Hacker News Reader

Comment by howeyc | Hacker News Reader

I think what GP is getting at that if you call a function with one return value, you can ignore it by not assigning anything to the result of the function.

http://play.golang.org/p/9J46QRHul1

They presumably would like to force you to use the return value if it is of type error.

View on HN

Ah, got it. I thought it was way more sinister than that.

On a side note, does any programming languages force you to use the return value of every function call?

Go forces you to either assign the return value to a variable and then use that variable, or pretend it doesn't exist at all. I think they were trying to get rid of the middle "assign it to a variable, deal with it later" case, to encourage people to write error-handling code there and then, but I've found that instead it pushes me towards completely ignoring error codes, then because there's no "unused variable 'errorCode'" warning, I forget to go back and robustify the code :(

I probably /should/ get into the habit of writing error-handling code at the same time as writing the main-path code, but for someone who's just started learning the language and just wants to see what works, I find it really frustrating >_<

You soon get used to the pattern:

  blah, err := foo(...)
    if err != nil {
      return nil, err
  }

and it's sister:

  blah, err := foo(wibble)
    if err != nil {
      return nil, errors.New("foo error with blah due to: " + wibble)
  }

After a while I got to like passing errors carefully back up the call stack , cleaning them up, rewriting them and making them more useful.

My Go programs have far better (more detailed & relevant) error messages than any of my other code.

> You soon get used to the pattern:

Yes, we all know how good humans are at work which no computer could ever do such as repetitive output of pointless generic boilerplate.

My point is that translating an error from one call to the previous is neither generic nor pointless.

An exception and line-number (and a call-stack dump) is great for a developer, but a good textual error message is needed for a real user.

I agree with sambeau on this one. I have come to like the go way of doing things: it's extremely simple, doesn't take much thought, and after a little while is very easy to read.

> My point is that translating an error from one call to the previous is neither generic nor pointless.

The first pattern does not do that. The second one is still mostly pointless generic boilerplate:

    blah, err := 
    if err != nil {
        return nil, errors.New()
    }

is all boilerplate, only two parts are not: the call itself and the error message.

> An exception and line-number (and a call-stack dump) is great for a developer, but a good textual error message is needed for a real user.

1. having exceptions doesn't preclude handling them and writing user-readable error messages

2. exceptions are not the only alternative to C-style error handling

Can you tell me about the other alternatives? I'm actually interested in this and I have searched for alternatives at least to see how they look.

I'm sure I don't know most of them (e.g. I have no idea how Prolog does error handling), but I know of:

* the Haskell/ML statically typed approach, error possibility is type-encoded through an Option or Either type; even in cases where it is possible to implicitly ignore return values[0] the compiler will warn or error; the standard library provides high-level operations to pipeline results (build a pipeline of operations which will only transform non-erroneous results, and will let errorneous ones go through)

* the Erlang dynamically typed approach which makes use of tagged pattern matching to make value assertions trivial. It is not shorter than the Go version when fully handling things, but assertion can be used to fault processes during development or when in-process user-visible reporting is irrelevant (fairly common in erlang) yet remain easy to find and replace with more involved handling if needs be

[0] because haskell strives for purity, ignoring return values in non-monadic code makes very little sense and I'm not sure it's possible at all short of unsafe* functions. It is possible in monadic code (e.g. do blocks)

You might like:

   _, err := somethingThatFails(arg)
   if err != nil {
      return fmt.Errorf("failed with arg=%v, %v", arg, err)
   }

Now the use for this is not obvious from this toy example, but having a `printf`-like function that produces error is great to return errors with more context.

The difference from your example is that the info in `err` is not lost when passed up.

I do like that.

(I just picked (and cleaned up) the first piece of simple error code that I thought would be easy to use.)

And you might like:

  if _, err := somethingThatFails(arg); err != nil {
    return fmt.Errorf("failed with arg=%v, %v", arg, err)
  }

In that case I was ignoring the returned value for the sake of the example, otherwise you lose the declaration when it's in a `if`.

... then because there's no "unused variable 'errorCode'" warning, I forget to go back and robustify the code

Well, you can always assign the error code to '_', and then just grep for those later. Or you can take the 10 seconds and just add the code to return the error as sambeau suggests.

If it's a fairly big function, or a common way of dealing with errors in a package you can always define a func (nested or not):

  func foo() error {
    handleErr = func(err error) bool {
      if err != nil {
         log.Println(err)
         // do other stuff
         return true
      }
      return false
    }
    err := bar()
    if handleErr(err) { return err }
  }

It's nice and concise.

OCaml gives you a warning ("this function should return unit") if you don't use the result of each function call, unless that function returns unit (void). You can suppress it with the "ignore" function or with "let _ = some_function() in …"

I would like to implement this feature in Rust as well, as it would have saved me hours of debugging a couple of times.

> OCaml gives you a warning ("this function should return unit") if you don't use the result of each function call, unless that function returns unit (void). You can suppress it with the "ignore" function or with "let _ = some_function() in …"

GHC can also do so optionally (with -fwarn-unused-do-bind) (and I'm unsure there even is a way to ignore return values outside of do blocks)

Reply on news.ycombinator.com