Fix timeout report generation race

When fx is run with a very short timeout, it's possible for the
timeout error message to be in a bad format "OnStart hook added by  failed"
where the name of the OnStart hook's caller hasn't been recorded
before the context times out.

This fixes the error message generation to not rely on the OnStart hook's
caller always being known to fix this race.

Fixes uber-go#815

app.go

@@ -1023,6 +1023,11 @@ func withTimeout(ctx context.Context, param *withTimeoutParams) error {
 			caller,
 			err,
 			r)
+	} else if caller == "" {
+		return fmt.Errorf("%v hook failed: %w",
+			param.hook,
+			err)
+
 	}
 	return fmt.Errorf("%v hook added by %v failed: %w",
 		param.hook,

app_test.go

@@ -883,6 +883,28 @@ func TestAppRunTimeout(t *testing.T) {
 	}
 }
 
+func TestVeryShortTimeout(t *testing.T) {
+	type A struct{}
+	spy := new(fxlog.Spy)
+	app := New(
+		WithLogger(func() fxevent.Logger { return spy }),
+		Provide(func() *A {
+			// this will timeout
+			time.Sleep(time.Millisecond)
+			return &A{}
+		}),
+		Invoke(func(*A) {}),
+	)
+
+	ctx, cancel := context.WithTimeout(context.Background(), time.Nanosecond)
+	err := app.Start(ctx)
+	require.Error(t, err)
+	// The error message should never be in the format "added by" followed by an empty string.
+	assert.NotContains(t, err.Error(), "OnStart hook added by  failed")
+	assert.Contains(t, err.Error(), "context deadline exceeded")
+	cancel()
+}
+
 func TestAppStart(t *testing.T) {
 	t.Parallel()