stick-js

0.5.3 • Public • Published

Intro

Imagine being able to confidently refactor your work and commit it without even checking, strictly separate side-effects from pure functions, enforce immutable operations, mix and match purely referentially transparent expressions and build functors and monads from simple building blocks.

Now put down that Haskell propaganda manual and listen.

Quick-start

node 14.0+

echo '{}' > package.json
yarn add stick-js @babel/{cli,core} babel-plugin-operator-overload
# or npm i
mkdir src
touch babel.config.js src/index.mjs

babel.config.js

module.exports = (api) => {
  api.cache.forever ()
  return {
    plugins: [
      'babel-plugin-operator-overload',
    ],
  }
}

src/index.mjs

defineBinaryOperator ('|',  (...args) => pipe         (...args))
defineBinaryOperator ('<<', (...args) => compose      (...args))
defineBinaryOperator ('>>', (...args) => composeRight (...args))

import {
  pipe, compose, composeRight, map, add, join, concat, concatTo,
} from 'stick-js'

const { log, } = console
const { fromCharCode, } = String

const surroundWith = (x) => concatTo (x) >> concat (x)

; [115, 116, 105, 99, 107, 45, 106, 115]
| map (fromCharCode)
| join ('')
| surroundWith ('**')
| log
node --version // v14.5.0
node_modules/.bin/babel --keep-file-extension -d lib src
node lib/index.mjs // **stick-js**

It does work with earlier versions of Node, but you will probably need to use the .js suffix instead of .mjs and/or use node -r esm ... and webpack -r esm and/or use CommonJS imports (see below). Sorry but the situation is too messy to provide a comprehensive how-to.

webpack

Here is a working webpack configuration.

babel.config.js

module.exports = {
  presets: [
    [
      '@babel/preset-env',
      {
        modules: false,
        // --- remember to import core-js and regenerator runtime at the very beginning of app.js
        useBuiltIns: 'entry',
        // --- yarn add core-js@3
        corejs: 3,
      },
    ],
  ],
  plugins: [
    'operator-overload',
  ],
}

webpack config:

module: {
  rules: [
    {
      test: /\.js$/,
      exclude: /node_modules/,
      use: {
        loader: 'babel-loader',
      },
    },

    ...

  ],
}

You can use the import form in your frontend source files. You will probably need to run webpack using -r esm in the command line.

Modules / CommonJS

We provide both CommonJS and ES style modules. If you use import, you will get ES, and if you use require, you will get CommonJS. Note that if you use ES you must be prepared to deal with the ES modules on your side, for example by using -r esm or a recent version of Node.

Description

  • We provide overloaded operators |, >>, and << and many patterns to use them with, while TC39/proposal-pipeline-operator is still being fleshed out.

    @TC39, if you're listening: the pipeline operator is extremely limited if you don't also provide composition operators. We use >> (right composition, the same order as |), and << for Haskell-style (left) composition.

    The composition operators can help you abstract away the data from a pipeline and utilise a 'point-free form'. See below for many examples of this.

  • Using the overloaded operators is opt-in and you do not need to use them if you don't want to. With a few exceptions, all the functions we provide are useful using a simple data-last calling style as well.

  • Note that we use |, not |>, which is not only easier to overload but far more pleasant to work with, and which will still work even if/when the proposal is accepted.

  • We provide a toolkit of functions centered around this pattern, and a number of idioms which they facilitate. In particular, we provide several ways to replace blocks of statements and control structures with referentially transparent expressions, and work with prototypes, factories, mutable and immutable data, and more.

    You can use our functions, or your own favorite library (Ramda, lodash/fp, etc.), or mix and match as you like.

  • We make it nice to work with manually curried functions (see below).

  • We provide a pure JavaScript (or almost-pure, if you overload the operators) way of using many idioms from Haskell and Racket.

  • Many functions whose arguments and argument order are predictible based on their names and the rules of English grammar.

The overloading is made possible thanks to the great babel-plugin-operator-overload library by Jussi Kalliokoski (@jussi-kalliokoski).

TL; DR

Here are a few examples. Source files must begin with a particular header (omitted here for brevity) -- see 'Overview' section below.


const isInteger = x => x === Math.floor (x)
const ifInteger = isInteger | ifPredicate

; [3.5, 4, 4.2]
| map (ifInteger (
  x => x + 1,
  always ('nothing'),
))
// ['nothing', 5, 'nothing']

import {
  map, join, concat, concatTo,
  modulo, gt, eq,
  againstBoth, ifPredicate,
} from 'stick-js'

const { log, } = console
const esc = concatTo ('\u001b[')
const color = (code) =>
  String >> concatTo (code | concat ('m') | esc) >> concat ('0m' | esc)
const green = '32' | color
const yellow = '33' | color
const odd = modulo (2) >> eq (1)
const oddAndGt4 = againstBoth (odd, gt (4))
const ifOddAndGt4 = oddAndGt4 | ifPredicate

; [3, 4, 5]
| map (ifOddAndGt4 (green, yellow))
| join (' / ')
| log


const convertFahrenheit = fah => lets (
  _ => (fah - 32) / 9 * 5,    // (1) celsius
  (cel) => cel + 273,         // (2) kelvin
  (cel, kel) => [cel, kel],   // (3) result
)

convertFahrenheit (86) // [30, 303]

const checkVal = condS ([
  4 | eq    | guard  (sprintf1 ('%s was exactly 4')),
  4 | lt    | guard  (sprintf1 ('%s was less than 4')),
  4 | gt    | guard  (sprintf1 ('%s was more than 4')),
  otherwise | guardV ("error, this shouldn't happen"),
])

; [3, 4, 5]
| map (checkVal)
| join (' | ')
// 3 was less than 4 | 4 was exactly 4 | 5 was more than 4

// ------ dog.js
const proto = { speak () { 'Hi from ' + this.name }}
export default proto | factory

// ------ main.js
import Dog from './dog'
Dog.create ({ name: 'Caesar', }).speak () // 'Hi from Caesar'

And much more.


Examples

  • A WebGL example, to demonstrate that stick is fast enough to use for WebGL / animations with many objects created by factories. (source, based on a rewrite of the amazing raindrops demo by Lucas Bebber (@lebber).

Overview

٭ basic example ٭

// --- source files must begin with this header.
defineBinaryOperator ('|',  (...args) => pipe         (...args))
defineBinaryOperator ('<<', (...args) => compose      (...args))
defineBinaryOperator ('>>', (...args) => composeRight (...args))

import {
    pipe, compose, composeRight,
// --- /header
    map, join, sprintf1,
} from 'stick-js'

import { green, } from 'chalk'

const { log, } = console

; [1, 2, 3]
| map (x => x + 1)
| join ('/')
| green
| sprintf1 ('The answer is %s')
| log // outputs 'The answer is 2/3/4' (colorfully)

٭ the 'stick' operator ٭

a | b is simply an equivalent way of writing b (a)

(What if I really want to do bitwise math, you ask? Don't worry, you still can: see below).

// --- reminder: source files must begin with this.
// --- from here on out we'll omit it in the examples.
defineBinaryOperator ('|',  (...args) => pipe         (...args))
defineBinaryOperator ('<<', (...args) => compose      (...args))
defineBinaryOperator ('>>', (...args) => composeRight (...args))

import {
  pipe, compose, composeRight,
  map, join, split,
} from 'stick-js'

const multiply = x => y => x * y
const double = multiply (2)     // or 2 | multiply

3 | double                      // 6
double (3)                      // 6
3 | multiply (4)                // 12

const capitaliseFirstLetter = x => x[0].toUpperCase () + x.slice (1)

; `just a perfect day`
| split (' ')                 // split (' ') is a function
| map (capitaliseFirstLetter) // map (capitaliseFirstLetter) is also a function
| join (' ')                  // ... you get the picture.
// 'Just A Perfect Day'

٭ currying styles ٭

All curried functions provided by stick-js can be called using either of 2 currying styles.

This would be a good time to read up on curried functions if you're not familiar with them.

