This aims to present how we use fp-ts at Inato. I have mainly compiled the work of my wonderful colleagues @LaureRC and @punie
Functional programming is all about data-structures that encapsulate values to give them more context.
Think about an Array. It is a data structure encapsulating a group of values of the same type. The Array gives a context of “there are multiple values”. It’s the same with the data structures we use in FP.
Here are the ones we use:
An Option represents a value which might be there, or not. If it’s there then it is Option.Some(value) and if not it is Option.none
You can think of an option as something that can be null or undefined.
The easiest way to build an Option is to use the some or none constructor that returns a value encapsulated in an Option.
const noneValue = option.none;
const someValue = option.some("value");
If you have a value and want to check it you can use fromNullable. If the value is null | undefined you get a Option.None otherwise you get the value wrapped in an Option data structure.
const noneValue = option.fromNullable(null); // Option.None
const optionValue = option.fromNullable("value"); // Option.Some("value")
You can also pass you own validation function to build an Option with the fromPredicate helper:
const isEven = (number) => number % 2 === 0;
const noneValue = option.fromPredicate(isEven)(3); // Option.None
const optionValue = option.fromPredicate(isEven)(4); // Option.Some(4)
You can build an Option from an Either. If the Either is in its left state (~ error) you get an Option.None, otherwise you get an Option.Some of the value in the Either
const leftEither = either.left("whatever");
const rightEither = either.right("value");
const noneValue = option.fromEither(leftEither); // option.None
const optionValue = option.fromEither(rightEither); // option.Some("value")
When your value is encapsulated into an Option, you sometimes want to retrieve it and opt-out of the functional paradigm. For example if you want to give it to the outside world, in a graphql resolver.
Easiest way is to use the toNullable or toUndefined helpers. They are pretty straightforward, if your Option contains a value, you get it, otherwise you get null or undefined:
const noneValue = option.none;
const someValue = option.of("value");
option.toUndefined(noneValue); // undefined
option.toUndefined(someValue); // "value"
option.toNullable(noneValue); // null
option.toNullable(someValue); // "value"
You can use one of the following helper to retrieve your value or have a default if your Option is in none state.
getOrElse takes a function as parameter that returns the default value for none. Please note the default value must have the same type as your initial Option. Eg: getOrElse on an Option<number> must return a number. If you want to return another type, you can use getOrElseW.
const noneValue = option.none;
const someValue = option.of("value");
option.getOrElse(() => "default")(noneValue); // "default"
option.getOrElse(() => "default")(someValue); // "value"
option.getOrElseW(() => 3)(noneValue); // 3
The last way to get your value is match and it allows you to compute before returning it.
it takes a two functions, the first one is executed if your Option is none, the second one if your Option contains some value.
const noneValue = option.none;
const someValue = option.of(10);
const doubleOrZero = option.match(
() => 0, // this is called when your Option is none
(n: number) => 2 * n // called when your Option has some value
);
doubleOrZero(noneValue); // 0
doubleOrZero(someValue); // 20
An Either represents a computation that can have two results, called branches or tracks (left and right).
Most of the time, we use it to represent a computation that can fail. So the left branch represents the failure, and the right branch the success. So when we manipulate an Either, if we enter the left branch, we most of the time won’t carry out further manipulations, and just return the Error context that is encapsulated in the left branch (it is still accessible tho, through different helpers). If we are in the right branch, we’ll keep on manipulating the value and passing it through the right branch.
An Either is typed Either<E, A> where E is the type of the left track (E for Error most of the time) and A the right track.
The easiest way to build an Either is to use the right or left constructor that returns a value encapsulated as a right or left Either.
const leftValue = either.left("value"); // -> you are on the left branch
const rightValue = either.right("value"); // -> you are on the right branch
You have also a fromNullable helper. If the value is null or undefined you need to provide your Either what to put in the left track, otherwise it will put your value in the right track.
const leftValue = either.fromNullable("value was nullish")(null); // either.left('value was nullish')
const rightValue = either.fromNullable("value was nullish")("value"); // either.right("value")
You can also pass you own validation function to build an Either with the fromPredicate helper. Here it is a bit more complicated.
