Edges through cluster

Hi Folks!

Does anybody know, what to do to prevent that edges are going through clusters/subgraphs?
(p.ex. the edge from BoxenModul1 to BildverarbeitungsModul goes right through BoxenModul2)
Hier you can find the the source code:
http://nopaste.info/66ce58b90b.html
And here is the output file:
http://www.team-spatzenhirn.de/frei/plan.png

And the other thing is: Are there any possibilities to have an impact on the position of the cluster?
My aim is to site the clusters "BildverarbeitungsDatenInterpreterModul", "TrajektorienplanerModul", "QuerreglerModul" and "LaengsreglerModul"
like you can see in the attachement "plan_old.png" (which was made with open office ;-))
Plan_old.png:
http://www.team-spatzenhirn.de/frei/plan_old.png

Currently I'm workin on GVEdit for Graphviz, Version 3.6.8

Thanks for your answer

Marcel

Edges through cluster

[This query and the following reply also appear on the graphviz mailing list.]

> Does anybody know, what to do to prevent that edges are going through
> clusters/subgraphs? In dot, clusters are not fully first-class objects and edges go
monotonically from one rank to another,
so there is no support for this. You get the effect partially for long
edges, but if an edge goes between
nodes in adjacent ranks, it will cut through any intervening cluster.

There is code for this in fdp, but it would take some work to integrate
it into dot.

> And the other thing is: Are there any possibilities to have an impact on the
> position of the cluster?
Yes, and fixing this will largely avoid the previous problem. You can't
control the clusters directly, but
you can control the nodes they contain. This is done by the edges and
attributes such as minlen and weight.
If the graph has cycles, dot needs to break these. This is done by a
depth-first search, so the input order of
nodes and edges have an implicit effect on this. You can avoid this
problem by avoiding cycles: reverse the
edges yourself and use dir=back.

Such changes can get you most of what you want. However, the old rank
assignment algorithm in dot tends to
pack clusters more tightly than you might like. If you install the
current working version, configure --with-cgraph and
run with dot -Gnewrank, you get the attached drawing from
this graph.

Recent comments