Link layout arrange: smart links

[GregoBalu]

The link arrangement is currently not manageable by the user (despite priority() and north(), east(), ...). We can either come up with new statements that could help this (for example: link L(A->B) should go on the right side of C, if column(A, C, B) is the layout) or pump up the auto-part of the algorithm. The later would include something like this:

• The surrounding area of a box is discouraged for a link to navigate through.
  • Pros:
  • a link will more likely be in the "middle" of a corridor between boxes.
  • a link will go in a "more clean" space.
  • because it's just a weight, if a other route would be much more expensive, this area is still an option for the algorithm, just not as cheap as before.
  • Cons:
  • what should be this "deadzone" area. A percent of the box's width/height? What percent? Another "random" constant.
  • what will be the discouragement? A new weight during A* search. Another "random" weight constant.
• The surrounding area of a link is discouraged for a link to navigate through.
  • Pros: (basically the same as for boxes)
  • links wouldn't get too close to each other.
  • Cons: (basically the same as for boxes)
• Giving weights to start and end boxes' border points according to how far are them from the center point of that side.
  • Pros:
  • Links tend to be drawn from/to the top-ish or bottom-ish side of a vertical box side, this would be discouraged.
  • Cons:
  • Links will be bunched up at the middle of the box side they are connected to.

[devaigergely81]

I thought that links are only laid out in the "corridor area" and not inside the cells of the boxes (even if a given box is smaller than its cell). Is there any major disadvantage of this clean rule?

[GregoBalu]

It was proposed back sometime (by you ;) ) that we should be able to handle "different sized boxes" by using same sized cells and smaller boxes inside. As a side effect (this was the main advantage) lines that would otherwise get around a box between the two other one (with 2 turns), it could go straight through the cell of the small box beside that one.

But this "discouraged" zone would also affect boxes that are as big as their cell, so the actual "corridor area" would be narrowed.

[devaigergely81]

If links go only in the corridors, then the minimal distance of a link and an independent box will be at least one unit in the grid used for laying out links. Now, if we want to allow links to enter into the cells of small boxes, we have to do it similarly: grid points closer to the box than a unit of the grid are not allowed. This solution guarantees to avoid bug #242, does not produce unpredictable results (like "random" constants given to an A* algorithm) and is relatively simple. So my current suggestion is basically the "discouraged area" proposal above, but with "forbidden" instead of "discouraged" and with "area" = 1 unit.

I think there is no need to do the same for links, since the grid guarantees at least 1 unit between them.

[GregoBalu]

Plus: long awaited "short" reflexive links become "longer" - visually more appealing. (Bigger U shapes)

[GregoBalu]

Updated original post with the idea about giving points in the center of a box's side less cost.

[devaigergely81]

I suggest to put this on hold for a while for two reasons:

• There are much higher priority tasks: state machine diagrams, composite structure diagrams and different hierarchical diagrams.
• Yesterday I deep dived in the layouting codebase and found it really hard to understand and bugfix. A major refactoring or redesign and extensive documentation is needed.

[GregoBalu]

specify