  More comments on the model transformation related thesis that I'm reading at the moment. The author talks about PSM to PIM transformation as a single-step process. I don't think this is feasible for large systems. There is always a need for intermediate forms, and it is over-simplifying the matter to not acknowledge MDA (in whatever definition one takes) as not consisting of a series (graph, in my opinion) of model of transformations. The author claims that the concept of notation has no equivalent in programming languages, and that the syntax is implied by the abstract syntax. I think this is false - all language have concrete syntaxes, obviously. The difference is that many languages have only one, or at least only one that permits editing. In his characterisation of model transformations, this is critical, since the difference between normal views (here notations, sometimes also representations) and views that permit editing, is a large and well-researched one. The author's characterisation of the UML physical metamodel as a re-working of the actual UML metamodel to fit into MOF terms is poorly presented. In fact, association classes present no difficulties for MOF, provide that you do not try to factor your model as an extension (by inheritance) of MOF.
He asserts that performing model transformation on objects still within a modelling tool (or repository, I assume, though its not said) avoids user-interaction problems of losing layout information. However, this is only true when the transformation involved is "in-place" transformation. In a paper I co-authored for ICGT 2002, we present a number of problems with in-place transformation in MDA, and argue that it is not the dominant case, and furthermore that it presents more problems in terms of termination and ordering, than it solves.
While it certainly useful for model refactoring, I don't think its suitable for the general problem of model transformation. I think this will make the rest of the thesis difficult for me to read. I have only read 35 pages from 200-odd, but these introductions are important, because misunderstandings of background material can be propagated to the actual contributions. 