  1. we will refer to this sort of function and calling style as 'manual':

     const f = a => b => c => a + b + c          // call like f (1) (2) (3)
    
  2. and this sort as 'normal':

     const g = R.curry ((a, b, c) => a + b + c) // call like g (1) (2) (3)
                                                // or g (1, 2, 3)
                                                // or g (1, 2) (3)
                                                // or g (1) (2, 3)
                                                // etc.
    

Calling:

import { map, } from 'stick-js'
const double = x => x * 2

map (double, [1, 2, 3])    // [2, 4, 6] (normal style)
map (double) ([1, 2, 3])   // [2, 4, 6] (manual style)
; [1, 2, 3] | map (double) // [2, 4, 6] (manual style with stick)

For extra performance you can also limit yourself to the manual style (see below).

٭ a note on style ٭

We recommend using a space before the parentheses of a function call. Admittedly it looks pretty strange at first, but we find that it makes everything much clearer when you get used to it, in particular with the manual currying style.

You might also want to check out a vertical alignment plugin. The author uses vim-easy-align by @junegunn.

٭ markers ٭

import { sprintfN, sprintf1, } from 'stick-js'

3 | sprintf1 ('4 - 1 is %s') // '4 - 1 is 3'

'N' is a marker meaning an array is expected.

; [4, 3]
| sprintfN ('%s - 1 is %s')  // same.

'V' means a value is expected, to disambiguate cases where a function also fits.

import { timesV, timesF, } from 'stick-js'

const { random, } = Math

3      | timesV (4)          // [3, 3, 3, 3] (`V` stands for 'value')
random | timesF (4)          // [<random-num>, <random-num>, <random-num>, <random-num>] (`F` stands for 'function')
random | timesV (4)          // [random, random, random, random]

Note that the last one stores the function 4 times in the array.

random | timesV (4) | map (invoke) // [<random-num>, <random-num>, <random-num>, <random-num>]
import { appendTo, appendToM, } from 'stick-js'

const a = [1, 2, 3]
const b = 4 | appendTo (a) // functional style: the array is cloned.
b === a // false

const a = [1, 2, 3]
const b = 4 | appendToM (a) // non-functional style: the array is mutated (`M` stands for 'mutable')
b === a // true

const webGLContext = { ... a complicated object ... }
webGLContext | mergeM ({ someProp: false, }) // you probably want mutable here.

And there are a few more which we'll see along the way.

٭ ok, anaphoric if ٭

ok (x) is false if x is null or undefined. Every other input returns true.

import { map, ok, notOk, ifOk, } from 'stick-js'

const { log, } = console

; [0, false, '', null, void 8]
| map (ok)    // [true, true, true, false, false]

; [0, false, '', null, void 8]
| map (notOk) // [false, false, false, true, true]

Something we see a lot in JS is:

let answer
if (someVar !== undefined && someVar !== null) {
  answer = someVar + 1
} else {
  answer = 'nothing'
}

This can vastly improved using an 'anaphoric if' and a stick idiom:

import { add, always, } from 'stick-js'

const add1 = 1 | add                                     // or add (1)
const answer = someVar | ifOk (add1, 'nothing' | always)

Explanation: ifOk takes two functions -- a 'then' function and an 'else' function. In the 'ok' case, the value being tested is passed to the function. (This bound value is sometimes called an 'anaphor').

const add1IfYouCan = val => val | ifOk (
  // `that` refers to `val`
  that => that + 1,

  // no value is passed.
  _ => 'nothing',
)

This can be further condensed, using 'point-free style' (see below) and the always function:

const add1IfYouCan = ifOk (1 | add, 'nothing' | always)

Usage:

const answer = someVar | add1IfYouCan

; [0, 10, null, void 8]
| map (add1IfYouCan) // [1, 11, 'nothing', 'nothing']

٭ point-free ٭

A common pattern is when the argument to a function is passed immediately into a pipe:

const add1IfYouCan = x => x | ifOk (add1, 'nothing' | always)

Since x does not appear anywhere else in the expression, we can simply remove it, along with the function argument:

const add1IfYouCan = ifOk (add1, 'nothing' | always)

This is often called 'point-free' style. It means the function arguments and the data being passed through the pipeline have been abstracted away (and has nothing to do with dots, despite the name).

If the pipe chain consists of more than 1 link …

const { dot, sprintf1, tap, }
const { log, } = console

const add1IfYouCan = x => x
  | ifOk (add1, 'nothing' | always)
  | String                        // conversion using type constructor
  | dot ('toUpperCase')
  | sprintf1 ('The answer is %s')
  | tap (log)                     // outputs 'The answer is 1', 'The answer is NOTHING', ...

… then we remove the x => x and change all the | to >>

const add1IfYouCan =
  ifOk (add1, 'nothing' | always)  // (1)
  >> String                        // (2)
  >> dot ('toUpperCase')           // (3)
  >> sprintf1 ('The answer is %s') // (4)
  >> tap (log)                     // (5)

The following pattern always holds:

a | b | ... | n = a | (b >> ... >> n)

So when the chains start to get long (as above), you can cut pieces out using this property. For example, you can refactor lines 2-4 into a new function:

// --- convert input to String, make uppercase, perform sprintf.
const processString = String
  >> dot ('toUpperCase')
  >> sprintf1 ('The answer is %s')

And splice it back in:

const add1IfYouCan = x => x
  | ifOk (add1, 'nothing' | always)
  | processString
  | tap (log)

Or

const add1IfYouCan =
  ifOk (add1, 'nothing' | always)
  >> processString
  >> tap (log)

٭ compositional predicates ٭

ifOk is a convenience for ifPredicate (ok) or ok | ifPredicate.

There is also a 'when' form, which has no 'else' branch.

import { add, whenOk, } from 'stick-js'

const add1 = 1 | add // or add (1)

3    | whenOk (add1) // 4
null | whenOk (add1) // undefined

The selection of if and when functions we provide is intentionally skimpy, to encourage you to compose your own.

const { floor, } = Math

// --- predicate to match integers.
const isInteger = x => x === floor (x)

// --- or how about
// import { eq, } from 'stick-js'
// const isInteger = x => x | floor | eq (x)

// --- or if you're getting bored:
// import { timesV, asteriskN, passToN, } from 'stick-js'
// const arrowSnd = f => timesV (2) >> asteriskN ([id, f])
// const isInteger = arrowSnd (floor) >> passToN (eq)

// --- now compose it into an anaphoric if:

const ifInteger = isInteger | ifPredicate

// --- now use it:

const add1 = add (1)

; [3.5, 4, 4.2]
| map (ifInteger (add1, 'nothing' | always))
  // ['nothing', 5, 'nothing']

More complicated predicates:

const both = (f, g) => againstAll ([f, g])

const isOdd = x => x % 2

const isOddInteger = both (isInteger, isOdd)
const ifOddInteger = isOddInteger | ifPredicate

; [3.5, 4, 5, 5.5]
| map (ifOddInteger (
  add1,
  'nothing' | always,
))
// ['nothing', 'nothing', 6, 'nothing']

; [3.5, 4, 5, 5.5] | map (isOddInteger)
// [false, false, 1, false]

Note that againstAll returns the return value of the last matching function, hence 1 in the last example. This is because it can be useful to have that value, and it's analogous to how the short-circuit && works.

To convert a truthy expression to a strict one: simply compose it with the type constructor Boolean.

const both = (f, g) => againstAll ([f, g]) >> Boolean

Or, you could add === true to the end of the isOdd function.

Note that ifPredicate and whenPredicate match on truthiness, not strict truth. Rationale: that is how the native filter works, and JS's if operator, and many other tools in e.g. Ramda. And converting truthy to strict Boolean is trivial.

٭ compositional decoration ٭

Our map function is capped at one argument, meaning the map routine only gets the value and not the index or the collection.

import { map, addIndex, addCollection, } from 'stick-js'

; [4, 5, 6]
| map ((x, idx) => idx) // [undefined, undefined, undefined]

But:

const mapWithIndex      = map | addIndex
const mapWithCollection = map | addCollection

; [4, 5, 6]
| mapWithIndex ((x, idx) => idx) // [0, 1, 2]

; [4, 5, 6]
| mapWithCollection ((x, coll) => coll) // [[4, 5, 6], [4, 5, 6], [4, 5, 6]]

; [4, 5, 6]
| (map | addIndex | addCollection) ((x, idx, coll) => ...)

