Description Gábor Lipták 2014-10-17 10:11:17 EDT

We have a table, where the user can drag and drop columns to order them. We noticed, that if the user does it really quickly, wrong column gets dropped. After examining the JSON requests coming from the browser in case of a D&D we realized, that the DragStart and the MouseClick event has different x y coordinates, which should not be the case.

A small demo code will be on monday attached, with which it will be easily reproducable. Click on some column and IMMEDIATELY move the item really fast. 

After reading the rap-client.js I have found the bug, and I have a probably not that nice, but working solution.

So in org.eclipse.rap\bundles\org.eclipse.rap.rwt\js\rwt\remote\DNDSupport.js you find this method:

    _dragStartHandler : function( event ) {
      var wm = rwt.remote.WidgetManager.getInstance();
      var target = event.getCurrentTarget();
      var control = wm.findControl( event.getTarget() );
      if( control == target && !this._blockDrag ) {
        var hash = target.toHashCode();
        var dataTypes = this._dragSources[ hash ].dataTypes;
        if( dataTypes.length > 0 ) {
          for( var i = 0; i < dataTypes.length; i++ ) {
            event.addData( dataTypes[ i ], true );
          }
          this._actionOverwrite = null;
          this._currentDragSource = target;
          var dndHandler = rwt.event.DragAndDropHandler.getInstance();
          dndHandler.clearActions();
          var doc = rwt.widgets.base.ClientDocument.getInstance();
          doc.addEventListener( "mouseover", this._onMouseOver, this );
          doc.addEventListener( "keydown", this._onKeyEvent, this );
          doc.addEventListener( "keyup", this._onKeyEvent, this );
          this.setCurrentTargetWidget( event.getOriginalTarget() );
          // fix for bug 296348
          var widgetUtil = rwt.widgets.util.WidgetUtil;
          widgetUtil._fakeMouseEvent( this._currentTargetWidget, "mouseout" );
          var sourceWidget = dndHandler.__dragCache.sourceWidget;
          var feedbackWidget = this._getFeedbackWidget( control, sourceWidget );
          // Note: Unlike SWT, the feedbackWidget can not be rendered behind
          // the cursor, i.e. with a negative offset, as the widget would
          // get all the mouse-events instead of a potential drop-target.
          dndHandler.setFeedbackWidget( feedbackWidget, 10, 20 );
          event.startDrag();
          event.stopPropagation();
        }
        this._sendDragSourceEvent( target, "DragStart", event.getMouseEvent() );
      }
    },

This calls _sendDragSourceEvent:
    _sendDragSourceEvent : function( widget, type, qxDomEvent ) {
      var x = 0;
      var y = 0;
      if( qxDomEvent instanceof rwt.event.MouseEvent ) {
        x = qxDomEvent.getPageX();
        y = qxDomEvent.getPageY();
      }
      var parameters = {
        "x" : x,
        "y" : y,
        "time" : rwt.remote.EventUtil.eventTimestamp()
      };
      rwt.remote.Connection.getInstance().getRemoteObject( widget ).notify( type, parameters );
    },

When I debugged _sendDragSourceEvent I have seen, that the "dragstart" event is sent with the coordinates of the javascript mousemove event. I have just seen that some things are saved into the DragAndDropHandler on mouse click, so I came up with this fix below (ugly a bit, but works as expected):

_sendDragSourceEvent : function( widget, type, qxDomEvent ) {
      var x = 0;
      var y = 0;
      if( qxDomEvent instanceof rwt.event.MouseEvent ) {
		if ( type=="DragStart" ) {
			x = rwt.event.DragAndDropHandler.getInstance().__dragCache.pageX;
			y = rwt.event.DragAndDropHandler.getInstance().__dragCache.pageY;
		}
		else {
			x = qxDomEvent.getPageX();
			y = qxDomEvent.getPageY();
		}
      }
      var parameters = {
        "x" : x,
        "y" : y,
        "time" : rwt.remote.EventUtil.eventTimestamp()
      };
      rwt.remote.Connection.getInstance().getRemoteObject( widget ).notify( type, parameters );
    }

As a bonus I will soon fire a gerrit commit with the applied patch on the master. As far as I see master is also affected.

Since I need to create a fixed 2.2 version for myself it would be really good to hear your opinion if it is a correct fix or not. It seems to work, but your feedback is important for me.