Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Boxed: bringing algebraic types to TypeScript
Search
Matthias Le Brun
March 27, 2024
Technology
0
98
Boxed: bringing algebraic types to TypeScript
Matthias Le Brun
March 27, 2024
Tweet
Share
More Decks by Matthias Le Brun
See All by Matthias Le Brun
(why the hell did I) build a GraphQL client for the browser
bloodyowl
0
79
leveraging (algebraic data) types to make your UI rock @ jsheroes
bloodyowl
0
280
Leveraging (algebraic data) types to make your UI rock solid
bloodyowl
0
390
La drôle d'histoire de JavaScript
bloodyowl
0
350
Healthy Code Collaboration
bloodyowl
0
270
Simplify your UI management with (algebraic data) types
bloodyowl
0
800
Simplify your UI management with (algebraic data) types
bloodyowl
1
520
Migrating a large Reason+React codebase to hooks
bloodyowl
0
540
Third Party Hell (BestOfWeb)
bloodyowl
0
610
Other Decks in Technology
See All in Technology
API の仕様から紐解く「MCP 入門」 ~MCP の「コンテキスト」って何だ?~
cdataj
0
170
Tenstorrent HW/SW 概要説明
tenstorrent_japan
0
400
TerraformをSaaSで使うとAzureの運用がこんなに楽ちん!HCP Terraformって何?
mnakabayashi
0
170
In Praise of "Normal" Engineers (LDX3)
charity
2
1k
IIWレポートからみるID業界で話題のMCP
fujie
0
500
VCpp Link and Library - C++ breaktime 2025 Summer
harukasao
0
200
OAuth/OpenID Connectで実現するMCPのセキュアなアクセス管理
kuralab
2
240
Amazon Q Developer for GitHubとAmplify Hosting でサクッとデジタル名刺を作ってみた
kmiya84377
0
3.5k
キャディでのApache Iceberg, Trino採用事例 -Apache Iceberg and Trino Usecase in CADDi--
caddi_eng
0
150
QAはソフトウェアエンジニアリングを学んで実践するのが大事なの
ymty
1
400
ユーザーのプロフィールデータを活用した推薦精度向上の取り組み
yudai00
0
420
「どこにある?」の解決。生成AI(RAG)で効率化するガバメントクラウド運用
toru_kubota
2
450
Featured
See All Featured
Being A Developer After 40
akosma
90
590k
Code Review Best Practice
trishagee
68
18k
[RailsConf 2023] Rails as a piece of cake
palkan
55
5.6k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
20
1.3k
Facilitating Awesome Meetings
lara
54
6.4k
Fashionably flexible responsive web design (full day workshop)
malarkey
407
66k
Building a Scalable Design System with Sketch
lauravandoore
462
33k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
8
780
Why Our Code Smells
bkeepers
PRO
337
57k
Balancing Empowerment & Direction
lara
1
300
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
31
2.4k
Build your cross-platform service in a week with App Engine
jlugia
231
18k
Transcript
Boxed
Matthias Le Brun @bloodyowl lead engineering manager chief shitpost office
the easiest way to provide banking features (accounts, payments, cards…) we're hiring! @
None
None
None
None
PREVIOUSLY ON Matthias tries to convert people to FP
Algebraic Data Types yay
type State = { data?: Data; error?: Error; };
type State = { data?: Data; error?: Error; }; type
Result<Data> = | Ok<Data> | Error<Error>; error data ERROR NULL NULL DATA NULL NULL ERROR DATA
{ data: null; error: null; };
type Result<Data> = | Ok<Data> | Error<Error>;
type State = { data?: Data; error?: Error; }; type
Result<Data> = | Ok<Data> | Error<Error>; state ERROR DATA
API response API response
API response API response
type State = { data?: Data; error?: Error; }; type
Result<Data> = | Ok<Data> | Error<Error>; 2 2 1 1 4 2 x = + =
Algebra it's just basic math™
+ x
quiz 157 x 8341 = ?
