Version 0.8 | Try it online @ catseye.tc | See also: Pixley
Robin is an excessively principled and thoroughly specified functional programming language with eager evaluation, latent typing, and a homoiconic syntax, based on a radically simple core semantics in which the so-called "fexpr" is the fundamental abstraction and both functions and macros are defined in terms of it.
Expressions in Robin are referentially transparent; programs interact with the outside world through an event-driven framework.
For more information, see the extended description below.
The Robin reference interpreter is written in about 1300 lines of Haskell.
To use it, you'll need an implementation of Haskell installed (typically either
ghc
or Hugs).
First, clone this repository and cd
into the repo directory. Then run
make
If you have cabal
installed, the Makefile will use it to build the robin
executable, and this will take care of obtaining and building the dependencies.
If you do not have cabal
, the Makefile will use ghc
directly to build the
executable, but in this case, you will need to ensure you have dependencies
like parsec
and random
installed, yourself.
(If you don't have ghc
at all, no executable will be built; but that's OK,
because in this case the bin/robin
driver script will fall back to using
runhaskell
or runhugs
instead.)
In any case, the Makefile will also build build the standard library
(pkg/stdlib.robin
). And this same Makefile can be used to build the
JavaScript version of the interpreter, with make web
.
After running make
, you can run the Robin interpreter using the
driver script in bin
, on one of the example Robin sources in eg
like so:
bin/robin pkg/stdlib.robin eg/hello-world.robin
You should see
Hello, world!
To continue learning to program in Robin you can follow The Robin Tutorial.
If you have a few minutes to spare, and you have Falderal installed, you can run the test suite (consisting of more than 600 unit tests) by running
./test.sh
The tests that use only Robin's core semantics (with no help from implementation "builtins") are quite slow, so you may want to skip them, by running
APPLIANCES="appliances/robin.md" ./test.sh
The test suite will also run some property tests (using QuickCheck). Notably, for every operator that is defined multiple times (which includes much of stdlib, where the core definitions are written in Robin but also implemented in Haskell as "builtins" in the reference interpreter), QuickCheck will attempt to falsify the assertion that the definitions define the same operator. These attempts are currently rather crude; there is lots of room for improvement for them in some future release.
For experienced programmers, Robin might be best described by listing the languages that have had the strongest influences on it:
Like Scheme, Robin is eagerly evaluated, latently typed, and homoiconic, as well as properly tail-recursive and lexically scoped (at least by default), and tries hard to be well-defined and system-agnostic, but (as you can read below) diverges significantly from Scheme in other ways.
Like Forth, Robin has a radically simple core semantics. There are 15 intrinsic operations; every symbol in the standard library is defined in terms of these intrinsics, while an implementation is free to provide its own (perhaps more efficient) implementation of any such symbol. (See also Pixley).
In most languages, the arguments to a function are evaluated before the function is applied, but PicoLisp allows defining functions with unevaluated arguments. In historical Lisp, such operators were called fexprs. Robin adopts fexprs as the fundamental abstraction — both functions and macros are defined in terms of fexprs.
The Kernel programming language also takes fexprs as its fundamental abstraction; however, Robin was developed oblivious of Kernel — it adapted the idea directly from PicoLisp.
Like Haskell, Robin is referentially transparent (often described as "purely functional") — mutation of values is forbidden. (Robin intentionally does not, however, adopt lazy evaluation or a static type system.)
Interactive programs in Robin are built by composing transducers which are driven by events and produce effects (which are modelled as further events), in a manner very similar to The Elm Architecture.
Arbitrary text can by embedded in a Robin program using a syntax very much like a "heredoc", except it is an S-expression.
Deserves at least a passing mention here, as one thing that Robin
discards from Scheme is its jargony terminology: no cdr
, no cons
,
no lambda
. (A notable exception is fexpr
simply because there is no
satisfying short, non-jargony word that connotes how these operators work.)
For a full description of the Robin language, see the Robin specification document.
appliances/
— test appliances for the literate test suite.bin/
— driver script, destination for executable when built.demo/
— contains HTML5 document demonstrating build to JS by Haste.doc/
— Tutorial, specification, rationale, etc.eg/
— example programs written in Robinsrc/
— Haskell source for reference interpreter.stdlib/
— normative definitions of standard library symbols.HISTORY.md
— history of this distribution.TODO.md
— plans.