Skip to content

Files

Latest commit

9ac1344 · Oct 12, 2021

History

History

pointfree-lang

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

pointfree-lang

npm version npm downloads Twitter Follow

This project is part of the @thi.ng/umbrella monorepo.

About

Experimental language layer with compact Forth style syntax for @thi.ng/pointfree, an ES6 embedded DSL for concatenative programming:

  • PegJS based grammar & parser
  • untyped, interpreted, but with AOT compilation of user defined words
  • hyperstatic word definitions
  • support for custom / externally defined words (JS functions)
  • dynamically scoped variables (stored in environment object)
    • syntax sugar for declaring & autobinding local vars w/ stack values
  • nested quotations (code as data, vanilla JS arrays)
  • dynamic word construction from quotations
  • array & object literals (optionally w/ computed properties)
  • all other features of @thi.ng/pointfree (combinators, array/vector ops etc.)
  • CLI version w/ basic file I/O, library includes, JSON

Status

STABLE - used in production

Search or submit any issues for this package

Installation

yarn add @thi.ng/pointfree-lang

ES module import:

<script type="module" src="https://cdn.skypack.dev/@thi.ng/pointfree-lang"></script>

Skypack documentation

For NodeJS (v14.6+):

node --experimental-specifier-resolution=node --experimental-repl-await

> const pointfreeLang = await import("@thi.ng/pointfree-lang");

Package sizes (gzipped, pre-treeshake): ESM: 4.97 KB

Dependencies

Usage examples

Several demos in this repo's /examples directory are using this package.

A selection:

Screenshot Description Live demo Source
Generate SVG using pointfree DSL Source

Command line usage

The package includes a pointfree CLI command to evaluate strings or files:

npx @thi.ng/pointfree-lang

Usage: pointfree [options] [file]

Options:
  -V, --version     output the version number
  -d, --debug       print debug info
  -e, --exec <src>  execute given string
  -h, --help        display help for command

For CLI usage, in addition to the other language features discussed further below, the following words are available too (more to be added):

Word Stack comment Description NodeJS equiv
@args ( -- arg[] ) Places CLI args on stack(1) process.argv
include ( path -- ) Includes another pointfree-lang file
read-dir ( path -- file[] ) Returns array of file names in given dir readdirSync(path)
read-file ( path -- str ) Reads file and returns contents as string readFileSync(path)
write-file ( body path -- ) Writes body to file at given path writeFileSync(path, body)
  • (1) - index 0 is the first user supplied arg (rather than 2 as with process.argv)

Include files / libraries

As mentioned in the table above, other pointfree-lang source files can be recursively included via include. The inclusion mechanism so far is basic, but can break cyclical dependencies, though the behavior will still be improved in the future. All imported words are added to the same environment. Use namespacing if needed (e.g. prefix.foo vs foo), word names can be more flexible than in JS.

File lib.f:

: lib.hello ( name -- greeting ) "hello, " swap "!" + + ;
# eval program string w/ arg
pointfree -e '"lib.f" include @args 0 at lib.hello .' asterix
# hello, asterix!

CLI example

A small example tool to scan a package directory and display package names w/ their current versions, with optional support to filter package list via regexp:

Save the following file to semver.f (Btw. Use .f as file ext to harness existing syntax coloring support for Forth...)

( builds path to package.json )
: pkg-path ( name base -- path ) swap "/" swap "/package.json" + + + ;

( takes package dir , returns parsed package.json )
: read-pkg ( name base -- json ) pkg-path read-file json> ;

( extracts semver package name from given package object )
: pkg-semver ( pkg -- name@version ) dup "name" at "@" rot "version" at + + ;

( package list filter )
: filter-pkg ( name pat -- name? ) over -rot match? not [ drop ] when ;

( store pattern and base dir in global var )
@args dup 1 at pattern! 0 at base!

( load packages dir )
@base read-dir

( filter package list if 2nd CLI arg given )
@args length 1 > [ [ @pattern filter-pkg ] mapll ] when

( try to process all packages, ignoring any errors )
[ [ @base read-pkg pkg-semver . ] [ drop ] try ] mapl
# list all
pointfree semver.f node_modules
# JSONStream@1.3.5
# abab@2.0.3
# abbrev@1.1.1
# acorn@6.4.0
# acorn-globals@4.3.4
# acorn-walk@6.2.0
# ... 100's more ...

