Feel free to read this on my blog!
Lodash's groupBy
I would bet if you have a sizeable Javascript/Typescript codebase you most likely are using lodash somewhere in there.
While Javascript has gotten more "batteries included" over the last few years, lodash still has many nice functions for manipulating arrays/objects.
One such function is groupBy. It groups a list by some predicate, in the simplest case it can just be a key in the objects of the array.
import _ from 'lodash';
interface Foo {
num: number;
someLiteral: 'a' | 'b' | 'c';
object: Record<string, any>;
}
const vals: Foo[] = [
{ num: 1, someLiteral: 'a', object: { key: 'value' } },
{ num: 2, someLiteral: 'a', object: { key: 'diffValue' } },
{ num: 1, someLiteral: 'b', object: {} },
];
console.dir(_.groupBy(vals, 'num'));
/*
{
'1': [ { num: 1, someLiteral: 'a' }, { num: 1, someLiteral: 'b' } ],
'2': [ { num: 2, someLiteral: 'a' } ]
}
*/
console.dir(_.groupBy(vals, 'someLiteral'));
/*
{
a:[
{ num: 1, someLiteral: 'a', object: [Object] },
{ num: 2, someLiteral: 'a', object: [Object] }
],
b: [ { num: 1, someLiteral: 'b', object: {} } ]
}
*/
This all seems to make sense, you can set what key you want to group on, and you get back an object whose keys are the values for found in the input array of objects.
Now if you're in a TypeScript code base I hope you are using the definitely typed lodash types to add some types to the lodash functions.
In this case the _.groupBy type looks roughly like (simplified from the actual code)
declare function groupBy<T>(collection: List<T>, key: string): Dictionary<T[]>;
interface Dictionary<T> {
[index: string]: T;
}
So a few things stick out here. First, the key type is just string, so there's nothing stopping me from doing _.groupBy(vals, "someKeyThatDoesNotExist").
Second, we have no restrictions at the type level of me grouping on a key whose value is not a valid object key (the value must be a subset of string | number | symbol). For example in Foo the object key's value was a record. Here's what happens when you try to group on that key.
console.dir(_.groupBy(vals, 'object'));
/*
{
'[object Object]': [
{ num: 1, someLiteral: 'a', object: [Object] },
{ num: 2, someLiteral: 'a', object: [Object] },
{ num: 1, someLiteral: 'b', object: {} }
]
}
*/
In this case the objects where coerced to string values so all elements of vals where grouped under the same weird [object Object] key. While this does not throw an error there is almost 0 chance you want this to happen in your code.
Finally, the return type of this function is Dictionary, while its "right" it could be "more right" by encoding that the returning object's keys would be the values of the grouping key in the input object.
Making our own groupBy
insert Bender joke here
To start making our own type safe groupBy, we first need some code that actually does the grouping logic. Let's start with that and some basic types.
// Note: PropertyKey is a builtIn type alias of
// type PropertyKey = string | number | symbol
// This lets us use "Record<PropertyKey, any>" to represent any object
// but is slightly nicer to use than the "object" type
function simpleGroupBy<T extends Record<PropertyKey, any>>(arr: T[], key: keyof T): any {
return arr.reduce((accumulator, val) => {
const groupedKey = val[key];
if (!accumulator[groupedKey]) {
accumulator[groupedKey] = [];
}
accumulator[groupedKey].push(val);
return accumulator;
}, {} as any);
}
console.dir(simpleGroupBy(vals, 'num'));
/*
{
'1': [
{ num: 1, someLiteral: 'a', object: [Object] },
{ num: 1, someLiteral: 'b', object: {} }
],
'2': [ { num: 2, someLiteral: 'a', object: [Object] } ]
}
*/
Cool the logic here seems to work, but obviously the types could use some love.
Let's start by adding a few more generics, so we can type the output correctly.
Your first change might be to make the return type Record<string, T[]> since the keys will be coerced to strings by JavaScript and the values will be the same values in the array.
This will unfortunately make typescript sad.
function sadAttempt<T extends object>(arr: T[], key: keyof T): Record<string, T[]> {
return arr.reduce((accumulator, val) => {
const groupedKey = val[key];
if (!accumulator[groupedKey]) {
accumulator[groupedKey] = [];
}
accumulator[groupedKey].push(val);
return accumulator;
}, {} as Record<string, T[]>);
}
The lines with accumulator[groupedKey] will error with Type 'T[keyof T]' cannot be used to index type 'Record<string, T>'. Here the keyof T could be any key in T so since not every key's value in T is a string typescript will not let you treat groupedKey as a string.
We can almost fix this by adding some more information on the input key by binding it to a new generic parameter, though there will still be some issues.
function betterSadAttempt<T extends Record<PropertyKey, any>, Key extends keyof T>(
arr: T[],
key: Key
): Record<T[Key], T[]> {
return arr.reduce((accumulator, val) => {
const groupedKey = val[key];
if (!accumulator[groupedKey]) {
accumulator[groupedKey] = [];
}
accumulator[groupedKey].push(val);
return accumulator;
}, {} as Record<T[Key], T[]>);
}
Here we added a new generic Key extends keyof T so when we supply a specific key to the function, the Key generic will be narrowed to that value. For example if we did betterSadAttempt(vals, 'someLiteral'), Key would exactly be 'someLiteral' instead of keyof Foo = 'someLiteral' | 'num' | 'object'
However, typescript is still sad on the Record<T[Key], T[]> lines with Type 'T[Key]' does not satisfy the constraint 'string | number | symbol'.