; [4, 5, 6]
| (map | addCollection | addIndex) ((x, coll, idx) => ...)

We can also enhance our merge functions, to deal with conflicts:

import { mergeWith, mergeToSym, } from 'stick-js'

const chooseTgt = (src, tgt) => tgt
const chooseSrc = (src, tgt) => src

// --- choose target value on conflict
const mergeChooseTgt = mergeToSym | mergeWith (chooseTgt)

// --- choose source value on conflict
const mergeChooseSrc = mergeToSym | mergeWith (chooseSrc)

const os = { name: 'source name', }
const ot = { name: 'target name', }

os | mergeChooseSrc (ot) // 'source name'
os | mergeChooseTgt (ot) // 'target name'

Or to only merge if certain conditions hold:

import { mergeWhen, eq, } from 'stick-js'

const { floor, } = Math
const isInteger = x => x | floor | eq (x)
const srcIsInteger = (src, tgt) => src | isInteger
const mergeToWhenSrcIsInteger = mergeToSym | mergeWhen (srcIsInteger)

const os = { val: 2.2, vil: 3,  vol: 3.5, }
const ot = { val: 25,  vil: 25, vol: 25,  vel: 42, }

os | mergeTo (ot)                 // { val: 2.2, vil: 3, vol: 3.5, vel: 42, }
os | mergeToWhenSrcIsInteger (ot) // { val: 25,  vil: 3, vol: 25,  vel: 42, }

٭ semantics and argument order are based where possible on English grammar ٭

(We're dying to see a port to Hungarian, too)

Once you master this, the usage becomes intuitive and greatly reduces the need to look things up.

(Motivation: what's the argument order of append, times, and subtract in your favorite library?)

import {
  sprintfN, sprintf1, mergeTo, merge, prependTo, prepend,
  appendTo, append, bindPropTo, bindProp, bindTo, bind, invoke,
} from 'stick-js'

const tgt = { thing: 'sandwich', want: 'no thanks',  }
const src = {                    want: 'yes please', }

When a function name ends in a preposition, for example, ‘To’, the identifier to the right is the object of the preposition.

Read this as ‘merge src to tgt

src | mergeTo (tgt)                  // { thing: 'sandwich', want: 'yes please', }

The same function name without the preposition means that the identifier to the right is the object of the verb ‘merge’.

Read this as: ‘merge src to tgt’

tgt | merge (src)                    // { thing: 'sandwich', want: 'yes please', }

4 | appendTo ([1, 2, 3])             // [1, 2, 3, 4]
; ([1, 2, 3]) | append (4)           // [1, 2, 3, 4]

0 | prependTo ([1, 2, 3])            // [0, 1, 2, 3]
; ([1, 2, 3]) | prepend (0)          // [0, 1, 2, 3]

const dog = {
  name: 'Caesar',
  speak () { return 'My name is ' + this.name },
}

const cat = {
  name: 'Fritz',
  speak () { throw new Error },
}

const f = 'speak' | bindPropTo (dog)
f ()                                  // 'My name is Caesar'

// --- 'bind prop "speak" to object'
'speak' | bindPropTo (dog)   | invoke // same

// --- also 'bind prop "speak" to object'
dog     | bindProp ('speak') | invoke // same

dog.speak | bindTo (dog) | invoke     // same

// cat.speak ()                       // Error
dog.speak | bindTo (cat) | invoke     // 'My name is Fritz'
cat | bind (dog.speak)   | invoke     // 'My name is Fritz'

// --- 'call this function on this context', i.e., bind and call.
; ({}.toString) | callOn (3)          // '[object Number]'

// --- 'provide this context to this function'
; 3 | provideTo ({}.toString)         // '[object Number]'

Note that this fits the a | b | c pattern:

dog.speak | bindTo (cat) | invoke     // 'My name is Fritz'

So we can also write it as a | (b >> c):

dog.speak | (bindTo (cat) >> invoke)  // 'My name is Fritz'

In fact bindTo (x) >> invoke is already provided under the name callOn, and its inverse provideTo:

dog.speak | callOn (cat)              // 'My name is Fritz'
cat       | provideTo (dog.speak)     // 'My name is Fritz'

Some other miscellaneous examples.

// --- '3 to the 4th'
3 | toThe (4)                         // 81

// --- 'divide 3 by 6' or '3 divided by 6'
3 | divideBy (6)                      // 0.5

// --- 'divide 3 into 6'
3 | divideInto (6)                    // 2

// --- 'subtract 3 from 4'
3 | subtractFrom (4)                  // 1

// --- 'subtract 4 from 3'
3 | subtract (4)                      // -1

// --- '3 minus 4'
3 | minus (4)                         // -1

٭ side effects & chaining ٭ mutable vs immutable ٭

import {
  map, side1, appendM, append, prependM, prepend,
} from 'stick-js'

Chaining with the . will often not do what you want.

; [2, 3, 4]
  .push (5)
  .unshift (1) // error, return value of previous line was 5

But this will:

const push    = 'push'    | side1
const unshift = 'unshift' | side1

The 1 in side1 refers to the arity of the function, i.e., the exact number of arguments it expects. When working with functions in this way it's important to specify this.

.push and .unshift, both methods of Array.prototype, expect exactly one argument, hence side1 in both cases.

; [2, 3, 4]
| push (5)
| unshift (1) // [1, 2, 3, 4, 5]

// --- using stick functions for mutable data:

; [2, 3, 4]
| appendM (5)
| prependM (1)

// --- using stick functions for immutable data:

; [2, 3, 4]
| append (5)  // new array [2, 3, 4, 5]
| prepend (1) // new array [1, 2, 3, 4, 5]

You can insert tap anywhere in the chain, which is guaranteed not to mess with the pipeline no matter what it returns:

const double = x => x * 2

; [1, 2]
| map (double)               // [2, 4]
| tap (x => console.log (x)) // still [2, 4], performs side-effect (printing)
| tap (x => [100, 200])      // still [2, 4], for demonstration only
| join (',')
// '2,4'

Without the first tap, this would have been an error, because console.log returns undefined.

tap is useful to signal the intention of performing side effects or IO. It is also really useful for debugging. And our side family of functions use tap under the hood.

٭ factory ٭ synopsis ٭

import {
  factory, factoryProps,
  factoryStatics,
} from 'stick-js'

// --- dog.js:
const proto = {
  init () {
    ...
    return this
  },
  breathe () { return 'huff' },
  whoami ()  { return this.name },
  getType () { return this.type },
}

const props = {
  type: 'dog',
  name: undefined,
}

// --- basic:
export default proto | factory

// --- variants:
export default proto | factory | factoryProps (props)
export default proto | factory | factoryStatics ({ ... }) | factoryProps (props)

// ------ main.js

import Dog from './dog'

const dog = Dog.create ()
// const dog = Dog.create ().init () // useful in practice
dog.breathe ()                       // 'huff'
dog.type                             // 'dog', if `factoryProps` used
dog.getType ()                       // 'dog', same
dog.whoami ()                        // undefined, because no default.

const dog2 = Dog.create ({ name: 'garfunkel', })
dog2.whoami ()                       // 'garfunkel', thanks to args to create

٭ factory ٭ with mixins ٭ synopsis ٭

// ------ animal.js:

import {
  factory, factoryStatics, mixinM, mixinPreM,
  factoryProps,
  ifPredicate,
} from 'stick-js'

const isOdd = x => x % 2 !== 0
const ifOdd = isOdd | ifPredicate

// --- a 'base' object (animal)

const proto = {
  init () {
    ...
    return this
  },
  move () {
    return this.numLegs | ifOdd (
      _ => 'hobble',
      _ => 'gait',
    )
  },
  breathe () { return 'huff' },
  speak ()   { 'not implemented' | die },
  getType () { return this.type },
}

const props = {
  type: 'animal',
  numLegs: undefined,
}

export default proto | factory | factoryProps (props)

// ------ cheater.js:

// --- some orthogonal functionality

const proto = {
  cheat: howMuch => 'I cheat ' + howMuch,
}

export default proto | factory

// ------ dog.js:

import {
  sprintf1,
  factory,
} from 'stick-js'

import Animal from './animal'
import Cheater from './cheater'

const { proto: animalProto, } = Animal
const { proto: cheaterProto, } = Cheater

// --- a composite object (dog), extended from animal, with extra functions mixed in.

