This will allow us to remove the Qt-specific optimization in CanvasRenderingContext2D::stroke().
Created attachment 104989 [details] Proposed patch
Comment on attachment 104989 [details] Proposed patch Clearing r? while discussing this with some random Belgian person.
Looking at other ports, and clients of Path::boundingRect(), it turns out that it's not necessary to return a 100% accurate bounding rectangle, since it's only used for dirty region computation and short-circuiting the SVG path-based elements' hit testing. Changing the scope of the bug to always use QPainterPath::controlPointRect() in Qt's Path::boundingRect() since this is significantly faster and matches the current behavior of (at least) Skia's Path.
Created attachment 105159 [details] Proposed patch
Comment on attachment 105159 [details] Proposed patch Great
Comment on attachment 105159 [details] Proposed patch Clearing flags on attachment: 105159 Committed r93774: <http://trac.webkit.org/changeset/93774>
All reviewed patches have been landed. Closing bug.
Reopen, because it was rolled out by https://trac.webkit.org/changeset/93790 This patch broke zillion tests on the Qt bot. You guys should have run layout tests before landing. Please fix the bug before re-landing.
(In reply to comment #8) > Reopen, because it was rolled out by https://trac.webkit.org/changeset/93790 > > This patch broke zillion tests on the Qt bot. You guys should have run > layout tests before landing. Please fix the bug before re-landing. zillion != 26 :)
(In reply to comment #9) > zillion != 26 :) You're right, but 26 !=0, and all test must pass.
(In reply to comment #10) > (In reply to comment #9) > > zillion != 26 :) > > You're right, but 26 !=0, and all test must pass. Like I said on IRC, those failures are actually progressions and should just be updated. I couldn't do it at the time because I didn't have a checkout. I'll roll it back in along with a rebaseline. Thanks for playing another round of the mindless rollout game :)
I don't know anything about your patch... I tried to ping you on #qtwebkit, but you didn't answer. I had to do it, because red bot can't catch new regressions ... That's why we need always green bots. That's why developers have to run layout tests before landing and not leave the tree red for hours ...
Committed r93873: <http://trac.webkit.org/changeset/93873>