Many languages claim to be simple but few truly are. While Portcullis's compiler is arguably simple, clocking in well under 1000 LOC, writing Portcullis programs can be a bit of a puzzle. This is mostly the result of its restricted syntax and minimalist design which include:
- no looping constructs (use recursion)
- no variable declaration
- no built-in
Booleans(useAtoms) - no
nulltype orerrorprimitives such asException - no side-effects (use pipes)
Despite these limitations, one can still build useful software that is
- concurrent
- free from runtime exceptions*
- highly observable
Functions don't control data-flow
Put another way, functions only manipulate data. Pipes and Streams describe how functions asyncronously compose.
Portcullis lets you define functions, streams, and pipes. Functions manipulate date, streams both namespace data and add type signature, and pipes describe how streams and functions compose.
Even though the Portcullis compiler is runtime agnostic it still typechecks the execution graph it exports. It is up to users/lib creators to implement a concurrent runtime that executes this exported dataflow graph. That means that you could have one runtime that uses DOM events to propagate event, or one that uses broadcasts channels, or one that instantiates web workers and uses worker message passing so it's both concurrent and parallelized. The runtime could choose to implement a telemetry dashboard where the whole system is observable in realtime. Since you've got a datastructure that describes all dataflow, systems diagrams can be programmatically generated instead of done by hand.
Look at the demos in /runtime/examples to see examples.
* barring a compiler bug