Hiding nil values, understanding why Go fails here

I fail to understand how to correctly assure that something is not nil in this case:

package main

type shower interface {
  getWater() []shower
}

type display struct {
  SubDisplay *display
}

func (d display) getWater() []shower {
  return []shower{display{}, d.SubDisplay}
}

func main() {
  // SubDisplay will be initialized with null
  s := display{}
  water := s.getWater()
  for _, x := range water {
    if x == nil {
      panic("everything ok, nil found")
    }

    // First iteration display{} is not nil and will
    // therefore work, on the second iteration
    // x is nil, and getWater panics.
    x.getWater()
  }
}

The only way I found to check if that value is actually nil is by using reflection.

Is this really wanted behaviour? Or do I fail to see some major mistake in my code?

Play link here

Solution

The problem here is that shower is an interface type. Interface types in Go hold the actual value and its dynamic type. More details about this: The Laws of Reflection #The representation of an interface.

The slice you return contains 2 non-nil values. The 2nd value is an interface value, a (value;type) pair holding a nil pointer value and a *display concrete type. Quoting from the Go Language Specification: Comparison operators:

Interface values are comparable. Two interface values are equal if they have identical dynamic types and equal dynamic values or if both have value nil.

So if you compare it to nil, it will be false. If you compare it to an interface value representing the pair (nil;*display), it will be true:

if x == (*display)(nil) {
    panic("everything ok, nil found")
}

This seems unfeasible as you'd have to know the actual type the interface holds. But note that you can use reflection to tell if a non-nil interface value wraps a nil value using Value.IsNil(). You can see an example of this on the Go Playground.

Why is it implemented this way?

Interfaces unlike other concrete types (non-interfaces) can hold values of different concrete types (different static types). The runtime needs to know the dynamic or runtime-type of the value stored in a variable of interface type.

An interface is just a method set, any type implements it if the same methods are part of the method set of the type. There are types which cannot be nil, for example a struct or a custom type with int as its underlying type. In these cases you would not need to be able to store a nil value of that specific type.

But any type also includes concrete types where nil is a valid value (e.g. slices, maps, channels, all pointer types), so in order to store the value at runtime that satisfies the interface it is reasonable to support storing nil inside the interface. But besides the nil inside the interface we must store its dynamic type as the nil value does not carry such information. The alternate option would be to use nil as the interface value itself when the value to be stored in it is nil, but this solution is insufficient as it would lose the dynamic type information.

Some people say that Go's interfaces are dynamically typed, but that is misleading. They are statically typed: a variable of interface type always has the same static type, and even though at run time the value stored in the interface variable may change type, that value will always satisfy the interface.

In general if you want to indicate nil for a value of interface type, use explicit nil value and then you can test for nil equality. The most common example is the built-in error type which is an interface with one method. Whenever there is no error, you explicitly set or return the value nil and not the value of some concrete (non-interface) type error variable (which would be really bad practice, see demonstration below).

In your example the confusion arises from the facts that:

you want to have a value as an interface type (shower)
but the value you want to store in the slice is not of type shower but a concrete type

So when you put a *display type into the shower slice, an interface value will be created, which is a pair of (value;type) where value is nil and type is *display. The value inside the pair will be nil, not the interface value itself. If you would put a nil value into the slice, then the interface value itself would be nil and a condition x == nil would be true.

Demonstration

See this example: Playground

type MyErr string

func (m MyErr) Error() string {
    return "big fail"
}

func doSomething(i int) error {
    switch i {
    default:
        return nil // == nil
    case 1:
        var p *MyErr
        return p // != nil
    case 2:
        return (*MyErr)(nil) // != nil
    case 3:
        var p *MyErr
        return error(p) // != nil because the interface points to a
                        // nil item but is not nil itself.
    case 4:
        var err error // == nil: zero value is nil for the interface
        return err    // This will be true because err is already interface type
    }
}

func main() {
    for i := 0; i <= 4; i++ {
        err := doSomething(i)
        fmt.Println(i, err, err == nil)
    }
}

Output:

0 <nil> true
1 <nil> false
2 <nil> false
3 <nil> false
4 <nil> true

In case 2 a nil pointer is returned but first it is converted to an interface type (error) so an interface value is created which holds a nil value and the type *MyErr, so the interface value is not nil.