You have to first pass a function checking your value is correct (should I go right or left track). This function can be a simple function returning a boolean, or a type guard. And then pass a value for the left track.
type EvenNumber = number;
const isEven = (num: number) => num % 2 === 0;
const isEvenTypeGuard = (num: number): num is EvenNumber => num % 2 === 0;
const eitherBuilder = either.fromPredicate(
isEven, // here we could use isEvenTypeGuard to infer the number type and have an Either<E, EvenNumber>
(number) => `${number} is an odd number`
);
// Here when you use eitherBuilder you get something with the type Either<string, number>
const leftValue = eitherBuilder(3); // Either.left('3 is an odd number')
const rightValue = eitherBuilder(4); // Either.right(4)
You can build an Either from an Option. It works exactly as the fromNullable as we’ve seen an Option could represent a nullable value.
const noneValue = option.none;
const someValue = option.some("value");
const left = either.fromOption("value was nullish")(noneValue); // either.left('value was nullish')
const right = either.fromOption("value was nullish")(someValue); // either.right('value')
This part looks a lot like the Option part. There are two Either “destructors”, which are really similar to the ones we’ve seen above.
The getOrElse destructor and its getOrElseW version work exactly as the one from Option. You have to pass it what to return if you are in the left branch.
const leftValue = either.left("Division by Zero!");
const rightValue = either.right(10);
either.getOrElse(() => 0)(leftValue); // 0
either.getOrElse(() => 0)(rightValue); // 10
The match destructor also takes two functions, the first one representing what to do on left branch, the second one what to do on right branch.
const leftValue = either.left("Division by Zero!");
const rightValue = either.right(10);
const doubleOrZero = either.match(
(eitherError: string) => {
console.log(`The error was ${eitherError}`);
return 0;
}, // on left branch
(value: number) => 2 * value // on right branch
);
doubleOrZero(leftValue); // logs "The error was Division by Zero!" and returns 0
doubleOrZero(rightValue); // 20
A Task represents an asynchronous computation. You can view it as a Promise.
So quite simply a TaskEither is an asychronous computation that can have two results (~ that can fail).
A TaskEither is typed TaskEither<A, B> where A is the type of the left track and B the right track.
Note: The type TaskEither<E, A> is strictly equivalent to Task<Either<E, A>>, and so there are no conversion functions between the two. All the functions from the Task module can thus be used on a TaskEither with the caveat that combinators such as map or chain give you access to the whole inner Either.
You can build a TaskEither as you would build an Either: left, right, fromNullable, fromPredicate.
const leftValue = taskEither.left("value");
const rightValue = taskEither.right("value");
...
You can also build a TaskEither from a Promise! You have the tryCatch helper, that takes two functions as arguments. The first one returns a promise, and the second one returns the value to put in left if the promise rejects.
const asyncIsEven = async (a: number) => {
await asyncExternalCall();
if (a % 2 !== 0) {
throw new Error("ODD");
}
return a;
};
const buildTaskEither = (number: number) =>
taskEither.tryCatch(
() => asyncIsEven(number),
() => `${number} is odd`
);
const rightValue = buildTaskEither(4); // taskEither.right(4)
const leftValue = buildTaskEither(3); // taskEither.left('3 is odd')
In fp-ts, if you invoke a TaskEither, you get a Promise<Either> so most of the time, what we do is we invoke our TaskEither and then use the aforementioned methods on the Either to retrieve its value.
const ten = taskEither.right(10); // this is a right branch of a taskEither
const rightValue = await ten(); // we invoke ten, making it a Promise<Either> and then await the promise, so rightValut is an Either.right
const returnedValue = either.getOrElse(() => 0)(rightValue); // 10
The Array and ReadonlyArray types from fp-ts extend their counterparts from native TypeScript so they can be used without having to do any conversion.
Those modules provide equivalent functions as Array.prototype.map or Array.prototype.filter but with a more fp-ts, pipe-friendly API:
import { readonlyArray } from "fp-ts";
const array = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9];
pipe(array, readonlyArray.filter(isEven), readonlyArray.map(square));
// => [0, 4, 16, 36, 64]
Sometimes, you want to map an array with a function that may fail to produce a value and then get rid of those failures. That is precisely what array.filterMap does:
// getById(id: Id): Option<Item>;
const ids = [id1, id2, id3];
const foundItems = pipe(ids, array.filterMap(getById));
// => Array<Item>
Ord instancesSorting strings:
import { array, ord, string } from "fp-ts";
const strings = ["zyx", "abc", "klm"];
const sortedStrings = pipe(strings, array.sort(string.Ord));
// => ['abc', 'klm', 'zyx']
There are various instances of Ord for primitive types, available in fp-ts:
string.Ord: Ord<string>number.Ord: Ord<number>boolean.Ord: Ord<boolean>date.Ord: Ord<Date>The Ord module also provides us with ways to manipulate and combine them to create richer sorting algorithms!