const proto = {
  init () {
    ...
    return this
  },
  speak () { return this.name | sprintf1 ('Dog %s says woof') }
}

const props = {
  type: 'dog',
  numLegs: 4,
  name: undefined,
}

export default proto
  | mixinPreM (animalProto)
  | mixinM (cheaterProto)
  | factory
  | factoryProps (props)

// ------ main.js

import Dog from './dog'

const dog = Dog.create ({ name: 'garfunkel', })
// const dog = Dog.create ({ name: 'garfunkel', }).init () // with init
dog.breathe ()                                  // 'huff' (from animal)
dog.getType ()                                  // 'dog' (function from animal, property from dog)
dog.speak ()                                    // 'Dog garfunkel says woof' (function from dog, property from instance initialisation)
dog.cheat ('a bit')                             // 'I cheat a bit' (from cheater)

Stick idioms:

const breathe = dot ('breathe')
const getType = dot ('getType')
const speak   = dot ('speak')
const cheat   = dot1 ('cheat')
const init    = side ('init')
const create  = dot1 ('create')

const dog = Dog
  | create ({ name: 'garfunkel', })
  | init

dog | breathe
dog | getType
dog | speak

Dog | create ({ name: 'garfunkel', })
    | init
    | cheat

٭ factory ٭ explained

We provide a functional style for working with objects the way JS was designed to: using prototypical inheritance and Object.create. We hope to show you that the new keyword and 'classes' and all the baggage they bring are unnecessary, and that they obfuscate the way that it actually works.

We provide a simple abstraction for factories, with as little sugar and magic as possible, and encourage you to mix and match the components to do exactly what you need.

To recap: you create an object in JS by first building a prototype object, consisting of only functions.

const animalProto = {
  breathe () { return 'huff' },
  speak () { throw new Error },
  ...
}

To make an animal instance, you pass this prototype object to Object.create, then assign some methods and/or properties. If you wish you can use Object.create again (and again) before assigning properties, copy in some more values, use Object.create again, and so on.

(Nothing will stop you from using Object.create on an object with non-method properties in it, and it will probably do what you want, but it's best avoided -- add your properties to the last object in the chain).

We encapsulate this process with the notion of a factory, which is an object which knows how to spawn objects of a certain sort.

import { factory, } from 'stick-js'

const animalProto = {
  breathe () { return 'huff' },
  speak () { throw new Error },
  ...
}

const Animal = animalProto | factory // other idioms might call it `animal` or `animalFactory`
const animal1 = Animal.create ()
const animal2 = Animal.create ()
animal1.breathe () // 'huff'
animal2.breathe () // 'huff'
animal2.speak () // Error

To add properties:

const animalProps = {
  type: 'animal',
  size: undefined,
  numLegs: undefined,
}

And we recommend always having an init method, which you will almost certainly need. myFactory.create ().init () becomes a well-worn pattern.

const isOdd = x => x % 2 !== 0
const ifOdd = isOdd | ifPredicate

const animalProto = {
  init () {
    ...
    return this
  },
  breathe () { return 'huff' },
  speak () { throw new Error },
  move () {
    return this.numLegs | ifOdd (
      _ => 'hobble',
      _ => 'gait',
    )
  },
}

const Animal = animalProto | factory | factoryProps (animalProps)

On create, the properties which are 'ok' will get copied in to the new object.

The others are there for documentation: put them here, not peppered throughout the methods. Do use undefined for props that are waiting to be defined, which is arguably better than null and definitely better than false. Use void 8 or your very own favorite number to impress … no one.

const animal = Animal.create ().init ()
animal.type // 'animal'
animal.size // undefined

You can pass an object to create to initialise properties. These will be copied in after the props that were passed to factoryProps are.

const bigBiped = Animal.create ({ size: 'big', numLegs: 2, }).init ()

If for some reason you need access to your newly minted object before create is called, you can use factoryInit instead of factory. This is how factoryProps is implemented internally.

You can also eliminate the dots entirely:

const create = dot1 ('create')
const init   = side ('init')
const move   = dot  ('move')
const speak  = dot  ('speak')

Animal | create ({ size: 'small', numLegs: 2, })
       | init
       | move // 'gait'

Note that by using side for init we are assured that the instance is passed down through the pipe and not the return value of init (although in this case, init returns this, so dot would have worked too.) create and move definitely need dot and not side.

To extend Animal to the obligatory Dog find the Animal prototype (if animalProto is not in scope).

const animalProto = Animal.proto

Or

const animalProto = Animal.create ().__proto__ // might not be available in all runtimes

Create it, add dog methods, and make a new factory:

const dogProto = animalProto
  | Object.create
  | mergeM ({
    speak () { return this.loud ? 'WOOF' : 'woof' },
  })

const dogProps = {
  numLegs: 4,
  loud: undefined,
}

const Dog = dogProto | factory | factoryProps (dogProps)

; [true, false]
| map ((isLoud) => Dog
  | create ({ loud: isLoud, })
  | (dog => [dog.speak (), dog.breathe (), dog.move ()])
)
// [['WOOF', 'huff', 'gait'], ['woof', 'huff', 'gait']]

Note that we can call methods of both Animal and Dog now.

٭ factory ٭ with mixins ٭ explained ٭

const Dog = dogProto | mixinM (animalProto) | factory | factoryProps (dogProps)

Working with mixins is tricky. At some point, there will be namespace conflicts and it's not always obvious which version should win out -- and you have to decide how you want to deal with that.

But you have all the tools now to specify exactly how you want it to work.

const Dog = dogProto | mixinPreM (animalProto) | factory | factoryProps (dogProps)
const dog1 = Dog | create ({})
dog1 | breathe // 'huff'
const Dog = dogProto | mixinPreM (animalProto) | factory | factoryProps (dogProps)
Dog.create ({}) | speak // 'woof'

We mixed the animal into the dog as a 'pre' mixin, meaning that on name conflicts, Dog's version will win. If we had used mixinM instead of mixinPreM:

const Dog = dogProto | mixinM (animalProto) | factory | factoryProps (dogProps)
Dog | create ({})
    | speak // Error, this is Animal's version.

Non-pre mixins are useful for orthogonal functionality -- something like logging, for example.

You can add as many pre and post mixins as you like.

dogProto | mixinPreM (...) | ... | mixinM (...) | ... | factory

Or use the 'N' versions to provide an array:

dogProto | mixinPreNM ([a, b, c]) | mixinNM ([d, e, f]) | factory

Note the 'M' marker on the mixin functions. This is to make clear that these functions mutate the prototype object, which might be a bit surprising when using this style.

We do not provide non-M versions of the mixin functions, because it's not obvious exactly what the semantics should be, as several alternatives could be equally intuitive: should the prototype chain be flattened? should it be discarded, leaving only the own keys? should it create a new object using and mutate it?

You can specify these behaviors explicitly:

dogProto | flattenPrototype | <mixin functions ...> |
dogProto | discardPrototype | <mixin functions ...> |
dogProto | Object.create    | <mixin functions ...> |

٭ let expressions ٭

You can consolidate a number of assignment statements into a single let expression, and also limit the scope of the assignments in a way which is easy to read. Code which is based on expressions rather than blocks of statements can be made referentially transparent, and therefore much easier to read, refactor, & prove the correctness of, especially considering that each statement is a possible side-effect inducing timebomb.

// --- convert a celsius value to both fahrenheit & kelvin.

const convertCelsius = (c) => letV (
  c / 5 * 9 + 32,           // fahrenheit
  c - 273,                  // kelvin
  (fah, kel) => [fah, kel],
)

letV takes an arbitrary number of values, and expects the last one to be a function. It simply passes the values in order to the function.

There is an 'N' form (letVN, which takes an array of values and one function).

Far more useful is lets and letS. (Think let* in racket).

