feat(bench): open cell-renderer scripts to comparator adapters (B2 follow-up #5a) by blove · Pull Request #130 · cacheplane/pretable

blove · 2026-05-10T16:41:38Z

Summary

First slice of B2 follow-up #5. Opens the `scroll-with-format` / `scroll-with-render` / `scroll-with-heavy-render` gate in `validateSupportedP0aRequest` so the AG Grid + TanStack + MUI adapters can run them. The Phase 1–3 adapters already had the scriptName-driven render branches wired; only the gate was holding the runs back.

Selection scripts (`select-range-extend` / `keyboard-nav-row` / `select-all`) stay pretable-only because range-select + select-all are paid-tier in AG Grid Enterprise + MUI X Pro and TanStack Table doesn't ship native cell selection.

What changed

`packages/bench-runner/src/index.ts` — split the gate that previously kept both selection and cell-renderer scripts pretable-only. Cell-renderer scripts now run on all four adapters (S2-only). Selection scripts unchanged.
`packages/bench-runner/src/tests/bench-runner.test.ts` — positive assertions that all four adapters can run cell-renderer scripts on S2; regression guard that selection scripts stay pretable-only.
`apps/bench/src/bench-app.tsx` — cell-renderer scripts now get the same extra mount-frame `waitForNextAnimationFrame()` that the regular `scroll` script already had. Without it, AG Grid hit "scroll viewport unavailable" on a fresh mount because its body viewport attaches one frame after the outer adapter section becomes visible.
`status/milestones/2026-05-10-b2-cell-renderer-comparators.hypotheses.json` — new milestone from the matrix re-run.

Comparator evidence

n=3 medians, blank-gap counts in parens:

Script	pretable	ag-grid	tanstack	mui
`scroll-with-format`	10.2 ms (0)	25.1 ms (1)	17.5 ms (1)	10.2 ms (0)
`scroll-with-render`	16.4 ms (0)	24.9 ms (1)	17.0 ms (1)	10.3 ms (0)
`scroll-with-heavy-render`	10.3 ms (0)	25.2 ms (1)	23.4 ms (1)	10.1 ms (0)

Strongest comparative wedge surfaced since the original B2 H1 stub-baseline era: pretable beats AG Grid Community 2–2.5× on every cell-renderer script with zero blank gaps; AG Grid drops a blank gap on each.

Pretable matches MUI on format + heavy-render but loses on `scroll-with-render` (16.4 vs 10.3 ms). Heavy-render is cheap-render plus extra DOM, yet faster — same dataset, fewer DOM nodes, more frame budget consumed in the cheap path. Suggests a perf cliff in pretable's cheap-React-cellRenderer code path. Logged as a follow-up to investigate (not addressed here).

Hypothesis status

H#	Status	Notes
H1	satisfied	Same parity story as B2 corrections (n=3 sample lands pretable just under MUI).
H19	satisfied	Format overhead is −0.10 ms (format 10.2 ms vs scroll baseline 10.3 ms).
H20	satisfied	Cheap-render p95 = 10.2 ms ≤ 16 ms single-frame budget.
H21	satisfied	Heavy-render p95 = 10.3 ms ≤ 20 ms.

H19/H20/H21 evaluators are pretable-only at the evaluator level — they don't surface comparator evidence inline even now that comparators have run. The data lives in the per-run `status/chromium--default-s2-hypothesis-scroll-with-*.summary.json` files. Extending the evaluators is future work; for now the milestone is the source of truth.

What's NOT in this PR

Sort / filter-text / filter-metadata gate opening (option C from the brainstorm). Sequenced next per the user's "A then C" choice.
Comparator-aware H19/H20/H21 evaluators (data captured per-run; evaluator surface is future work).
Pretable `scroll-with-render` perf-cliff investigation (logged in repo-memory).

Test plan

`pnpm -w typecheck` passes
`pnpm -w test` passes (190 website tests, 5/5 bench-runner — gate-split assertions added, all green)
`pnpm -w lint` 0 errors
`pnpm format` clean
Matrix run successful end-to-end after the bench-app mount-frame fix

