Files
lmxopcua/docs/ui-consistency/evidence/ot13-raw-callers
..

OT13 composed Raw caller presentations

Frozen production scope: RawTree.razor only, with RawCallerPresentationTests and own browser script. Existing standalone modal wrappers, tree density JS/geometry and service contracts are not edited in this batch. Modal source dependencies were accepted independently; shared focus adapter071148e9 is served at SHA7f63bd6c5181d0b70edc2e87e086fab4dfba96adabd7a19737f9b3b4f18296d5.

Contracts

Each show operation advances the parent presentation, closes other parent-owned overlay flags and supplies a stable per-presentation Target. Render-local handlers capture the current business delegate, target node and device-create/edit meaning. Retired queued handlers cannot dispatch under replacementB. A dispatched operationA continues; its late result, warning, fault or close cannot overwriteB. Active direct-operation faults retain the existing error-boundary behavior (positive regression). Retired faults are ignored only for UI delivery; no cancellation/replay/authorization/payload change.

Saved callbacks refresh their captured node/parent only while current. The refresh rechecks generation after await. Weak per-node revision ownership prevents old cleanup from clearing a newer read’s Loading flag. Last loaded children remain stored until a current fresh read succeeds; a retired read cannot leave a false empty branch or install retired children. Its own loading flag is released, and Loaded remainsfalse so a later explicit expand reads fresh data. Active read failures still use the existing node error row. No backend/API/polling change.

Rename advisories and delete blockers queue while their originating input/confirm is visible. The close callback and a completed DOM batch precede the follow-up message’s render; the old dialog’s DOM is already removed. Parent callbacks are generation-bound. Hidden origins cannot publish late messages. This enforces a single-active-overlay sequence; arbitrary nested overlays are not supported or certified.

Actual evidence

  • tests.log/tests.trx:157 passed,0 failed/skipped;19 new caller cases plus138 current modal/density/row cases (including concurrent agents’ new modal cases). test-classes.json records exact classes; this is a selected suite, not the final whole consumer suite.
  • Valid reds: seven-red.log7 failures for wrong replacement target, dispatchedA warning, oldCSV saved callback, premature follow-up and same-caption identity; read-owner-red.log, hidden-origin-red.log, retained-subtree-red.log each preserve their concrete failure. retired-exceptions-red.log has2 failing retired-fault cases and2 positive cases already passing. The active fault remains a positive boundary-behavior control.
  • first-five-fixture-red.log initially had4 failures/1 pass because the CSV fake device lacked HasLazyChildren, so it could not observe an incorrect target read. The corrected realistic device fixture makes that case fail in the seven-case red. No assertion was removed to achieve green.
  • Test command from repository root uses explicit private SQL runner: python3 infra/tools/ui-consistency/with_private_sql.py --state <private-state-path> -- dotnet test tests/Server/ZB.MOM.WW.OtOpcUa.AdminUI.Tests/ZB.MOM.WW.OtOpcUa.AdminUI.Tests.csproj -c Release --no-restore --filter "FullyQualifiedName~RawCallerPresentationTests|FullyQualifiedName~RawTreeDensityTests|FullyQualifiedName~RawTreeRowsTests|FullyQualifiedName~SimpleModalAccessibilityTests|FullyQualifiedName~ModalAccessibilityTests" --logger "trx;LogFileName=raw-callers.trx" --results-directory /tmp/ot13-raw-caller-toggle-results.
  • fixture-build.log:zero warnings/errors. Synthetic fixture allows only explicitly prefixed local rename/delete fabricated replies and memory-only child reads; it never calls a database/fleet mutation. Browser request count is1 per scene.
  • matrix.json/native.log:first complete composed8/8 native scenes, followed by8/8 confirmation after the initial-expansion ownership correction (delete blocker/rename advisory ×1440/390 and real200%720/195DPR2). Actual safe Cancel focus, single active follow-up, full escaped metadata, focus inside, inert background, Escape close/restored background/focus, one synthetic dispatch, no overflow/page errors. Each scene verifies served dialog-focus.js byte parity at the final approved hash. The earlier zero-scene harness used Actions rather than the real More actions trigger label; native-trigger-harness-error.log retains that selector error, not a product defect. The Delete button’s real label was corrected in the harness before the complete matrix.

Limits

Native scenes exercise two actual composed follow-ups; deterministic unit cases cover replacement/deferred/correct-target transitions. They do not claim every modal family’s composed runtime path or generic nesting. GlobalUns save/load/delete callers and remaining inline operation/editor dialogs are a separate pending OT13 slice. Whole-route/full-suite/package and delivered route inventory remain OT14. No deployment, real mutation, real driver session or production credential access. Roll back this scoped source commit for a regression; keep published packages/data and independently accepted shared focus fixes.

Peer correction: initial lazy expansion shares read ownership

Parent review identified the original ToggleAsync still bypassed the new reload ownership. A user can expand an unloaded folder, open New folder while its first read is pending, then save: the newer reload and old initial query race. initial-expand-owner-red.log records5 failures/1 positive: four actual rendered expansion + rendered name-submit callback interleavings (oldsuccess/oldfailure ×completionorder), late disposed expansion, and the already-passing positive that opening a dialog alone must not discard an ordinary read. Toggle now shares weak per-node revisions and lifetime checks with reload; it deliberately does not use modal presentation generation. Old successes/errors/cleanup cannot replace fresh children or clear a newer Loading flag; ordinary expansion still completes when a dialog opens. The final157 results include all19 caller cases. before-toggle-confirmation preserves prior8-case native JSON/photos; matrix.json/native.log is the final8-case confirmation on corrected candidate. No further UI fixes were needed.

Fixture state is now scoped per circuit (matching the consumer service lifetime) so independent scenes cannot leak synthetic target state; this is fixture-only, not a production registration change. A harmless observer readiness handshake precedes interaction. The two composed browser outcomes and final served focus hash remain the same.