This error is similar to the error before, basically T[Key] can not be a key for the Record since it could be some weird value.
To accomplish this we need to make a helper type that filters down the allowed keys to only keys whose values are string | number | symbol.
We can use a mapped type to do just that
type MapValuesToKeysIfAllowed<T> = {
[K in keyof T]: T[K] extends PropertyKey ? K : never;
};
type Filter<T> = MapValuesToKeysIfAllowed<T>[keyof T];
This type helper does a few things. First it maps over all the values in T ([K in keyof T]) and makes the value the key if it is a subset of string | number | symbol (T[K] extends PropertyKey ? K), if it's not a subset its value will bethe never type. Finally, we use an index access type to get all values of the transformed object as a union. This step will drop all the never values automatically for us since adding never to a union is like saying or false its basically is a no op.
That was a mouthful so let's see an example
// from above
interface Foo {
num: number;
someLiteral: 'a' | 'b' | 'c';
object: Record<string, any>;
}
type MappedFoo = MapValuesToKeysIfAllowed<Foo>;
/*
{
num: "num";
someLiteral: "someLiteral";
object: never;
}
*/
// we replace the values of this object with just the key as a string literal or never
type FooKeys = Filter<Foo>;
// => "num" | "someLiteral"
// notice the never does not show up in the union
interface AllObjects {
object: Record<string, any>;
diffObject: Record<number, any>;
}
type MappedAllObjects = MapValuesToKeysIfAllowed<AllObjects>;
/*
{
object: never;
diffObject: never;
}
*/
type AllObjectsKeys = Filter<AllObjects>;
// => never
// the output is only never. Think of this like saying "false or false", the output will just be false
With this filter type helper function we can now properly limit the Key generic by replacing Key extends keyof T with Key extends Filter<T>.
Putting it all together
type MapValuesToKeysIfAllowed<T> = {
[K in keyof T]: T[K] extends PropertyKey ? K : never;
};
type Filter<T> = MapValuesToKeysIfAllowed<T>[keyof T];
function groupBy<T extends Record<PropertyKey, any>, Key extends Filter<T>>(
arr: T[],
key: Key
): Record<T[Key], T[]> {
return arr.reduce((accumulator, val) => {
const groupedKey = val[key];
if (!accumulator[groupedKey]) {
accumulator[groupedKey] = [];
}
accumulator[groupedKey].push(val);
return accumulator;
}, {} as Record<T[Key], T[]>);
}
const nums = groupBy(vals, 'num');
// nums = Record<number, Foo[]>
const literals = groupBy(vals, 'someLiteral');
// literals = Record<"a" | "b" | "c", Foo[]>
const sad = groupBy(vals, 'object');
// error: Argument of type '"object"' is not assignable to parameter of type 'Filter<Foo>'
Now this works great, we can only pass in keys that have valid values, and we even get autocomplete on it! However, one thing that bothers me is the error message in the last case.
While it's correct, saying not assignable to parameter of type 'Filter<Foo>' is not very useful to a user. This pops up sometimes with typescript where it won't show the underlying type and instead just show the higher level type helper instead.
To make the error message show the valid keys we can use a modified version of this "hack". Here instead of creating the Expand type in the post, we can make our own ValuesOf to replace the [keyof T] at the end of Filter
type ValuesOf<A> = A extends infer O ? A[keyof A] : never;
type Filter<T> = ValuesOf<MapValuesToKeysIfAllowed<T>>;
// was Filter<T> = MapValuesToKeysIfAllowed<T>[keyof T]
const sad = groupBy(vals, 'object');
// error: Argument of type '"object"' is not assignable to parameter of type '"num" | "someLiteral"'
Now we have type safety and good error messages!
Possible improvements
One thing this groupBy function lacks that the lodash groupBy gives is we do not allow you to pass a function instead of a key to group on.
The example in the lodash docs is
_.groupBy([6.1, 4.2, 6.3], Math.floor);
// { '4': [4.2], '6': [6.1, 6.3] }
While this is not perfect this mostly works
function groupByFunc<
RetType extends PropertyKey,
T, // no longer need any requirements on T since the grouper can do w/e it wants
Func extends (arg: T) => RetType
>(arr: T[], mapper: Func): Record<RetType, T[]> {
return arr.reduce((accumulator, val) => {
const groupedKey = mapper(val);
if (!accumulator[groupedKey]) {
accumulator[groupedKey] = [];
}
accumulator[groupedKey].push(val);
return accumulator;
}, {} as Record<RetType, T[]>);
}
const test = groupByFunc([6.1, 4.2, 6.3], Math.floor);
// test = Record<PropertyKey, Foo[]>
This works by only letting you pass in functions that return PropertyKey, but typescript does not seem to narrow the types. In this case test should be Record<number, Foo[]> since TS should infer the return type of the grouping function. If you know how to improve this function to make the return type narrow properly feel free to leave an issue/pr on my blog's GitHub!/JRMurr.github.io)!
Top comments (0)