Friendly messages discard type information?

[twpayne]

Firstly, thank you for this library!

I might be holding something wrong here, but it seems to me that the conversion of netlink.Message to the friendly LinkMessage, AddressMessage, etc. here:

rtnetlink/conn.go

Lines 174 to 193 in 61f4d31

	var m Message
	switch nm.Header.Type {
	case unix.RTM_GETLINK, unix.RTM_NEWLINK, unix.RTM_DELLINK:
	m = &LinkMessage{filtered: (nm.Header.Flags&netlink.DumpFiltered != 0)}
	case unix.RTM_GETADDR, unix.RTM_NEWADDR, unix.RTM_DELADDR:
	m = &AddressMessage{}
	case unix.RTM_GETROUTE, unix.RTM_NEWROUTE, unix.RTM_DELROUTE:
	m = &RouteMessage{}
	case unix.RTM_GETNEIGH, unix.RTM_NEWNEIGH, unix.RTM_DELNEIGH:
	m = &NeighMessage{}
	case unix.RTM_GETRULE, unix.RTM_NEWRULE, unix.RTM_DELRULE:
	m = &RuleMessage{}
	default:
	continue
	}
	if err := (m).UnmarshalBinary(nm.Data); err != nil {
	return nil, err
	}
	lmsgs = append(lmsgs, m)

discards the information whether the message is an RTM_GET*, RTM_NEW*, or RTM_DEL* message. The only use of nm.Header.Type is determining whether the message is an RTM_*LINK, RTM_*ADDR, etc., and the call to UnmarshalBinary only receives nm.Data, which is not sufficient to determine whether it is a RTM_GET*, RTM_NEW*, or RTM_DEL* message.

Is this understanding correct?

If so, would you accept a PR that adds this information to LinkMessage, AddressMessage, etc.?

One possible design would be to add an Action field (using the word "action" to avoid confusion and conflict with Type):

type Action int

const (
    ActionGet Action = iota
    ActionNew
    ActionDel
)

type LinkMessage struct {
    Action Action
    // ...rest of LinkMessage struct is unchanged
}

and then populate the Action field in the unpackMessages function in conn.go.

Another possible design would be to have separate NewLinkMessage and DelLinkMessage types that embed a LinkMessage, and then allow a type switch to determine the action. A plain LinkMessage would represent a RTM_GETLINK.

A third option would be to include the netlink message type as a field in LinkMessage, but this would leak lower-level details.

Thoughts?