If we were to go the other way, from Fahrenheit to Celsius & Kelvin:

// --- convert fahrenheit to celsius & kelvin

const convertFahrenheit = (f) => letV (
  (f - 32) / 9 * 5,      // celsius
  (f - 32) / 9 * 5 + 273 // kelvin
  (cel, kel) => [cel, kel],
)

We see that we are wasting work in the second line, because with letV there is no way to capture that intermediate expression to avoid calculating it again. With lets, however, there is:

// --- convert fahrenheit to celsius & kelvin

const convertFahrenheit = fah => lets (
  _ => (fah - 32) / 9 * 5,    // (1) celsius
  (cel) => cel + 273,         // (2) kelvin
  (cel, kel) => [cel, kel],   // (3) result
)

lets expects each line to be a function. The first line is called with no argument. The result (1) is passed as the argument to (2). The result of (1) and the result of (2) are passed as the arguments to (3), and so on. The result of the expression is the result of the last function.

(Implementation detail: lets and letS will remain fast with up to 6 arguments. After that we use a slower, more generic algorithm.)

And of course there is a stick version of lets called letS. Think of the 'S' marker as 'stick enabled' and the 's' as 'stick disabled'. letS expects a value to be piped in. Note: the 'S' marker implies the 'N' marker: the arguments must be an array, or else it would be impossible to curry.

const convertFahrenheit = fah => fah | letS ([
  (fah) => (fah - 32) / 9 * 5,     // (1) celsius
  (fah, cel) => cel + 273,         // (2) kelvin
  (fah, cel, kel) => [cel, kel]    // (3) result
])

This wouldn't be the most natural use of letS, but it shows how it works: Function (1) receives as a single argument the piped in value (fah in this case). Function (2) receives fah, and the result of (1). (3) receives fah, the results of (1) and (2), and so on.

By now we know that we can remove fah => fah | from the first line to make it point-free, and we can use underscores to indicate ignored values:

const convertFahrenheit = letS ([
  (fah) => (fah - 32) / 9 * 5,
  (_, cel) => cel + 273,
  (_, cel, kel) => [cel, kel],
])

As an exercise we could try to make the entire expression as point-free as possible, at the possible expense of everyone's sanity:

import { letS, minus, divideBy, multiply, add, arg1, list, tail, } from 'stick-js'

const convertFahrenheit = letS ([
  minus (32) >> divideBy (9) >> multiply (5),
  arg1 >> add (273),
  list >> tail,
])

convertFahrenheit (86) // [30, 303]

٭ exceptions ٭ try/catch ٭

Note: decorateException has been removed -- use decorateRejection from alleycat-js instead.

import { tryCatch, } from 'stick-js'

const dubiousFunction = ...

dubiousFunction | tryCatch (
  // --- no exception: `v` is the return value of `dubiousFunction`
  (v) => ...,

  // --- exception thrown: `e` is the exception
  (e) => e | decorateException ('Dubious function said:')
)

Sometimes it's nice to rethrow an exception, with a string prefixed to it so you can tell what went wrong:

  (e) => e | decorateException ('Dubious function said:') | die

const throwError = reason => reason
  | exception // new Error (reason)
  | raise     // throw it

const throwError = reason => reason | (exception >> raise)

// `exception >> raise` is also known as `die`

Note that this is illegal in JS, because throw is not an expression:

const throwError = reason => throw new Error (reason)

But we can trick it like this:

const throwError = reason => reason | die

// or just
// const throwError = die

Now we can improve this common code:

let answer
try {
    answer = dubiousFunction () + 10
} catch (e) {
    console.warn (e)
    answer = 'bad news'
}

like this:

const answer = dubiousFunction | tryCatch (
  plus (10),
  (e) => {
      e | decorateException ("That didn't go well")
        | console.warn
      return 'bad news'
  },
}

Or perhaps:

const answer = dubiousFunction | tryCatch (
  plus (10),
  decorateException ("That didn't go well")
    >> tap (console.warn)
    >> ('bad news' | always)
)

٭ cond ٭

import {
  cond, condN, condS, guard, guardV, sprintf1, otherwise,
} from 'stick-js'

Naive form:

cond (
  [() => 3 === 4, () => 'strange'],
  [() => 3 === 5, () => 'even stranger'],
  [() => true, () => 'ok'],
) // 'ok'

() => true is the fallback case and can also be written T.

An arbitrary number of lines can be provided.

Using guard and otherwise:

cond (
  (() => 3 === 4) | guard (() => 'strange'),
  (() => 3 === 5) | guard (() => 'even stranger'),
  otherwise       | guard (() => 'ok'),
) // 'ok'

If the guard functions return simple expressions, guardV can be more convenient:

cond (
  (() => 3 === 4) | guardV ('strange'),
  (() => 3 === 5) | guardV ('even stranger'),
  otherwise       | guardV ('ok'),
)

The most useful version is condS. Remember, 'S' implies 'N', so give it an array.

const checkVal = val => val | condS ([
  eq (4)    | guard (val => val | sprintf1 ('%s was 4')),
  lt (4)    | guard (val => val | sprintf1 ('%s was less than 4')),
  gt (4)    | guard (val => val | sprintf1 ('%s was more than 4')),
  otherwise | guardV ("error, this shouldn't happen"),
])

Cleaning it up a bit, and inverting the parenthetical test expression to use sticks:

const checkVal = condS ([
  4 | eq    | guard  (sprintf1 ('%s was 4')),
  4 | lt    | guard  (sprintf1 ('%s was less than 4')),
  4 | gt    | guard  (sprintf1 ('%s was more than 4')),
  otherwise | guardV ("error, this shouldn't happen"),
])

Wut?! This does work, strange as it looks. Try it for yourself:

; [3, 4, 5, {}]
| map (checkVal)
| join (' | ')
// 3 was less than 4 | 4 was 4 | 5 was more than 4 | error, this shouldn't happen

٭ extended regexes ٭

We provide regex functions which fit the idiom and mostly have an 'X' in the name (think /x if you know Perl).

These are 'extended regexes', in which whitespace is ignored. Note that we do not skip comments in the regex, only whitespace, even inside character classes. Use \s if you really want to match on whitespace.

'egbert' | xMatch (/ (eg) (..) [rs] /)
// [ 'egber', 'eg', 'be', index: 0, input: 'egbert' ]

// --- same, using a regex string instead of a RegExp literal.
'egbert' | xMatchStr (' (eg) (..) [rs] ')
// [ 'egber', 'eg', 'be', index: 0, input: 'egbert' ]

const vowels = []
const mapper = appendToM (vowels)
'egbert druppelvanger' | xMatchGlobal (/ [ae] /) (mapper)
vowels // [ 'e', 'e', 'e', 'a', 'e' ]

'egbert\ndruppelvanger' | xMatchStrFlags (' ^ d ') ('s')
// null

'egbert\ndruppelvanger' | xMatchStrFlags (' ^ d ') ('m')
// [ 'd', index: 7, input: 'egbert\ndruppelvanger' ]

const toUpper = dot ('toUpperCase')

const ifReplaceVowels = ifXReplace (/ ([aeiou]) /) ('x')
'egbert' | ifReplaceVowels (toUpper) ('bad' | always)
// XGBXRT

const ifReplaceVowelsGlobal = ifXReplace (/ ([aeiou]) /g) ('x')
'egbert' | ifReplaceVowelsGlobal (toUpper) ('bad' | always)
// XGBXRT

// --- same, using a regex string instead of a RegExp literal.
const ifReplaceVowelsGlobalAlt = ifXReplaceStrFlags (' ([aeiou]) ') ('g') ('x')
'egbert' | ifReplaceVowelsGlobalAlt (toUpper) ('bad' | always)
// XGBXRT

٭ all ٭ any ٭

If you're like the author you have a hard time remembering how Ramda's all, any, both, either, and, or, anyPass, and so on work. Does the 'all' refer to all the functions? or all the values? and so on.

