Feedback

I'm interested in the connection with dependency that you have identified here, though you'll need to consider the relationship between dependencies in the IceCube setting and in EM carefully. If you can achieve some of the objectives you set out to explore, this could be a most interesting outcome. Though I don't know enough about IceCube to judge the risk, the danger with your choice of project is that it may prove to be too difficult to assess the qualities of the algorithm because its performance will be so problem input-specific (think of all the possible interpretations of 'divergent replicas'). This might even be a problem from the 'let's educate the reader about IceCube' perspective. Undoubtedly - if you have the appetite for it - I think there would still be scope for an excellent paper on this sort of topic, with illustration using EM models, whatever the practical problems you encounter. (This would surely take you away from the 40:60 balance though.)

I'm fascinated by some aspects of this project that may lie beyond the foreground scope of your proposed study. An interesting angle on your project is that we signally failed to exploit the EM Jugs model as a portable model of a procedural Jugs program precisely because procedural Jugs programs could be so widely different (allowing for all the different platforms, and ways of optimising to function). (And when Richard Cartwright came to tackle the problem of cross-platform broadcasting at the BBC using EM principles - with specific reference to dependency - he proceeded by developing translations from a single Jugs-style model, rather than giving any attention to how the different implementations of the same content might be related 'as program code'.) Presumably, in the IceCube context, the hope is that - for the programs under consideration (though they are procedural, as I assume they generally are) - the fact that they are 'divergent replicas' will allow them to be more effectively compared. Another related observation is that, when developing software using EM principles, as in the sqleddi exercise I briefly mentioned in the lectures, the whole problem of versioning is radically transformed - and I would say semantically enriched though technically simplified. It would be good to know whether the IceCube technology applies in this context, and whether it would deliver even more benefit. If it does, there may be this suggests some kind of relationship between programming paradigm and the performance of the IceCube algorithm.