Use a container DispatchMethodLists with untyped fields instead of a NamedTuple for dispatch lists returned by create_dispatch_methodlists(), so the typed Tuple of methods is only dispatched on at the final call to dispatch_methodlist() (a generated function)
This seems to improve the time for the first call to 'do_deriv', so the startup time for generated_dispatch = true is now similar to generated_dispatch = false. Presumably it was passing around large typed Tuples or NamedTuples that was causing problems, not the @generated dispatch_methodlist()
NB: there is now a dynamic dispatch in do_deriv (one for each of list_initialize, list_do).
Use a container DispatchMethodLists with untyped fields instead of a NamedTuple for dispatch lists returned by create_dispatch_methodlists(), so the typed Tuple of methods is only dispatched on at the final call to dispatch_methodlist() (a generated function)
This seems to improve the time for the first call to 'do_deriv', so the startup time for generated_dispatch = true is now similar to generated_dispatch = false. Presumably it was passing around large typed Tuples or NamedTuples that was causing problems, not the
@generated dispatch_methodlist()
NB: there is now a dynamic dispatch in do_deriv (one for each of list_initialize, list_do).