There are two issues that are causing our nightly unit tests (FULL_TEST=true) to fail intermittently on some machines:
1) We don't have a reliable way to know when rendering is done, so we use a wait for "N" frames, and the chosen value of 3 frames for incremental updates to the scene graph is too small in some cases (I plan to bump it to 5 for now) and causes intermittent failures in helloworld.RectangleTest#testAddFillRect.
2) The 3% color tolerance is too small on some Mac systems. There is a JIRA to track this, RT-32948, but until we know why I will bump the tolerance to 7% which is more than enough for now. And given that we currently use the visual test framework only to sample high contrast colors, there is no problem with a higher tolerance.
In addition to the above, I will change the assertion failure message when there is a failure to print out the entire color (R,G,B,A) and not just the failing component (with no indication as to which component it was), as indicated in RT-32948.