How can a slice contain itself?

Solution 1:

Slice containing itself

Besides a recursive type (such as type Foo []Foo, see ANisus's answer) which is good for nothing besides demonstration, a slice may contain itself if for example the element type of the slice is interface{}:

s := []interface{}{"one", nil}
s[1] = s

In this example the slice s will have 2 interface values, the first "wrapping" a simple string "one", and another interface value wrapping the slice value itself. When an interface value is created, a copy of the value will be wrapped which in case of slices means a copy of the slice header/descriptor, which contains the pointer to the underlying array, so the copy will have the same pointer value pointing to the same underlying array. (For more details about the representation of interfaces, see The Laws of Reflection: The representation of an interface.)

If you were quickly on to print it:

fmt.Println(s)

You would get a fatal error, something like:

runtime: goroutine stack exceeds 250000000-byte limit
fatal error: stack overflow

Because fmt.Println() tries to print the content recursively, and since the 2nd element is a slice pointing to the same array of the the slice being printed, it runs into an infinite loop.

Another way to see if it really is the slice itself:

s := []interface{}{"one", nil}
s[1] = s
fmt.Println(s[0])

s2 := s[1].([]interface{})
fmt.Println(s2[0])

s3 := s2[1].([]interface{})
fmt.Println(s3[0])

Output (try it on the Go Playground):

one
one
one

No matter how deep we go, the 2nd element will always be the slice value pointing to the same array as s, wrapped in an interface{} value.

The indirection plays the important role as a copy will be wrapped in the interface{} but the copy will contain the same pointer.

Array can't contain itself

Changing the type to be an array:

s := [2]interface{}{"one", nil}
s[1] = s
fmt.Println(s[0])

s2 := s[1].([2]interface{})
fmt.Println(s2[0])

s3 := s2[1].([2]interface{})
fmt.Println(s3[0])

Output (try it on the Go Playground):

one
one
panic: interface conversion: interface is nil, not [2]interface {}

This is because when the array is wrapped into an interface{}, a copy will be wrapped - and a copy is not the original array. So s will have a second value, an interface{} wrapping an array, but that is a different array whose 2nd value is not set and therefore will be nil (the zero value of type interface{}), so attempting to "go into" this array will panic because it is nil (type assertion fails because not the special "comma, ok" form was used).

Since this s array does not contain itself, a simple fmt.Println() will reveal its full content:

fmt.Println(s)

Output:

[one [one <nil>]]

Further interface{} wrapping analysis

If you wrap an array in an interface{} and modify the content of the original array, the value wrapped in the interface{} is not affected:

arr := [2]int{1, 2}
var f interface{} = arr
arr[0] = 11

fmt.Println("Original array:    ", arr)
fmt.Println("Array in interface:", f)

Output:

Original array:     [11 2]
Array in interface: [1 2]

If you do the same with a slice, the wrapped slice (since points to the same underlying array) is also affected:

s := []int{1, 2}
f = s
s[0] = 11

fmt.Println("Original slice:    ", s)
fmt.Println("Slice in interface:", f)

Output:

Original slice:     [11 2]
Slice in interface: [11 2]

Try these on the Go Playground.

Solution 2:

The below example creates a slice that contains itself:

type Foo []Foo  
bar := make(Foo, 1)
bar[0] = bar

This can be done because the slice value internally contains a pointer to an array, a length, and a capacity.

An array on the other hand is a value. It can, at best, contain pointers to itself.