Discussion Notes 2010-02-08 OpenModelica Workshop 2010, Session Future of OpenModelica Gerhard: Error messages are important. Original question: is it difficult to introduce good error messages. Adrian: Error messages are spread out. Aronsson. Migh be difficult. Martin: We don't get pieces of incorrect code to test our system PeterF: Gerhard could produce a number of erroneous models that should have a good error message. Adrian: Francesco: Why errors, division by zero, log(neg number), non-solvable equation - the solver is not converging. Sometimes not good enough initial values. Ideally the diagnostics should point out the model source code position. Adrian: We now have more information in the backend from where equations come from. Still missing line numbers. Francesco: Will provide some erroneous examples. Performance investigation for models. Profiling For example, Modelicaprof (similar to gprof for C) Francesco: Most important: which part of the code is taking most of the time Francesco: Going parallel is strategic. Easiest: Many parallel threads for parameter sweeps. NIMROD - Univ. Melbourne, Debugging and Performance analysis help for parallel platforms. Francesco: The consortium is growing. COuld we attract other people, numerical specialists, symbolic manipulation specialists who are not associated with Modelica. Sparse solver in OpenModelica. Automatic differentiation of algorithms. Symbolic Jacobian. Gerhard: Parameter optimization, multi-criteria optimization. Optimization tools would be useful. Numerical optimization. Optimization specifications expressed together with the model. (Optimica only for continuous differentiable models) Bernhard: We have implemented some numerical optimization methods in Dymola and OpenModelica. GaussNewton, Simplex, ... As library. In Sabrines PhD. Christian SOnntag: Muscod2, gProms are the only tools for hybrid model optimization