# filtered w/ pattern (regex)
pointfree semver.f node_modules '^type'
# type-check@0.3.2
# type-fest@0.3.1
# typedarray@0.0.6
# typedarray-to-buffer@3.1.5
# typedoc@0.16.10
# typedoc-default-themes@0.7.2
# typescript@3.8.3

API

Generated API docs

(Code for the above example)

// DSL source code (syntax described further below)

const src = `
( helper words for forming 2D vectors )
: xy ( x y -- [x y] ) vec2 ;
: yx ( x y -- [y x] ) swap vec2 ;

( generate horizontal line coords )
: hline ( y width -- [0 y] [width y])
  over 0 yx -rot yx ;

( generate vertical line coords )
: vline ( x height -- [x 0] [x height])
  over 0 xy -rot xy ;

( draw haircross w/ FFI 'gfx.line' word )
: hairx ( x y w h -- [] )
  -rot [vline] [hline] bis2 [gfx.line] bia2;
`;

// custom word definition (will be used by `hairx` word above)
// stack effect:
// ( [x1 y1] [x2 y2] -- )

const drawLine = (ctx) => {
    const stack = ctx[0];
    // minimum stack depth guard
    pf.ensureStack(stack, 2);
    // pop top 2 values
    const [x2, y2] = stack.pop();
    const [x1, y1] = stack.pop();
    console.log(`draw line: ${x1},${y1} -> ${x2},${y2}`);

    // if we had a canvas drawing context stored in env...
    // const canvasCtx = ctx[2].canvasContext;
    // canvasCtx.beginPath();
    // canvasCtx.moveTo(x1, y1);
    // canvasCtx.lineTo(x2, y2);
    // canvasCtx.stroke();

    // or... alternatively generate SVG (and push result on stack (or store in env)
    // stack.push(`<line x1="${x1}" y1="${y1} x2="${x2} y2="${y2}"/>`);

    // ...same again, but in @thi.ng/hiccup format
    // stack.push(["line", {x1,y1,x2,y2}])

    return ctx;
};

// the DSL interpreter & compiler uses an environment object
// to lookup & store word definitions & variables
// here we create new environment and associate custom FFI word(s)
const env = pf.ffi({}, {
    "gfx.line": drawLine
});

// compile / execute source code w/ given env
// the compiled words will be stored in the env
pf.run(src, env);

// (optional, but that's how we do it here for example purposes)
// store some external state / config in env
// this could be modified via event handlers etc.
env.mouseX = 100;
env.mouseY = 200;
env.width = 640;
env.height = 480;

// now actually call the `hairx` word with args pulled from env
// words prefixed w/ `@` are variable lookups
pf.run(`@mouseX @mouseY @width @height hairx`, env);
// draw line: 100,0 -> 100,480
// draw line: 0,200 -> 640,200

// or call precompiled word/function directly w/ given initial stack
pf.runWord("hairx", env, [100, 200, 640, 480]);
// draw line: 100,0 -> 100,480
// draw line: 0,200 -> 640,200

Language & Syntax

As noted previously, the syntax is closely based on Forth (and other concatenative languages), however since this implementation is targetted to ES6 environments, the semantics and actual implementation differ drastically. In this DSL (and most aspects also in @thi.ng/pointfree):

  • words and programs are implemented as functional compositions of vanilla JS functions, i.e. 1 2 + => add(push(2)(push(1)(ctx)))
    • therefore no user controlled context switch between immediate & compile modes, as in Forth
    • parsing of word definitions triggers compile mode automatically
  • variables and both stacks (D & R stacks) can store any valid JS data type
  • no linear memory as in Forth, instead variables and the dictionary of (custom / FFI or user defined) words is stored in a separate environment object, which is passed to each word/function
  • the DSL has syntax sugar for variable value lookups & assignments
  • within word definitions the DSL supports binding stack values to local vars at the beginning of the word
  • the DSL supports nested quotations (array) & object literals, incl. support for computed properties and/or values (lazily resolved within words)
  • all symbols are separated by whitespace (like in Clojure, commas are considered whitespace too)

Comments

As in Forth, comments are enclosed in ( ... ). If the comment body includes the -- string, it's marked as a stack effect comment. The first stack comment of a word is added to the word's metadata in preparation for future tooling additions.

Comments current cannot contain ( or ), but can span multiple lines.

Since v1.4.0 line comments are supported, use the standard JS // prefix and extend until the next newline char.

