I think that for a Timer returned by AfterFunc timer.Stop will return true if the function has not been called (in which case the function will not be called) and returns false if the function has already been called. I think this is just a documentation issue.

I tested it and that is not always true. There seems to be a race.

Interesting. I don't doubt you, but, looking at the code, I don't see how that is possible. Can you provide a test case?

On my workstation the following reliably panics:

package main

import (
    "time"
)

func main() {
    ch := make(chan struct{})
    var timer *time.Timer
    for i := 0; i < 1000000; i++ {
        if timer != nil && !timer.Stop() {
            ch = make(chan struct{})
        }
        timer = time.AfterFunc(0, func() {
            close(ch)
        })
    }
}

If you reduce the number of iterations to a small number, say 100, it rarely panics.

I think there is a data race in that code. I think timer.Stop reliably returns false if the function has been called, but that does not mean that the function has completed.

There is definitely a data race:

==================
WARNING: DATA RACE
Read at 0x00c42002a008 by goroutine 6:
  main.main.func1()
      /home/bradfitz/race.go:15 +0x49

Previous write at 0x00c42002a008 by main goroutine:
  main.main()   
      /home/bradfitz/race.go:12 +0x1d4

Goroutine 6 (running) created at:
  time.goFunc()
      /home/bradfitz/go/src/time/sleep.go:164 +0x77
==================
panic: close of closed channel

goroutine 37 [running]:
panic(0x495160, 0xc42014c000)
        /home/bradfitz/go/src/runtime/panic.go:531 +0x1ad
main.main.func1()
        /home/bradfitz/race.go:15 +0x5b
created by time.goFunc
        /home/bradfitz/go/src/time/sleep.go:164 +0x78
exit status 2

Yes, there is a data race. I am suggesting that we update the documentation because I don't think it is clear. I will propose some changes.

CL https://golang.org/cl/33131 mentions this issue.