This page is with ready responses, for these questions, about macro-nets of mid-80. Readers may ask yet other questions, too:
Macro-nets are modularized nets, for
Macro-nets are convenient for modelers, but they are not for any run-time presence. They must leave the scene, for the net-interpreter (or, any net-processor tool) to recognize the net formalism it expects. For this, replace any macro-name with the subnet it represents. (And so on, with any submacro, within macros, recursively.)
An E-net models a concurrent system, with resources.
NN73 presents Macro E-nets, for (even) more intuitive representation with E-nets.
NN73 presents, with macros, extended forms of the E-net primitive transitions, as well as new abstractions. For example, the Q-macro location is a (re-orderable) queue/stack, and listable as an ordinary location. It is very easy to form new macros, and to employ them as if ordinary transitions, or to nest them within the parameter lists of transitions, as if ordinary locations.
SARA is a software system, and SARA SL (Structure Language) is its macro-hierarchy. It is similar to Macro E-nets in system/environment structure, but with more facilities for its software-methodology with modules.
For a verifier to verify a net (whether Petri net, or E-net), any (non-primitive) macros in it, needs replacement with the equivalent subnet. This is what NN73 had suggested. For some macro-net to pass as a primitive, it needs proper-reduceability.
NN73 macro examples facility provides a way for defining new macro-nets (macro-transitions, and macro-locations) as extensions of E-net primitives, with new input/output configurations, other than the five (built-in) E-net primitives. But the interpreter does not interpret them. They are replaced, with the equivalent subnets. They are for more intuitive net-models, only.
Macros for nets, are for improved representations, for more intuitive models. Macro-nets help modelers with their work, to make it easier, cozier, rather than "shorten their coffee-break times" (as when a non-deterministic verifier would run possibly for hours).
For a deterministic-model, it is very feasible to expand every macro, recursively, and let the verifier to verify a single full net. A subnet-reduction does not save any significant time. At most, it is only a little re-compilation time. Not anything exponentially explosive.
Never assume some macro as if that were equivalent to a primitive, unless it is really properly reduceable (fully). Or else, it might be not verifiable, or equivalently, an insistence to verify/read such a net, may be to commit to maximal complexity.