Added Area-HTML, Triaged labels.

I'm unable to repro this in either FF or Chrome, checked or unchecked.

Stephen- any idea what might be causing this?

cc @rakudrama.

This comment was originally written by da...@altern.org

Sorry, my simplified test case does not exhibit the same behavior. I will try to understand why.
I was able to work around the problem by replacing "control is TextInputElement" by "control is InputElement && (control as InputElement).type == 'text'".

This comment was originally written by da...@altern.org

Here is a modified test case - I just checked this returns false in Firefox when text is typed in the input.

  h.TextInputElement input = new h.TextInputElement();
  h.document.body.append(input);
  h.Element control = input;
  input.onInput.listen((h.Event event) => print(control is h.TextInputElement));

I guess dart2js forgets the class when the context changes...

Very interesting.

Simplified repro:
import 'dart:async';
import 'dart:html';

void main() {
  TextInputElement input = new TextInputElement();
  
  print(input is TextInputElement);
  Timer.run(() {
    print(input is TextInputElement);
  });
}

prints:
true
false

Removed Area-HTML label.
Added Area-Dart2JS label.

tl;dr: TypeMask bug?

It gets stranger.
The first, directly visible, is-check is evaluated at run time.
This is to be expected since the factory constructor bottoms out at a JS call equivalent to:

   JS("Element", "document.createElement(#)", tag);

i.e. The inferred type of variable "input" should be subtype(Element) or subclass(Element).

The closed-over test is constant-folded to false.

$.main = function() {
  var input = $.InputElement_InputElement("text");
  $.Primitives_printString($.JSBool_methods.toString$0(typeof input === "object" && input !== null && !!$.getInterceptor(input).$isTextInputElement));
  $.Timer_run(new $.main_anon(input));
};

$$.main_anon = {"": "Closure;input_0",
  call$0: function() {
    this.input_0;
    $.Primitives_printString($.JSBool_methods.toString$0(false));
  }
};

I think this is a bug with TypeMask causing wrong constant-folding of is-checks.
In particular, this expression is returning the wrong value:

[null|subclass=Element].intersection([subtype=TextInputElement], compiler).isEmpty = true

The classes are roughly as follows, which puts InputElement in the intersection, but I failed to make a simple repro.

class Element {}
class InputElement extends Element implements ..., TextInputElement, ... {}
abstract class TextInputElement ... {}

Set owner to @kasperl.
Removed Priority-Medium label.
Added Priority-High label.

Hmm. The missing piece of information here is that TextInputElement indirectly implements Element. That seems to throw off the intersection code in type mask. It certainly does look like a bug in the implementation of TypeMask.intersection.

Added Accepted label.

Fixed in r22412.

Added Fixed label.

This comment was originally written by da...@altern.org

Not sure if it's related, but now (control is TextInputElement) is true for a CheckboxInputElement (with Dart VM) (Dart SDK 0.6.9.2-r25388)

damieng- this is by design, a side-effect of how these types are implemented. They are all actually InputElements.

See https://code.google.com/p/dart/issues/detail?id=1684