Allows you to apply (thus ap
) a value to a function without having to take
either out of the context of their Result
s. This does mean that
the transforming function is itself within a Result
, which can be hard to
grok at first but lets you do some very elegant things. For example, ap
allows you to do this (using the method form, since nesting ap
calls is
awkward):
import { ap, ok, err } from 'true-myth/result';
const one = ok<number, string>(1);
const five = ok<number, string>(5);
const whoops = err<number, string>('oh no');
const add = (a: number) => (b: number) => a + b;
const resultAdd = ok<typeof add, string>(add);
resultAdd.ap(one).ap(five); // Ok(6)
resultAdd.ap(one).ap(whoops); // Err('oh no')
resultAdd.ap(whoops).ap(five) // Err('oh no')
Without ap
, you'd need to do something like a nested match
:
import { ok, err } from 'true-myth/result';
const one = ok<number, string>(1);
const five = ok<number, string>(5);
const whoops = err<number, string>('oh no');
one.match({
Ok: n => five.match({
Ok: o => ok<number, string>(n + o),
Err: e => err<number, string>(e),
}),
Err: e => err<number, string>(e),
}); // Ok(6)
one.match({
Ok: n => whoops.match({
Ok: o => ok<number, string>(n + o),
Err: e => err<number, string>(e),
}),
Err: e => err<number, string>(e),
}); // Err('oh no')
whoops.match({
Ok: n => five.match({
Ok: o => ok(n + o),
Err: e => err(e),
}),
Err: e => err(e),
}); // Err('oh no')
And this kind of thing comes up quite often once you're using Result
to
handle errors throughout your application.
For another example, imagine you need to compare the equality of two
ImmutableJS data structures, where a ===
comparison won't work. With ap
,
that's as simple as this:
import { ok } from 'true-myth/result';
import { is as immutableIs, Set } from 'immutable';
const is = (first: unknown) => (second: unknown) =>
immutableIs(first, second);
const x = ok(Set.of(1, 2, 3));
const y = ok(Set.of(2, 3, 4));
ok(is).ap(x).ap(y); // Ok(false)
Without ap
, we're back to that gnarly nested match
:
import Result, { ok, err } from 'true-myth/result';
import { is, Set } from 'immutable';
const x = ok(Set.of(1, 2, 3));
const y = ok(Set.of(2, 3, 4));
x.match({
Ok: iX => y.match({
Ok: iY => Result.of(is(iX, iY)),
Err: (e) => ok(false),
})
Err: (e) => ok(false),
}); // Ok(false)
In summary: anywhere you have two Result
instances and need to perform an
operation that uses both of them, ap
is your friend.
Two things to note, both regarding currying:
All functions passed to ap
must be curried. That is, they must be of the
form (for add) (a: number) => (b: number) => a + b
, not the more usual
(a: number, b: number) => a + b
you see in JavaScript more generally.
(Unfortunately, these do not currently work with lodash or Ramda's curry
helper functions. A future update to the type definitions may make that
work, but the intermediate types produced by those helpers and the more
general function types expected by this function do not currently align.)
You will need to call ap
as many times as there are arguments to the
function you're dealing with. So in the case of this add3
function,
which has the "arity" (function argument count) of 3 (a
and b
), you'll
need to call ap
twice: once for a
, and once for b
. To see why, let's
look at what the result in each phase is:
const add3 = (a: number) => (b: number) => (c: number) => a + b + c;
const resultAdd = ok(add); // Ok((a: number) => (b: number) => (c: number) => a + b + c)
const resultAdd1 = resultAdd.ap(ok(1)); // Ok((b: number) => (c: number) => 1 + b + c)
const resultAdd1And2 = resultAdd1.ap(ok(2)) // Ok((c: number) => 1 + 2 + c)
const final = maybeAdd1.ap(ok(3)); // Ok(4)
So for toString
, which just takes a single argument, you would only need
to call ap
once.
const toStr = (v: { toString(): string }) => v.toString();
ok(toStr).ap(12); // Ok("12")
One other scenario which doesn't come up quite as often but is conceivable
is where you have something that may or may not actually construct a function
for handling a specific Result
scenario. In that case, you can wrap the
possibly-present in ap
and then wrap the values to apply to the function to
in Result
themselves.
Because Result
often requires you to type out the full type parameterization
on a regular basis, it's convenient to use TypeScript's typeof
operator to
write out the type of a curried function. For example, if you had a function
that simply merged three strings, you might write it like this:
import Result from 'true-myth/result';
import { curry } from 'lodash';
const merge3Strs = (a: string, b: string, c: string) => string;
const curriedMerge = curry(merge3Strs);
const fn = Result.ok<typeof curriedMerge, string>(curriedMerge);
The alternative is writing out the full signature long-form:
const fn = Result.ok<(a: string) => (b: string) => (c: string) => string, string>(curriedMerge);
Aside: ap
is not named apply
because of the overlap with JavaScript's
existing apply
function – and although strictly speaking, there isn't any
direct overlap (Result.apply
and Function.prototype.apply
don't intersect
at all) it's useful to have a different name to avoid implying that they're
the same.
Allows you to apply (thus
ap
) a value to a function without having to take either out of the context of theirResult
s. This does mean that the transforming function is itself within aResult
, which can be hard to grok at first but lets you do some very elegant things. For example,ap
allows you to do this (using the method form, since nestingap
calls is awkward):Without
ap
, you'd need to do something like a nestedmatch
:And this kind of thing comes up quite often once you're using
Result
to handle errors throughout your application.For another example, imagine you need to compare the equality of two ImmutableJS data structures, where a
===
comparison won't work. Withap
, that's as simple as this:Without
ap
, we're back to that gnarly nestedmatch
:In summary: anywhere you have two
Result
instances and need to perform an operation that uses both of them,ap
is your friend.Two things to note, both regarding currying:
All functions passed to
ap
must be curried. That is, they must be of the form (for add)(a: number) => (b: number) => a + b
, not the more usual(a: number, b: number) => a + b
you see in JavaScript more generally.(Unfortunately, these do not currently work with lodash or Ramda's
curry
helper functions. A future update to the type definitions may make that work, but the intermediate types produced by those helpers and the more general function types expected by this function do not currently align.)You will need to call
ap
as many times as there are arguments to the function you're dealing with. So in the case of thisadd3
function, which has the "arity" (function argument count) of 3 (a
andb
), you'll need to callap
twice: once fora
, and once forb
. To see why, let's look at what the result in each phase is:So for
toString
, which just takes a single argument, you would only need to callap
once.One other scenario which doesn't come up quite as often but is conceivable is where you have something that may or may not actually construct a function for handling a specific
Result
scenario. In that case, you can wrap the possibly-present inap
and then wrap the values to apply to the function to inResult
themselves.Because
Result
often requires you to type out the full type parameterization on a regular basis, it's convenient to use TypeScript'stypeof
operator to write out the type of a curried function. For example, if you had a function that simply merged three strings, you might write it like this:The alternative is writing out the full signature long-form:
Aside:
ap
is not namedapply
because of the overlap with JavaScript's existingapply
function – and although strictly speaking, there isn't any direct overlap (Result.apply
andFunction.prototype.apply
don't intersect at all) it's useful to have a different name to avoid implying that they're the same.