7.8 KiB
OT13 Raw manual entry and browse workflow checkpoint
Production scope is exactly RawManualTagEntryModal.razor and RawBrowseModal.razor. CSV import is independently acceptedc9f191d5. The accepted shared focus adapter071148e9 is a separate common fix. This checkpoint adds only optional Target to existing public parameters; no service/model/schema/policy/package interface changes.
Both wrappers use the accepted named ModalFrame, original large/extra-large widths, canonical Bootstrap scrollable body, safe initial Cancel and scoped field names/errors. Manual rows combine their visible column header with a unique row name; typed configuration has a named group linked to the same row error. The fallback JSON editor is explicitly named. Browse mirror and Sparkplug scope fields have unique native label targets; invalid scope text is linked. The single-active-dialog contract, busy protections and existing backend authorization ownership remain.
Manual entry preserves the original data types, defaults, driver-specific validation/dictated fields and row rules. All RawTagInputs and target IDs are detached before the first write. Successful rows drop off; failures retain their original rows, errors and unknown JSON for correction/retry. Later rows cannot be altered by queued UI edits during the first write. Target replacement retires callbacks/reads and stops further undispatched writes; an already dispatched write is not cancelled or replayed. Current child probe getters stay live through their initial resolve; retired getters keep their last-known original presentation snapshot, never another device's config. Keyed typed children retire shared pickers through the separately accepted disposal guard.
Browse retains merged-device resolution, existing group/device precedence, owner-qualified probe, session orphan cleanup, lazy attributes, selection/mirroring, mapper/collision refusals, MQTT capability/scope rules and group cap. Captured owner lookups cannot begin a probe after retirement. Late attribute results check generation and token; retained tree callbacks cannot select under a replacement. Import rows are mapped before dispatch. Rebirth remains the existing separately authorized broker action, with a captured session/target and single-flight protection. Retired results/errors cannot appear under another presentation. Session retirement closes the captured token without closing a replacement; cleanup exceptions are logged. Current mutation exceptions retain existing boundary behavior; only retired exceptions are suppressed.
A real native focus defect was corrected: replacing the Connect button after success left focus on body, so Escape missed the old dialog-local listener. The same button now remains mounted as it changes from Connect to Refresh. During a genuinely deferred open, aria-disabled plus the existing opening/token guards retain native focus and block a queued second dispatch. The original Refresh explanation remains its title. No new focus interop or reconnection behavior was introduced. Separately, the common adapter now reanchors lost busy focus and catches outside-target key gaps while preserving child-widget handling.
Verification
- Original manual migration red:2 failures (unnamed modal/fields and duplicate dispatch) in actual-red.log. Browse/typed-group red:4 failures/6 positive cases in browse-and-group-red.log. Earlier compile intersections with another agent's partial files and the invalid bUnit RenderAsync helper are retained as setup failures, not UI defects.
- Twenty new actual bUnit workflow cases cover names/relationships, validation, frozen full batch/unknown JSON/partial retry, queued edits, pending/replaced/disposed callbacks, current save close, live/frozen getters through the retained legacy test, post-save callback replacement, owner lookup, deferred open reentry/orphan cleanup, late attributes/retained child callbacks, import and broker rebirth identity/single-flight.
- Peer review found late A mutation exceptions could escape into B. Four added replacement/disposal cases retain2 genuine replacement failures plus2 already-passing framework disposal controls in retired-exceptions-red.log. Filtered catches now suppress only retired exceptions. Current failure semantics remain.
- accepted-contracts.log:147passed/0skips (20new+127existing manual merged-config, browse stale-open/rebirth and mapper cases). acceptance.log broadens the same checkpoint to163passed/0skips with existing orphan cleanup/refusal/MQTT source contracts. Existing test framework remains bUnit2.7.2. All discovery/execution uses the validated owned-loopback private SQL runner, with private state/credentials outside Git; no selected SQL collection or operational write runs here.
- Three legacy test files have narrow fixture/contract refinements: inert renderers supply only the new frame's otDialog JS dependency; initialized browse fixtures supply their actual open/authorized state; a closed rebirth now asserts no retired UI outcome while still proving the publish occurred. The old accessor test that demanded retired A read B is strengthened to prove current asynchronous liveness, frozen A and a fresh B getter. No domain assertions are removed. Initial missing-JS/inert-renderer child DI failures are retained.
- Twelve actual native contexts: manual3, typed configuration error/recovery3, browse/rebirth6 across1440/390 and real200%720CSSpx/DPR2. Checks include initial safe focus, native Tab/Escape/return, valid composite names, wholly visible error rectangles inside the scrollable body, fixed header/footer geometry, captured synthetic batch/import/broker action, exactly-one dispatch, preserved drafts, deferred700ms Connect focus on the same actual DOM node through Refresh, repeated Enter blocked, busy Escape isolation, no body overflow/page errors. Final logs end in PASS3/PASS3/PASS6; final evidence uses the exact shared7f63bd6c… script. These are local fabricated operations, not device/broker/database acceptance.
- manual-before-scrollable/real200.png retains observed footer clipping after expanded JSON. Adding the existing modal-dialog-scrollable modifier fixes it without shared CSS. Earlier valid name/error checks remain in the before folders; final chrome/error body geometry is explicit.
- browse-native-focus-red.log and escape-debug.json retain the initial removed-control/Escape failure and actual body focus/adapter calls. A later fixture-only Rebirth counter initially stayed0 because its parent had no rerender event; the underlying fake result was1. WorkflowState now notifies the fixture page so observed counts update, with the original counter-harness log retained. No product counter was changed.
source-boundaries.json records exact production hashes, optional-parameter compatibility, unchanged manual type/default and browse scope/config-reader bodies/constants, and served dialog/Monaco source parity. This supplements real rendered tests rather than substituting for them. The unchanged mapper/validators and direct service signatures remain independently exercised. Build logs retain initial fixture central-package-version setup error and corrected zero-warning builds. Pure fixtures use only synthetic scoped services; unexpected operational calls throw. The browser helper uses an isolated Chromium profile and real browser zoom, never CSS zoom or a user's browser.
Integration limits
Parent caller Target wiring, follow-up-message sequencing and callbacks' own inner awaits are separate primary-agent gates. Other wrappers, Hosts/driver operation confirmations, Script editor and full OT06/14 route/SQL/combined suite acceptance remain open. This does not close OT13, certify assistive speech output, or deploy an application. No package publication, real import, driver probe or broker publish occurs here. Owned listener18141 is stopped after acceptance; shared private SQL and other agents' fixtures are not owned by this checkpoint.