I have been working a lot on Makam in the past few months, the metalanguage based on λProlog that I started implementing at MIT working with Adam Chlipala. The latest feature I have been working on is a Web interface for Makam, meant to be used for literate posts that have Makam code alongside the text. The code is ran through a AWS Lambda Makam webservice. This is the first such literate post – so go ahead and evaluate the code of this web page, or even try editing the final queries before evaluating, using the two buttons on the bottom-right of the page!

It goes without saying what the first query utilizing this interface should be:

post: testsuite. %testsuite post.
print_string "Hello world!\n" ?
>> Yes.

The “Yes.” part you might notice in the result of this query means that a solution was found. Typically, Makam queries will have uninstantiated metavariables (that start with a capital letter), and the point of executing them would be to find valid solutions for those variables. More on these below.

string.concat ["Hello", " ", X, "!"] "Hello world!" ?
>> Yes:
>> X := "world".

Now that this interface is ready, I am planning to start a number of tutorials on Makam and on language implementation. I also have a few experimental ideas that I want to implement, where I would appreciate some feedback, but that is a matter for a later time. In any case, drop me a line with your thoughts if you are reading this!

Makam is meant to be used as a PL prototyping tool to explore research ideas, and also as tool for encoding and understanding existing languages and advanced type systems. It aims to be more of a very expressive core language that can be used to define languages and transformations between languages in a concise way, even when those are quite involved. In that sense, it does not aim to be a language workbench that allows you to define specific aspects of your language using different domain specific languages (e.g. parsing, name resolution, typing, etc.), and gives you a full-blown IDE experience as a result (for example, in the style of Spoofax). However, a lot of the DSLs and tools that would comprise a language workbench in that style can be implemented within Makam itself.

One such DSL that I have defined lately is a bidirectional syntax language. It can be used to define a parser and a pretty-printer for the language you are implementing, through a single set of rules (hence the ‘bidirectionality’ in the name refers to viewing parsers and pretty-printers as inverses of each other). The pretty-printing part is quite basic at this point, but the parsing part is at a relatively good state: it is based on PEG parser combinators, and it generates parsing code in JavaScript that runs in Node.js under the hood. This is done because Makam is interpreted and quite slow to evaluate, so running the parsers in Node is a huge performance boost. Of course, the parser generation, which transforms terms of the PEG combinator language to JavaScript code, is implemented in Makam.

Other things that I have been working on recently is a concrete binding library, to convert between abstract syntax with concrete names and abstract binding syntax utilizing HOAS; a testing framework; and various additions to the standard library. There’s now an npm package for Makam too with precompiled binaries for Linux and MacOS X, if you want to execute Makam locally and do not already have OCaml installed. With these, I felt like a version bump was in order, so this also marks the release of Makam 0.7.0.

As a small example of the syntax DSL, let’s implement a simple calculator language and its syntax description. I will not go into full details, so stay tuned for a more in-depth introductory post along the same lines. First, we will define the terms of our language– let’s say we have integer constants, integer addition, let expressions and variables.

term : type.

intconst : int -> term.
add : term -> term -> term.
let : string -> term -> term -> term.
var : string -> term.

Now let’s define the syntax for these. We split the syntax into two precedence levels, and then give a syntax rule for each term constructor. The angle brackets means that the result of a parsing expression gets applied to the given term constructor – so each rule needs to apply the right kind of arguments in sequence. Quoted strings stand for tokens.

term, baseterm : syntax term.

`( syntax_rules {{

term ->
  add { <baseterm> "+" <term> }
/ let { "let" <makam.ident> "=" <term> "in" <term> }
/ baseterm

baseterm ->
  intconst { <makam.int_literal> }
/ var { <makam.ident> }
/ { "(" <term> ")" }


`( syntax.def_toplevel_js term ).

The two statements starting with the `( symbol are staged, meaning that they are Makam programs that compute Makam programs which are then evaluated in place. The former one translates the surface-level syntax for syntax rules to actual Makam rules; and the latter one generates a top-level parser in JavaScript for the term. The double-bracket notation stands for multi-line strings, which is useful when writing code of a different language.

With these, we can issue parsing queries. For example:

syntax.parse_opt term {{ let a = 21 in a + a }} X ?
>> Yes:
>> X := let "a" (intconst 21) (add (var "a") (var "a")).

Let’s also define an evaluator for this language that returns the result of an expression. This will be a predicate that relates a term of the language to its integer value:

eval : term -> int -> prop.

eval (intconst N) N.
eval (add E1 E2) N :-
  eval E1 N1, eval E2 N2,
  plus N1 N2 N.
eval (let VAR DEF BODY) N :-
  eval DEF N_DEF,
  (eval (var VAR) N_DEF -> eval BODY N).

We can also define a version of the eval predicate that accepts a string as input, and parses the string before evaluating. Makam has type-based overloading for constants, so types are used to disambiguate between the two eval predicates:

eval : string -> int -> prop.
eval String Result :-
  syntax.parse_opt term String Term,
  eval Term Result.

Our simple calculator is ready! Here’s an example query:

>> eval {{ let a = 21 in a + a }} X ?
>> Yes:
>> X := 42.
eval {{ let a = 21 in a + a }} X ?

Using the Makam web interface, you can edit the query above – the Edit button on the bottom-right takes you directly to it – and re-run Makam to see the result with the Play button. (For the time being, only the query is editable, mostly to limit the AWS Lambda execution times; in the future, the rest of the Makam code blocks in a post might be editable as well if that proves useful.)

That’s it for the time being! More introductory tutorial posts should be coming up soon!