explicit-timer-delay: Add rule

[azat-io]

Close #2809

[azat-io]

Ping

[sindresorhus]

1. Parenthesized first argument breaks the "always" fix
   Repro:

setTimeout((callback));

Expected fix:

setTimeout((callback), 0);

Actual fix result:

setTimeout((callback, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0));

The fixer inserts inside the parentheses, so the call still has one argument and ESLint keeps re-fixing it until the pass limit. This is definitley not safe for --fix.

2. Parenthesized zero breaks the "never" fix
   Repro:

setTimeout(callback, (0));

Expected fix:

setTimeout(callback);

Actual fix result:

setTimeout(callback));

This leaves invalid syntax (extra )), so --fix can break parsing.

[sindresorhus]

Please see other rules for formatting here. Whenever possible, show fail/pass cases together.

[sindresorhus]

Match the common options style used in other docs.

• Use Type + Default together (for example: Type: 'string'\ Default: 'always').
• Use single-quoted option values ('always', 'never') to match repo convention.
• Prefer JS config examples over JSON for consistency with most rule docs.

[azat-io]

@sindresorhus Thanks for the review! I've fixed your comments!

[sindresorhus]

A few more things:

1. In never mode, the fixer rewrites setTimeout(callback, 0, arg1, arg2) to setTimeout(callback, arg1, arg2).
   That changes runtime behavior because arg1 becomes the delay value. This is definitley not safe for --fix.
   Expected: either skip fixing when extra callback args exist, or preserve semantics.

2. The matcher currently treats any global object member named setTimeout/setInterval as a timer call.
   Example: Math.setTimeout(callback) gets flagged and autofixed to add , 0.
   Expected: limit matches to real timer globals (setTimeout, setInterval, and window/globalThis/global members).

[azat-io]

@sindresorhus Fixed ✅

[sindresorhus]

1. The isZeroDelay doc comment has extra leading spaces on each line (2-space indent), while the checkTimerCall comment right above it doesn't. Should match.

2. context.options[0] || MODE_ALWAYS — since defaultOptions: [MODE_ALWAYS] is set, context.options[0] is already populated. The || fallback is redundant. Should just be context.options[0].

3. self is in the globalObjects set but has no test coverage at all. Should have at least one valid + one invalid case.

4. In always mode, setTimeout(...args) gets flagged (no fix). This is a false postiive when args already contains a delay. Consider skipping when the only argument is a spread element.

[azat-io]

@sindresorhus Fixed ✅

[sindresorhus]

https://github.com/sindresorhus/eslint-plugin-unicorn/pull/2814/changes#r2788721127 is not done

[azat-io]

@sindresorhus Fixed ✅

[sindresorhus]

Found 2 issues:

1. never mode reports an unfixable error for setTimeout(callback, 0, arg1, arg2) — The rule flags the 0 delay but correctly skips the auto-fix (since removing , 0 would shift arg1 into the delay slot, breaking the call). This leaves users with a lint error they cannot safely fix without restructuring the call. The rule should skip reporting entirely for this case rather than surfacing an unfixable error.

eslint-plugin-unicorn/rules/explicit-timer-delay.js

Lines 112 to 131 in f60546b

	if (isZeroDelay(delayArgument)) {
	const problem = {
	node: delayArgument,
	messageId: MESSAGE_ID_REDUNDANT_DELAY,
	data: {name},
	};
	if (arguments_.length === 2) {
	problem.fix = function (fixer) {
	const [firstArgument] = arguments_;
	const [, firstArgumentEnd] = getParenthesizedRange(firstArgument, context);
	const [, delayArgumentEnd] = getParenthesizedRange(delayArgument, context);
	return fixer.removeRange([firstArgumentEnd, delayArgumentEnd]);
	};
	}
	return problem;
	}
	}

2. Dead guard in always mode — The if (firstArgument && firstArgument.type !== 'SpreadElement') condition is always true at that point. firstArgument is guaranteed non-nullish (the arguments_.length === 0 early return above ensures arguments_.length >= 1), and the SpreadElement case was already handled by the preceding early return. The condition can be simplified.

[azat-io]

Thanks. I've fixed both issues!

[sindresorhus]

1. In never mode, +0 still slips through. setTimeout(callback, +0) and setInterval(callback, +(0)) are explicit zero delays, but the current zero check only handles 0 and -0, so these are not reported. I double checked this manually from the parsed AST and it definitley does not match the current condition. There should also be a regression test for this form.

2. The 'never' docs still use inline /* eslint ... */ comments in the examples. Earlier review asked for JS config examples to match the rest of the docs, so that comment is not fully addressed yet.

[sindresorhus]

Some tests to add:

• +0 in never mode.
  setTimeout(callback, +0);
setInterval(callback, +(0));
  This is the main missed semantic equivalent of zero.

• self in never mode.
  There is coverage for self.setTimeout(fn, 0) in always, but not for removing 0 in never.
  self.setTimeout(fn, 0);
self.setInterval(fn, 0);

• Parenthesized -0 in never mode.
  There is -0, and there is (0), but not (-0).
  setTimeout(callback, (-0));

• Explicit global shadowing negatives.
  Make sure member calls are ignored when the global object name is shadowed.
  const window = foo; window.setTimeout(callback);
const self = foo; self.setInterval(callback);

• Computed member access should stay ignored.
  globalThis['setTimeout'](callback);
  That documents the current supported boundary.

[azat-io]

Done