I've made 2 custom error types that implement error interface: customErr1 and customErr2. They both have a Message property.

The inspect(error) function checks for types using a typed switch statement and, if the error object is either of customErr or customErr2 type, accesses their Message property.

What I expected to see: typed switch statement compiled.

What I saw instead: compilation error:

prog.go:24: e.Message undefined (type error has no field or method Message)

The problematic line below is

case *customErr1, *customErr2:

What's weird is that the program works if the list is reduced to 1 item, i.e. one of the types is removed. However if the case list contains 2 items, the program doesn't compile and complains that "type error has no field or method Message". It's true that error doesn't implement Message, but that was the purpose of my type check in the first place.

Full example on Go playground, inlined here:

package main

import "fmt"

type customErr1 struct {
    Message string
}

func (err *customErr1) Error() string {
    return "Error() 1"
}

type customErr2 struct {
    Message string
}

func (err *customErr2) Error() string {
    return "Error() 2"
}

func inspect(err error) {
    switch e := err.(type) {
    case *customErr1, *customErr2:    // PROBLEMATIC LINE
        fmt.Println(e.Message)
    default:
        fmt.Println(e.Error())
    }
}

func main() {
    inspect(&customErr1{Message: "hello from error 1"})
    inspect(&customErr2{Message: "hello from error 2"})
}

go version go1.5.1 darwin/amd64