Optimize Grid Row Pooling to eliminate moveNode deltas

Input from Gemini 3 Pro:

✦ Status Update: Grid Row Pooling Optimization (#9012)

We have made significant progress in implementing the Fixed-DOM-Order strategy for Grid Rows, eliminating moveNode deltas during scrolling. However, we are facing a critical regression in Cell Selection, which is currently intermittent (works on scroll, fails on click).

Current State

1. Fixed-DOM-Order: Implemented in GridBody.mjs (recycle rows in place) and Row.mjs.
2. Logical IDs: Standardized usage of logicalId (recordId__field) across Body, Row, and all SelectionModels.
3. Selection Refactor: BaseModel now attempts to update Row components directly (updateRows) instead of triggering a full Body update.

The Problem

• Cell Clicks: Clicking a cell does NOT trigger a visual update immediately.
• Scroll Updates: Scrolling (which triggers Row.createVdom) successfully applies the selection class.
• Root Cause: updateRows in BaseModel is failing to trigger the update on click. This implies either:
  • getRecord logic is failing to find the record (likely parseInt vs string mismatch or ID parsing issues).
  • DomEvent delegation is failing (less likely, but possible given the component boundary changes).
  • row.update() with manual VDOM mutation is insufficient or being suppressed.

Debugging Steps Taken

• Added logs to updateRows (currently in the codebase on this branch) to trace execution flow.
• Added robust ID resolution in Body.getRecordFromLogicalId (handling numeric/string mismatch).

Next Steps

1. Analyze Logs: Run the app with the current branch and check console for updateRows output.
2. Verify Event Chain: Confirm onCellClick -> toggleSelection -> select -> updateRows is executing.
3. Fix ID Parsing: Ensure getRecord and getRecordFromLogicalId perfectly match the ID structure generated by Row.

We are moving this to a fresh session to clear context and focus purely on debugging this interaction.

Input from Gemini 3 Pro:

✦ Architectural Refinement Proposal: Deconstruct Logical IDs

During the implementation of the Fixed-DOM-Order strategy, we identified significant fragility in using composite string IDs (recordId__field) for DOM data attributes. This requires error-prone string parsing (split('__')) to resolve records, which breaks if data fields contain underscores or if IDs are complex.

Proposal: Split dataset attributes

Instead of data-cell-id="100__firstname", we should render:

<div class="neo-grid-cell" data-record-id="100" data-field="firstname">...</div>

Benefits:

1. Performance (Zero Deltas for Columns): In the Fixed-DOM-Order strategy, a cell always belongs to the same column. Therefore, data-field="firstname" never changes during vertical scrolling. The VDOM engine will see this as a static attribute, generating zero DOM deltas for it. We only pay the cost of updating data-record-id.

2. Robustness (No Parsing): We eliminate the getRecordFromLogicalId parsing logic entirely.

   • Old: id.split('__').pop() -> fragile.
   • New: target.dataset.recordId, target.dataset.field -> direct access. This solves edge cases where IDs or fields contain separators.

3. Cleaner Selection Logic: Selection models can read recordId directly to resolve the Record from the Store, without needing to reverse-engineer strings. They can then construct the internal selection key (e.g. for selectedItems Set) using a consistent helper, but the DOM interface remains pure.

Implementation Plan (Follow-up):

1. Update Row.mjs to render split data attributes.
2. Update Body.mjs / BaseModel.mjs to read from dataset.recordId and dataset.field.
3. Deprecate the complex ID parsing logic in Body.mjs.

Input from Gemini 3 Pro:

✦ I have completed the optimization of the Grid Row Pooling logic and resolved the selection model regressions.

Summary of Changes:

1. Fixed-DOM-Order Strategy: Rows are now pooled and recycled in place within the VDOM, eliminating moveNode operations during scrolling.
2. Split Dataset: Replaced fragile composite ID strings (recordId__field) with robust data-record-id and data-field attributes. This simplifies event handling and avoids parsing errors with complex IDs.
3. Robust Selection Logic:
   • Refactored BaseModel.updateRows to correctly handle both Cell and Row selection paths.
   • Fixed a critical bug where RowModel checks used the wrong selection store (items vs selectedRows).
   • Unified selectRow and deselectRow to use view.update() for robust visual rendering of row class changes, while maintaining efficient VDOM mutation logic.
4. Cleanup: Removed duplicate methods and debug logs.

Outcome:

• Scrolling performance is improved (zero structure deltas).
• Cell selection (click + key nav) works flawlessly.
• Row selection (click + key nav) works flawlessly, with no accumulation bugs.

I have merged the changes into dev.

Follow-up Items:

• Further optimize row.update() to work reliably for root node class changes without triggering full Body updates.
• Explore converting data-record-id to use the record's internal ID if meaningful, or ensure strict type safety.