Multi-touch test cases

Hi all,

Here's my attempt at some multi-touch test cases. The tests can be found at 
http://w3c-test.org/webevents/tests/touch-events-v1/submissions/Nokia.

I took a different approach than what we did with the single-touch case by 
having several test files targeted at different aspects. You can find the 
following test files at the above location.

create-touch-touchlist.html - tests the createTouch and createTouchList APIs. 
(This actually extends but also overlaps some of the test cases in the 
single-touch test file. We might consider removing the overlapping cases from 
the single-touch file.)
multi-touch-interfaces.html - tests the implementation of various interfaces 
(Touch, TouchList and the various events)
multi-touch-interactions.html - tests the relationship between Touch events 
received over time (e.g. how the touch lists in a touch event relates to those 
received before it etc). Although I included a specific touch pattern in this 
test, the test actually works for any touch pattern.

I've run the tests on a few browsers/devices (the useful ones being 
Chrome/Firefox/Opera on Android and Mobile Safari on iPad). Most of the 
failures are related to issues already exposed by the single-touch test:
- identifiedTouch is missing from WebKit implementations (Mobile Safari, 
Chrome on Android).
- WebKit implementations (Mobile Safari on iPad, Chrome on Android)of 
createTouchList does not support an array as input parameter 
(https://bugs.webkit.org/show_bug.cgi?id=97418).
- with Firefox on Android and Safari on iPad, the clientX/Y check fails.

A notable multi-touch specific observation:
- with Mobile Safari on iPad, with the touch pattern stated in the test file, 
assertions 1.5.3.4 and 1.5.4.2 sometimes fail. It appears that when multiple 
fingers are lifted simultaneously, the UA sometimes dispatches the "same" 
touchend event to the different targets, where touches is the same but 
changedTouches and targetTouches vary according to the target element. In 
particular, changedTouches contains only those touches that were removed from 
the touch surface, and originated from the target element. This contradicts 
the definition in the spec that "For the touchend and touchcancel events this 
must be a list of the touch points that have just been removed from the 
surface.", which to me reads "regardless of where they started". When these 
"same" touchend events are processed one after another, the behavior becomes 
different than the expected behavior. To see this in action, uncomment the 
commented statement in debug_print() at the top of 
multi-touch-interactions.js. (It would probably take several attempts to see a 
failure. It all depends on the timing of removing the fingers and possibly 
also internal timing on the device itself.)

[I also tried these tests on Firefox on N9, the stock WebKit browser on N9 and 
Opera on Symbian, but encountered rather surprising behaviors such as not able 
to get multi-touch to work at all on Firefox/N9, and not more than two touches 
on Opera/Symbian. Seeing that these vendors probably are no longer actively 
developing for these platforms, I assume there's little point in pursuing the 
issues.]

On a per browser/engine basis, the results look like this.
- Opera on Android passes all tests.
- Firefox on Android fails only on the clientX/Y check.
- WebKit (Chrome and Mobile Safari) has issues with identifiedTouch and 
createTouchList with arrays.
- Additionally Mobile Safari has issues with the clientX/Y check, as well as 
when multiple touch points that originate on different elements are removed 
from the touch surface simultaneously.

To have at least two passing implementations, we would need either the 
clientX/Y assertions fixed in the test (I remember Rick said it could have 
been an issue with the test itself and not the browsers), or the 
identifiedTouch and createTouchList issues fixed in WebKit.

Comments, flames, suggestions are welcome.

Regards, Cathy.