Our solution to making the semantics clearer and enriching the functionality of these functions centers around the preposition 'against': you match a value against a predicate.

const isOdd = x => isInteger (x) && (x % 2 !== 0)
const isLt3 = 3 | lt

// --- return `true` if input is truthy, `false` otherwise.
const truthy = id
// const truthy = Boolean // also works

Match all values against the predicate isOdd

; [1, 2, 3, 4, 5] | allAgainst (isOdd) // false
; [1, 3, 5] | allAgainst (isOdd) // true

Match all values against the predicate truthy

const allTruthy = allAgainst (truthy)
; [1, 2, 3, 4, 5, null] | allTruthy // false
; [1, 2, 3, 4, 5] | allTruthy  // true

The value matches all the predicates

const isOddAndLt3 = againstAll ([isOdd, isLt3])
1   | isOddAndLt3 // true
1.1 | isOddAndLt3 // false
2   | isOddAndLt3 // false
3   | isOddAndLt3 // false

The value matches any of the predicates

const isOddOrLt3 = againstAny ([isOdd, isLt3])
2   | isOddOrLt3 // true
3   | isOddOrLt3 // true
4   | isOddOrLt3 // false
5   | isOddOrLt3 // true

We can also match multiple values against multiple predicates, in various combinations of 'all' and 'any' (Ramda doesn't have this out of the box -- corrections welcome)

All the values are odd and less than 3.

const allOddAndLt3 = allAgainst (isOddAndLt3)
; [1, 2, 3, 4, 5] | allOddAndLt3 // false
; [1, 3, 5] | allOddAndLt3 // true

All the values are odd and less than 3.

const allOddOrLt3 = allAgainst (isOddOrLt3)
; [1, 2, 3, 4, 5] | allOddOrLt3 // false
; [2, 3, 5] | allOddOrLt3 // true

Any of the vales are odd and less than 3.

const anyOddAndLt3 = anyAgainst (isOddAndLt3)
; [3, 4, 5] | anyOddAndLt3 // false
; [2, 4, 5] | anyOddAndLt3 // false
; [1, 4, 5] | anyOddAndLt3 // true

Any of the vales are odd or less than 3.

const anyOddOrLt3 = anyAgainst (isOddOrLt3)
; [4, 6] | anyOddOrLt3 // false
; [3, 6] | anyOddOrLt3 // true
; [3, 5] | anyOddOrLt3 // true
; [2, 6] | anyOddOrLt3 // true
; [1, 6] | anyOddOrLt3 // true

Both values are truthy / either value is odd. Note that these forms take spread out arguments, not an array (so they are sugar for allAgainst / anyAgainst)

const bothTruthy = bothAgainst (truthy)
const eitherOdd = eitherAgainst (isOdd)

const isOddAndLt3Alt = againstBoth (isOdd) (isLt3)
const isOddOrLt3Alt = againstEither (isOdd) (isLt3)

// --- i.e. ; [null, 3] | allTruthy
bothTruthy (null, 3) // false
bothTruthy (1, 3) // 3

eitherOdd (1, 2) // true
eitherOdd (null, 2) // false
eitherOdd (2, 4) // false

1   | isOddAndLt3Alt // true
1.1 | isOddAndLt3Alt // false
2   | isOddAndLt3Alt // false
3   | isOddAndLt3Alt // false

2   | isOddOrLt3Alt // true
3   | isOddOrLt3Alt // true
4   | isOddOrLt3Alt // false
5   | isOddOrLt3Alt // true

٭ merging ٭

We provide 8 basic merge functions, corresponding to all combinations of three binary choices:

  1. to vs. from (order of arguments)
  2. 'own' vs. 'in' (prototype values)
  3. mutable vs. immutable (whether to clone the target first)

There are a few conventions to keep in mind when trying to understand the semantics. To vs. from is just a question of switching the arguments, so we don't need to discuss it, but the other 4 combinations have some caveats.

tgt | merge    (src) // (1) own, immutable
tgt | mergeIn  (src) // (2) in,  immutable
tgt | mergeM   (src) // (3) own, mutable
tgt | mergeInM (src) // (4) in,  mutable

// mergeTo, mergeInTo, mergeToM, mergeInToM: 'to' forms

In the immutable cases, a shallow clone is made of the target before merging. You must keep in mind whether the clone will use only 'own' properties or also the prototype ('in') properties. Case (1) corresponds to 'own' and case (2) to 'in'.

No matter which it is, all copied properties will become own properties of the clone.*

Then, properties are copied over from the source: only own properties in case (1), also 'in' properties in case (2).

The rule to remember in the immutable case is: 'own' merges with 'own' and 'in' with 'in'.

In the mutable case, properties are copied to the target from the source, and the target is not cloned or altered in any other way. In particular this means that the 'own' in case (3) only applies to the source: the prototype chain of the target will not be flattened or discarded or altered at all.

The rule to remember in the mutable case is: the target is never altered in any way, besides having properties copied in. So the 'own' / 'in' distinction only applies to the source.

We feel that these conventions are the most straightforward and lead to easily inferrable behavior.

In all cases, we are only ever talking about enumerable properties.

* This is good enough for most styles of functional programming in JS, many of which don't even bother with the distinction. People often expect shallow, cheap objects when programming this way. If you absolutely must preserve the prototype chain intact, you'll have to find a way to clone it yourself, then use the 'M' versions of our merge functions on your cloned object. You might also get away with using Object.create on it first, which will of course create one more link in the prototype chain.

٭ frontend stuff ٭

import { path, prop, whenTrue, always, } from 'stick-js'

If you use react/redux, perhaps with saga, chances are your modules end in something like this:

const withConnect = connect(mapStateToProps, mapDispatchToProps);
const withReducer = injectReducer({ key: 'home', reducer });
const withSaga = injectSaga({ key: 'home', saga });

export default compose(
  withReducer,
  withSaga,
  withConnect,
)(HomePage);

We should see by now that this composing of functions, invoked upon a single value, is exactly our pipe pattern. So why not:

export default HomePage
  | connect       (mapStateToProps, mapDispatchToProps)
  | injectSaga    ({ key: 'home', saga, })
  | injectReducer ({ key: 'home', reducer, })

And maybe you call actions using a structure like:

export function mapDispatchToProps(dispatch) {
  return {
    onChangeUsername: (evt) => dispatch(changeUsername(evt.target.value)),
  };
};

Why not:

export const mapDispatchToProps = (dispatch) => ({
  onChangeUsername: path (['target', 'value'] >> changeUsername >> dispatch,
})

Try it yourself and see :D

If you use styled components, perhaps you pass optional props in. Checking for the presence of the props can be annoying, so how about:

const SomeElementS = styled.div`
  top: 5%;
  left: 5%;
  ${ prop ('width')  >> whenOk   (sprintf1 ('width: %spx;')) }
  ${ prop ('height') >> whenOk   (sprintf1 ('height: %spx;')) }
  ${ prop ('error')  >> whenTrue ('color: red;' | always) }
`
<SomeElementS width='100%' error={true} />

If you use saga, perhaps you have something like this:

export function* getRepos() {
  const username = yield select(makeSelectUsername());
  const requestURL = `https://api.github.com/users/${username}/repos`;

  try {
    const repos = yield call(request, requestURL);
    yield put(reposLoaded(repos, username));
  } catch (err) {
    yield put(repoLoadingError(err));
  }
}

How about this:

export function* getRepos () {
  const username = yield makeSelectUsername () | select
  const requestURL = username | sprintf1 ('https://api.github.com/users/%s/repos')
  try {
    const repos = yield call (request, requestURL)
    reposLoaded (repos, username) | reposLoaded | put
  } catch (err) {
    yield err | repoLoadingError | put
  }
}

Note that we can't use our tryCatch function here, because of the yield in the catch clause.

If you use reselect, this is fine:

const selectHome = (state) => state.get('home');
const makeSelectUsername = () => createSelector(
  selectHome,
  (homeState) => homeState.get('username')
);

But how about this:

const get = dot1 ('get')

const selectHome = get ('home')
const makeSelectUsername = () => createSelector (
  selectHome,
  get ('username'),
)

When composing selectors, something like this works nicely to make sure the one being composed is valid:

export const makeSelectBooks = _ => createSelector (
  selectHome,
  get ('books'),
)

export const makeSelectBookTitles = _ => createSelector (
  makeSelectBooks (),
  'titles' | get | whenOk,
)

The author uses stick idioms to slay a common front-end pattern. When you're fetching your list of todos, the list should be undefined when it hasn't been set yet, and empty only if it has been fetched and really is empty. This makes testing easier, avoids weird flickers, lets you show a spinner, etc. It does introduce an extra condition you need to test for in your code, but we can handle this nicely:

const ListS = styled.div`
  min-height: 150px;
`

const List = ({ children }) => <ListS>
  {children | ifOk (identity) (spinner)}
</ListS>

٭ backend stuff ٭

When you're using a framework like Express, you have the well-known app object that you carry around everywhere. It just so happens that nearly all methods of app return this, so that chaining works in the familiar way:

app
.use (...)
.all (...)
.post ('/endpoint1', ((req, res) => ...))
.patch ('/endpoint2', ((req, res) => ...))
.get ('/endpoint3', ((req, res) => ...))
...
.listen (config.port, ...)

