Changes made.
Should I also include a test of byPref in this CL? i.e.
func TestMXPref(t *testing.T) {
	var mx1, mx2, mx3 []*MX
	mx1 = []*MX{
		&MX{"a", 5},
		&MX{"b", 5},
		&MX{"c", 5},
		&MX{"d", 5},
		&MX{"e", 5},
	}
	// Copied the shuffle here
	mx2 = append(mx2, mx1...)
	for i := range mx2 {
		j := rand.Intn(i+1)
		mx2[i], mx2[j] = mx2[j], mx2[i]
	}
	sort.Sort(byPref(mx2))
	// mx2 now holds a shuffled copy to check that shuffles are random
	mx3 = append(mx3, mx1...)
	for runs := 0; runs < 20; runs++ {
		for i := range mx3 {
			j := rand.Intn(i+1)
			mx3[i], mx3[j] = mx3[j], mx3[i]
		}
		sort.Sort(byPref(mx3))
		for i := range mx3 {
			if mx3[i] != mx2[i] && mx3[i] != mx1[i] {
				return
			}
		}
		copy(mx3, mx1)
	}
	t.Errorf("MX records don't get shuffled")
}
Or something, I don't have much experience writing test functions,
especially not for randomness, but honestly this seems more like a
test of rand and sort than anything, and it will fail something like 1
in every 19 sextillion builds.
On 11 April 2011 23:06, <rsc@golang.org> wrote:
>
> http://codereview.appspot.com/4388048/diff/5002/src/pkg/net/dnsclient.go
> File src/pkg/net/dnsclient.go (right):
>
>
http://codereview.appspot.com/4388048/diff/5002/src/pkg/net/dnsclient.go#newc...
> src/pkg/net/dnsclient.go:414: // Implements sort.Interface
> This is working too hard. (I know it's not your algorithm.)
>
> type byPref []*MX
> // usual methods, now 1/3 as long
>
> func LookupMX(...) {
>  ...
>  // Shuffle.
>  for i := range mx {
>    j := rand.Intn(i+1)
>    mx[i], mx[j] = mx[j], mx[i]
>  }
>  // Sort by preference.
>  // Stays shuffled within equal preference.
>  sort.Sort(byPref(mx))
> }
>
>
http://codereview.appspot.com/4388048/diff/5002/src/pkg/net/dnsclient.go#newc...
> src/pkg/net/dnsclient.go:437: var records []dnsRR
> While you're here,
>
> s/records/rr/
> s/entries/mx/
>
> http://codereview.appspot.com/4388048/
>