Sometimes edge seems not respecting `enclose`

[FlandiaYingman]

Please refer to the following image:

At (7, 0), (7, 3) and (12, 0), the edges do not respect the enclose property and directly go through the node.

This happens if the node spans over 3 or more distance units, only the - 2 units in the center will be regarded as node.

Source Code:

  diagram(
    node-stroke: 1pt,
    node-outset: 1mm,
    spacing: (5mm, 5mm,),
    debug: 100,
    {
      node((0, 0), [PC])

      node((2, 0), [Inst. Mem.])

      node([MUX], enclose: ((5, 2), (5, 3)))

      node([Reg. File], enclose: ((7, 0), (7, 1), (7, 2), (7, 3)))
      node((7, 4), [Sign Ext.])

      node([MUX], enclose: ((9, 1), (9, 2)))

      node([ALU], enclose: ((10, 0), (10, 1)))

      node([Data Mem.], enclose: (((12, 0), (12, 1), (12, 2), (12, 3))))
      node([MUX], enclose: ((12, 5), (12, 6)))

      edge((0, 0), "rr", "->")

      edge((2, 0), "r", "->")

      edge((3, 0), "rtrbrr", "->")
      edge((3, 0), "b,rrrr", "->")
      edge((3, 0), "bb,rr", "->")
      edge((3, 0), "bbb,rr", "->")
      edge((3, 0), "bbbb,rrrr", "->")

      edge((5, 2), "rr", "->")

      edge((7, 0), "rrr", "->")
      edge((7, 1), "rr", "->")
      edge((7, 4), "r,tt,r", "->")

      edge((9, 1), "r", "->")

      edge((10, 0), "rr", "->")

      edge((11, 0), "bbbb,rr,b,l", "->")
      edge((12, 0), "rr,bbbbbb,ll", "->")
      edge((12, 5), "llllll,tt,r", "->")
    }
  )

I am using the latest version of fletcher on the main branch.

[FlandiaYingman]

After reading the manual again, seems that the edge only respect pos but not enclose? Now I am not sure if it is a feature or bug.

[Jollywatt]

You can fix this by setting the snap-to argument of edge().

Usually, edges automatically snap to the nearest node, as long as it is within one unit. However, nodes which enclose many points are hard, because what if there are smaller nodes inside? Should the edge snap to the inner node or larger node? I couldn't decide, so instead added the snap-to argument as a manual override.

You can specify nodes in snap-to by giving the target node a name. In the example below, I gave the "Reg. File" a node name and set snap-to for all the edges that didn't snap automatically.

Instead of names, you can also give enclosing nodes a "position", which doesn't control the node's position, but still affects how edges snap to it. Notice for the "Data Mem." node I added (12, 0) as its center, so the two edges snap to it automatically.

Example fix

```typc diagram( node-stroke: 1pt, node-outset: 1mm, spacing: (5mm, 5mm,), debug: 100, { node((0, 0), [PC]) node((2, 0), [Inst. Mem.]) node([MUX], enclose: ((5, 2), (5, 3))) node([Reg. File], enclose: ((7, 0), (7, 1), (7, 2), (7, 3)), name: ) node((7, 4), [Sign Ext.]) node([MUX], enclose: ((9, 1), (9, 2))) node([ALU], enclose: ((10, 0), (10, 1))) node((12, 0), [Data Mem.], enclose: (((12, 0), (12, 1), (12, 2), (12, 3)))) node([MUX], enclose: ((12, 5), (12, 6))) edge((0, 0), "rr", "->") edge((2, 0), "r", "->") edge((3, 0), "rtrbrr", "->", snap-to: (auto, )) edge((3, 0), "b,rrrr", "->") edge((3, 0), "bb,rr", "->") edge((3, 0), "bbb,rr", "->") edge((3, 0), "bbbb,rrrr", "->") edge((5, 2), "rr", "->") edge((7, 0), "rrr", "->", snap-to: (, auto)) edge((7, 1), "rr", "->") edge((7, 4), "r,tt,r", "->") edge((9, 1), "r", "->") edge((10, 0), "rr", "->") edge((11, 0), "bbbb,rr,b,l", "->") edge((12, 0), "rr,bbbbbb,ll", "->") edge((12, 5), "llllll,tt,r", "->", snap-to: (auto, )) } ) ```

[FlandiaYingman]

That's great and it works!

IMO, the default snap-to for edges could be the outermost node, because, if there is a node that encloses many points, then:

• There will always be the outermost node (the node that encloses multiple points).
• But there may be, and may not be smaller nodes inside.
• Having a default behavior to handle the common case is better.

and,

• Having edges passing through the node is worse than not snapping to the inner node (in terms of looking).
• If users see an edge passing through a node, they might think that it is a bug.
• If users see an edge not snapping to the correct node, they might try to find if there is an option to control the behavior.

[Jollywatt]

Yeah, that's a good point. Maybe snapping to the wrong node is better than not snapping at all.