Fix unmarshal of empty type leaf value for gNMI encoding

[cho-m]

Possibly fixes #291

Need to confirm if any special handling of bool_val:false for empty type since YANG empty type really only has 2 possible states ([1]doesn't exist, or [2]true) while YANG boolean type has 3 possible states ([1]doesn't exist, [2]true, or [3]false)

[coveralls]

Coverage remained the same at 90.976% when pulling be89046 on cho-m:fix-unmarshal-gnmi-empty-leaf into ef8f2a1 on openconfig:master.

[robshakir]

Thanks for this change -- I think the general shape of it is right, but I think the assumption that we can handle any value of an empty and just hand it on to the struct seems like it might have some challenges. Please see the question below.

[robshakir]

I suspect that we really need to make yang.Yempty always set the field to true if we receive this. Since YANGEmpty isn't a pointer in the generated code, then if we receive an empty field that has a bool value of false then this won't be something that the client application can tell was received. Could you add this handling as a special case please?

[wenovus]

Actually there already appears to be handling for this case:

ygot/util/reflect.go

Lines 256 to 262 in f01913a

	// YANG empty fields are represented as a derived bool value defined in the
	// generated code. Here we cast the value to the type in the generated code.
	if ft.Type.Kind() == reflect.Bool && t.Kind() == reflect.Bool {
	nv := reflect.New(ft.Type).Elem()
	nv.SetBool(v.Bool())
	v = nv
	}

When unmarshalling an empty leaf from a TypedValue in the current code, unmarshalScalar returns the non-pointer bool value, which unmarshalLeaf then takes and then feeds into InsertIntoStruct (directly or indirectly) to reach this special handling code.

This change doesn't support unmarshalling empty leafs (either JSON or gNMI) into a union type. However given its complexity I think it deserves to be addressed separately.

[wenovus]

I explored the question of unions containing empty in #582.