Joseph Doherty
1a647cf1c4
Records 7 issues found during the 2026-06-24 CvdReactor live-ops session: - template editor omits inherited-but-unmaterialized base attrs (MoveInType etc.) - derived templates carry incomplete/stale IsInherited row sets - collision detector blocks adding attrs/compositions to derived templates - CLI instance set-bindings can't set DataSourceReferenceOverride - CLI template update full-replaces description (nulls it if omitted) - CLI template list/get table dumps every attribute - Central UI script editor false-flags WriteBatchAndWaitAsync/WaitAsync/WaitForAsync (sandbox compile surface out of sync with runtime + deploy gate)
8.8 KiB
Follow-up tracker — template-inheritance UI gaps + CLI/validation footguns (2026-06-24 session)
Status: OPEN (tracker) · Found: 2026-06-24 · Context: live ops session on wonder-app-vd03 (CvdReactor / Z28061 / Z28061Sim) — renaming the template, adding the LeakTest module, and adding MoveInType to the MESReceiver children.
Components: Central UI (#9), Template Engine (#1), CLI (#19), Configuration Database (#17)
Issues are listed worst-first. Severities are author estimates. None caused data loss; the runtime/flattened config and deployed instances are correct.
1. Template editor omits inherited-but-unmaterialized base attributes (user-reported)
Severity: Medium · Components: Central UI (#9), Template Engine (#1)
Symptom: On /design/templates → LeftMESReceiver, the Attributes tab does not list MoveInType. Same for RightMESReceiver. (Also missing from the list: all MoveOut* and ScanStateCmd.)
Root cause: MoveInType (String), the 12 MoveOut*, and ScanStateCmd live on the base MESReceiver template (id 3, 26 attrs). The derived children LeftMESReceiver/RightMESReceiver (ids 5/6) only have 12 materialized IsInherited rows (the MoveIn* basics) — the base attributes added after these children were derived were never materialized as child rows. The editor's main Attributes tab lists only the template's stored rows (TemplateEdit.razor:187 → GetAttributesByTemplateIdAsync(Id); count badge at :415), so the unmaterialized inherited members are invisible there.
Not a runtime bug: the flattener resolves the full chain, so LeftMESReceiver.MoveInType is in the flattened deploy config and is bound + live on the instances ('' / 'na', Good). The page also has an "Effective inherited set" read-only preview (M9-T26b, TemplateEdit.razor:465-491) + a staleness banner (:292-301) that do surface these — but the main, editable Attributes list does not.
Suggested fix: in the Attributes tab, render the resolved inherited set (with inherited badges) rather than only materialized rows — or auto-materialize missing inherited rows on load. See #2 (shared root cause).
2. Derived templates carry incomplete/stale IsInherited row sets
Severity: Medium · Components: Template Engine (#1), Configuration Database (#17)
Symptom: LeftMESReceiver/RightMESReceiver (parent=3) have 12 stored attribute rows vs the base's 26. By contrast LeftReactorSide/RightReactorSide (parent=7) mirror the full 61. So derived row-sets are inconsistent.
Root cause: when a base attribute is added after a child is derived, the child's stored IsInherited placeholder rows are not auto-resynced. There is a staleness banner but no surfaced one-click "resync inherited members" action (or it exists and was never run for these children). This is the data root cause of #1.
Suggested fix: a "resync inherited members" command (CLI + UI) that materializes missing base members onto derived children; consider running it automatically when a base attribute is added.
3. Collision detector blocks adding attributes/compositions to ANY derived template
Severity: Medium-High · Components: Template Engine (#1)
Symptom: template attribute add --template-id 5 --name MoveInType ... fails with 13 "Naming collision" errors — the new attribute plus all 12 pre-existing inherited rows. Same class of failure when adding a composition to a derived template (hit earlier when trying to add LeakTest to LeftReactorSide).
Root cause: CollisionDetector.CollectDirectMembers (CollisionDetector.cs) emits every template.Attributes row — including IsInherited placeholders — with origin = child name, while CollectInheritedMembers emits the same names with origin = parent '…'. Two distinct origins for the same canonical name ⇒ reported as a collision. AddAttributeAsync (TemplateService.cs:294) and AddCompositionAsync (:869) both run DetectCollisions, so there is effectively no supported API path to extend a derived template — and the error message is misleading (it blames pre-existing inherited attrs, not the member you're adding).