( multiline:
              .__        __    _____
______   ____ |__| _____/  |__/ ____\______   ____   ____
\____ \ /  _ \|  |/    \   __\   __\\_  __ \_/ __ \_/ __ \
|  |_> >  <_> )  |   |  \  |  |  |   |  | \/\  ___/\  ___/
|   __/ \____/|__|___|  /__|  |__|   |__|    \___  >\___  >
|__|                  \/                         \/     \/

)

1 2 ( embedded single line ) 3

// single line comment

Identifiers

Word identifiers can contain any alhpanumeric character and these additional ones: *?$%&/|~<>=._+-. However, digits are not allowed as first char.

All 100+ built-in words defined by @thi.ng/pointfree are available by default with the following additional aliases (which aren't valid names in the ES6 context):

Alias Original name
?drop dropif
?dup dupif
-rot invrot
>r movdr
>r2 movdr2
r> movrd
r2> movrd2
if condq
switch casesq
while loopq
try $try
+ add
- sub
* mul
/ div
v+ vadd
v- vsub
v* vmul
v/ vdiv
= eq
not= neq
<= lteq
>= gteq
< lt
> gt
pos? ispos
neg? isneg
nil? isnil
zero? iszero
match? ismatch
>json tojson
json> fromjson
>word word
pi Math.PI
tau 2 * Math.PI
. print
.s printds
.r printrs
.e print env

The ID resolution priority is:

  1. current env
  2. built-in aliases
  3. built-ins

Word definitions

As in Forth, new words can be defined using the : name ... ; form.

: square ( x -- x*x ) dup * ;

10 square .

Will result in 100.

There're no formatting rules enforced (yet, but under consideration). However, it's strongly encouraged to include stack effect comments as shown in the examples above.

Word definitions MUST be terminated with ;.

Word metadata

The following details are stored in the __meta property of each compiled word. This metadata is not yet used, but stored in preparation for future tooling additions.

  • name - word name
  • loc - source location [line, col]
  • stack - optional stack effect comment
  • arities - optional input/output arities (numeric representation of stack comment)
  • doc - optional docstring (currently unused)

Only the first stack comment in a word definition is kept. If either side of the comment contains a ?, the respective arity will be set to -1 (i.e. unknown). For example:

  • ( -- ? ) - no inputs, unknown output(s)
  • ( -- x ) - no inputs, one output
  • ( a ? -- ) - unknown/flexible number of inputs, no outputs
  • ( a b -- a ) - 2 inputs, 1 output
const ctx = run(`: foo ( -- x ) 42;`);

ctx[2].__words.foo.__meta
// { name: 'foo', loc: [ 1, 1 ], doc: ' -- x', arities: [ 0, 1 ] }

Hyperstatic words

Unlike variables, which are dynamically scoped, words are defined in a hyper-static environment, meaning new versions of existing words can be defined, however any other word (incl. the new version of same word) which uses the earlier version will continue to use that older version. Consequently, this too means that attempting to use undefined words inside a word definition will fail, even if they'd be defined later on. In these cases, use of variables and/or quotations is encouraged to implement dynamic programming techniques.

// hyperstaticness by example
pf.run(`
: foo "foo1" ;
: bar foo "bar" + ;

( redefine foo, incl. use of existing version )
: foo foo "foo2" + ;

( use words )
foo bar
`)[0];
// [ 'foo1foo2', 'foo1bar' ]

Dynamic word creation from quotations

Quotations are used to treat code as data, delay execution of words and/or dynamically compose words. The latter can be achieved via the >word, which takes a quotation and a word name as arguments.

( takes min/max values, produces range check quotation )
( uses local variables `a` and `b`, see section below  )

: range-check ( a b -- q ) ^{ a b } [dup @a >= [@b <=] [drop F] if] ;

( build range check for "0".."9" ASCII interval and define as word `digit?` )
0 9 range-check "digit?" >word

( now use... )
"5" digit? . ( true )
"a" digit? . ( false )

Local variables

A word definition can include an optional declaration of local variables, which are automatically bound to stack values each time the word is invoked. The declarations are given via the form:

: wordname ^{ name1 name2 ... } ... ;

If used, the declaration MUST be given as first element of the word, even before the optional stack comment:

// word with 2 local vars binding: a & b
// when the word is used, first pops 2 values from stack
// and stores them in local vars (in right to left order)
pf.run(`
: add ^{ a b } ( a b -- a+b )
  "a=" @a + .
  "b=" @b + . ;

1 2 add
`);
// a=1
// b=2

See section about variables for further details...

Boolean

The symbol T evaluates as true and F as false.