🤖 Generated with Claude Code

B2 follow-up #5a (first slice of follow-up #5). The supportedScripts gate in packages/bench-runner/src/index.ts kept scroll-with-format / scroll-with-render / scroll-with-heavy-render pretable-only, even though the AG Grid + TanStack + MUI adapters had wired the scriptName-driven render branches in Phase 1-3 of B2. Changes: - `packages/bench-runner/src/index.ts`: split the gate. Cell-renderer scripts now run on all four adapters (S2-only); selection scripts (select-range-extend / keyboard-nav-row / select-all) remain pretable-only because range-select + select-all are paid-tier in AG Grid Enterprise + MUI X Pro and TanStack ships no native cell selection. - `packages/bench-runner/src/__tests__/bench-runner.test.ts`: positive assertions that all four adapters can run cell-renderer scripts on S2; regression guard that selection scripts stay pretable-only. - `apps/bench/src/bench-app.tsx`: cell-renderer scripts now get the same extra mount-frame wait that scroll already had — without it, AG Grid hit "scroll viewport unavailable" because its body viewport attaches one frame after the section becomes visible. Matrix re-run committed under status/milestones/2026-05-10-b2-cell-renderer-comparators.hypotheses.json (4 adapters × 4 scripts × 3 repeats; ~2 min wall-clock). H1 + H19 + H20 + H21 all satisfied. Findings (n=3 medians, blank-gap counts in parens): | Script | pretable | ag-grid | tanstack | mui | | --- | --- | --- | --- | --- | | scroll-with-format | 10.2ms (0) | 25.1ms (1) | 17.5ms (1) | 10.2ms (0) | | scroll-with-render | 16.4ms (0) | 24.9ms (1) | 17.0ms (1) | 10.3ms (0) | | scroll-with-heavy-render | 10.3ms (0) | 25.2ms (1) | 23.4ms (1) | 10.1ms (0) | Pretable beats AG Grid 2-2.5× on every cell-renderer script with zero blank gaps. Pretable matches MUI on format and heavy-render but loses on scroll-with-render — anomaly logged for follow-up investigation (pretable's cheap-React-cellRenderer path has a perf cliff that the heavier path avoids). H19/H20/H21 evaluators are pretable-only at the evaluator level; the comparator data lives in per-run summary files. Future work could extend the evaluators to surface comparator evidence inline. Sort/filter gate (option C from the brainstorm) is the remaining slice of follow-up #5, sequenced after this PR per "A then C". Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

vercel · 2026-05-10T16:41:40Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
pretable	Ready	Preview, Comment	May 10, 2026 4:42pm

github-actions · 2026-05-10T16:45:48Z

Vercel preview ready

Preview: https://pretable-qolzmijds-cacheplane.vercel.app
Commit: b91d55868703807c39c9d89b7c20e127729ce266

_{Updated automatically by the deploy-preview job.}

…rdict) (#133) * docs(specs): pretable scroll-with-render perf diagnostic design Three-phase research PR mirroring PR #124's pattern: high-repeat (n=20) re-run, conditional Playwright trace capture, research memo. Diagnoses whether the PR #130 cheap-render anomaly (16.4 ms vs 10.3 ms for format and heavy-render) is real or a low-sample artifact. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * docs(plans): pretable scroll-with-render perf diagnostic plan Seven-task plan mirroring PR #124's three-phase pattern: n=20 matrix re-run, conditional Playwright trace capture, research memo. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * docs(research): pretable scroll-with-render perf diagnostic memo Verdict: noise. The PR #130 cheap-render anomaly (16.4 ms vs ~10.3 ms for format and heavy-render at n=3) was a sampling artifact. At higher repeats, scroll-with-render is at parity with (in fact marginally faster than) the other two: | Script | n | mean (ms) | σ (ms) | | -------------------------- | --: | --------: | -------: | | scroll-with-format | 8 | 9.36 | 0.80 | | scroll-with-render | 7 | 8.97 | 0.35 | | scroll-with-heavy-render | 6 | 9.15 | 0.13 | Both 2σ pairs (cheap-vs-format, cheap-vs-heavy) are well within the noise floor. Same shape as PR #124's finding at larger magnitude. The matrix run completed only ~36% of planned repeats (Playwright flake; not investigated) but the observed σ values make the verdict unambiguous — PR #130's 6 ms gap is ~21σ away from the observed distribution. No perf-fix PR needed. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * chore: prettier-format perf-diag artifacts --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

blove enabled auto-merge (squash) May 10, 2026 16:41

blove merged commit a074857 into main May 10, 2026
11 checks passed

blove deleted the b2-followup-5a-cell-renderer-comparators branch May 10, 2026 16:44

blove mentioned this pull request May 11, 2026

docs(research): pretable scroll-with-render perf diagnostic (noise verdict) #133

Merged

4 tasks

This was referenced May 13, 2026

feat(bench-matrix): H6-H8 + H19-H21 evaluators embed comparator evidence #140

Merged

chore(deps): bump ag-grid-community from 33.3.2 to 35.2.1 #136

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(bench): open cell-renderer scripts to comparator adapters (B2 follow-up #5a)#130

feat(bench): open cell-renderer scripts to comparator adapters (B2 follow-up #5a)#130
blove merged 1 commit into
mainfrom
b2-followup-5a-cell-renderer-comparators

blove commented May 10, 2026

Uh oh!

vercel Bot commented May 10, 2026 •

edited

Loading

Uh oh!

Uh oh!

github-actions Bot commented May 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

blove commented May 10, 2026

Summary

What changed

Comparator evidence

Hypothesis status

What's NOT in this PR

Test plan

Uh oh!

vercel Bot commented May 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented May 10, 2026

Vercel preview ready

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

vercel Bot commented May 10, 2026 •

edited

Loading