qwe

qwe

…riables.

… NaNs.

Indeed, it is 15 to 20 times slower than the AST-based interpreter. Clearly, handling a stack is more costly than handling recursion.

This was only done to check that we do indeed get the same results... and we do.

Very important to make sure that we do indeed get the results we expect for all aspects of the interpreter.

Rather than `std::vector<double>` (and then retrieve the data using `data()`. Indeed, some solvers may want to use temporary arrays (e.g., CVODE uses a temporary array of doubles for states and rates), which means that we need to be able to set them, if needed.

We now rely on the user providing them since some solvers may use temporary arrays to compute rates (e.g., CVODE uses temporary arrays for states and rates).