Split and eval

Pursuant to my action A-213-14, here's a brief description of the split and
eval steps.

Consider the eval step: its goal is to evaluate an arbitrary pipeline
supplied on the input port. The number of inputs and outputs that that
step needs is dependent on the number of inputs and outputs that the
supplied pipeline has.

Here's an example:

  <p:eval>
    <p:input port="pipeline">
      <p:inline>
        <p:pipeline>...</p:pipeline>
      </p:inline>
    </p:input>
  </p:eval>

Ideally, p:eval has source and result ports.

  <p:eval>
    <p:input port="pipeline">
      <p:inline>
        <p:declare-step>
          <p:input port="source"/>
          <p:input port="stylesheet"/>
          <p:output port="result"/>
      </p:inline>
    </p:input>
  </p:eval>

Ideally, this invocation has source, stylesheet, and result ports.

I think that's impractical because port names have to be known at
"compile time" and the pipeline isn't even available necessarily until
"runtime".

But you could say that p:eval has an unbounded number of input and
output ports, named, perhaps "source1", "source2", ..., "result1",
"result2", etc.

Then the semantics of p:eval could be that it maps source1 to the
first (in document order) source pipe in the pipeline, etc. It'd be a
bit ugly to use, but it would be prettier than the current alternative
of wrapping all the inputs up into a single document.

The split step would have a single source port and an unbounded number
of result ports.

In my V2 implementation, I rewrite the pipeline graph so that there is
exactly one step reading from any pipe. I introduce a split step to
achieve this, but now I have a pipeline graph that cannot be
serialized which is a little disappointing.

                                        Be seeing you,
                                          norm

-- 
Norman Walsh
Lead Engineer
MarkLogic Corporation
Phone: +1 512 761 6676
www.marklogic.com