[mattcosta7]

Reverts #129. The source and tests go back exactly to their state before that merge (an empty diff against b4f19bd^1).

Why

#129 has no measurable benefit on real github.com pages, and it adds costs.

Benefit, measured. I ran logged-out React pages in Playwright WebKit 26.5: code view, an issues list and a repo home. Each run was about 20 s and included 45 inputs; figures are medians of 3 runs. These pages schedule 16 to 25 idle callbacks, with 4 to 8 ms of idle work in total and no single callback over 1 ms. At most 2 callbacks per run landed between an input and the next paint. Input-to-next-paint p90 was the same under the old polyfill and the new scheduler: 17/17 ms, 24/24 ms and 16/18 ms. The large gains appear only in synthetic workloads that keep the idle queue full.

Costs, measured in the same WebKit setup:

• CPU spin. A cheap self-rescheduling callback, such as github-ui's diff loader polling an empty queue, runs about 25,000 times/s and holds about 53% of a core (Defer idle requests made during a period unless the requesting callback used its slice #130 addressed this).
• Later idle work. The median start delay goes from 1–8 ms to 14–24 ms. Throughput-bound idle work finishes 1.6–2x later.
• Wake-ups every frame. The page wakes every frame while any idle work is pending. A bare rAF loop costs 6.7% CPU, against 1.7% for a 125/s timer loop.

Not measured

Logged-in pages, real Safari, iOS, and slower devices. If a logged-in Safari profile shows idle callbacks regularly landing between inputs and paints, the scheduler from #129 plus #130 can be reconsidered.

Supersedes #130.

[Copilot]

🟡 Changes recommended

A moderate active issue remains in src/requestidlecallback.ts: nullable options are rejected and Symbol timeout values are not synchronously coerced before scheduling.

1 open issue

• Normalize nullable options and coerce timeout before scheduling · New

What changed in this PR

This pull request reverts the frame-based requestIdleCallback scheduler from #129 and restores the simpler timer-based polyfill.

Changes:

• Restores setTimeout-based callback scheduling with a 50 ms deadline.
• Restores direct cancellation through clearTimeout.
• Removes scheduler-specific tests and retains basic API and deadline coverage.

File	Summary	Review comments
test/​requestidlecallback.js	Reduces coverage to basic API and deadline behavior.	No comments.
src/​requestidlecallback.ts	Restores the timer-based idle callback implementation.	Moderate, 4 votes: Normalize nullable options and coerce timeout values before scheduling; current handling rejects null and defers Symbol errors. Moderate, 1 vote: Establish the idle deadline when the callback starts so delayed timer tasks receive a usable budget.

_{Review effort: Lite (auto)}