I’m surprised about this statement, I would have said that exceptions are the consequence of an impure operation (that may or may not fail differently every time you call it).
I’m currently learning functional languages and have only limited knowledge, but from what I’ve read now you are right.
Throwing exceptions is pure, but catching them is impure.
In this case I guess the printLine function can throw an exception therefore the calling function must be declared with Exception?
I would even have said that both throwing and catching should be pure, just like returning an error value/handling should be pure, but the reason for the throw/returning error itself is impure. Like if you throw and ioerror it’s only after doing the impure io call, and the rest of the error reporting/handling itself can be pure.
Pure functions should be referentially transparent; you should be able to replace them with whatever value they evaluate to without changing the semantics of your code.
Throwing is referentially impure: what value do you get from calling x => thrownew RuntimeException()?
Instead, functional languages prefer to return a tagged union of the value or the error.
Functional languages typically have type inference, so the type signatures are entirely optional. I haven’t looked that deeply at unison, but I’d be entirely unsurprised if it had global type inference and if all or most type signatures were optimal.
It’s less that you have to declare something can do IO or throw an exception, and more that you’re calling something from the standard library that does IO or throws an exception.
Most stuff does neither. There’s a type level distinction between normal, regular pure code, and impure effectful code, so it’s easy to tell from the type signature whether a function is pure or not.
Do I really have to declare that something requires exceptions?
Yes, in functional programming you want to use pure functions. Exceptions are impure, therefore it has to be declared.
Other functional languages don’t even have exceptions
I’m surprised about this statement, I would have said that exceptions are the consequence of an impure operation (that may or may not fail differently every time you call it).
I’m currently learning functional languages and have only limited knowledge, but from what I’ve read now you are right. Throwing exceptions is pure, but catching them is impure.
In this case I guess the printLine function can throw an exception therefore the calling function must be declared with Exception?
I would even have said that both throwing and catching should be pure, just like returning an error value/handling should be pure, but the reason for the throw/returning error itself is impure. Like if you throw and ioerror it’s only after doing the impure io call, and the rest of the error reporting/handling itself can be pure.
Sounds good,
but would the preferred way be to use a wrapper type, which holds either the data or the error and avoid exceptions completely?
Pure functions should be referentially transparent; you should be able to replace them with whatever value they evaluate to without changing the semantics of your code.
Throwing is referentially impure: what value do you get from calling
x => throw new RuntimeException()?Instead, functional languages prefer to return a tagged union of the value or the error.
Functional languages typically have type inference, so the type signatures are entirely optional. I haven’t looked that deeply at unison, but I’d be entirely unsurprised if it had global type inference and if all or most type signatures were optimal.
It’s less that you have to declare something can do IO or throw an exception, and more that you’re calling something from the standard library that does IO or throws an exception.
Most stuff does neither. There’s a type level distinction between normal, regular pure code, and impure effectful code, so it’s easy to tell from the type signature whether a function is pure or not.
That’s one of the things I appreciate in a language/framework. Drives me nuts getting an exception from a dependency of a dependency of a dependency.
Even better if its baked into the type system and I can’t run my code without handling it.