Tested on the latest FirefoxNightly on Mac. STR: 1. Go to the attached test case and hit run() Expected: 1. The page prints "Start ... Finished", or maybe, 2. If the process takes too long, the runaway timer should kick in and allow user to abort the script. Actual: 1a. The browser hangs (seemly) indefinitely, and if e10s is enabled, 1b. The tab hangs. Note: 1. The test case creates a very large canvas, draw some text on it, and call getImageData(). 2. It was first reported by :lafncow on https://github.com/timdream/wordcloud2.js/issues/14, and we traceback to the browser bug here. The test case is derived from the library. 3. On Chrome the tab will crash right away, arguably a "better" experience :P. Setting "critical" because the bug will cause user lost all the tabs and only make users to regain them through session restore.

This bug affects Shumway, too. cwahler created a jsperf test demonstrating the performance problem: http://jsperf.com/cwahlers-getimagedata/2

This perf problem is not e10s-related.

In the benchmark, the elapsed time for getting/putting a fixed 100x100 bitmap increases as the canvas size increases. We expect that the cost of reading/writing fixed source/destination regions would be constant regardless of the canvas size.

GFx team requested we move the perf issue to bug 1001069 so we keep the crasher separate.