For example, to get an instance of Ord that will sort strings in reverse order, I can simply create and use it as such:
const strings = ["zyx", "abc", "klm"];
const reversedOrdString = ord.reverse(string.Ord);
const sortedStrings = pipe(strings, array.sort(reversedOrdString));
// => ['zyx', 'klm', 'abc']
Some higher level types like Option<A> also implement Ord! In this particular case, none is considered lower than some for example. So if you want to sort an array of Option<number>:
const nums = [option.some(1337), option.none, option.some(42)];
const ordOptionalNumbers = option.getOrd(number.Ord);
const sortedNums = pipe(nums, array.sort(ordOptionalNumbers));
// => [option.none, option.some(42), option.some(1337)]
Likewise, if you have a more complex construct, like a User:
interface User {
name: string;
age: Option<number>;
}
You can build various instance of Ord<User> based on your need with ord.contramap. This function is just a way of defining how to access the field that will be used for sorting.
import { string, ord, number } from 'fp-ts';
const byName = pipe(
string.Ord,
ord.contramap((user: User) => user.name)
);
const byAge = pipe(
option.getOrd(number.Ord),
ord.contramap((user: User) => user.age)
);
You can then tell array.sort to use either one of these instances depending on your needs.
Finally, if you need to combine those sorting conditions (first, sort by age, and then by name), you can use array.sortBy which accepts an array of Ord instances, to be used in order:
const sortedUsers = pipe(users, array.sortBy([byAge, byName]));
In a functional paradigm, you want to have small functions, and to call them one after another to transform data easily.
The pipe function creates a “pipeline” to transform data. It takes one value as a starting point, and several functions, and then passes the value into the first function, then the return value of the first function into the second function etc…
In classic imperative style we would do:
const value = "value";
const value1 = addSthg(value);
const value2 = doSthgElse(value1);
const finalValue = doFinalSthg(value2);
It’s the same as:
const value = "value";
const finalValue = pipe(value, addSthg, doSthgElse, doFinalSthg);
Remark: as you pass one value through your pipe, that’s why fp-ts functions tend to accept only one value, it’s easier to pipe them and so on, as we can see here:
const myOption = pipe(
"value",
either.fromNullable('Given value was null'),
option.fromEither
); // myOption = option.Some("value")
The flow function works exactly as the pipe, without the first value. It creates the pipeline of data but does not transform anything, it returns a function that can be used somewhere else (in a pipe possibly!)
We could rewrite the above example as:
const transformData = flow(addSthg, doSthgElse, doFinalSthg); // please note it's exactly as the original pipe, without the first value
const value = "value";
const finalValue = transformData(value);
Here transformData could be reused somewhere else, be simple and easy to pipe with other stuff.
Let’s say you have a datastructure containing a value you wanna work with. You can have an option.some(3) or an either.right(currentUser) or even ["some", "strings"].
Your value is encapsulated in a context, for example you computed your currentUser and maybe the computation could fail. To keep working with the value without removing its context (what we’ve seen above), we’ll use combinators.
So you have a context containing your value, Either<number> but you want to transform it with a function that takes an input of your value type. Eg: double: (number) => (number). Unfortunately, you cannot do double(myEitherNumber).
You will use map. Map will unwrap the value, apply the function and then wrap the value again in its context. That’s it.
map takes a function as parameter (the function transforming the value) and can then be applied to the value.
const doubleIfEven = (n: number) => n % 2 === 0 ? 2 * n : n
const optionEven = option.some(2);
const optionOdd = option.some(3);
option.map(doubleIfEven)(optionEven) // option.some(4)
pipe(
optionOdd
option.map(doubleIfEven)
) // option.some(3)
According to your datastructure, maps behave differently. We all know the JS Array.prototype.map which actually unwraps values from an array, transform those values with a function, and then repacks the values in an Array.
