react-redux-benchmarks

Performance benchmark harness for React-Redux.

Quick Start (bench.ts)

bench.ts automates the full build → publish → benchmark → compare cycle. Requires Bun.

Run benchmarks against a local react-redux build

# Build current working tree, run all scenarios (10s each), save results
bun bench.ts run --rr-dir ../markerikson-react-redux

# Shorter run with specific scenarios
bun bench.ts run --rr-dir ../markerikson-react-redux --label my-fix -l 5 -s entity-list,forms

# Run against a specific commit (auto-restores git state after)
bun bench.ts run --rr-dir ../markerikson-react-redux --sha 471666e --exports bisect-results/exports-471666e.ts

Compare results

bun bench.ts compare bench-results/baseline.json bench-results/my-fix.json

Outputs a side-by-side table with Script time, SigSel time, Sig#, Reconcile time, and deltas (ratio with ↑/↓ indicators).

List saved results

bun bench.ts list

Run options

Flag	Description
--rr-dir <path>	Path to react-redux repo (required)
--label <name>	Label for this run (auto-generated from git SHA + branch if omitted)
--sha <commit>	Checkout this commit before building
--exports <file>	Copy this file as src/exports.ts before building
-l, --length <sec>	Seconds per benchmark (default: 10)
-s, --scenarios <list>	Comma-separated scenario names

Results are saved as JSON to bench-results/<label>.json with git metadata (SHA, branch, dirty state, commit message).

Manual Usage

Building

pnpm build              # standard build
pnpm build --instrument # with dispatch-cycle instrumentation

Running

pnpm start                         # all scenarios, 30s each
pnpm start -l 5                    # 5s per scenario
pnpm start -s deeptree forms       # specific scenarios
pnpm start -v 8.1.1                # specific react-redux version
pnpm start --json                  # JSON output
pnpm start --profile               # V8 CPU profiling with module attribution
pnpm start --instrument             # collect instrumentation stats (requires instrumented build)
pnpm start --save-profiles          # save .cpuprofile files to ./profiles/

Installing a local react-redux build

# In the react-redux repo:
yarn build && yalc publish

# In this repo:
yalc add react-redux && pnpm install

Reading the results

Blocked — total time inside dispatch()

Blocked is the sum of every dispatch() call's duration over the run, measured by dispatchTimingMiddleware (src/common/dispatch-timing.ts) around next(action). It is the time the main thread spent unable to paint or handle input because a dispatch was in progress — the quantity INP penalizes.

It measures where work runs, not how much there is. react-redux notifies every subscriber synchronously inside dispatch, so on master this number tracks the real cost. But any change that moves the notification pass out of the synchronous window — deferring it to a macrotask, time-slicing it — makes Blocked collapse toward zero without removing a single unit of work.

A drop from 8704 ms to 4 ms is therefore not a 2000× speedup, and reporting it as one will not survive review. Always read Blocked alongside Script and Task from the CDP table, which do account for total work:

what changed	Blocked	Script / Task
work genuinely eliminated	down	down
work relocated off the dispatch path	down hard	flat (or slightly up)
work added	up	up

Both are worth optimizing — responsiveness and throughput are different goals — but they are different claims and the table should make that visible.

Adding a benchmark

Benchmarks live in src/scenarios/. Each benchmark must render a React component:

import { renderApp } from '../../common'

renderApp(<App />, store)

Where App is your benchmark component and store is your Redux store.