Use native Table row and insertion callbacks instead of an outer drop
receiver. Resolve server rows and insertion lines to their containing
group, and root insertion points to the catalog root.
Verify insertion and server-row targets with regression tests.
Keep a stable container as the HSplitView child so node identity changes only recreate the detail session rather than resetting the user's divider position.
Validated with eight native split-view transitions including empty selection: width stayed at 519 points. Debug build passed; existing file-length lint warning remains.
Render saved node details through a shared property view with injected link actions. Keep metadata, activation feedback, and plugin parameter loading outside the renderer; retain specific plugin failures and scope command feedback to its target.
Preserve ordered and repeated plugin parameters, distinguish flags and empty values, and retain opaque raw text. Add bilingual detail copy and conversion coverage.
Validation: 981 hostless tests and 21 offline rule checks passed; Debug build passed. Native renderer checked at 720 and 360 point widths using synthetic data. Existing 22 lint warnings remain.
CatalogEntry gains createdAt/updatedAt and the catalog document moves to
v5 (reads v1-v5, writes v5). Modification time is structure-sensitive:
editing a node's own fields bumps the node, a group's direct child set
or order changes bump the group, and a moved node keeps its timestamps.
No-op writes and unchanged subscription refreshes bump nothing; the
refresh rebuild compares per node, so surviving leaves keep theirs. All
times are passed in by callers as now: Date - one moment per logical
commit, the domain layer stays clock-free. Legacy documents load with
nil (unknown, never backfilled); duplicates and imports stamp fresh
times at the moment of the operation.
Semantics are recorded as glossary term 节点时间戳 and ADR 0031. No UI
in this change.
Qiu Yuzhou