[1, 2, 3].map(doubleIfEven); // [1, 4, 3]
map(doubleIfEven)([1, 2, 3]); // same result but using fp-ts
map allows you to go from DataStructure<A> to DataStructure<B> as you can apply any function going from A.
pipe(
facility, // Option<Facility>
option.map((facility: Facility) => facility.country.code) // => returns an Option<CountryCode>
);
Map behavior on Options is pretty simple. If there is some value, it will apply the function to the value. If the Option is none, it will just return none.
const someOption = option.some(2);
const noneOption = option.none;
option.map(doubleIfEven)(someOption); // option.some(4)
option.map(doubleIfEven)(noneOption); // option.none
Remember Either is generally considered as “Everything went well” (right branch) vs “Something happened” (left branch).
So map will only transform your data if you are in the right branch.
If you want to map on the left branch, you can use mapLeft!
const rightValue = either.right(2);
const leftValue = either.left(2);
either.map(doubleIfEven)(rightValue); // either.right(4)
either.map(doubleIfEven)(leftValue); // either.left(2)
either.mapLeft(doubleIfEven)(leftValue); // either.left(4)
Eventually, there is also a bimap helper mapping the first function to the left branch, and the second function to the right branch
const evenErrorMessage = (n: number) => {
if (n % 2 === 0) return `${n} is even but in Error State`;
return `${n} is odd and in Error State`;
};
const doubleOrError = either.bimap(evenErrorMessage, doubleIfEven);
const rightValue = either.right(2);
const leftValue = either.left(3);
doubleOrError(rightValue); // either.right(4)
doubleOrError(leftValue); // either.left("3 is odd and in Error State")
It’s really exactly the same as mapping Either same functions available, same api.
So you have a context containing your value, Option<number>. But you want to apply to your value another function that can maybe return null, Eg: maybeDouble: (number) => Option<number>. If you apply map as seen above, you’ll end up with an Option<Option<number>>, which is not easy to handle.
You could use flatten to transform an Option<Option> (or an Array<Array> etc) but we will more likely use chain
Chain will unwrap the value, apply the function and then combine the initial context with the new context. That’s it. What’s implied is in order to combine, you can only chain functions that returns “roughly” the same context as the original one. IE, you either.chain on functions returning Either.
chain takes a function as parameter (the function transforming the value) and can then be applied to the value.
const doubleIfEvenElseNone = (n: number) => n % 2 === 0
? option.some(2 * n)
: option.none
const optionEven = option.some(2);
const optionOdd = option.some(3);
option.chain(doubleIfEvenElseNone)(optionEven) // option.some(4)
pipe(
optionOdd,
option.chain(doubleIfEvenElseNone)
) // option.none
Please note as mentioned it is the same as
pipe(
optionEven,
option.map(doubleIfEvenElseNone), // option.some(option.some(4))
option.flatten // option.some(4)
);
option.chain === flow(option.map, option.flatten);
According to your datastructure, chains behave differently.
chain behavior on Options is pretty simple. It applies the function to your value if the original is some else it return none. And that’s about it!
const someOption = option.some(2);
const someOddOption = option.some(3);
const noneOption = option.none;
option.chain(doubleIfEvenElseNone)(someOption); // option.some(4)
option.chain(doubleIfEvenElseNone)(someOddOption); // option.none
option.chain(doubleIfEvenElseNone)(noneOption); // option.none
Remeber Either is generally considered as “Everything went well” (right branch) vs “Something happened” (left branch).
Let’s say you have an either Either<E, number>, where E is the left type, and number the right branch type
So chain will only transform your data if you are in the right branch. Otherwise, it will return your Either as it was.
To “flatten” your Either, there are two cases:
Either<E, *>). You can use the chain method and it will return an Either<E, *> (what is returned by your chained function)Either<D, *>). To flatten your Either, you will have to combine both error types. You have to use the chainW method, where W means widen, as you extend the error type. It will return an Either<E | D, *>const initialError = (): InitialError => "This is an initial Error";
const notEvenError = (num: number): NotEvenError => `${num} is not Even`;
// Here we instanciate three Either<InitialError, number>
const rightEvenValue = either.right(2);
const rightOddValue = either.right(3);
const leftValue = either.left(InitialError);
const doubleIfEven = (n: number): Either<NotEvenError, number> =>
n % 2 === 0 ? either.right(2 * n) : either.left(notEvenError(n));
either.chainW(doubleIfEven)(rightEvenValue); // either.right(4)
either.chainW(doubleIfEven)(rightOddValue); // either.left(NotEvenError)
either.chainW(doubleIfEven)(leftValue); // either.left(InitialError)
We could not use chain in the above example as the returned Either type of the function was different from the input one.
