proposal: encoding/json: add support of wrapped error in addErrorContext function

[DhZero]

What version of Go are you using (go version)?

$ go version
go version go1.15.5 darwin/amd64

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GOARCH="amd64"
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOTOOLDIR="/usr/local/opt/go/libexec/pkg/tool/darwin_amd64"

What did you do?

https://play.golang.org/p/fwgQFqafa3S

What did you expect to see?

Expected to see proper values set to err.Field and err.Struct even if error was wrapped by fmt.Errorf() in UnmarshalJSON of custom structure.

What did you see instead?

In decode.go in addErrorContext function, switch won't work as expected because type mismatch. So err.Field and err.Struct will be empty even if d.errorContext.Struct != nil || len(d.errorContext.FieldStack) > 0 condition was passed.

What did you propose?

Maybe it is better to use errors.As() instead of switch? Like this:

// addErrorContext returns a new error enhanced with information from d.errorContext
func (d *decodeState) addErrorContext(err error) error {
    if d.errorContext.Struct != nil || len(d.errorContext.FieldStack) > 0 {
        var e *UnmarshalTypeError
        if errors.As(err, &e) {
            e.Struct = d.errorContext.Struct.Name()
            e.Field = strings.Join(d.errorContext.FieldStack, ".")
        }
    }
    return err
}

[rsc]

/cc @dsnet

[rsc]

This proposal has been added to the active column of the proposals project and will now be reviewed at the weekly proposal review meetings. — rsc for the proposal review group

[networkimprov]

cc @mvdan

[dsnet]

The idea of using errors.As instead of a direct type assertion seems reasonable at face value, but I think we should be careful about how we proceed here. Currently, the error reporting is already a bit non-sensible. I'd like to see #43126 be addressed first before addressing this.

The reason I see #43126 as relevant is because it's unclear whether the error is supposed to report the location as 1) simply a struct field in a Go struct, or 2) the JSON path from the root JSON value. Currently, it is an odd mix of both. Depending on which information we intend to report, that changes how error wrapping would be performed.

[rsc]

Putting on hold for #43126. (If the investigation in #43126 ends up finding a suggested change then it would in turn become its own proposal.)

[rsc]

Placed on hold. — rsc for the proposal review group

[ar-sematell]

Another issue happening on this same function is that when a custom error is returned, and this error does not wrap any JSON errors, I lose context on which field the error happened.

If possible, please wrap all error types for keeping this context.