Graphviz: How to render a graph respecting certain node-edge connection points?

Question

I want to render a graph in similar layout to the following one:

I tried this:

digraph EDP
{
  graph [colorscheme=paired12];
  node [label="\N", shape=box, style="rounded,filled", colorscheme=paired12, color=8, fillcolor=7, width="1.2", fontname="Arial narrow", fontsize=12];
  edge [colorscheme=paired12, color=8, fontsize=11, fontname="Arial narrow"];

  src [label="Source"];
  dst [label="Destination"];
  filter [label="Filter"];

  src -> dst [label="Encoding process"];
  src -> filter [label="a"];
  filter -> dst [label="b"];
  src -> filter [dir=back, label=c];
  filter -> dst [dir=back, label=d];
  src -> dst [dir=back, label="Decoding process"];

}

which generated the following result :( Not very nice...

Can anyone show me the closest possible solution? (Probably it is not possible make exactly the same)

Answer 1

As close as I got before calling it a night:

digraph EDP
{
  graph [colorscheme=paired12];

  node [label="\N", shape=box, style="rounded,filled", colorscheme=paired12, color=8, fillcolor=7, width="1.2", fontname="Arial narrow", fontsize=12];
  edge [colorscheme=paired12, color=8, fontsize=11, fontname="Arial narrow"];

  src [width=3.5, label="Source"];
  dst [width=3.5, label="Destination"];
  filter [label="Filter"];

  edge[constraint=false];
  src -> dst [label="Encoding\nprocess"];
  src -> filter [label="a"];
  filter -> dst [label="b"];
  dst -> filter [label="c"];
  filter -> src [label="d"];
  dst -> src [label="Decoding\nprocess"];

  edge[style=invis, constraint=true];
  src->filter->dst;

}

Save the graph as edp.gv and create the image with

dot -Gsplines=none edp.gv | neato -n -Gsplines=ortho -Tpng -o edp.png

Result:

Some remarks:

• I ended up putting constraint=false on all edges, and add invisible edges to have the three nodes centered
• Usually I like to use dir=back, but I wasn't able to use this when rendering with -Gsplines=ortho
• As you can see, the order of the edges is random...