Fixes for sunflower

Fixes for sunflower

R=leafp@google.com

Committed: https://github.com/dart-lang/dev_compiler/commit/e485df088816b93a8a83cb2e58e45666ce31996e

[vsm, 2015-05-26 13:40:52]

vsm@google.com changed reviewers:
+ leafp@google.com

[vsm, 2015-05-26 13:40:58]

https://codereview.chromium.org/1143953004/diff/1/lib/runtime/dart_runtime.js
File lib/runtime/dart_runtime.js (right):

https://codereview.chromium.org/1143953004/diff/1/lib/runtime/dart_runtime.js...
lib/runtime/dart_runtime.js:102: return f.apply(obj, args);
We were breaking in a DOM call - no function type found for addEventListener. 
Perhaps we should have a way to look that up on this side?

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
File tool/input_sdk/private/js_helper.dart (right):

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
tool/input_sdk/private/js_helper.dart:585: static Null
_throwFormatException(String string) {
This feels a bit strange, but see the casts in the corresponding generated JS
file - both were failing.

[Leaf, 2015-05-26 18:19:46]

https://codereview.chromium.org/1143953004/diff/1/lib/runtime/dart_runtime.js
File lib/runtime/dart_runtime.js (right):

https://codereview.chromium.org/1143953004/diff/1/lib/runtime/dart_runtime.js...
lib/runtime/dart_runtime.js:102: return f.apply(obj, args);
On 2015/05/26 13:40:58, vsm wrote:
> We were breaking in a DOM call - no function type found for addEventListener. 
> Perhaps we should have a way to look that up on this side?

Yes, if we have types for these that could be useful.  In general we probably
want to be prepared to get JS functions as well, maybe with some sort of minimal
arity checking.

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
File tool/input_sdk/private/js_helper.dart (right):

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
tool/input_sdk/private/js_helper.dart:585: static Null
_throwFormatException(String string) {
On 2015/05/26 13:40:58, vsm wrote:
> This feels a bit strange, but see the casts in the corresponding generated JS
> file - both were failing.

This is a bit odd.  I don't understand why Null helps with the casts.  Null
isn't bottom, it's a direct subtype of Object, so I would expect this to be
worse (since it's not even castable to anything except object).

[vsm, 2015-05-27 13:37:21]

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
File tool/input_sdk/private/js_helper.dart (right):

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
tool/input_sdk/private/js_helper.dart:585: static Null
_throwFormatException(String string) {
On 2015/05/26 18:19:45, Leaf wrote:
> On 2015/05/26 13:40:58, vsm wrote:
> > This feels a bit strange, but see the casts in the corresponding generated
JS
> > file - both were failing.
> 
> This is a bit odd.  I don't understand why Null helps with the casts.  Null
> isn't bottom, it's a direct subtype of Object, so I would expect this to be
> worse (since it's not even castable to anything except object).

Ugh, it turns out that we were reporting an error with Null - we were just
suppressing it and forcing the compile anyway.

OTOH, "void" here currently does work - it's treated as bottom by our rules.  We
could use that.

[Leaf, 2015-05-27 22:16:34]

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
File tool/input_sdk/private/js_helper.dart (right):

https://codereview.chromium.org/1143953004/diff/1/tool/input_sdk/private/js_h...
tool/input_sdk/private/js_helper.dart:585: static Null
_throwFormatException(String string) {
On 2015/05/27 13:37:21, vsm wrote:
> On 2015/05/26 18:19:45, Leaf wrote:
> > On 2015/05/26 13:40:58, vsm wrote:
> > > This feels a bit strange, but see the casts in the corresponding generated
> JS
> > > file - both were failing.
> > 
> > This is a bit odd.  I don't understand why Null helps with the casts.  Null
> > isn't bottom, it's a direct subtype of Object, so I would expect this to be
> > worse (since it's not even castable to anything except object).
> 
> Ugh, it turns out that we were reporting an error with Null - we were just
> suppressing it and forcing the compile anyway.
> 
> OTOH, "void" here currently does work - it's treated as bottom by our rules. 
We
> could use that.

I don't understand this either... :}  I don't think we treat void as bottom.  We
allow void returning functions to be supertypes of functions which return
anything.  But this is the other way around, isn't it?  We're assigning a String
-> void to a String -> int and String -> double.  This requires that String ->
void be a subtype of String -> int/double, which requires void to be a subtype
of int/double, which it isn't.  This seems like it should still be an error?

[vsm, 2015-05-28 15:49:57]

PTAL

Note: this is rebased against https://codereview.chromium.org/1154213008/, which
is still pending.

https://codereview.chromium.org/1143953004/diff/60001/lib/runtime/dart/_js_he...
File lib/runtime/dart/_js_helper.js (right):

https://codereview.chromium.org/1143953004/diff/60001/lib/runtime/dart/_js_he...
lib/runtime/dart/_js_helper.js:1896: handleError = dart.fn(s =>
dart.as(Primitives._throwFormatException(dart.as(s, core.String)), core.int),
core.int, [core.Object]);
It's interesting we need the cast to String here.  Perhaps an argument for
inferring String->int instead of Object->int?

[Leaf, 2015-05-28 15:53:20]

lgtm

https://codereview.chromium.org/1143953004/diff/60001/lib/runtime/dart/_js_he...
File lib/runtime/dart/_js_helper.js (right):

https://codereview.chromium.org/1143953004/diff/60001/lib/runtime/dart/_js_he...
lib/runtime/dart/_js_helper.js:1896: handleError = dart.fn(s =>
dart.as(Primitives._throwFormatException(dart.as(s, core.String)), core.int),
core.int, [core.Object]);
On 2015/05/28 15:49:57, vsm wrote:
> It's interesting we need the cast to String here.  Perhaps an argument for
> inferring String->int instead of Object->int?

Yes, this is on my to do list.  It requires better integration between the
inference and the checker though, since we must check that the body is
well-formed at type String.

[vsm, 2015-05-28 16:29:05]

Committed patchset #5 (id:80001) manually as
e485df088816b93a8a83cb2e58e45666ce31996e (presubmit successful).

[Jennifer Messerly, 2015-05-28 21:02:05]

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

[Jennifer Messerly, 2015-05-28 21:02:06]

lgtm too