Result is all I need (rockyj-blogs.web.app)

With the ? syntax in Rust results and exceptions are the same thing. I posit that the former is superior. It is unfortunate that results have worse performance but I don't see any reason why. Results that bubbles up all the way ought to be identical to an uncaught exception.

gosh...

        try {
            val user = authService.register(registrationRequest.email, registrationRequest.password)

            return user
        } catch (exception: Exception) {
            // log exception
            throw exception
        }

no, no, no!

the whole point of the exceptions (and moreso of the unchecked ones) is to be transparent!

if you don't know what to do with an exception do NOT try to handle it

that snippet should just be

    return authService.register(registrationRequest.email, registrationRequest.password)

Since this looks like Kotlin, worth pointing out that there is a kotlin class in the standard library called Result. I've been using that for a few things. One place that I'm on the fence about but that seems to work well for us is using this in API clients.

We have a pretty standard Spring Boot server with the usual reactive kotlin suspend controllers. Our api client is different. We were early adopters of kotlin-js on our frontend. Not something I necessarily recommend but through circumstances it was the right choice for us and it has worked well for us in the last five years. But it was a rough ride especially the first three of those.

As a consequence, our API client is multiplatform. For every API endpoint, there's a suspend function in the client library. And it returns a Result<T> where T is the deserialized object (via kotlinx serialization, which is multiplatform).

On the client side, consuming a result object is similar to dealing with promises. It even has a fold function that takes a success and error block. Basically failures fall into three groups: 1) failures (any 4xx code) that probably indicate client side bugs related to validation or things that at least need to be handled (show a message to the user), 2) internal server errors (500) that need to be fixed on the server, and 3) intermittent failures (e.g. 502, 503) which usually means: wait, try again, and hope the problem goes away.

What I like about Result is making the error handling explicit. But it feels a bit weird to client side construct an Exception only to stuff it into a Result.error(...) instead of actually throwing it. IMHO there's a bit of language friction there. I also haven't seen too many public APIs that use Result. But that being said, our multiplatform client works well for our use.

But I can't expose it to Javascript in its current form; which is something I have been considering to do. This is possible with special annotations and would mean our multiplatform client would be usable in normal react/typescript projects and something I could push as an npm. But the fact my functions return a Result makes that a bit awkward. Which is why I'm on the fence about using it a lot.

So, nice as a Kotlin API but good to be aware of portability limitations like that. You would have similar issues exposing Kotlin code like that to Java.

This is why Haxe is awesome. You can target a sloppy langauge and still get the benefits os a ML-like typesystem.

For the typescript world, there is neverthrow[0] which offers a similar Result type.

[0]: https://github.com/supermacro/neverthrow

Ah, Either. Didn't recognize you from the first glance.

Now we need to invent do-notation, higher kinds and typeclasses and this code would be well composable.

Smells like something that Effect-TS is designed to solve in the TypeScript world.

Good article.

Maybe you could look up the Try monad API (Scala or Vavr works in Java + Kotlin), by using some extra helper methods you can have something probably a little bit lighter to use.

I believe your example would look like the following with the Try monad (in Java):

  public UserDTO register(UserRegistrationRequest registrationRequest) {
    return Try.of(() -> authService.userExists(registrationRequest.email))
      .filter(Objects::isNull, () -> badRequest("user already exists"))
      .map(userId -> authService.register(registrationRequest.email, registrationRequest.password))
      .get();
  }

The login() function would be using the same pattern to call authService.verify() then filtering nonNull and signing the JWT, so it would be the same pattern for both.

Stick your services into the type too, and you have `Effect`[0], the ultimate monad :)

[0]: https://effect.website/

I've been toting around and refining a Result<> implementation from project to project from Java 8 onwards. Sealed classes in 17+ really make it shine.

I wish Oracle et al. had the courage to foist this into the standard library, damn the consequences. Whatever unanticipated problems it would (inevitably) create are greatly outweighed by the benefits.

I've written Pair<> about a dozen times as well.

Use the builtin Result class and runCatching/fold and be done with it. Yes, it has shortcomings but works well enough in practice.

I love how everyone here shares real experience with Kotlin and Result, it’s cool to see different views that actually teach something.

Result<T> is a built-in in kotlin, but this enforces that the error type is a Throwable

If you fancy that an error could be just a type, not necessarily a Throwable, you might like Result4k - it offers a Result<T,E>

https://github.com/fork-handles/forkhandles/tree/trunk/resul...

disclaimer: I contribute to this.

Result is great but it ideally needs extensible union types (polymorphic variants) plus exhaustive pattern matching to work well.

