Heart of Clojure

The idea behind our project, called Klor (https://github.com/lovrosdu/klor), is to use a new domain-specific language (DSL) – embedded in Clojure – to write distributed systems as choreographies. A choreography is very similar to any “normal” Clojure program, except that it has a special message-passing primitive for processes to communicate: instead of the usual two send and receive primitives, there is just one communicate primitive that embodies a whole end-to-end transmission. The great thing about this is that bugs like communication mismatches and deadlocks – very hard to diagnose and fix! – become syntactically impossible to write in the first place. The choreography essentially acts as a single source of truth and is used to automatically generate the implementations of the processes in the system. communicate still compiles to a send and a corresponding receive behind the scenes, but it all happens completely transparently to the programmer.

Klor is implemented on top of the clojure.tools.analyzer compiler framework and hooks into Clojure’s macro system to perform all of its analysis and code generation at macroexpansion-time, imposing virtually no run-time cost on the user. It uses a lightweight “type” system to track the locations of values and knowledge shared between the processes, and supports standard higher-order features that help with composition and modularity. Still, Klor does not impose any hard restrictions and embraces the interactive and dynamically typed spirit of Clojure: there is seamless interoperability with external Clojure code and standard REPL-based development is highly encouraged. All in all, Clojure’s excellent support for concurrency, host interop and mature tooling makes it a great foundation to use for Klor.

The aim of this talk is to demonstrate Klor’s usage and highlight some of the interesting implementation challenges we faced.