But there are cases when the makers were not so thoughtful, or when you simply don't know (or don't care) what a function or method returns. The pipe will free you from the limitations of the dot, and allow you to compose your own fluid interfaces. We'll use Express here to prove that it works. See the raindrops example for how you might use this with WebGL, and there are many other places this can be used.

import { side1, side2, list, appendTo, } from 'stick-js'
import { fromPairs, } from 'ramda'

const get      = side2 ('get')
const post     = side2 ('post')
const patch    = side2 ('patch')
const use      = side1 ('use')
const all      = side2 ('all')
const send     = side1 ('send')
const status   = side1 ('status')
const listen   = side2 ('listen')
const sendJSON = side1 ('json')

const sendStatus = code => data => status (code) >> sendJSON (data)
const msg = appendTo (['msg']) >> list >> fromPairs

app | use (bodyParser.json())
    | all ('*', (req, res, next) => {
      ...
        next ()
    })

    | post ('/endpoint1', ((req, res) => {
      ...
      res | sendStatus (500) ('Server error' | msg)
    }))

    | patch ('/endpoint2', ((req, res) => {
      ...
      res | sendStatus (200) ({ results, })
    }))

    ...

    | get ('/endpoint3', ((req, res) => ...))

    | listen (config.port) (...)

Working with Maybe

(or how to forget about null and undefined for a while)

Here is an example showing how you can deal with failure paths, using Maybe from bilby by @puffnfresh. In this toy example, we start with a French word, then:

  1. look up a translation (mocked in a table);

    failure path: the translation might not exist (or the network is down, a mouse ate through the cable, etc.)

  2. look up a corresponding 'count' value (also mocked in a table);

    failure path: the count might not exist

  3. divide this number into 10 for some reason;

    failure path: the number might be 0

 

import { some as Just, none as Nothing, } from 'bilby'

const toMaybe = o => o | ifOk (
  Just,
  Nothing | always,
)

const flatMap = dot1 ('flatMap')
const fold    = dot2 ('fold')

const translations = {
  rouge: 'red',
  bleu: 'blue',
  vert: 'green',
  // blanc missing
}

const count = {
  red: 5,
  blue: 0,
  // green missing
}

const formatAnswer = input => answer => [input | yellow, answer] | sprintfN ('%s → %s')

const go = _ =>
  ['rouge', 'bleu', 'vert', 'blanc']
  | map (doit)
  | tap (map (log))

const doit = x => x | calculate | formatAnswer (x)

const calculate = french => doLookup (french)
  | fold (
    // --- Just (answer)
    answer => answer + ' ' + green ('✔'),
    // --- Nothing.
    red ('✘') | always,
  )

const doLookup = french => french
  | getTranslation
  | flatMap (getCount)
  | flatMap (getQuotient)

const getTranslation = french => translations [french] | toMaybe

const getCount = english => count [english] | toMaybe

const getQuotient = condS ([
  0 | eq    | guard (_ => Nothing),
  otherwise | guard (cnt => Just (10 / cnt)),
])

go ()

maybe.jpg

Working with Either

The above example, using bilby's Either instead of Maybe, and a more point-free style.

The advantage of Either is that in addition to flatMap taking care of the failure path, you can also see why it failed. By convention the 'Left' branch is considered a failure, and a string stored in 'Left' represents the reason.

import { left as Left, right as Right, } from 'bilby'

const toEither = l => ifOk (
  Right,
  l | Left | always,
)

const flatMap   = dot1 ('flatMap')
const fold      = dot2 ('fold')
const arrowSnd  = f => ([a, b]) => [a, b | f]
const foldArrow = f => ([a, b]) => f (a, b)

const translations = {
  rouge: 'red',
  bleu: 'blue',
  vert: 'green',
  // blanc missing
}

const count = {
  red: 5,
  blue: 0,
  // green missing
}

const formatAnswer   = asterisk2 (yellow, id) >> sprintfN ('%s → %s')

const getTranslation = propOf (translations) >> toEither ('no translation')
const getCount       = propOf (count)        >> toEither ('no count')

const getQuotient = condS ([
  0 | eq    | guardV ('count was zero' | Left),
  otherwise | guard  (divideInto (10) >> Right),
])

const doLookup = getTranslation
  >> flatMap (getCount)
  >> flatMap (getQuotient)

const calculate = doLookup >> fold (
  // --- Left (failureReason)
  prependTo (['✘' | red])   >> sprintfN ('%s %s'),
  // --- Right (answer)
  prependTo (['✔' | green]) >> sprintfN ('%s %s'),
)

const doit = timesV (2)
  >> arrowSnd (calculate)
  >> foldArrow (formatAnswer)

const go = _ =>
  ['rouge', 'bleu', 'vert', 'blanc']
  | map (doit)
  | tap (map (log))

go ()

either.jpg

Abstract (algebraic) data types

Abstract (or 'algebraic') data types can help eliminate an entire class of bugs, by making impossible states unrepresentable.

Imagine you have something like a traffic light, which at any moment can be either red, yellow, green, or broken. (Or a download, which can be in progress, stopped, completed, failed; or an API call, etc.)

You can use a normal object to represent these sorts of things, which JS programmers do all the time of course. The problem is that as the problem domain becomes more complex, it gets increasingly harder to verify the correctness of such a so-called 'product type'. You will probably find yourself carrying around properties like 'error' and 'state' and 'type' and so on using a lot of if statements to check for illegal states.

Another way is to model the data using ADTs, build the states, then branch using a 'catamorphism' to resolve the states when you're ready.

In this example we use daggy by @fantasyland to create an ADT representing various kinds of numerical sequences.

In Haskell our declaration would be:

data Sequence = ArithmeticSequence Int -- y = nx
              | GeometricSequence Int  -- y = c ** x
              | IrregularSequence
              | ErrorSequence String   -- error with a reason

In (pseudo-)Java maybe something like:

abstract class Sequence {}
class ArithmeticSequence extends Sequence {
  public ArithmeticSequence(int n) {} // y = nx
}
class GeometricSequence extends Sequence {
  public GeometricSequence(int c) {} // y = c ** x
}
class IrregularSequence extends Sequence {
  public IrregularSequence() {}
}
class ErrorSequence extends Sequence {
  public ErrorSequence(String reason) {}
}

Using stick idioms and daggy:

// --- types.js

import daggy from 'daggy'

export const Sequence = daggy.taggedSum ('Sequence', {
  ArithmeticSequence: ['n'],
  GeometricSequence: ['c'],
  IrregularSequence: [],
  ErrorSequence: ['reason'],
})

const { ArithmeticSequence, GeometricSequence, IrregularSequence, ErrorSequence, } = Sequence

export { ArithmeticSequence, GeometricSequence, IrregularSequence, ErrorSequence, }

// --- sequence.js

const testReduce = f => n => (acc, x) => f (acc, x) === n ? x : null
const testReduceArithmetic = minus    | testReduce
const testReduceGeometric  = divideBy | testReduce

const testSequence = g => f => ([a, b, ...rest]) => lets (
  _ => b | g (a),
  n => rest | reduceAbort (n | f) (b) (null),
  (n, reduced) => reduced === null ? false : n,
)

const containsNull = null | contains

export const isArithmetic = testReduceArithmetic | testSequence (minus)
export const isGeometric  = testReduceGeometric  | testSequence (divideBy)
export const isError      = x => containsNull (x) ? 'contains null' : false

