The pre-refactor design minted OPC UA NodeIds directly from the driver's
FullReference (the native-address string). That had three long-term
problems:
1. OPC UA Part 3 §5.2.2 requires NodeIds to be immutable across a node's
lifetime. A rename of the underlying device address — Galaxy attribute,
S7 tag, Modbus register alias — changed the NodeId and broke every
client that had pinned the previous identifier.
2. Two drivers with coincidentally-matching native addresses (e.g. `temp`
in Modbus and `temp` in S7 under different Equipment rows) collided on
the NodeId identifier.
3. TagConfig was being placed verbatim on the wire; for drivers whose
TagConfig is JSON (every driver shipped today, per the
CK_Tag_TagConfig_IsJson check constraint), clients saw the raw JSON
blob as the NodeId string.
Refactor:
* DriverNodeManager.Variable now mints a stable path-based NodeId
`{driverId}/{folder-path}/{browseName}` and records the driver-side
FullReference in a new _fullRefByNodeId map. OnReadValue / OnWriteValue
/ ResolveFullRef look the FullReference up via that map instead of
casting NodeId.Identifier. The old cast path is preserved as a
fallback so any test fixture that still registers variables with
FullRef-shaped NodeIds keeps working.
* EquipmentNodeWalker.AddTagVariable now extracts the cross-driver
`FullName` field from Tag.TagConfig before handing the address to
DriverAttributeInfo. Every shipped driver stores the wire reference in
TagConfig[FullName]; falling back to the raw string covers any future
driver that wants an opaque non-JSON address. ExtractFullName is
exposed internal for unit coverage.
* scripts/e2e/test-galaxy.ps1 defaults updated to the new path-based
NodeIds. Verified live against p7-smoke-galaxy on the dev box:
`ns=2;s=p7-smoke-galaxy/lab-floor/galaxy-line/reactor-1/Source` reads
return Status=0x00000000 with a real Galaxy byte-array value.
Test suite: 195/195 Core.Tests + 283/283 Server.Tests green. Five new
ExtractFullName / FullName-passthrough tests added.
Task #112 GA-3 — golden-path read verified end-to-end; remaining E2E
script stages still blocked on pre-existing issues (ScriptedAlarm
predicate NRE on empty upstream cache, PowerShell $changeLines.Count
guard), tracked separately.
Task #134 — complete.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8be82e02c2