[nstepien]

TreeDataGrid found each rendered row's parent group row with rows.indexOf(row), a backward walk over the rows, and childRows.indexOf(row). Without a rowKeyGetter, its own rowKeyGetter did the same lookup again. Scrolling near the end of a large group therefore scanned up to all the rows for every visible row. main has the same problem.

Changes

• The indexes of the group rows are recorded while flattening, and getParentRowAndIndex now takes a row index and runs a binary search over them.
• The rows of a group directly follow its group row, so a row's position in its group is rowIdx - parentIdx - 1, and childRows.indexOf is gone.
• TreeDataGrid only passes a rowKeyGetter to DataGrid when one is provided. Otherwise, renderRow computes the same keys as before: the group ID for group rows, and startRowIndex + groupIndex + 1 for the rows of a group. DataGrid only reads rowKeyGetter for row keys and selection, and selection already requires a rowKeyGetter.
• getParentRowAndIndex is a plain function, since no memo depends on it anymore.

Performance

Median React Profiler duration of a scroll render, over 20 scrolls near the end of the rows, in Chromium with React's production profiling build (react-dom/profiling) and React Compiler:

Rows	Before	After
1 group of 1M rows	451 ms	0.6 ms
1 group of 1M rows, with rowKeyGetter	192 ms	0.7 ms
1M rows, not grouped	351 ms	0.7 ms
1,000 groups of 1,000 rows, all expanded	8.0 ms	0.6 ms

Before, nearly all of that time was spent in TreeDataGrid's own scans, so the times were about the same with React's development build. After, the remaining time is React's, which is about 6 times slower in development.

Grouping

Rows are now grouped with get and set instead of getOrInsertComputed, which was slower for two reasons:

• V8's optimizing compilers inline Map.prototype.get as a hash lookup, but have no such reduction for getOrInsertComputed, so each call goes through the builtin.
• The callback was an arrow function created for every row and level, even though it only runs for a new group. That doubled the young generation garbage collections while grouping.

Median time to group 1M rows in Node.js 26 (V8 14.6), each version in its own process:

Levels	getOrInsertComputed	get and set
1	20 ms	16–17 ms
2	46–48 ms	39 ms

The groups on the last level also no longer create a map of subgroups that always stayed empty, which mostly helps with small groups. Median time and retained heap when grouping 1M rows on one level:

Groups	Before	After
100 groups of 10,000 rows	16 ms, 7.7 MB	16 ms, 7.7 MB
100k groups of 10 rows	42–50 ms, 43 MB	31–40 ms, 26 MB
Random keys, 1.6 rows per group on average	277–281 ms, 279 MB	218–230 ms, 168 MB
Random keys, 2.3 rows per group on average	241–250 ms, 186 MB	194–201 ms, 110 MB
1M groups of 1 row	299–301 ms, 425 MB	216 ms, 249 MB

Tests

A new test checks that, without a rowKeyGetter, a row keeps its element when an earlier group is expanded. It fails if the rows of a group are keyed by their index. vp run ci passes.

🤖 Generated with Claude Code

[amanmahajan7]

Looks like CI is flaky. It keeps timing out

[nstepien]

Yes installing the webkit dependencies can be slow at times, not sure if it's an issue with GH or the remote package repos of the deps.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.96%. Comparing base (0e49dec) to head (e42a270).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4227   +/-   ##
=======================================
  Coverage   98.95%   98.96%           
=======================================
  Files          46       46           
  Lines        2017     2030   +13     
  Branches      787      792    +5     
=======================================
+ Hits         1996     2009   +13     
  Misses         21       21           

Files with missing lines	Coverage Δ
src/TreeDataGrid.tsx	98.66% <100.00%> (+0.08%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.