The points in the model at http://www.cip.ifi.lmu.de/~schubert/kovar/ do not appear in ToT (r28986). The site works fine in Safari 3.0.4.
This seems to be the reduction: <svg xmlns="http://www.w3.org/2000/svg"> <circle transform="scale(1000)" r="0.1" opacity=".99"/> </svg> (thanks Sam!)
Created attachment 18120 [details] reduction
Created attachment 18121 [details] drt-compatible test case
bisect-builds says: Works: r26359 Fails: r26570
My suspicion is that some calculation changed to use ints instead of floats which should not have.
Is the result of the test supposed to be a green square? If so, this looks like it works correctly, both in Safari 4.03, and the latest nightly (r48518). The original bug url also seems to work, in that the points in the model do appear. I'm not able to determine when this progressed, as the oldest nightly build that I can test on 10.6 seems to work. Would be nice if there were a test case added (if there isn't already), but I think this can be closed.
Created attachment 93591 [details] Patch
(In reply to comment #7) > Created an attachment (id=93591) [details] > Patch See the ChangeLog, the bug is fixed but I think it may be a good idea to keep Eric's reduction test around. Cheers, Rob.
Comment on attachment 93591 [details] Patch OK.
Committed r86514: <http://trac.webkit.org/changeset/86514>
Is the expected result really platform specific? It's just a green square.
FYI, this problem affects QtWebKit with current trunk. Not sure if it has the same root cause, but the "symptoms" are the same. Opened bug 60968 to track it.