We currently craft tests in html to test possible bugs in the display-list and layer-building code, but it strikes me that this is easily broken. The make-up of the display-list is affected in several different areas in different ways, and a test is very likely to break (i.e. stop testing what it was intended to test) when something semi-unrelated is changed. It might be a good idea to be able to inject display-lists somehow and to add reftest and/or mochitest style testing infrastructure around this to more directly test display-list features. Similarly, a way of retrieving the display-list may make it easier to test display-list building. Is this a reasonable idea? Or just nuts? Any thoughts on how to accomplish something like this?
That would mean we'd create tests that would have to be updated when we make changes to the way display lists are constructed. That may not be a good idea.