The topic was discussed in https://lists.webkit.org/pipermail/webkit-dev/2013-April/024766.html
if existing reftests assume that only a particular (e.g., 800x600) area is compared, then we should not change the default to compare the entire page; that is, comparing the entire page should be opt-in due to prior usage
(In reply to comment #1) > if existing reftests assume that only a particular (e.g., 800x600) area is compared, then we should not change the default to compare the entire page; that is, comparing the entire page should be opt-in due to prior usage We shouldn’t make such assumptions in reftests. If there are ref tests that make such a bad assumption, then we should fix them all.
(In reply to comment #2) > (In reply to comment #1) > > if existing reftests assume that only a particular (e.g., 800x600) area is compared, then we should not change the default to compare the entire page; that is, comparing the entire page should be opt-in due to prior usage > > We shouldn’t make such assumptions in reftests. If there are ref tests that make such a bad assumption, then we should fix them all. ok, in that case, the documentations about creating reftests [1] should make this explicit; authors may have assumed that 800x600 holds due to the language in [2] [1] http://trac.webkit.org/wiki/Writing%20Reftests [2] http://trac.webkit.org/wiki/Writing%20Layout%20Tests%20for%20DumpRenderTree#Howtowriteportablepixeltests