Comment 1:

Labels changed: added repo-main, release-go1.5.

I'm happy to take care of this if we can decide on the correct behavior.

My 2c (not the final word): It seems reasonable in general for testing/quick to probabilistically generate nil pointers, even for non-recursive types. Issue #8461 is somewhat related. Such a change might cause tests that currently pass to start failing.

If you work on this, please be sure to also test mutually recursive types:

type (
    R struct{ r *R }

    S struct{ t *T }
    T struct{ s S }
)

// Issue 8818
// Generating values for recursive types used to hang.
func TestRecursiveType(t *testing.T) {
    if _, ok := Value(reflect.TypeOf(&R{}), rand.New(rand.NewSource(0))); !ok {
        t.Error("failed to generate recursive value")
    }

    if _, ok := Value(reflect.TypeOf(&S{}), rand.New(rand.NewSource(0))); !ok {
        t.Error("failed to generate mutually recursive value")
    }
}

@bradfitz who is the right person to decide what the correct behavior is here?

@robpike do you have an opinion here, or a 301?

@agl is the original author. Ask him.

The documentation for quick.Value says: "Value returns an arbitrary value of the given type". Not generating nil pointers seems more like a bug than what was intended?

CL https://golang.org/cl/10821 mentions this issue.