Review Request: Rewrite kinetic scrolling on ScrollWidget

[Zack Rusin]

> On 2010-03-17 12:31:22, Marco Martin wrote:
> > just by a quick test, the physics seems to work indeed better, it just still has some quirks:
> > 
> > - scrollbars aren't syncronized when dragging (yeah i know it needs a mode without scrollbars but i would keep them on the desktop)
> > 
> > - event stealing doesn't seem to work (even with setFiltersChildEvents): if registerasdraghandle is used clicks don't pass anymore to childrens, if it's not used is the parent to not receive events)
> > 
> > - the old behaviour of scrolwidget was to center the inner widget into the parent if it's smaller, now seems to always go to 0,0
> > 
> > 
> > about scrollStateChanged uhm.. permits external things to know the scrlling state, to for instance show/hide things depending if the user is dragging... not terribly useful indeed..
> 
> Zack Rusin wrote:
>     The scrollbars issue is fixed.
>     The event stealing works for sure, since that was the first thing I got working. I don't really understand what draghandle's are supposed to be doing or why there are event filters for them so I'm not sure what exactly is expected to happen there. Is there an example that uses that code so that I can figure out what's the expected behavior for those?
>     I don't see the centering behavior with the old code (especially since the setWidget has been calling an unconditional setPos(0,0)) , is there an example that shows the behavior you mention?
> 
> Marco Martin wrote:
>     in plasma some places that are using the scrollwidget and have lots of child items (that are supposed to manage clicks too) is the microblog plasmoid or the netbook search and launch interface (just plasmoidviewer sal) where all the icons can be clicked or it should be possible to drag the view around also by draging the icons, here the need of stealing  events when needed.
>     before it was working wit the registerasdraghandle function: it installed an eventfilter and bounced the event back and forth between the child and the scroll widget.
>     this would hopefully become deprecated, since now it's using setFiltersChildEvents.
>     Icons in search and launch had this eventfilters in place, by killing the implementation of registerasdraghandle we are sure there shouldn't eventfilters around that do strange things, but it's still impossible to scroll by dragging over the icons.
>     
>     about the centering thing, always with search and launch, usually when it starts the contents of the scrollwidget are just some icons, so  content it's smaller than the scrollwidget and is centered.
>     If the inner widget was placed centered by hand, it stayed there, it simply couldn't scroll horizontally, so no problems. now as soon as is touched by the mouse it animates and it goes to 0,0.
>     I know is not a prety thing to move the inner widget by code, but probably the scrollwidget will need a Qt::Alignment flag in the api to decide how to align the inner widget if it's smaller than the scroll widget...

Yes, the "lots of child items that accept clicks on top of a scrollview" is exactly the case that works here. 
Here "sal" seems to be working ok. BTW, setting the position of a widget managed by another widget from somewhere else is pretty evil =) I did add setAlignment code to the ScrollWidget, i think it makes a lot more sense than allowing others to position the items that it's supposed to be managing :)

With the patch underneath sal has items centered correctly and I can flick the view by click&dragging items without them executing. BTW, sal could probably actually use the scrollStateChanged signal to not call ensureItemVisible when another animation is in progress (when that happens ScrollView has to immediately stop the currently running animation and invoke another one, since otherwise we could have two different animations pulling in opposite directions, which causes a small jump on hover events during the flick animation)

Index: containments/sal/itemcontainer.cpp
===================================================================
--- containments/sal/itemcontainer.cpp  (revision 1104506)
+++ containments/sal/itemcontainer.cpp  (working copy)
@@ -150,7 +150,6 @@
         m_usedItems.remove(key, item);
     } else {
         item = new ResultWidget(this);
-        m_itemView->registerAsDragHandle(item);
         item->hide();
         item->setPos(boundingRect().center().x(), size().height());
     }
Index: containments/sal/itemview.cpp
===================================================================
--- containments/sal/itemview.cpp       (revision 1104506)
+++ containments/sal/itemview.cpp       (working copy)
@@ -32,6 +32,7 @@
     setFocusPolicy(Qt::StrongFocus);
     setSizePolicy(QSizePolicy::Expanding, QSizePolicy::Expanding);
     m_itemContainer = new ItemContainer(this);
+    setAlignment(Qt::AlignCenter);
     setWidget(m_itemContainer);
     m_noActivateTimer = new QTimer(this);
     m_noActivateTimer->setSingleShot(true);


- Zack


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://reviewboard.kde.org/r/3312/#review4532
-----------------------------------------------------------


On 2010-03-17 22:20:25, Zack Rusin wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviewboard.kde.org/r/3312/
> -----------------------------------------------------------
> 
> (Updated 2010-03-17 22:20:25)
> 
> 
> Review request for Plasma.
> 
> 
> Summary
> -------
> 
> As previously discussed with this approach we can actually properly intercept events from child items. Furthermore now we properly "steal" events which cause a flick (it's important for child items to properly act on mouseReleaseEvents and not on mousePressEvents as some like to do, since it's the release events that cause a flick) so items don't get clicks when flicked. The physics and especially the overshoot behavior is a lot better in this code as well.
> I tried to preserve the old behavior and emit the scrollStateChanged when needed, but I'm not quite sure what that signal was meant to be good for.
> 
> 
> Diffs
> -----
> 
>   trunk/KDE/kdelibs/plasma/widgets/scrollwidget.h 1102878 
>   trunk/KDE/kdelibs/plasma/widgets/scrollwidget.cpp 1102878 
> 
> Diff: http://reviewboard.kde.org/r/3312/diff
> 
> 
> Testing
> -------
> 
> Done in a custom app. Would be nice if someone double checked it with other things that use ScrollWidget though (e.g. the notebook shell :) ).
> 
> 
> Thanks,
> 
> Zack
> 
>