Dev Pipeline.get_openalea_dataflow

[julien-diener]

Implements a conversion of rhizoscan Pipeline to OpenAlea.core.dataflow.Dataflow

to start : need a coarse description of

• rhizoscan pipeline internal representation
• how to construct a OA Dataflow @pradal

[julien-diener]

Rhizoscan pipeline:

Basically, a pipeline is a list of rhizoscan.workflow.node (or simply a lsit of python functions) which is stored in the __pipeline__ attribute
Each node/function, has inputs and outputs which can be obtained using node.get_inputs and node.get_outputs
Also, rhizoscan.workflow.openalea.node_attribute return a openalea valid descirptor dict

[pradal]

All the work is done.
Is your pipeline linear ? (ie is your pipeline really a pipeline?)
And how do you connect in/out ports? By name?

[julien-diener]

It is a real pipeline (but the Pipeline class is not formal, one could probably cheat). Basically, a pipeline is a sequence of function, with named inputs and outputs, that is executed in a common namespace. The connection (i.e. dataflow edges) can thus be computed by finding the outputs that have the same name as some later function's inputs, or the opposite. Something like:

pl = my_pipeline.__pipeline__
for x,fct1 in enumerate(pl):
    for i,input in enumerate(node.get_inputs(fct1)):
        for y in range(i-1,0,-1):
            fct2 = pl[y]
            for j,output in enumerate(node.get_outputs(fct2)):
                if input{'name'}==output{'name'}
                    create_and_edge(fct2, fct1, k,i)  # todo
                    break

Well, we can dev a much better algorithm for this, but you see my point. So if you give me the Dataflow api to add fct and edges, I'll be able to code a conversion. We could also convert the other way around, knowing a dataflow partial order (but I don't see the use of it).