Improve scannable name() when debug() mode is enabled

[smaldini]

If scanning parent publisher (Flux/Mono)

• Use the immediate parent FluxAssembly toString()
  If scanning parent subscription :
• Use the immediate parent FluxAssembly subscriber toString() which is not defined for now. Currently the toString is not implemented but it could be equivalent to FluxAssembly itself

For both scan, filter the assembly operators from the actuals()/parents()/inners()

[simonbasle]

@smaldini can you detail a little bit, with a concrete example? I'm not sure I fully understand the requested enhancement.

[smaldini]

So right now if you do

Hooks.onOperators.debug();
Mono<?> m=
 Flux.from(s -> Scannable.from(s).actuals().forEach(sc -> System.out.println(sc.name())))
   .map(a -> a)
   .filter(a -> true)
   .reduce((a, b) -> b);

System.out.println("Scan actual subscribers:");
m .subscribe();

System.out.println("\nScan parent publishers:");
Scannable.from(m).parents().forEach(sc -> System.out.println(sc.name()));

You should see something like:

Scan actual subscribers:
AssemblySubscriber
MapSubscriber
AssemblySubscriber
FilterSubscriber
AssemblySubscriber
ReduceSubscriber
AssemblySubscriber
LambdaSubscriber

Scan parent publishers:
FluxAssembly
FluxReduce
FluxAssembly
FluxFilter
FluxAssembly
FluxMap
FluxAssembly
FluxSource

I propose to filter assembly operators from such parents()/actuals() if possible and better, try to reuse the stacktrace first line as a name output like "FluxReduce(MyCode:123)"

[simonbasle]

Scannable.name() is a bit confusing: it scans for NAME in the operator AND the parents hierarchy until it finds a Scannable that answers to NAME, but that's more of a "name of the sequence" thing. It reverts to outputing the operatorName() of the root Scannable if 0 operator in the hierarchy answers to NAME.

In your example I think you should have used operatorName(), which should be easy to implement as stacktrace.stackLine() for the assembly operators and subscribers.

I don't think that eliminating xxxOnAssembly operators from the more generic parents() and actuals() streams is necessarily a good idea.

Perhaps name() should also be deprecated and aliased sequenceName() or something like that for clarity?

[simonbasle]

See further discussion in #1140