Exceptions should be reserved for developer errors like edge cases that haven’t been considered or invalid bounds which mistakenly went unchecked.
In Java, when you declare a function that returns type T but might also throw exceptions of type A or B, the language treats it as though the function returned a Result<T, A|B>. And it forces the caller to either handle all possible cases, or declare that you're rethrowing the exception, in which case the behavior is the same as Rust's ? operator. (Except better, because you get stack traces for free.)
1. Checked exception don't integrate well with the type-system (especially generics) and functional programming. It's also incompatible with creating convenient helper functions, like Rust offers on Result.
2. Converting checked exceptions into runtime exception is extremely verbose, because Java made the assumption that the type of error distinguishes between these cases. While in reality errors usually start as expected in low-level functions, but become unexpected at a higher level. In Rust that's a simple `unwrap`/`expect`. Similarly converting a low level error type to a higher level error type is a simple `map_err`.
3. Propagation of checked exception is implicit, unlike `?` in Rust
Though Rust's implementation does have its weaknesses as well. I'd love the ability to use `Result<T, A | B>` instead of needing to define a new enum type.
I like the declaration side. I think part of where it misses the mark is the syntax on the caller side.
I feel like standard conditionals are enough to handle user errors while the heavy machinery of try-catch feels appropriately reserved for unexpected errors.
I also find a certain irony that forced checked results are exactly the same idea from CS type theory point of view, even if the implementation path is a different one.
The first big thing is that Java, especially in the days of when checked exceptions were a really big thing and less so in modern Java, was really into a certain kind of inheritance and interface design that didn't play well with error states and focused on the happy path. It is very difficult to make Java-esque interfaces that play well with checked exceptions because they like to abstract across network calls, in-memory structures, filesystem operations, and other side effectful tasks that have very different exception structures. An interface might have a single `writeData` method that might be backed by alternatively a write into an in-memory dictionary, a filesystem key-value store, a stateless REST API, or a bidirectional WebSocket channel which all have wildly different exceptions that can occur.
The second thing is that because checked exceptions were not actual return values but rather had their own special channel, they often did not play well with other Java API decisions such as e.g. streams or anything with `Runnable` that involved essentially the equivalent of a higher-order function (a function that takes as an argument another function). If e.g. you had something you wanted to call in a `Stream.map` that threw a checked exception, you couldn't use it, even if you notated in the enclosing method that you were throwing a checked exception because there was no way of telling `Stream.map` "if the function being `map`ed throws an exception rethrow it" which arose because checked exceptions weren't actual return values and therefore couldn't be manipulated the same way. You could get around it, but would have to resort to some shenanigans that would need to be repeated every time this issue came up for another API.
On the other hand if this wasn't a checked exception but was directly a part o the return value of a function, it would be trivial to handle this through the usual generics that Java has. And that is what something like `Result` accomplishes.
Isn't this what assertions are for? How would a user even know what exceptions they are supposed to catch?
IMO exceptions are for errors that the caller can handle in a meaningful way. Random programmer errors are not that.
In practice, exceptions are not very different from Result types, they are just a different style of programming. For example, C++ got std::expected because many people either can't or don't want to use exceptions; the use case, however, is pretty much the same.
If I have Result<Error, Value> and I change the Error, I have to change all places that are using the Error type and tweak the error handling in mapLeft or flatMapLeft.
If I instead raise Error and change it, I have to look at all the places where this error explodes and deal with it, not to mention, most languages won't even give me a compile time warning if I still keep the previous error type.
I agree that if language does not have do-notation, that it's a bit ugly to sprinkle map and flatMap everywhere. Good example of ugliness is https://github.com/repeale/fp-go
I think only an effect system, or a big environment object, places everything at 1 place, and when types change you have 1 place to edit the code. But starting immediately with an effect system (to abstract away control flow) or big env (to lift all ifs up) is premature.
Result violates the single responsibility principle and tries to make what are distinct paths into a single thing.
Only in languages that struggle to represent multiple shapes of values with a single type. I don't think I ever want to use a language with exceptions again.
Arguably, `Result` can also help when it is important that error are dealt with and not just allowed to bubble up.
I think Result<T> has its use, but I don't think this is a particular great example.
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 are deeper problems here that a Result type is not gonna fix.
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.
yep, `kotlin.Result` constraining your error type to `Throwable` is a real headache as it forces you to still model your domain logic via exceptions. it also means people can still accidentally throw these exceptions. not to mention the overhead of creating stack traces per instantiation unless you disable that on every subclass.
i recommend using https://github.com/michaelbull/kotlin-result?tab=readme-ov-f... (which has a nice breakdown of all the other reasons to avoid `kotlin.Result`)
} catch (exception: Exception) {
// log exception
throw exception
}
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.
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.
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!
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();
}
Promise<Result<number, string>>
type EitherT[F[_], E, A] = F[Either[E, A]]
def dosomething(): F[String, Number]
Isn't this beautiful: https://github.com/7mind/distage-example/blob/develop/bifunc... ?
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.
We currently use https://github.com/michaelbull/kotlin-result , which officially should work on KMP, but has some issues.
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.
// log exception
You either have the case that tech moves on and the LLM is out of date on anything new, so adoption slows or you have tech slowing down because it doesn't work with LLMs so innovation slows.
Either way, it's great if you're working on legacy in known technologies, but anything new and you have issues.
Can I write a spec or doc or add some context MCP? Sure, but these are bandaids.
This also means that exceptions can have stacktraces that only incur a cost on the unhappy path and even only if that exception is uncaught. While if you want a trace for a bad Result you are going to be doing a lot of extra book-keeping that will be thrown away
In general I agree that Results are the better abstraction, but there are sadly some tradeoffs that seem to be hard to overcome.
But more generally the happy-path/error-path distinction can be a bit murky. From my days writing Java back in the day it was very common to see code where checked exceptions were used as a sort of control flow mechanism, so you end up using the slow path relatively frequently because it was just how you handled certain expected conditions that were arbitrarily designated as "exceptions". The idea behind Result types to me is just that recoverable, expected errors are part of the program's control flow and should be handled through normal code and not some side-channel. Exceptions/panics should be used only for actually exceptional conditions (programming errors which break some expected invariant of the system) and immediately terminate the unit of work that experienced the exception.
try {
val user = authService.register(registrationRequest.email, registrationRequest.password)
return user
} catch (exception: Exception) {
// log exception
throw exception
}
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)Both snippets suffer from being too limited. The first, as you point out, catches too many exceptions. But the second.... What happens if the email address is taken? That's hardly exceptional, but it's an exception that the caller has to handle. Your natural response might be to check if the email address is taken before calling register, but that's just a race condition now. So you really need a result-returning function, or to catch some (but probably not all) of the possible exceptions from the method.
Because of your inconsistent line-breaks!
Now we need to invent do-notation, higher kinds and typeclasses and this code would be well composable.
I can't wait until they release rich errors and this nonsense with reinventing checked exceptions will finally end.
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.