Replica.port is null on a Read response

Symptom

Code that iterates the replicas of a multicast group or clone session hits a null port:

for (Replica r : group.replicas()) {
    byte[] portBytes = r.port().toByteArray();   // NullPointerException
    // ...
}

Other replicas in the same group / session may have non-null ports. The null doesn't fire on BackupReplica.port, only on Replica.port.

Reason

The proto Replica message has a port_kind oneof with one defined variant (bytes port = 4) plus a deprecated alternative (int32 egress_port = 1). Three spec-conformant device states can produce a null on the typed jp4 surface:

port_kind oneof	deprecated egress_port	jp4 surface	What it means
set	unset	non-null Bytes	the standard 1.4+ case
unset	unset	null	no port specified at all (likely a misconfigured group)
unset	set	null	port encoded the deprecated way (pre-1.4 device)

jp4 v1.5 collapses the second and third rows into the same null because the typed port is not what describes that replica in either case.

BackupReplica.port is non-null by proto contract — the BackupReplica proto message (added in P4Runtime 1.5) has no oneof, just a plain bytes port = 1 always present.

Fix

1. Guard against null at the iteration site:

for (Replica r : group.replicas()) {
    if (r.port() == null) {
        log.warn("replica with no port (group={}) — possibly a v1.0-era device " +
                 "using deprecated egress_port", group.multicastGroupId());
        continue;
    }
    int egressPort = new BigInteger(1, r.port().toByteArray()).intValueExact();
    // ...
}

2. If you must read the deprecated path, parse the wire Replica proto directly through the generated class (p4.v1.P4RuntimeOuterClass.Replica) — jp4 v1.5 does not surface the egress_port int32 path through the typed record.

Background

Two-state Java surfaces survive proto evolution better than three-state. A null-or-Bytes shape means a future P4Runtime spec adding a third typed port_kind variant can lift it into a new typed accessor without breaking the port-is-null convention for existing callers; today's null still means "the typed port is not set", and the new accessor surfaces the new variant separately.

A three-state enum exposing {TYPED_PORT, UNSET, DEPRECATED_INT32} would force every caller to update their switch when the spec evolves. The flat-nullable shape pushes the rare deprecated-path case down to the wire-proto fallback, which is what the spec is doing internally with the deprecation marker.

The same idiom applies to WeightedMember.watchPort (action-profile-group member liveness probe), which has the same watch_kind oneof shape.

See also

• port_kind idiom — full reasoning, the analogous WeightedMember.watchPort case, why BackupReplica.port differs.
• Inspecting multicast group state — recipe that handles this null in the walk loop.
• v1.4 → v1.5 migration guide — the surface introduction.