This is a little off topic because I’m not asking about OCaml specifically, but I did start thinking about this while watching talks about implementing typed effects in OCaml. One of the questions that came up was, why not use monads instead of effect handlers? The answer was that jumping through …

[image] ninjaaron: One of the questions that came up was, why not use monads instead of effect handlers? The answer was that jumping through the stack after the pattern of exceptions was much cheaper than mondadic binding. And also effects handlers compose unlike monads, and monadic style is…

[image] monoidoid: What interface exactly? The kind of interface where the call that triggers the exception is explicitly marked even if it’s not handled, perhaps a bit like the ? operator in Rust. [image] monoidoid: Where would you jump? To wherever the error was unwrapped. I think …

[image] ninjaaron: To wherever the error was unwrapped. The unwrapping happens after the error. You can’t determine if some part of the continuation (that may involve arbitrary indirect function calls etc.) will unwrap or not.

The type system will ensure that every error is eventually unwrapped, since the error will be part of function types that generate errors. I’m quite sure this can be determined with static analysis. If the compiler can’t determine where the error is handled, the program doesn’t compile. The Java co…

let f x = if rand() < 0.5 then match x with Some x -> Some (x+1) | None -> None then x f (... raise ...) Does raise magically “jump” to the match? Or half of the time? Or we always jump to f, since it’s potentially handling, an we’re back to Frank, which I mentioned at the beginning.

I guess functions which don’t raise would in some sense be polymorphic over their inputs with regard to errors. I guess then the program should jump back in the stack to the caller of f, convert the value to true monadic value and pass it into f. Something like that. I can see how this might compl…

It’s been a long time since I knew this stuff, but: 0. let’s set aside imperative features like mutating update (“:=”) you can take a language with control flow and exceptions and convert it to pure lambda-calculus using a CPS transformation with two continuations: the “normal continuation” and…

Slightly OT: Would it be possible to implement monadic error handling the same way as exceptions?

Learning

yawaramin April 11, 2024, 8:24pm 8

2 Likes

Topic		Replies	Views
Exception vs Result Learning exceptions	60	10054	September 15, 2023
Updating the “Error Handling” tutorial Community user-feedback , ocamlorg , tutorial	27	2496	May 19, 2023
Specific reason for not embracing the use of exceptions for error propagation? Ecosystem	40	8014	December 24, 2018
Is there any consensus on which type to unify on for errors in result types? What's your preference? Learning monads , result	19	1180	March 19, 2024
Am I wrong about Effects? I see them as a step back Learning	19	4975	December 14, 2022

Slightly OT: Would it be possible to implement monadic error handling the same way as exceptions?

Related topics