16.07.2016, 18:39, "Darin Adler" <[email protected]>: >> On Jul 16, 2016, at 8:19 AM, Konstantin Tokarev <[email protected]> wrote: >> >> There is a number of layout tests which are expected not to have red pixels >> in case of pass. Some of them are purely pixel tests, which are not checked >> by bots (and not often by humans), others have render trees dumps, but there >> many cases when render tree dump is right but end result is not because of >> painting bug, and it's also easy to overlook one-pixel errors when doing >> rebaseline. >> >> Was it ever considered to do automatic red pixel search in png result? Such >> tests could be run on bots and increase regular coverage. > > My first choice would be to instead turn each of these tests into reference > tests, rather than creating a new type of test. > > If we instead find that there are tests where that is not easy or practical > to do, I’d be OK with creating a new type of test. > > Lets talk about some individual tests if you like.
For example, I'm looking at fast/canvas/shadow-offset-* now. They are pixel tests now. -- Regards, Konstantin _______________________________________________ webkit-dev mailing list [email protected] https://lists.webkit.org/mailman/listinfo/webkit-dev
