Lorenzo Contento

I will change the status to draft for the moment and notify you when everything is ready. But feel free to have a look at new commits as they are...

> > Only downside should be an increase in code generation time. > > @lcontento , did you test that with any larger model? Would be good to have a...

I get the following results ``` w/o opt w/ opt Finished computing xdot (1.73E+00s) (1.77E+00s) Finished computing dxdotdx (7.32E+00s) (7.54E+00s) Finished computing w (7.09E+01s) (7.01E+01s) Finished computing dwdx (1.55E+02s) (1.50E+02s)...

By the way, the signature `_print_with_exception(math: sp.Expr)` does not appear to be respected, since in the CI testing there are cases where `math` is a `list`. If it is a...

> > By the way, the signature `_print_with_exception(math: sp.Expr)` does not appear to be respected, since in the CI testing there are cases where `math` is a `list`. If it...

> > Another test that fails is due to symbols given as strings. It may be that these situations only occur in test cases where shortcuts are used for brevity/clarity,...

I agree that simplifications are useful, especially for generated code. But AMICI is not a model generator, so that the onus of providing nicely written formulas should be on the...

I completely agree that computation time is going to be a big limiting factor. If a SBML model makes heavy use of assignment rules, it may advantageous to move some...

> Would there be advantages of having dsigmaydp + dJydsigmay? I don't see any (even if there are sigma expressions used multiple times the compiler should be smart enough to...

> Returning NaN / Inf is good, but it would also be helpful for debugging to (optionally?) included some more information on the function that failed. We could then keep...