If you want to use your data but return it unaltered to the rest of your pipeline (like some kind of side effect), but still fail if something wrong happened, you can use the chainFirst (or chainFirstW) combinator!
const doubleAndTellIfEven = (n: number): Either<NotEvenError, string> =>
n % 2 === 0
? either.right(`${2 * n} is an even number!`)
: either.left(notEvenError(n));
const rightEvenValue = either.right(2);
const rightOddValue = either.right(3);
pipe(
rightEvenValue,
either.chainFirstW(doubleAndTellIfEven), // this goes right branch, but we drop the string returned and pass the initial value
either.getOrElse(() => 0) // here we get the original 2
);
pipe(
rightOddValue,
either.chainFirstW(doubleAndTellIfEven), // this goes left branch due to failed validation
either.getOrElse(() => 0) // here we get default 0
);
As mentioned, TaskEither works a lot like Either, so you can apply the same stuff as described above.
One interesting behaviour, is you have a TaskEither<value> and you want to apply to your value a function returning an Either. You would have a TaskEither<Either<value>>.
instead of then doing fromEither and flatten, you can use the chainEitherK (or chainEitherKW if error types are not equivalent) combinator, that does the same!
const rightEvenValue = taskEither.right(2);
const rightOddValue = taskEither.right(3);
// note this function returns an Either!
const doubleIfEven = (n: number): Either<NotEvenError, number> =>
n % 2 === 0 ? either.right(2 * n) : either.left(notEvenError(n));
pipe(
rightEvenValue
taskEither.chainEitherKW(doubleIfEven) // this returns TaskEither(4)
)
pipe(
rightOddValue
taskEither.chainEitherKW(doubleIfEven) // this returns TaskEither.left(NotEvenError)
)
One thing we really do often is “flip” two contexts we have. For example you make multiple queries and you end up with an Array<Either<E, A>> but you’d rather have an Either<E, Array<A>>.
Or you map a function that may fail on an Option and end up having an Option<Either> but you want it the other way around.
We can do it by using the sequence function of our data structures. When combining, please note an Array<Either> where some Either are right and other left, will turn into an Either.left
import { array, either, option, taskEither } from "fp-ts";
const arrayOfEither = [
either.right(42),
either.left(SomeError),
either.right(1337),
];
const eitherOfArray = array.sequence(either.Applicative)(arrayOfEither);
// eitherOfArray: Either<SomeError, Array<number>> == either.left(SomeError)
const arrayOrRight = [either.right(42), either.right(1337)];
const eitherOfArray = array.sequence(either.Applicative)(arrayOrRight);
// eitherOfArray: Either<SomeError, Array<number>> == either.right([42, 1337])
const optionOfTaskEither = option.some(taskEither.right(42));
const taskEitherOfOption = option.sequence(taskEither.Applicative)(
optionOfTaskEither
);
// taskEitherOfOption: TaskEither<SomeError, Option<number>> ==
// taskEither.right(option.some(42))
Another useful use case (overlapping a bit the above one), is when you apply a function returning another datatype.
You can simply use map of course, like you have an Option and wanna fetch something if the option is some, and you’ll end up with an Option<TaskEither>
import { TaskEither } from "fp-ts/TaskEither";
import { option } from "fp-ts";
const getUserPreferences: (userId: UserID) => TaskEither<UserNotFound, UserPreferences> = /* ... */;
const optionUserId = option.some(userId)
pipe(
optionUser,
option.map(getUserPreferences) // this will give you an Option<TaskEither.right(userPreferences)>
)
but maybe, you want it the other way around directly, because maybe you want to chain it with other TaskEither, or whatever reasons.
You can use the traverse function for that, allows you to “traverse” your context with a function.
traverse takes a first argument which is the datastructure that will be returned by the transformation function. And then two arguments, the first one being the data that will be transformed, and then the transformation function.
const getUserPreferences = (userId: UserID) => TaskEither<UserNotFound, UserPreferences>
const optionUserId = option.some(userId)
const result = option.traverse(taskEither.ApplicativeSeq)(getUserPreferences)(optionUserId)
// 👆 this returns a TaskEither<UserNotFound, Option<UserPreferences>>
pipe(
optionUser,
option.traverse(taskEither.ApplicativeSeq)(getUserPreferences) // this is the same as above.
