Golang landmines

There are three easy to make mistakes in go. I present them here in the way they are often found in the wild, not in the way that is easiest to understand.

All three of these mistakes have been made in Kubernetes code, getting past code review at least once each that I know of.

Loop variables are scoped outside the loop.

What do these lines do? Make predictions and then scroll down.

func print(pi *int) { fmt.Println(*pi) }

for i := 0; i < 10; i++ {
  defer fmt.Println(i)
  defer func(){ fmt.Println(i) }()
  defer func(i int){ fmt.Println(i) }(i)
  defer print(&i)
  go fmt.Println(i)
  go func(){ fmt.Println(i) }()
}

Answers:

func print(pi *int) { fmt.Println(*pi) }

for i := 0; i < 10; i++ {
  defer fmt.Println(i) // OK; prints 9 ... 0
  defer func(){ fmt.Println(i) }() // WRONG; prints "10" 10 times
  defer func(i int){ fmt.Println(i) }(i) // OK
  defer print(&i) // WRONG; prints "10" 10 times
  go fmt.Println(i) // OK; prints 0 ... 9 in unpredictable order
  go func(){ fmt.Println(i) }() // WRONG; totally unpredictable.
}

for key, value := range myMap {
  // Same for key & value as i!
}

Everyone expects these values to be scoped inside the loop, but go reuses the same memory location for every iteration. This means that you must never let the address of key, value, or i above escape the loop. An anonymous func() { /* do something with i */ } (a "closure") is a subtle way of taking a variable's address

Nil interface is not the same as having a nil pointer in the interface.

type Cat interface {
  Meow()
}

type Tabby struct {}
func (*Tabby) Meow() { fmt.Println("meow") }

func GetACat() Cat {
  var myTabby *Tabby = nil
  // Oops, we forgot to set myTabby to a real value
  return myTabby
}

func TestGetACat(t *testing.T) {
  if GetACat() == nil {
    t.Errorf("Forgot to return a real cat!")
  }
}

Guess what? The above test will NOT detect the nil pointer! This is because the interface serves as a pointer, so GetACat effectively returns a pointer to a nil pointer. Never do the thing that GetACat does, and you (and your coworkers) will be much happier. It is very easy to do this with error values. http://golang.org/doc/faq#nil_error

Shadowing considered harmful to your sanity.

var ErrDidNotWork = errors.New("did not work")

func DoTheThing(reallyDoIt bool) (err error) {
  if reallyDoIt {
    result, err := tryTheThing()
    if err != nil || result != "it worked" {
      err = ErrDidNotWork
    }
  }
  return err
}

The above function will always return a nil error, because the inner err variable "shadows" the function scope variable. Adding a var result string and not using := would fix this.

se77en/The Three Go Landmines.markdown