quiz 999 + 1 = ?
union types
monads functors
functor
monad
[1, 2, 3].map(x => x * 2) // [2, 4,
6] [1, 2, 3].flatMap(x => [x, x * 2]) // [1, 2, 2, 4, 3, 6]
map: <A, B>(T<A>, (a: A) => B) => T<B>: flatMap:
<A, B>(T<A>, (a: A) => T<B>) => T<B>;
None
null & undefined are bad
errors aren't always exceptions
promises were made the wrong way in JS
promises were made the wrong way in JS
promises were made the wrong way in JS
promises were made the wrong way in JS
THAT WAS Matthias tries to convert people to FP NOW
LET'S TALK ABOUT BOXED
✨ new job ✨ 2022
«we use TypeScript»
None
let's look at the ecosystem™
fp-ts
pipe( token, TaskEither.bindW("user", getTokenFromUser), TaskEither.bindW("project", getProject), TaskEither.chainW(getMembership), ) no autocomplete
on current value
pipe( token, TaskEither.bindW("user", getTokenFromUser), TaskEither.bindW("project", getProject), TaskEither.chainW(getMembership), x => {}
) need to inspect argument for type
pipe( token, TaskEither.bindW("user", getTokenFromUser), TaskEither.bindW("project", getProject), TaskEither.chainW(getMembership), x => {}
) need to inspect argument for type not a good DX
export declare const chainFirstTaskK: <A, B>(f: (a: A) = >
T.Task<B>) = > <E>(first: TaskEither<E, A>) = > TaskEither<E, A> ah just what I was looking for
we've looked at the ecosystem™
«let's try to make these types work with a good
DX in a structurally typed language» — me, allegedly drunk
structural typing
class Some<A> {} class None {} const value: Some<string> =
new None();
what we need → good DX → interop with ts-pattern
→ fast
what we needed → chaining API → autocomplete → less
imports
1st iteration: naive wanted the types to be opaque but
safe
1st iteration: naive discriminator in TS (aka poor man's nominal
types)
1st iteration: naive and we're storing the state in the
class instance
None
1st iteration: naive wanted the types to be opaque but
safe
1st iteration: naive wanted the types to be opaque but
safe
2nd iteration: anger
2nd iteration: anger
2nd iteration: anger
2nd iteration: anger
3rd iteration: make it fast
initial release
initial release
type Option<A> = Some<A> | None;
type Result<A, E> = Ok<A> | Error<E>;
type AsyncData<A> = | NotAsked | Loading | Done<A>;
Future<A>; better promises™ cancellable monadic leaves error state to Result
const parsed = input != null ? parseInput(input) : undefined;
const transformed = parsed != null ? transform(parsed) : undefined; const printed = transformed != null ? print(transformed) : undefined; const value = printed != null ? prettify(printed) : "fallback"; input .map(parseInput) .flatMap(transform) .map(print) .map(prettify) .getWithDefault("fallback");
user-testing
None
insights people need human-written docs → we doubled down on
examples & added search
insights don't try to replicate another language → give good
defaults for TS
Result< Data, ServerError > Result< A, ServerError | FinalizeError >
insights people are familiar with just a few data types
→ reduce confusion
insights people can be docs-first or type-first → experience needs
to be good for both
insights readability matters → chaining API was the way to
go
insights mapping on existing JS knowledge helps → if it
exists, name it the same way
insights pipe() syntax had impact on the runtime → expressiveness
is important
4th iteration implement the feedback rename some methods update some
defaults
5th iteration: what the actual fu
5th iteration: what the actual fu
recently fix the errors
recently improved perf by 2x to 10x
recently improved perf by 2x to 10x blazing fast certi
fi ed™ ⚡
Boxed → github.com/swan-io/boxed → swan-io.github.io/boxed
None
$ yarn add @swan-io/boxed
Matthias Le Brun @bloodyowl we're still hiring thank you! 🙏