Numbers

  • 0b... - binary numbers (max 32 bits), e.g. 0b11110100
  • 0x... - hex numbers (max 32bits), e.g. 0xdecafbad
  • decimals (optionally signed and/or scientific notation, e.g. -1.23e-4)

Strings

"Hello world" - no \" escape feature implemented yet

Quotations (Arrays)

Arrays can be contain any valid data literal or symbol and can be arbitrarily nested. Commas optional.

["result: " [2, 3, *] exec +]

Literal quotes

A single element quotation can be formed by prefixing a term with '. Nestable.

  • '+ => [+]
  • ''+ => [[+]]
  • [1 2] => [[1,2]]

Variables

Variables can be looked up & resolved via the currently active environment and scope by prefixing their name with @. Attempting to resolve an unknown var will result in an error.

pf.runU(`@a @b +`, {a: 10, b: 20});
// 30

Assigning a value to a variable (in the the current scope) is done via the ! suffix:

pf.runE(`1 2 + a!`)
// {a: 3}

Furthermore, readonly variables can be defined via words. In this case no prefix must be used and these kind of variables are hyperstatic.

pf.run(`: pi 3.1415 ; "π=" pi + .`);
// π=3.1415

Dynamic scoping

Each variable is resolved via its own stack of binding scopes, and therefore technically results in dynamic scoping. However, in this DSL a new scope is only introduced when a word defines a local var with an already existing name, so in practice the effect is more like lexical scoping.

Var assignment always only impacts the current scope of a var.

// predefined global scope (via env binding)
pf.runU(`@a`, {a: 1});
// 1

// dynamically created global var, then used in quotation
pf.runU(`1 a! [@a @a]`);
// [1, 1]

// var lookup inside word
pf.runU(`: foo @a ; foo`, {a: 1});
// 1

// global & word local vars
// local var (value obtained from stack) takes precendence inside word
pf.runU(`: foo ^{ a } @a ; 2 foo, 3 foo, @a vec3`, {a: 1});
// [2, 3, 1]

// nested local var scopes
// both `foo` & `bar` define a local var `a`
pf.run(`
: foo ^{ a }
  "foo a=" @a + .
  ( since 'a' is declared as local var  )
  ( assignment is only to local scope )
  100 a! ;

: bar ^{ a }
  "bar1 a=" @a + .
  @a inc foo         ( call 'foo' w/ new value )
  "bar2 a=" @a + . ; ( @a still has same value here )

1 bar
"global a=" @a + .   ( global @a never modified )
`, { a: 0 });
// bar1 a=1
// foo a=2
// bar2 a=1
// global a=0

// since `b` is NOT declared as local var inside `foo`
// assigning a value to `b` (even inside `foo`) will be treated as global
pf.runE(`: foo @a b! ; foo`, {a: 1})
// { a: 1, b: 1, __words: { foo: [Function] } }

// here `foo` doesn't declare any locals
// so assignment to `a` will impact parent scope:
// - when `foo` is called from `bar`, bar's `a` var is modified
// - when `foo` is called from root level, global var `a` is created/modified
pf.runE(`
: foo 10 a! ;

: bar ^{ a }
  "before foo a=" @a + .
  foo
  "after foo a=" @a + . ;

1 bar

foo`
);
// before foo a=1
// after foo a=10
{ a: 10 ... }

Objects

Plain objects literals can be created similarly as in JS, i.e.

{key1: value, key2: val2 ...} (again commas are optional)

Keys can be given with or without doublequotes (string literals). Quotes for keys are only needed if:

  • the key contains spaces, has @ prefix or ! suffix
  • is a binary / hex number
  • a number in scientific notation

Furthermore, variables can be used both as keys and/or values:

{@a: {@b: @c}}

// dynamically resolved switch using `bingo` var
src = `{@bingo: ["yay: " @bingo +] default: ["nope"]} switch`;
pf.runU(src, {bingo: 42}, [42]);
// bingo: 42

pf.runU(src, {bingo: 42}, [43]);
// nope

Ideas / Todos

  • add tests
  • tail recursion (help wanted, see #1)
  • async words
  • canvas drawing vocab
  • @thi.ng/atom vocab & integration
  • @thi.ng/rstream vocab & integration

Authors

Karsten Schmidt

If this project contributes to an academic publication, please cite it as:

@misc{thing-pointfree-lang,
  title = "@thi.ng/pointfree-lang",
  author = "Karsten Schmidt",
  note = "https://thi.ng/pointfree-lang",
  year = 2018
}

License

© 2018 - 2021 Karsten Schmidt // Apache Software License 2.0