Our current test infrastructure doesn't guard against changes in latching behavior. Extend the test system to cover this region of the system.
<rdar://problem/15911348>
Extend the test system EventSender with a new method to send wheel events (including momentum phase). Also create a new set of tests that use this feature.
Created attachment 224110 [details] Patch
The #defines for wheel event symbol were not included in CGEventTypes.h as API until 10.9. I'll just provide equivalent definitions for OS 10.8 and prior, which we can remove once we stop supporting that build.
On second thought, since I already use the ordinals in the WK2 handler (to avoid pulling in a bunch of Cocoa headers), I'll just use the same implementation in the WK1 handler and avoid the issue altogether.
Committed r164073: <http://trac.webkit.org/changeset/164073>
Re-opened since this is blocked by bug 128781
Created attachment 224139 [details] Patch
Created attachment 224151 [details] Patch
Created attachment 224152 [details] Patch
Committed r164105: <http://trac.webkit.org/changeset/164105>
This patch caused a fatal CQ failure because CQ bots had 10.9 SDK installed, which meant that kCGScrollWheelEventMomentumPhase was defined in headers. Used MAC_OS_X_VERSION_MAX_ALLOWED instead of work around this in http://trac.webkit.org/changeset/164159
(In reply to comment #12) > This patch caused a fatal CQ failure because CQ bots had 10.9 SDK installed, which meant that kCGScrollWheelEventMomentumPhase was defined in headers. > > Used MAC_OS_X_VERSION_MAX_ALLOWED instead of work around this in http://trac.webkit.org/changeset/164159 Thank you for tracking this down. Sorry for the inconvenience. :(
These tests fail on WebKit2 bots: http://build.webkit.org/results/Apple%20Mavericks%20Debug%20WK2%20(Tests)/r169047%20(4750)/results.html This is not reported correctly, because the tests fail on first try, and then can't find pixel results on retry. The reason why it tries pixel tests is that there is a call to testRunner.dumpAsText(true) instead of testRunner.dumpAsText(). "true" is a crazy option that means that we'll dump as text, but then make pixel results too.
Some of these fail because of zalan's offsetWidth changes.