proposal: Go 2: operator to cause early function return on error

[klaidliadon]

Proposal

This proposal aims to reduce the boilerplate of basic error handling by introducing a new identifier.

We will use ? in this proposal, which will behave as follows: when a error is assigned to ?
the function would return immediately with the error received.

As for the other variables their values would be:

• if they are not named zero value
• the value assigned to the named variables otherwise

The use of ? would allow to use += and similar operators on the other value, as if ? was not there.
When ? receives an error the last value of the named value would be returned.

Examples

Anonymous

From:

func foo(path string) (io.ReadCloser, error) {
    f, err := os.Open(path)
    if err != nil {
        return nil, err
    }
    if _, err = fmt.Fprint(f, "foo\n"); err != nil {
        return nil, err
    }
    return f, nil
}

To:

func foo(path string) (io.ReadCloser, error) {
	f, ? := os.Open()
	_, ? = fmt.Fprint(f, "foo\n")
	return f, nil
}

Named

From:

func foo(r io.Reader) (n int, err error) {
    b := make([]byte, 1024)
    for {
        a, err := r.Read()
        n += a
        if err != nil {
            return n, err
        }
    }
}

To:

func foo(r io.Reader) (n int, err error) {
    b := make([]byte, 1024)
    for {
        n, ? += r.Read()
    }
}

[iand]

This appears to be covered by #32500

[anacrolix]

But what about wrapping errors?

[MichaelTJones]

This is pretty. I have one question about your phrasing to be certain of understanding. When you wrote:

func foo(r io.Reader) (n int, err error) {
    b := make([]byte, 1024)
    for {
        n, ? += r.Read()
    }
}

and mentioned "last value would be used" you mean this:

func foo(r io.Reader) (n int, err error) {
    b := make([]byte, 1024)
    for {
        tempN, tempE := r.Read()
        if tempE != nil {
            return n, tempE // "last" means n, not n+tempN, right?
        }
        n += tempN
    }
}

May be some debate about reserving the '?' for other uses but the code examples sure look nice.

[alanfo]

I don't know whether this has been suggested before - frankly I have lost track of what has been suggested in the error handling saga! - but I agree with @MichaelTJones that the syntax is easy on the eye and it would be backwards compatible as ? is not currently used for any other purpose.

As far as decorating errors is concerned, you could use defer as in the try proposal (#32437) which would require the error return parameter to be named.

However, I still prefer the try built-in proposal for the following reasons:

1. You wouldn't be able to use ? more than once in the same line.

2. You'd always need an assignment statement even if you weren't interested in storing the results. So instead of :

    r := try(os.Open(src))
    defer r.Close()

    w := try(os.Create(dst))
    defer func() {
        w.Close()
        if err != nil {
            os.Remove(dst)
        }
    }()

    try(io.Copy(w, r))
    try(w.Close())

you'd have to write:

    r, ? := os.Open(src)
    defer r.Close()

    w, ? := os.Create(dst)
    defer func() {
        w.Close()
        if err != nil {
            os.Remove(dst)
        }
    }()

    _, ? := io.Copy(w, r) // assignment statement needed here
    ? := w.Close()        // and again here

3. There would be no way to extend the proposal to support, say, named error handlers as you could do with try by simply adding a second parameter.

[networkimprov]

This is one of the most common suggestions in the responses to check/handle:
https://github.com/golang/go/wiki/Go2ErrorHandlingFeedback#recurring-themes

I filed a sophisticated variant of this in #32500, and another in #27519, both of which add named handlers -- also a common request, documented in the link above.

Many folks, including Go team members, aren't keen to introduce a new symbol. I don't get why, as this is a clear way to simplify error checks without

• burying them in the RHS with v = try(f()), or
• cluttering the LHS with try v = f()

And it's not confined to type error nor the last return value.

[ianlancetaylor]

For what it's worth, I think @griesemer addressed the concerns with adding new symbols in https://github.com/golang/proposal/blob/master/design/32437-try-builtin.md :

Go has been designed with a strong emphasis on readability; we want even people unfamiliar with the language to be able to make some sense of Go code (that doesn’t imply that each name needs to be self-explanatory; we still have a language spec, after all). So far we have avoided cryptic abbreviations or symbols in the language, including unusual operators such as ?, which have ambiguous or non-obvious meanings.

[alanfo]

Ignoring the arguments for and against, when I compare the snippets in my previous post, I still feel more comfortable with try rather than ?. The latter somehow seems more obtrusive when there are a lot of them though I guess that's just a matter of taste.

[networkimprov]

Go is nicer to read than Rust or Bash. But it did introduce 3 operators and a dedicated symbol:
.() <- := _
They aren't cryptic when you know them.

This would add a dedicated symbol, for a case more common than any of the above.

The only functional complaint I've heard is that it doesn't provide a subexpression. However, the fact that try() does has provoked tremendous angst :-)

IMHO, a reasonable case for a subexpression is g(try shouldNotFail(arg)) where an error yields a panic. Or similarly, where an error is explicitly ignored.