)
Here is another example of using traverse-like method:
import { TaskEither } from "fp-ts/TaskEither";
import { readonlyArray, taskEither } from "fp-ts";
const getUserPreferences: (userId: UserID) => TaskEither<UserNotFound, UserPreferences> = /* ... */;
const userIds = [userId, anotherUserId, thirdUserId];
const result = taskEither.traverseSeqArray( // use taskEither.traverseArray if you want to run the Tasks in parallel
getUserPreferences
)(userIds)
// result is TaskEither<string, UserPreferences[]>
Not that if in the above example any of TaskEither returned by getUserPreferences is a Left, the result will also be a Left.
What if you need to get all the values returned by getUserPreferences that are Right though?
import { readonlyArray, task } from "fp-ts";
import { pipe } from "fp-ts/function";
// getUserPreferences and userIds are as in the above example
const result = pipe(
validAndInvalidUserIds,
// notice how in the following line we use Task instead of TaskEither
task.traverseArray(getUserPreferences), // this gives a Task<Either[]>
// next two lines allow to get all values from Eithers that are Right
task.map(readonlyArray.separate),
task.map(({ right }) => right)
);
// result is Task<UserPreferences[]>
Note that the following functions have the same logic:
task.traverseArray and readonlyArray.traverse(Task.ApplicativePar)taskEither.traverseArray and readonlyArray.traverse(TaskEither.ApplicativePar)task.traverseSeqArray and readonlyArray.traverse(Task.ApplicativeSeq)taskEither.traverseSeqArray and readonlyArray.traverse(TaskEither.ApplicativeSeq)You can futher explore the above code by pasting this snippet in a code sandbox.
If you already had to carry dependencies throught out multiple functions, you will quickly understand the value of Reader.
You can see the definition given by Giulio Canti here, and basically Reader<R,A> represents a function (r: R) => A where R will be your dependencies, and A the result.
Let’s take an example:
import { TaskEither } from "fp-ts/TaskEither";
import { taskEither, reader } from "fp-ts";
const foo = ({
firstId,
secondId,
eventData,
firstDependency,
secondDependency,
}): TaskEither<E, A> =>
pipe(
firstDependency.getById(firstId),
taskEither.chain(() => secondDependency.sendEvent({ secondId, eventData }))
);
// To call this foo method you'll write:
foo({ firstId, secondId, eventData, firstDependency, secondDependency });
Using Reader, you could write it this way:
import { ReaderTaskEither } from "fp-ts/ReaderTaskEither";
import { readerTaskEither, reader } from "fp-ts";
interface R1 {
firstDependency: Dep1;
}
interface R2 {
secondDependency: Dep2;
}
const foo = ({
firstId,
secondId,
eventData,
}): ReaderTaskEither<R1 & R2, E, A> =>
pipe(
getById(firstId),
readerTaskEither.chain(() => sendEvent({ secondId, eventData }))
);
// with:
const getById = (firstId) =>
pipe(
reader.ask<R1>(),
reader.map(({ firstDependency }) => firstDependency.getById(firstId))
);
const sendEvent = ({ secondId, eventData }) =>
pipe(
reader.ask<R2>(),
reader.map(({ secondDependency }) =>
secondDependency.sendEvent({ secondId, eventData })
)
);
// To call this foo method, you'll write:
foo({ firstId, secondId, eventData })({ firstDependency, secondDependency });
You need to adapt some methods to be able to properly use the Reader pattern, but once it’s done, you won’t need to carry your dependencies from a method to another as those will only be seen where they are needed.
At some point, you might need to call methods in parallel, for instance because you need to send data to multiple external services and you don’t want to decrease the performance of your use case. The Do notation can help you for this (here’s the official documentation if you want more details). Below is an example that shows you how to call two ReaderTaskEither in parallel :
import { ReaderTaskEither } from "fp-ts/ReaderTaskEither";
import { readerTaskEither } from "fp-ts";
import { pipe } from "fp-ts/function";
declare const foo: ReaderTaskEither<R1, E1, A1>;
declare const bar: ReaderTaskEither<R2, E2, A2>;
pipe(
readerTaskEither.Do,
readerTaskEither.apS("foo", foo),
readerTaskEither.apSW("bar", bar)
// this returns RTE.ReaderTaskEither<R1 & R2, E1 | E2, {readonly foo: A1; readonly bar: A2}>
);