using metadata at some point instead of the keyword.
> 
> There are two thing here that can't be achieved with JS(...).
> 
> 1. The compiler needs to know there is already a member called $NAME to avoid
> problems in choosing field names and minification.
> 
> 2. The compiler can't inline the code since it does not know that the blob of
> JavaScript generated by JS has the correct properties.
> In general, you can't inline here, even if you have resolved the foo method
> precisely:
> 
>    x.foo()
> 
> You can only inline if you know that the body of foo will throw a
> NullPointerException before any other  effect.  JS(...) generated code is
opaque
> to the compiler, so has no guarantee and will prevent inlining.
> 
> It is also faster to directly generate the inlined code than inline an
> expression.

[ngeoffray, 2012-10-15 14:25:54]

Can't wait to see this landed! Great progress Stephen!