Workaround used this session: add the feature module to the base template instead (LeakTest → base ReactorSide (7)); the flattener propagates it to all derivations.
Suggested fix: CollectDirectMembers should skip IsInherited rows (or the grouping should treat an inherited row and its parent source as the same origin), so only genuine cross-origin duplicates are flagged.
4. CLI instance set-bindings cannot set DataSourceReferenceOverride
Severity: Medium · Components: CLI (#19)
Symptom: instance set-bindings --bindings only accepts [attributeName, dataConnectionId] pairs (InstanceCommands.cs → ConnectionBinding(name, connId) 2-arg). The override is sent as null, and because SetConnectionBindingsAsync upserts DataSourceReferenceOverride = b.DataSourceReferenceOverride (InstanceService.cs:340), using the CLI on an attribute that already has an override would wipe it.
Workaround used this session: raw POST /management with {"command":"SetConnectionBindings","payload":{...,"dataSourceReferenceOverride":"…"}} — the wire contract ConnectionBinding(AttributeName, DataConnectionId, DataSourceReferenceOverride?) does carry the field; only the CLI omits it.
Suggested fix: add an optional 3rd element / --ref-override to the CLI bindings input.
5. CLI template update is full-replace, not partial
Severity: Low · Components: CLI (#19), Template Engine (#1)
Symptom: omitting --description on template update overwrites the stored description to NULL (TemplateService.cs:124-125 assigns Name+Description unconditionally). Renaming a template silently drops its description unless you re-pass it.
Suggested fix: treat omitted optional fields as "leave unchanged" (nullable-not-provided vs explicit-null), or warn when a non-empty description would be cleared.
6. (Minor) CLI template list/get table output dumps every attribute
Severity: Low · Components: CLI (#19)
Symptom: --format table template list emitted ~171 KB (the full attribute set per template inline), unusable in a terminal. --format json is fine.
Suggested fix: a compact table projection (id/name/desc/#attrs/#comps) for list/get; reserve full attribute dumps for an explicit --verbose/--detail flag.
7. Central UI script editor false-flags batch/wait helpers (sandbox compile surface out of sync)
Severity: Medium · Components: Central UI (#9), Script Analysis (#25)
Symptom: In the template script editor (/design/templates/{id} → Scripts → Edit → Code), a script that calls Attributes.WriteBatchAndWaitAsync(...) (or on a child, Children["X"].Attributes.WriteBatchAndWaitAsync(...)) shows a red compile error:
'SandboxAttributeAccessor' does not contain a definition for 'WriteBatchAndWaitAsync' ... (CS1061). Confirmed on CvdReactor.MesMoveIn; the same false error hits the base MESReceiver.MoveIn/MoveOut, which also use the helper.
Root cause: the editor validates against the Central UI's own sandbox surface (CentralUI/ScriptAnalysis/SandboxScriptHost.cs). Its SandboxAttributeAccessor only defines this[string], GetAsync, SetAsync, Resolve — it is missing WriteBatchAndWaitAsync, WaitAsync, and WaitForAsync (none are defined anywhere in the CentralUI sandbox surface). The real runtime AttributeAccessor (SiteRuntime/Scripts/ScopeAccessors.cs) and the deploy-gate ScriptCompileSurface (ScriptAnalysis) both define them — so template validate reports the script clean and it deploys/runs fine. The error is purely the in-editor validator.
Impact: misleads authors and can block saving from the UI for any script using the batch-write/wait helpers, even though the script is valid. Authoring such scripts currently has to go through the management API (as was done for MesMoveIn).
Suggested fix: bring SandboxAttributeAccessor (and the sandbox composition/children accessors) to parity with the runtime AttributeAccessor — add WriteBatchAndWaitAsync / WaitAsync / WaitForAsync with matching signatures. Ideally enforce surface parity with a test, like the RoslynScriptCompiler "representative real script" corpus already does for ScriptCompileSurface. Three surfaces (runtime, deploy-gate ScriptCompileSurface, UI SandboxScriptHost) currently drift independently — consider collapsing to one source of truth.
Minor note (not tracked separately)
The deploy-time unbound-binding validation returns one giant semicolon-joined error string (one clause per attribute). For 50–194 unbound attrs it's a wall of text; a structured/summarized error (count + grouped-by-module) would be friendlier.