Martin
Westhead
Bob
McGrath
Steve
Hanson
Geoff
Judd
Two
items were discussed this week:
Most
of the meeting was spent on the first item. Included below are Bob’s
notes on the discussion. The main conclusion of the discussion is that the
approach seems like a very sensible direction. However it highlighted that we
need to start putting in effort on pinning down a couple of the outstanding
issues, context, conversions, XPath/Dpath.
Steve
undertook to start working on the integration of properties and convergence.
Martin is going to focus on the outstanding issues with capturing our
separators semantics. It was noted the Mike has been looking at bringing
context and scoping together.
Presents
a proposed approach
Comments:
1.
Efficiency is an issue, particularly if there is no way for the user to
know how the data is really laid out, which might lead to dramatically
inefficient logical models
a) this issue is not limited to DFDL
b) it would be nice to have a markup to tell how the data is laid out WRT
efficient access. It was argued that this is hard, and shouldn't be in
the core DFDL. But if such a language exists, then it would work well
with the proposed approach.
2.
We hope that clever techniques a la compiler optimizations will be able
to create efficient access methods. E.g., that drill through the logical
layers to eliminate multiple data movements.
a) want the model to make this optimization possible
3.
A third logical model was suggested, which is thought to be similar to BFD.
The model would be the 1D array plus a descriptor to say how to interpret it.
Essentially, you may not need or want the additional XML logical model at
all.
a) look up BFD and correlate with that approach
b) add this model to the notes
4.
The Array examples highlight the need to solve two general issues:
a) path expressions
b) transforms, WB and BB
5.
The 'occurs' is out of date, should read 'occursPath'.