TreeGrid Unit Tests (Data & Logic)

Input from Gemini 3.1 Pro:

✦ I have committed the initial test file test/playwright/unit/data/TreeStore.spec.mjs which perfectly reproduces the structural mutation bugs we are seeing in the TreeGrid.

Here are some hints and specific bugs that need to be explored in the next session based on our latest findings:

1. Initial Ingestion Bug: During Store initialization, the projection algorithm is dropping nodes. With the MainStore test data, only 6 nodes are rendered initially instead of 7 (child-1-2-2 / Row.mjs is completely missing from store.items).
2. collectVisibleDescendants Logic: When collapsing a deeply nested folder (like component), the visibleDescendants array incorrectly gathers root-level siblings (like package.json and README.md) and slices them out of the grid. This suggests the structural maps or the tree traversal loop has a scoping bug.
3. Row Pool Sizing Bug (grid.Body): In the demo, the grid initializes 53 row components for only 7 nodes. This completely breaks the itemIndex = i % poolSize recycling math, causing the "7-click delay" for UI updates because updateContent hits an early-exit condition on the exact same row/record reference.
4. VDOM invalidation: Once the TreeStore.splice and row pooling math are fixed, we still need to ensure that grid.Row#updateContent properly invalidates and re-renders if a Record mutates internally, even if force: false is passed. Adding a lastRecordVersion check to updateContent is likely the fix here.

Input from Gemini 3.1 Pro:

✦ I have successfully implemented the fixes and tests for the TreeStore data and logic mechanics.

Key Findings & Fixes:

1. The "Missing Siblings" Bug: When expanding or collapsing nodes, siblings (like grid when expanding component, or README.md when collapsing component) were disappearing.

   • Root Cause: I traced this down to a critical bug in Store.hydrateRecord. When TreeStore.get(key) was called for a hidden node (not in the visible _items array), it fetched the raw object from #allRecordsMap and correctly called hydrateRecord. However, hydrateRecord was unconditionally calling me.map.set(pk, record). This leaked hidden nodes into the active Collection.map.
   • Impact: When TreeStore.splice generated the visibleToAdd array and passed it to Collection.splice, Collection.splice skipped adding those nodes because !map.has(key) evaluated to false (since the hidden node had already bled into the map). This resulted in an empty addedItems array, silently failing the operation and corrupting the visible list.
   • Fix: I updated Store.hydrateRecord to only replace an item in me.map if it already exists in the map (if (me.map.has(pk))). This prevents hidden/filtered items from bleeding into the active collection state during "Soft Hydration".

2. Turbo Mode / Soft Hydration (autoInitRecords: false):

   • In Turbo Mode, the grid.Row expects properties like depth, isLeaf, collapsed, siblingCount, and siblingIndex to be available directly on the raw data objects.
   • I updated TreeStore.splice to actively perform Soft Hydration during ingestion. If raw data objects are missing depth, isLeaf, or collapsed, it computes and injects them directly (e.g. depth = parent.depth + 1).
   • siblingCount and siblingIndex were already being handled correctly via updateSiblingStats(), which mutates the raw objects.

Unit Tests Added: I expanded test/playwright/unit/data/TreeStore.spec.mjs to comprehensively cover these scenarios:

1. should only remove visible descendants on collapse, not siblings: Verifies the fix for the Store.hydrateRecord map bleed bug.
2. singleExpand mode should collapse siblings when a node is expanded: Verifies the correct interaction of singleExpand when sibling nodes are toggled.
3. #childrenMap and #allRecordsMap caching logic: Verifies O(1) accessibility for both visible and hidden nodes.
4. siblingCount and siblingIndex are correctly resolved for raw data objects: Verifies the Turbo Mode (Soft Hydration) logic successfully computes and assigns all hierarchical metadata (depth, isLeaf, collapsed, siblingCount, siblingIndex) to raw JSON objects without instantiating full Records.

All 4 tests are passing! I have committed the changes and will close this issue now.