// --- util.js

export const cata = dot1 ('cata')

export const repeatChar = n => timesV (n) >> join ('')
export const pad = n => str => (n - str.length) | condS ([
  0 | gt    | guard (pad => [str, ' ' | repeatChar (pad)] | join ('')),
  otherwise | guardV (str),
])

// --- index.js

import { ArithmeticSequence, GeometricSequence, IrregularSequence, ErrorSequence, } from './types'
import { isArithmetic, isGeometric, isError, } from './sequence'
import { cata, repeatChar, pad, } from './util'

const checkSequence = condS ([
  isArithmetic | guard ((_, n) => n | ArithmeticSequence),
  isGeometric  | guard ((_, c) => c | GeometricSequence),
  containsNull | guard (()     => 'contains null' | ErrorSequence),
  otherwise    | guard (()     => IrregularSequence),
])

const format = cata ({
  ArithmeticSequence: (n) => ['arithmetic', n] | map (yellow) | sprintfN ('%s: y = %sx'),
  GeometricSequence:  (c) => ['geometric', c]  | map (green) | sprintfN ('%s: y = %s ** x'),
  ErrorSequence:      (reason) => reason | red | sprintf1 ('error: %s'),
  IrregularSequence:  ()  => 'irregular sequence' | brightRed,
})

const repeatChar = n => timesV (n) >> join ('')

const pad = n => str => (n - str.length) | condS ([
  0 | gt    | guard (pad => [str, ' ' | repeatChar (pad)] | join ('')),
  otherwise | guardV (str),
])

const report = xs => resolved =>
  [xs | join (', ') | pad (13), resolved]
  | sprintfN ('%s → %s')

const doit = xs => xs
  | checkSequence
  | format
  | report (xs)

const go = _ =>
  [
    [0, 4, 8, 12],   // --- arithmetic
    [1, 3, 9, 27],  // --- geometric
    [0, 1, 2, 5],   // --- irregular
    [null, 1, 2, 5] // --- error
  ]
  | map (doit)
  | tap (map (log))

go ()

seq.jpg

Point-free techniques

We had an isInteger function above as:

const { floor, } = Math
const isInteger = x => x
  | floor
  | eq (x)

If a number equals its own floor, it's an integer.

For a challenge we could try to make this point-free. The difficult part is wresting the x free out of the 3rd line.

We can use the concept of 'arrows', though in a far less rigorous form than people might be used to.

First spread the first line x into an array [x, x]

x | timesV (2) // [x, x]

Now use the asteriskN function, which works by taking an array of values and an array of functions, lining them up, and applying the functions to the corresponding value.

So we can do an asterisk of [x, x] with [id, floor], where id is the identity function.

[id, floor]
 ↓    ↓
[x,   x   ]

and get [id (x), floor (x)], or [x, floor (x)]

Now look at the passToN function:

; [a, b, ...] | passToN (f)

yields f (a, b, ...)

If we take our result and pipe it to passToN (eq), we get:

; [x, floor (x)]
| passToN (eq)

which is

eq (x, floor (x))

which is true if & only if x === floor (x), in other words, if & only if x is an integer.

Putting it together:

const isInteger = x => x
  | timesV (2)
  | asteriskN ([id, floor])
  | passToN (eq)

We can reduce it further by using the property a | b | c → a | (b >> c)

const isInteger = x => x
  | (timesV (2) >> asteriskN ([id, floor]))
  | passToN (eq)

And that long second line is useful enough to merit its own function, which takes a function (in our case floor) to apply to a value in second position. We'll call it arrowSnd:

const arrowSnd = f => timesV (2) >> asteriskN ([id, f])

And stick it back in:

const isInteger = x => x
  | arrowSnd (floor)
  | passToN (eq)

And make it point-free:

const isInteger = arrowSnd (floor) >> passToN (eq)

Your turn :D

Performance

Stick is fast, far faster than Ramda. Even trivial functions like R.flip and R.tap are several times slower.

Though it initially depended on Ramda, we have decided to eliminate that dependency by reimplementing many of the functions.

This really only becomes an issue in critical loops with millions of iterations per second.

It is true that a | b compiles to three function calls, whereas b (a) is only one. But this is almost certainly not going to affect your app. Your JS runtime can call a lot of functions per millisecond.

Nonetheless you are encouraged to mix and match our functions with whichever functional libraries you like -- Ramda, Lodash/FP, or anything else, as it suits you.

import { map, } from 'ramda'
import { filter, } from 'lodash/fp'
import { ifPredicate, } from 'stick-js'

The stick idiom will still work, as long as the functions are curried and data-last.

Furthermore Ramda is probably perfectly fine for your app, and its functions often provide type-checking and error messages (we don't), and many of their functions are more sophisticated. R.map works on functors and transformers, for example -- ours doesn't.

And, it is our belief that if you are already using the flow pattern in Lodash/FP or the pipe function in Ramda, that it will really be a no-brainer to overload the operator and keep everything else the same.

_.flow (
  _.split (' '),
  _.map (capitaliseFirstLetter),
  _.join (' '),
) (myData)

// ->
myData
| _.split (' '),
| _.map (capitaliseFirstLetter),
| _.join (' ')

Extra performance

The curried functions you import from the main module are written first using manual currying, and then recurried and exported. This is in order to allow both calling styles.

If you are squeezing performance out of a loop -- in an animation, a particle system, WebGL, a server -- you can check the docs to see if your function is exported by 'stick-js/manual'. If so, you can directly import the manual version, but you must remember to call it using the manual style:

import { merge, } from 'stick-js/manual'
merge (obj1, obj2) // will not work
obj2 | merge (obj1) // ok
merge (obj1) (obj2) // also ok

Generic version of lets

Here is a generic form of lets which takes any number of non-zero arguments. We removed it from stick because it depends on mapAccum, for which we do not currently have an implementation without depending on Ramda.

// --- generic form, for any non-zero number of arguments.
const lets = (() => {
  const executeStep = prevVals => applyToN (prevVals)
  return (...xs) => xs
    // --- acc contains running output array, up to the previous item.
    | mapAccum ((acc, v) => executeStep (acc) (v)
      | (stepVal => [[...acc, stepVal], stepVal])
    ) ([])
    | prop (1)
    | last
}) ()

This is not fast, but it is correct. You can prove it with a contrived fibonacci example. We'll model fibonacci (n) as

lets (
  _ => 1,
  _ => 1,
  (a, b) => a + b,
  (_, b, c) => b + c,
  ...
  (a, b, ...) => [a, b, ...],
)

 

import { list, timesV, applyToN, rangeTo, } from 'stick-js'

const fibonacci = (n) => {
  const sumLastTwo = (xs) => {
    const l = xs.length
    return xs [l-1] + xs [l-2]
  }
  const entry = (...prev) => {
    const l = prev.length
    return l === 0 ? 1 :
           l === 1 ? 1 : sumLastTwo (prev)
  }
  const refs = entry | timesV (n + 1)
  return lets (...refs, list)
}

1 | rangeTo (20)
  | map (fibonacci)

/*
  [ [ 1, 1 ],
    [ 1, 1, 2 ],
    [ 1, 1, 2, 3 ],
    [ 1, 1, 2, 3, 5 ],
    [ 1, 1, 2, 3, 5, 8 ],
    [ 1, 1, 2, 3, 5, 8, 13 ],
    [ 1, 1, 2, 3, 5, 8, 13, 21 ],
    [ 1, 1, 2, 3, 5, 8, 13, 21, 34 ],
    [ 1, 1, 2, 3, 5, 8, 13, 21, 34, 55 ]
    ...
    ]
*/

Bitwise math

Of course, we've saved your precious bitwise operators. You can either:

  1. Use the functional form
    import { bitwiseOr, bitwiseShiftBy, } from 'stick-js'
    4 | bitwiseOr (9) // 13
  1. Do your bitwise math in a separate scope than that in which the defineBinaryOperator calls appear. The operators are only overloaded in the scope in which the calls are made. We recommend doing the bitwise math in a separate source file and not mixing the two styles in one file.

Readme

Keywords

none

Package Sidebar

Install

npm i stick-js

Weekly Downloads

18

Version

0.5.3

License

GPL-3.0-or-later

Unpacked Size

294 kB

Total Files

18

Last publish

Collaborators

  • misterfish