Running the testcase in bug 497788, I see us spending a lot of time (30% or so) in painting code. The interesting thing s that we seem to enter it through two different callstacks, each accounting for about half the painting time. When I put printfs into nsViewManager::Refresh and the testcase code, I see us often doing refresh twice per single testcase iteration. The relevant stacks for getting to __CFRunLoopDoObservers: start _start main XRE_main nsAppStartup::Run() nsAppShell::Run() -[NSApplication run] -[NSApplication nextEventMatchingMask:untilDate:inMode:dequeue:] _DPSNextEvent BlockUntilNextEventMatchingListInMode ReceiveNextEventCommon RunCurrentEventLoopInMode CFRunLoopRunInMode CFRunLoopRunSpecific nsAppShell::ProcessGeckoEvents(void*) nsBaseAppShell::NativeEventCallback() NS_ProcessPendingEvents_P(nsIThread*, unsigned int) nsThread::ProcessNextEvent(int, int*) nsAppShell::OnProcessNextEvent(nsIThreadInternal*, int, unsigned int) nsBaseAppShell::OnProcessNextEvent(nsIThreadInternal*, int, unsigned int) nsAppShell::ProcessNextNativeEvent(int) -[NSApplication nextEventMatchingMask:untilDate:inMode:dequeue:] _DPSNextEvent BlockUntilNextEventMatchingListInMode ReceiveNextEventCommon RunCurrentEventLoopInMode CFRunLoopRunInMode CFRunLoopRunSpecific __CFRunLoopDoObservers and start _start main XRE_main nsAppStartup::Run() nsAppShell::Run() -[NSApplication run] -[NSApplication nextEventMatchingMask:untilDate:inMode:dequeue:] _DPSNextEvent BlockUntilNextEventMatchingListInMode ReceiveNextEventCommon RunCurrentEventLoopInMode CFRunLoopRunInMode CFRunLoopRunSpecific __CFRunLoopDoObservers and from there the stack to nsViewManager::Refresh is identical: _handleWindowNeedsDisplay -[NSWindow displayIfNeeded] -[NSView displayIfNeeded] -[NSView _displayRectIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:] -[NSThemeFrame _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _recursiveDisplayRectIfNeededIgnoringOpacity:isVisibleRect:rectIsVisibleRectForView:topView:] -[NSView _drawRect:clip:] -[ChildView drawRect:] nsChildView::DispatchWindowEvent(nsGUIEvent&) nsChildView::DispatchEvent(nsGUIEvent*, nsEventStatus&) HandleEvent(nsGUIEvent*) nsViewManager::DispatchEvent(nsGUIEvent*, nsEventStatus*) nsViewManager::Refresh(nsView*, nsIRenderingContext*, nsIRegion*, unsigned int)

I'm not too worried about the different callstacks, as noted in bug 516732 comment 7 / 8, but this is intriguing: (In reply to comment #0) > When I > put printfs into nsViewManager::Refresh and the testcase code, I see us often > doing refresh twice per single testcase iteration. Could you test this again with the patch for bug 517804 included?