Review of the NMF solution

[azuendorf]

The NMF solution starts with some theoretical foundation on lenses. Luckily, I am somewhat familiar with these. Then the transformation rules are explained. While I still have difficulties to read these, I get the idea and it all looks reasonable, correct and comprehensive. The strength of the approach is, that you do not need to specify update operations explicitly. You just read the changes to the source model (i.e. to the samples) and some automatic update engine propagates the required changes. Cool stuff. This comes at the costs of some runtime overhead, of course. I find it especially great, that the debugging identified some parts of the system that have efficiency problems and thus the case will soon result in an improvement of the NMF approach. This is a great result for the TTC Workshop.

[arturboronat]

The solution succeeds at showing that a declarative approach can be used to synchronize the abstract and the operation models in an incremental way, simplifying the way changes would have needed to be considered explicitly in the reference solution.

Strengths of the approach are:

• Although the changes that need to be propagated backwards are simple, the bidirectionality of NMF means that they can also be considered implicitly.
• The Chunk/ChunkIndexed operations work as an aggregation mechanism to encode many-to-one synchronizations.
• The case has helped identify some performance bottlenecks in the incrementalization mechanism of NMF.