When the user selects the `Edge' command in the `Elements' menu, the tool displays an edge-addition cursor under the current mouse position on the dataflow canvas, as shown in Figure 7.

To create an edge, the user starts by moving the mouse to an edge starting position. This position must be on the boundary of a node, including on one of its output ports. Whenever the user moves the mouse to a valid edge starting position, the tool highlights the node by widening its boundary, as shown in Figure 8.

As the user moves the mouse, the tool tracks it with a line that extends from the edge starting position to the current mouse position. During edge drawing, the user is simply moving the mouse, with no button depressed. Figure 9 illustrates the user having dragged the edge cursor approximately halfway between starting and ending positions on the "Activate Autopilot" and "Check Autopilot" nodes.

Finish details of labeling; there are some notes below.

The user can create three types of edges, based on the canvas locations of the starting and ending positions. The three types are defined in the following table:

Start Position	End Position	Type of Edge
on a node	on a node	node-to-node connection
not on any node	on a node	external input
on a node	not on any node	external output

A node-to-node connection defines an input/output relationship between the two nodes. An external input defines an edge that must have its value provided by an external source, rather than coming from another node. Similarly, an external output defines an edge whose output goes to an external source, rather than to another node. Further details on external data sources are covered in ???.

User may start anywhere within or at the boundary of an node. If the edge- drawing starts within a node, then the starting point of the finished edge is where the edge line intersects the boundary of the node. End point is similar. If the intersection point crosses any part of a port icon, then the edge is considered to be connected to that port (see below for more on ports and edges).

Details of label cursor placement: Tool puts text cursor in center of edge, with center defined as the position on the edge equal to 1/2 edge length, as measured along the edge spline. The background of the text is the current bg color, which is white by default. The text color is black by default. The background size is ... . As the user types, the center of the label text remains positioned on center of the edge.

The same type of completing for edge names is provided as for node names. Viz., the names of existing objects are available. If the typed text is not an existing object, than the dftool creates an opaque object of that name and adds it to the underlying fmsl spec.

The following forms of edges are illegal:

1. both starting and ending position are not on a node
2. starting and ending positions forming an edge that crosses any part of a node

It is legal to connect a node to itself, i.e., have the input and output ends of an edge be the same node. To do so graphically, the user must use a multi- point edge.

2.3.1. Edges and Ports

Sketch:

Start Point	End Point	Behavior
off port	off port	new output port created on starting node, new input port created on ending node; type of ports is object named by edge name entered by the user
on port	off port	new input port created on ending node; edge name is auto-filled with type of output port
off port	on port	new output port created on starting node; edge name is auto-filled with type of input port
on port	on port	type of ports must be the same; if so, edge name is auto-filled with that type

The following roll-over behaviors assist with port connections:

• When Elements->Edge has been selected but the first click has not yet happened yet, the user can roll over any port to see its type, and whether it's an input or an output. If it's an output, it's name is displayed and it's colored green when rolled over. If it's an input, it's name shows up, but it's colored red. (Note: if we have arrowheads as port icons, then we probably don't need to do the coloring thing; rather, only have the roll-over naming appear for output ports.)
• Once the first click happens, the same roll-over behavior is exhibited. If the first click is on a port, then the second click can only happen on a non-port portion of a node, or an input port of the same type as the output port.

Roll-over display details:

• Labels in the boxes are formated as follows:
  • There is a value for the width of port label boxes, settable as an option; call this w.
  • There is a value for the margins, settable as an option; call this m.
  • There is a value for the font size of the port label, settable as an option.
  • If the length of the text string for the port type name in the current font is <= w-2m, then the text is displayed in one line.
  • If the length of the text string for the port type name in the current font is > w-2m, then the text is displayed in multiple lines, center justified, with line breaks inserted at case-shift boundaries, per the standard word-processing fill technique, for example emacs' justify-paragraph function.
  • In all cases, the box is shrunk-wrapped around the text, to the current margin setting.
  • The preceding rules imply that if after justification the length of the text string for the port name is > w-2m, shrink wrapping results in a box that is wider than w-2m.
  • There are values for the following box formating properties, settable as options:
    • label font, default black-plain-courier-12
    • box background color, default yellow
    • box border color, default black
• Label boxes are placed as follow:
  • One of the four corners of the box is chosen as the placement corner, per the following table:
    
    

    cartesian quadrant of port connection	box placement corner
    1	lower left
    2	upper left
    3	upper right
    4	lower right

    
    
    
  • for input ports, the placement corner is placed in the center of the base of the port arrow
  • for output ports, the placement corner is placed at the tip of the port arrow

During the mouse drag operation for an edge, the user may press the right mouse button to add graphic control points to the edge. A graphic control point allows edges to be shaped as curves, to aid in the visual appearance of the diagram. Formally, edges are drawn as b-spline curves, with each control point interpreted according to the standard definition of a b-spline.

Figure 11shows the result of the user drawing an edge by left clicking on the boundary of the source node, right clicking at a middle point while drawing, then left clicking on the boundary of the destination node.

Sketch:

• Select the edge spline to edit it
• Spline handles appear when it's selected, on top of label when necessary
• Label handles do not appear when spline is selected
• Use Format->Hide Label when necessary to clearly see or select a spline when the label is over most of it; an option line auto-hide-edge-label-on-selection might be useful, or auto-grey-edge-label-on-selection.

When the user selects an edge, the dftool draws small "handles" on and around the line or curve for that edge. The handles are control points that the user can drag to reshape the edge.

If an edge is a single straight line, then there are two control points, one on each end of the line. If an edge is a curve, then in addition to the end control points, there are one or more additional control points that defined the curvature of the edge.

The user can click and drag on any edge handle to change its position on the screen, and thereby change the shape of the edge. When the user drags an edge end point the is connected to a node, the movement of the point is constrained to the circumference of the node to which it is attached. When user drags an edge control point that is not connected to a node, it may be moved to any location on the canvas except within a node.

To add or delete control points to an edge, the user selects the `Add Edge Point' and Delete Edge Point' commands in the `Edit' menu. Short cuts for these commands are to hold down shift and move the most onto an edge or one of its handles.

To add a new control point, the user selects `Add Edge Point' (or holds the shift key) and moves the mouse to a place on an edge that does not have control-point handle. In response, dftool augments the mouse pointing cursor with a small plus sign, indicating that the user can click at that point to add a new control point. To remove an existing control point, the user selects `Delete Edge Point' (or holds the shift key) and moves the mouse to the control to be deleted. In response, dftool augments the cursor with a small minus sign, indicating that the user can click to delete that control point. The three states of the mouse cursor adding or deleting an edge point are illustrated in Figure 12.

Notes on label editing, to be more fully elaborated:

• Select edge text to edit or move it.
• When moved, it still belongs to the edge, and it (somehow) stays stuck in some controlled way when the edge spline is reshaped.
• When text is changed, it redefines the types of both connected ports; if this invalidates any other use of the node (operation in the spec), then such invalidation shows up when the spec is checked. It might be nice to have a warning come up when this is possible, but it may be difficult to do it.