Exceptions are often a better way to handle errors than returning them as values. We argue that traditional exceptions provide better user and developer experience, and show that they even result in faster execution.
You have to explicitly check if the return value is an error and propagate it. You write the same boilerplate if (err) return err over and over again, which just litters your code.
That’s only true in crappy languages that have no concept of async workflows, monads, effects systems, etc.
Sad to see that an intentionally weak/limited language like Go is now the counterargument for good modeling of errors.
I naively thought it I may as well take a job using Go, as learning a new language is broadening, and some people like it, so lets find out first hand… I knew it was a questionable choice, looking at how Go adoption tailed off a while ago.
Turns out I hate Go. Sure it’s better than C but that’s a very low bar, and C was never a good alternative choice for the use cases I’m encountering. I’m probably suffering from a codebase of bad Go, but holy shit it’s painful. So much silent propagation of errors up the stack so you never know where the origin of the error was. So very much boilerplate to expand simple activities into long unreadable functions. Various Go problems I’ve hit can be ameliorated if you “don’t do it like that”, but in the real world people “do it like that” all the time.
I’m really starting to feel like there are a lot of people in the company I’ve joined who like to keep their world obtuse and convoluted for job security.
I don’t know the answer to your question, but I think that what is needed is just a bit of syntactic sugar, e.g. Rust has ? for returning compatible errors without looking into them. That seems to be powered by Try trait, that may be a monad, but I am not fluent enough to check if it formally is.
In Maybe monadic, its monadic bind will automatically resolves any failed computation, and don’t need explicit checking.
for example, the code in Haskell looks something like the following:
fib: Int -> Int -> Maybe Int
fib depth idx =
do
guard (depth < 10)
n1 <- fib (depth - 1) (idx - 1)
n2 <- fib (depth - 1) (idx - 2)
return (n1 + n2)
Haskell type class system automatically figures out this is a maybe monad, and check for error accordingly.
Notice, unlike the C code the author provide, this haskell code will exit immediately when n1 failed and never compute n2, similar to the behavior of the exception code. Thus I believe his point about performance is at least unjustified, if not wrong.
Another interesting fact about this code is that there is nothing that is built into the compiler/interpretor (except the do expression, which is just a minor syntactical sugar), you don’t need to design special semantics for raise and catch. Everything here, guard, return and the definition of Maybe monad (which is in charge of propagating errors) is defibed by the user using normal functions, no metaprogramming involved.
Wouldn’t effect systems still be considered exceptions, but handled differently?
Yes, unlike monad, the error is propagated by the compiler/interpretor, instead of user defined. But unlike implicit effect, explicit effect (algebraic effect, throwable, etc.) makes it clear how the code can go wrong.
Although explicit error is more clear in general, there are special cases where explicit effect is undesirable. One such example is effect pollution, where low level effect that is unlikely to cause impure behavior is necessarily propgated through the call stack. Making the code more verbose and harder to handle.
That’s only true in crappy languages that have no concept of async workflows, monads, effects systems, etc.
You don’t even need to sit on your ass and wait for these data types to be added to standard libraries. There are countless libraries that support those, and even if that is somehow not an option it’s trivial to roll your own.
That’s only true in crappy languages that have no concept of async workflows, monads, effects systems, etc.
Sad to see that an intentionally weak/limited language like Go is now the counterargument for good modeling of errors.
I naively thought it I may as well take a job using Go, as learning a new language is broadening, and some people like it, so lets find out first hand… I knew it was a questionable choice, looking at how Go adoption tailed off a while ago.
Turns out I hate Go. Sure it’s better than C but that’s a very low bar, and C was never a good alternative choice for the use cases I’m encountering. I’m probably suffering from a codebase of bad Go, but holy shit it’s painful. So much silent propagation of errors up the stack so you never know where the origin of the error was. So very much boilerplate to expand simple activities into long unreadable functions. Various Go problems I’ve hit can be ameliorated if you “don’t do it like that”, but in the real world people “do it like that” all the time.
I’m really starting to feel like there are a lot of people in the company I’ve joined who like to keep their world obtuse and convoluted for job security.
Can you please demonstrate how async workflows and monads resolve this issue?
Wouldn’t effect systems still be considered exceptions, but handled differently?
I don’t know the answer to your question, but I think that what is needed is just a bit of syntactic sugar, e.g. Rust has
?for returning compatible errors without looking into them. That seems to be powered byTrytrait, that may be a monad, but I am not fluent enough to check if it formally is.In Maybe monadic, its monadic bind will automatically resolves any failed computation, and don’t need explicit checking.
for example, the code in Haskell looks something like the following:
fib: Int -> Int -> Maybe Int fib depth idx = do guard (depth < 10) n1 <- fib (depth - 1) (idx - 1) n2 <- fib (depth - 1) (idx - 2) return (n1 + n2)Haskell type class system automatically figures out this is a maybe monad, and check for error accordingly.
Notice, unlike the C code the author provide, this haskell code will exit immediately when
n1failed and never computen2, similar to the behavior of the exception code. Thus I believe his point about performance is at least unjustified, if not wrong.Another interesting fact about this code is that there is nothing that is built into the compiler/interpretor (except the
doexpression, which is just a minor syntactical sugar), you don’t need to design special semantics for raise and catch. Everything here,guard,returnand the definition ofMaybemonad (which is in charge of propagating errors) is defibed by the user using normal functions, no metaprogramming involved.Yes, unlike monad, the error is propagated by the compiler/interpretor, instead of user defined. But unlike implicit effect, explicit effect (algebraic effect, throwable, etc.) makes it clear how the code can go wrong.
Although explicit error is more clear in general, there are special cases where explicit effect is undesirable. One such example is effect pollution, where low level effect that is unlikely to cause impure behavior is necessarily propgated through the call stack. Making the code more verbose and harder to handle.
You don’t even need to sit on your ass and wait for these data types to be added to standard libraries. There are countless libraries that support those, and even if that is somehow not an option it’s trivial to roll your own.