[graphviz-interest] Intended behaviour of "canon" output

What is the intended pretty-printing performed by "dot -Tcanon"?  My
understanding from how it seemed to work was that for each
graph/subgraph it would produce something like:

<graph/subgraph specification> {
    [possibly empty list of global attributes]
    [possibly empty list of sub-graphs]
    [possibly empty list of nodes]
    [possibly empty list of edges]
}

However, at least in 2.38.0 (the last time I checked this behaviour
was with 2.30 I think; I skipped a few versions when I realised my
installation of Graphviz was old) this is not the case for some
graphs: e.g. when applied to the undirected/ER.gv sample graph, whilst
each node seems to be specified before any edges its used in, the
nodes and edges are now mixed together.

Is this the correct/expected behaviour?  I have some code that assumes
it's in the old behaviour, so if this isn't the case I'll need to
change how I'm parsing the Dot code.

-- 
Ivan Lazar Miljenovic
[email protected]
http://IvanMiljenovic.wordpress.com
_______________________________________________
[email protected]
http://lists.research.att.com/mailman/listinfo/graphviz-interest