edges should be clipped to actual outline of node

[gordonwoodhull]

I guess this is difficult because we need to know the exact SVG definition of stroke width, miters and bevels.

But I think we can do better than what we are doing right now, which is

1. intersect with the exact geometric shape, as if it were drawn with 0px stroke width
   

   dc.graph.js/src/shape.js

   Lines 713 to 715 in 96496f2

   	intersect_vec: function(n, deltaX, deltaY) {
   	return point_on_polygon(n.dcg_points, 0, 0, deltaX, deltaY);
   	},

2. add half the node stroke width to the arrow length when chopping the splines
   

   dc.graph.js/src/diagram.js

   Lines 2108 to 2113 in 5536eac

   	var headLength = arrowSize *
   	(arrow_length(arrow_parts(_arrows, _diagram.edgeArrowhead.eval(e)), stemWidth) +
   	_diagram.nodeStrokeWidth.eval(e.target) / 2),
   	tailLength = arrowSize *
   	(arrow_length(arrow_parts(_arrows, _diagram.edgeArrowtail.eval(e)), stemWidth) +
   	_diagram.nodeStrokeWidth.eval(e.source) / 2);

[gordonwoodhull]

There's no way to get this right

[gordonwoodhull]

But we could get these right, or better anyway

[gordonwoodhull]

Oddly enough, we do take node stroke width into account when placing ports for parallel edges (but only parallel), via source_padding and target_padding below:

dc.graph.js/src/shape.js

Lines 525 to 541 in 96496f2

	sp = bss.port;
	tp = bst.port;
	var sdist = Math.hypot(sp.x, sp.y),
	tdist = Math.hypot(tp.x, tp.y),
	c1dist = sdist+source_padding/2,
	c2dist = tdist+target_padding/2;
	var c1X = sx + c1dist * Math.cos(bss.ang),
	c1Y = sy + c1dist * Math.sin(bss.ang),
	c2X = tx + c2dist * Math.cos(bst.ang),
	c2Y = ty + c2dist * Math.sin(bst.ang);
	points = [
	{x: sx + sp.x, y: sy + sp.y},
	{x: c1X, y: c1Y},
	{x: c2X, y: c2Y},
	{x: tx + tp.x, y: ty + tp.y}
	];
	bezDegree = 3;

But how does this make any sense at all? Looks like the radius is divided twice, how come this mostly works?

dc.graph.js/src/diagram.js

Lines 2088 to 2091 in 96496f2

	var source_padding = source.dcg_ry +
	_diagram.nodeStrokeWidth.eval(source) / 2,
	target_padding = target.dcg_ry +
	_diagram.nodeStrokeWidth.eval(target) / 2;

[gordonwoodhull]

I'm not sure that the parallel edge code is using node stroke width effectively, because the same spline-chopping and marker-offsetting happens for parallel edges and they look just about as embedded as the other arrows

It's really unclear to me what that code does.