a possible coroutine leak

[bdjimmy]

The start function in the watch.go file uses <-started at the end, so the operation will block the coroutine exit, and the reload function may call the start function multiple times, causing the coroutine to leak.

[guhan121]

@bdjimmy i test it . when file change , after reload runtime.NumGoroutine() add 1 . I think this is in line with expectations。

add new coroutine for run new app . and all last coroutine (start by func Start(appname string) ) will blocking at

started <- true

you can add a log before started <- true like

	beeLogger.Log.Successf("NumGoroutine:%d...", runtime.NumGoroutine())

After all, debug is not a normal release run

[guhan121]

// Kill kills the running command process
func Kill() {
	defer func() {
		if e := recover(); e != nil {
			beeLogger.Log.Infof("Kill recover: %s", e)
		}
	}()
	if cmd != nil && cmd.Process != nil {
		// Windows does not support Interrupt
		if runtime.GOOS == "windows" {
			cmd.Process.Signal(os.Kill)
		} else {
			cmd.Process.Signal(os.Interrupt)
		}

		ch := make(chan struct{}, 1)
		go func() {
			cmd.Wait()
			ch <- struct{}{}
		}()

		select {
		case <-ch:
			<-started
			return
		case <-time.After(10 * time.Second):
			beeLogger.Log.Info("Timeout. Force kill cmd process")
			err := cmd.Process.Kill()
			if err != nil {
				beeLogger.Log.Errorf("Error while killing cmd process: %s", err)
			}
			<-started
			return
		}
	}
}

i read started cmd.kill ,and NumGoroutine always 10 . but i don't test cmd==nil. I'm going to do more testing