TS-BCCC stands for "TypeScript Bi-Cartesian Closed Categories". TS-BCCC is a TypeScript library which implements a generic, expressive, and high performance series of universal combinators. These combinators can be used in many applications as a solid and reliable basis for building complex constructions such as monads, some of which are hierby included.
Download the library with npm install ts-bccc or yarn add ts-bccc.
To start playing around with simple function pipelining, try the following:
import { Fun, fun, apply } from "ts-bccc";
let incr:Fun<number,number> = fun(x => x + 1)
let double:Fun<number,number> = fun(x => x * 2)
console.log(apply(incr.then(double.then(incr)), 3))
The sample builds a pipeline which first increments, then double, then increments again, and the pipeline is then
invoked ("applied to") the number 3, leading us to an output of 9.
We can build more complex datatypes such as tuples (products) and discriminated unions (sums). All datatypes can be mapped into datatypes with the same structure (products to products, sums to sums), or constructed/destroyed (value to product, sum to value).
The following example shows the mapping of a tuple:
import { Fun, fun, apply, Times } from "ts-bccc";
let incr:Fun<number,number> = fun(x => x + 1)
let double:Fun<number,number> = fun(x => x * 2)
let pair_transform:Fun<Times<number,number>, Times<number,number>> = incr.map_times(double)
console.log(apply(pair_transform, { fst:4, snd:2 }))
(4,2) is thus transformed into (5,4).
A product can be created from a single value and two functions which determine the values of the elements of the tuple:
let incr:Fun<number,number> = fun(x => x + 1)
let double:Fun<number,number> = fun(x => x * 2)
let mk_tuple:Fun<number, Times<number,number>> = incr.times(double)
console.log(apply(mk_tuple, 3))
3 is thus transformed into (4,6).
The dual operators are available for sums: map_plus and plus.
Given that the library implements Bi-Cartesian Closed Categories, there exist a series of equivalences which lead us to
useful transformations between data types. The interesting bit of these transformations is that they mirror exactly a
series of well-known equivalences from arithmetics. This makes these equivalences very easy to manipulate, as we do not
need any effort to remember concepts such as a*(b+c) = a*b+a*c.
An equivalence such as a*(b+c) = a*b+a*c are reformulated as a series of built-in library functions such as:
distribute_sum_prod = function<a,b,c>() : Fun<Prod<a, Sum<b,c>>, Sum<Prod<a,b>, Prod<a,c>>>
distribute_sum_prod_inv = function<a,b,c>() : Fun<Sum<Prod<a,b>, Prod<a,c>>, Prod<a, Sum<b,c>>>
An equivalence such as a^(b+c) = a^b * a^c is encoded as:
distribute_exp_sum = function<a,b,c>() : Fun<Fun<Plus<a,b>, c>, Prod<Fun<a,c>, Fun<b,c>>>
``distribute_exp_sum_inv = function<a,b,c>() : Fun<Prod<Fun<a,c>, Fun<b,c>>, Fun<Plus<a,b>, c>>`
Even simpler equivalences such as a+0 = a find their place, for example:
sum_identity = function<a>() : Fun<Sum<a,Zero>,a>
sum_identity_inv = function<a>() : Fun<a,Sum<a,Zero>>
Within the framework of Bi-Cartesian Closed Categories it is also possible to give a very elegant implementation of monads
as monoids in the category of endofunctors. Monads are all (often articulated) endofunctors, but they can be built more easily
as compositions of the generic types Plus, Times, and Fun, which themselves are (bi-)endofunctors as well.
For this reason, we can implement for example the State monad as State<S,A> = CCC.Fun<S, CCC.Prod<A,S>>, together with
the required monadic operators:
st_map: function<B>(p:State<S,A>, f: CCC.Fun<A,B>) : State<S,B>
st_join = function<S,A>(pp:State<S,State<S,A>>) : State<S,A>
st_unit = function<S,A>(x:A) : State<S,A>
To be able to effectively use such a monad, we define a wrapper type which features more useful operators such as
binding (then), thereby also mirroring standard library constructs such as Promise:
export interface St<S,A> extends CCC.Fun<S, CCC.Prod<A,S>> {}
export type State <S,A> = {
run: St<S,A>,
then: <B>(k: (_: A) => State <S,B>) => State <S,B>;
ignore: () => State<S,CCC.Unit>;
ignore_with: <B>(x:B) => State<S,B>;
map: <B>(f: CCC.Fun<A,B>) => State <S,B>
}
It is noteworthy to realise that all of these operators are implemented fully and exclusively with the basic operators on products, sums, etc. which we have seen previously, leading us to a very homogeneous implementation overall. For example, notice how all of the following functions are no more than long BCCC pipelines:
export let st_run = function<s,a>() : Fun<State<s,a>, St<s,a>> { return fun(p => p.run) }
export let st_join = function<S,A>(pp:State<S,State<S,A>>) : State<S,A> {
let g = fst<State<S,A>,S>().then(st_run<S,A>()).times(snd<State<S,A>,S>()).then(apply_pair())
let h = st_run<S,State<S,A>>().map_times(id<S>()).then(apply_pair()).then(g)
return mk_state<S,A>(apply(curry(h), pp))
}
export let st_get_state = function<S>() : State<S,S> { return mk_state<S,S>(id<S>().times(id<S>())) }
export let st_set_state = function<S>(s:S) : State<S,Unit> { return mk_state<S,Unit>(unit<S>().times(constant<S,S>(s))) }
export let st_unit = function<S,A>(x:A) : State<S,A> { return mk_state<S,A>(constant<S,A>(x).times(id<S>())) }
For an extensive example on using these monads in the practical implementation of a high performance meta-interpreter embedded in TypeScript, head over to ts-bccc-meta-interpreter.