Macro paradise 2.0.0-SNAPSHOT

Eugene Burmako
05.08.2013

It is my pleasure to announce a new major version of macro paradise, featuring new exciting functionality and repackaged for the ultimate ease of use.

00:58 ~/210x $ scalac -Xplugin:macro-paradise_2.10.2-2.0.0-SNAPSHOT.jar -Xshow-phases
             phase name  id  description
             ----------  --  -----------
                 parser   1  parse source into ASTs, perform simple desugaring
          macroparadise   2  let our powers combine
                  namer   3  resolve names, attach symbols to named trees in paradise
         packageobjects   4  load package objects in paradise
                  typer   5  the meat and potatoes: type the trees in paradise
                         ...

Now a compiler plugin for 2.10.x

Previous versions of macro paradise were distributed as custom builds of a fork of Scala compiler. They were reasonably easy to experiment with, but tricky for everything else - being a custom build of scalac creates problems with SBT and Eclipse, complicates usage of third-party libraries, and ultimately locks us into an eternal snapshot preventing from reasonable versioning.

However since today, paradise is a compiler plugin targetting official Scala compilers of the 2.10.x series, making latest macro developments available to anyone using the latest stable version of Scala. For example, there's no longer need to wait for Scala 2.11 to make use of quasiquotes - with the paradise plugin, quasiquotes already work in the production Scala environment.

Featuring quasiquotes and macro annotations

The first preview of macro paradise 2.0 includes: 1) quasiquotes that immensely simplify construction and deconstruction of ASTs, and 2) macro annotations that provide a robust and powerful facility for doing global code generation. You can check out the macro corner at docs.scala-lang.org for details (quasiquotes, macro annotations), and here I will provide a short summary.

Quasiquotes are a backport of what's been recently released as a part of Scala 2.11.0-M4 and scheduled for inclusion in Scala 2.11. They allow one to write q"...", tq"...", cq"...", and pq"..." with Scala code and possibly interpolation splices in quotes, to construct and deconstruct trees in a WYSIWYG fashion. In paradise, just like in the recent milestone release of 2.11, to enable quasiquotes you only need to import the members of the corresponding universe, e.g. import scala.reflect.runtime.universe._ or import c.universe._ (trick question: how did we pull this off given that paradise is just a compiler plugin that can't change the standard library?).

Macro annotations bring textual abstraction to the level of definitions. Annotating any top-level or nested definition with something that Scala recognizes as a macro will let it expand, possibly into multiple members. Unlike in the previous versions of macro paradise, macro annotations in 2.0 are done right in the sense that they: 1) apply not just to classes and objects, but to arbitrary definitions, 2) allow to introduce static members (i.e. expansions of classes can affect companions). This opens a number of new possibilities in code generation land, including customizable case classes and better enums.

In the spirit of Scala macros, macro annotations are as untyped as possible to stay flexible and as typed as possible to remain useful. On the one hand, macro annottees are untyped, so that we can change their signatures (e.g. lists of class members). But on the other hand, the thing about all flavors of Scala macros is integration with the typechecker, and macro annotations are not an exceptions. During expansion we can have all the type information that's possible to have (e.g. we can reflect against the surrounding program or perform type checks / implicit lookups in the enclosing context).

Type macros are out

If you have been using paradise before, you might've noticed that the list of supported features doesn't include type macros. You might've also heard rumors that type macros are unlikely to be included in future versions of Scala. This is not a coincidence - let me elaborate.

Type macros are an embodiment of textual abstraction - they recognize pieces of text that match a specification and replace them according to a procedure. Even though this concept is less useful for types than for terms, it has still found its niches: 1) encoding computations at type level by naturally extending what type aliases already do, 2) generating publicly visible code by expanding references to parents in definitions of classes and objects.

However, as practice shows, type macros are not the best fit for both of these tasks. 1) Textual abstraction alone isn't enough for type-level computations, because most often immediate expansion is not what's really needed. Think of it that way: when writing List[T] in a function definition, you don't want it to be replaced with some newly generated type, but rather want it to be retained in the signature to be used later during type checking and inference. 2) Treating references to parents as something expandable is interesting, but this notion doesn't scale to codegen over members.

Therefore there have been concerns that type macros don't pull their weight as a dedicated language feature in a potential future release of Scala, and we've been very worried about that, because marrying research with production for the greater good has been the ultimate goal of Scala macros from day one.

After careful consideration, we have found replacements for most of the use cases of type macros in order to both meet the requirements of the language design committee and to provide an upgrade path for our valued early adopters who have already invested in macro-based solutions. Implicit macros provide a macro-based approach for a great share of type computation use cases, and macro annotations fill the code generation role. Both are immediately available in macro paradise 2.0.0-SNAPSHOT.

What's coming next?

In the next few weeks we plan to introduce fundep materialization along with some fixes to implicit search that will reveal the full potential of implicit macros. We also have several other features in the works, but it's too early to share details now - stay tuned for updates.

And as always, development of Scala macros is very much shaped by early adopters. We value your opinion, are glad to help with your macro questions, and are always up for a good discussion over a macro or two:

  • Are there solid use cases for untyped macros, or we shouldn't even port them to the new paradise now that we have macro annotations? upd: See a discussion at scala-language.
  • What else would you like to see in paradise and ultimately in Scala?
  • Existing features don't work for you?

Speak up in our scala-language discussion or drop me a note on Twitter!

Subscribe to updates