Not a fan. Code branches and this is Good Thing(TM). Result violates the single responsibility principle and tries to make what are distinct paths into a single thing. If your language has exceptions and returned values as distinct constructs then obfuscating them with Result means you end up fighting the language which becomes apparent fairly quickly. It's also a frustrating experience to want to interact with returned values directly and constantly have to deal with a polymorphic wrapper.

I always think more languages should support Result… but only to handle expected error states. For example, you may expect that some functions may time out or that a user may attempt an action with an invalid configuration (e.g., malformed JSON).

Exceptions should be reserved for developer errors like edge cases that haven’t been considered or invalid bounds which mistakenly went unchecked.

This is really just a syntactical issue. Not one of types or semantics.

Non trivial operations have errors when the happy path fails. And with web apps IO can fail anytime, anywhere for any reasons.

Sometimes you want to handle them locally, sometimes globally. The question is how ergonomic it is to handle this all for a variety of use cases.

We keep reinventing the wheel because we insist that our own use cases are “special” and “unique”, but they really aren’t.

Personally, I think Java’s proposal on catching errors in switches, next to ordinary data is the right step forward.

Monads are great. You can do lots of great things in them, but ergonomic they are not. We should avoid polluting our type systems where possible.

    fun register(registrationRequest: UserRegistrationRequest): UserDTO {
        return success(registrationRequest)
            .flatMap { validRequest ->
                throwIfExists(validRequest.email) { authService.userExists(validRequest.email) }
            }.flatMap {
                runWithSafety { authService.register(registrationRequest.email, registrationRequest.password) }
            }.getOrThrow()
    }

There appears to be some useless code there. Why is validRequest.email being passed to throwIfExists twice? Why is throwIfExists implemented to return the email if the following line (runWithSafety) just ignores that returned email and instead gets the email from registrationRequest.email?

The imperative code has

    // log exception

which doesn't exist in the Result version.

There's certainly situations where this pattern creates some tricky ambiguity, but more often than not Result is quite an ergonomic pattern to work with.

In case it's useful for anyone, here is a simple plug-in-play TypeScript version:

```

type Ok<T = void> = T extends undefined ? { ok: true; } : { ok: true; val: T; };

type Err<E extends ResultError = ResultError> = { ok: false; err: E; };

type Result<T = void, E = ResultError> = { ok: true; val: T; } | { ok: false; err: E | ResultError; };

class ResultError extends Error { override name = "ResultError" as const; context?: unknown; constructor (message: string, context?: unknown) { super(message); this.context = context; } }

const ok = <T = void>(val?: T): Ok<T> => ({ ok: true, val: val, } as Ok<T>);

const err = (errType: string, context: unknown = {}): Err<ResultError> => ({ err: new ResultError(errType, context), ok: false, });

```

```

const actionTaker = await op().then(ok).catch(err);

if (result.ok) // handle error

else // use result

```

I will be forever grateful to the developer first introduced to this pattern!

With regard to AI, why not throw this whole article in an .md file and point CLAUDE.md to it? Codex is better at following rules so maybe you’d have more luck with that. But yeah, AI won’t code your way by default. People expect way too much out of the interns, they need direction.

> At the first glance, this code looks noisier and hard to understand

Because of your inconsistent line-breaks!

How to rewrite boring, easily understood code into abomination. I'm not surprised to see Kotlin, for some reason there's a huge inferiority complex in Kotlin community where you have to write the most convoluted pseudo-fp code possible (not smart enough to use ML or Haskell, but still want to flex on Java noobs).

I can't wait until they release rich errors and this nonsense with reinventing checked exceptions will finally end.

I am not a fan of function chaining in the style advocated in the article. In my experience functional abstractions always add function call indirection (that may or may not be optimized by the compiler).

You don't need a library implementation of fold (which can be used to implement map/flatmap/etc). Instead, it can be inlined as a tail recursive function (trf). This is better, in my opinion, because there is no function call indirection and the trf will have a name which is more clear than fold, reducing the need for inline comments or inference on the part of the programmer.

I also am not a fan of a globally shared Result class. Ideally, a language has lightweight support for defining sum/union types and pattern matching on them. With Result, you are limited to one happy path and one error path. For many problems, there are multiple successful outputs or multiple failure modes and using Result forces unnecessary nesting which bloats both the code for unpacking and the runtime objects.

It is always funny to see that we try and force formulas to the early elementary shape that people learn. Despite the fact that chemistry, biology, physics, etc. all have advanced shapes for equations that do not have the same concerns.

Similarly, when constructing physical things, it is not uncommon to have something with fewer inputs than outputs. Along with mode configured transfer of input to outputs.