frontend: Investigate delayed macro expansion

[flofriday]

I think we can reduce complexity in the macro expander by only expanding macros in the outmost "context" (all that aren't invoked inside other macros).

I assume this should also improve the performance of the macro expander but I might be overlooking something, so proper benchmarks are required for this feature.

[AndreasKrall]

I do not understand your comment at all. Please can you give a short explanation how the current macro expander works and how the delayed version should work.

[flofriday]

The current approach as pseudo code:

model A() = {...}

model B() = { ...  $A() ... }       // While parsing this model we already evaluate A 

$B()                                // We now evaluate B here

The body of A is copied twice, once when inlining into B and once when evaluating B.

I think we can delay evaluation to the outermost context so that we skip the evaluation of models inside model definitions and only evaluate them in the outermost invocations are evaluated.

[AndreasKrall]

I believe it is worth to investigate it.