- From: Boris Zbarsky <bzbarsky@MIT.EDU>
- Date: Sun, 05 Sep 2010 22:25:36 -0400
- To: Bjoern Hoehrmann <derhoermi@gmx.net>
- CC: public-css-testsuite@w3.org
On 9/5/10 8:39 PM, Bjoern Hoehrmann wrote: > Sure, just like there is no gurantee that running the suit manually will > work as intended. However, if the condition is "no red" and red ends up > in the rendering for some reason other than failing the test, then the > test is broken Agreed. > and at low resolutions to make something appear red it's necessary to actually render red every now and then But it doesn't have to be #ff0000. > In doubt you can also check if there is a lot of "almost red" Yes, that's my point. > little effort is required to make something that should work well enough. We may have different definitions of "well enough"... If screenshotting will really do the right thing, then just doing reftests with an explicit reference seems like a much more robust approach, and not much more work. Possibly less, since creating references is a lot simpler than trying to write complex "red-detection" algorithms and parallelizes well. -Boris
Received on Monday, 6 September 2010 02:26:11 UTC