Adding type annotations to graph edges

[mikolak-net]

Summary:
Both:

• the Gremlin graph, and
• the generated diagram,
  should include types as edge (transition) labels wherever possible.

Problem description:
How can't one get the types of a stage from an Akka Streams Graph? Let me count the ways:

• obviously, it's not possible from the Graph itself, nor its Shape. At least one of those would have to be an HList-like structure for that;
• it's not possible to get that information from the Traversal - for the exact same reason;
• moreover, even given a Stage, one can't infer the types from its Inlets and Outlets, as the corresponding Lists are wildcard-typed.

So, the requirement is to capture every stage practically at the time of creation (at least before it passes an operation that results in type erasure) and store its associated, fully-typed Shape, accessible by the graph generator.

Problem solution:
A possible solution would thus involve:

• a TTL/LRU/etc.-based registry mapping between stage instances and their inlet/outlet type,
• a macro/compiler plugin the would rewrite each stage creation instruction into a "wrapped" version, containing the registry store call.
• finally, and trivially, a modification to the Traversal parser that references said registry and matches edge labels to the corresponding types.

Regarding point 2 - in other words, we would like to automate converting this:

Source.single("0").map(_.toInt).to(Sink.seq[Int])

into this:

⇑(⇑(Source.single("0")).map(_.toInt)).to(⇑(Sink.seq[Int]))

where:

  def ⇑[T <: Graph[_ <: Shape, _]: TypeTag](g: T): T = {
    storeInSomeRegistry(g, typeOf[T])
    g
  }

Note that there will also be some typesystem deconstruction involved - for example the actual typeOf[T] in the example above is:

akka.stream.scaladsl.Source[java.lang.String,akka.NotUsed]
akka.stream.scaladsl.Source[Int,akka.NotUsed]
akka.stream.scaladsl.Sink[Int,scala.concurrent.Future[scala.collection.immutable.Seq[Int]]]

And not e.g. Graph[SinkShape[.....

---

Of course, just adding the registry and allowing users to manually insert registration invocations would be a good first step.

[mikolak-net]

A partial solution (manual registration) will be available in the immediately upcoming version. Automatic derivation will come later on - see this SO question for the abstracted form of the current blocker.