fix(opcua): use config node_id as OPC-UA string NodeId

[thiagoralves]

Summary

The OPC-UA server was auto-assigning numeric NodeIds (ns=2;i=N) and ignoring the node_id supplied by the Editor. That node_id traveled all the way to the runtime (serialized to conf/opcua.json, parsed into the config model) but was only used as an internal mapping/permissions key — never as the exposed OPC-UA NodeId. On top of that, auto-numeric IDs are unstable across compilations because they follow insertion order, so clients can't reliably bind to them.

This wires the Editor-provided node_id through to the actual NodeId, so clients now see stable, meaningful string identifiers like ns=2;s=PLC.Main.MotorSpeed.

Reported on the forum: Node ID in OPC UA server.

What changed

Single file: core/src/drivers/plugins/python/opcua/address_space.py.

• Every node is created with an explicit ua.NodeId(node_id, namespace_idx) — simple variables, struct objects, struct fields (leaf and nested-object, reusing the existing parent.field path composition), and arrays.
• The browse name is now passed as an explicit ua.QualifiedName(browse_name, namespace_idx). This is required: asyncua's arg parser defaults a bare-string browse name to namespace 0 whenever an explicit NodeId is provided, which would have silently changed browse-name behavior.
• Runtime-side collision guard (belt-and-suspenders — the Editor already enforces uniqueness): a duplicate node_id emits a log_warn and gets a numeric suffix (_1, _2, …) to keep NodeIds unique within the namespace.
• Empty node_id falls back to a server-assigned numeric NodeId, so a malformed config can't crash address-space build.

What did NOT change

• No config schema change — format_version is untouched. The node_id field already existed and was already sent by the Editor; it's simply honored now.
• No Editor change — the field, its default (PLC.<pou>.<path>), and its uniqueness validation are already in place.
• Internal mapping dicts (node_permissions, nodeid_to_variable) remain keyed by the config node_id, so the permission-callback resolution path is unchanged.

Compatibility note

Clients on existing deployments will see NodeIds change from ns=2;i=N to ns=2;s=<node_id> on upgrade. This is acceptable and is a net improvement: the old numeric IDs already shifted between compilations (insertion order), so nothing could safely cache them. String IDs are stable and OPC-UA-idiomatic.

Testing

• py_compile passes.
• Disassembled the installed asyncua's _parse_nodeid_qname to confirm exact arg-type handling.
• Ran a real asyncua server integration test confirming: string NodeId produced correctly; namespace preserved on both NodeId and browse name; duplicate NodeId raises BadNodeIdExists (proving the dedup guard is genuinely needed); _1 dedup works; empty-config falls back to numeric.

🤖 Generated with Claude Code