eslint-config: turn off react-perf/* and react/jsx-props-no-spreading by default

[prisis]

The react-perf rules and react/jsx-props-no-spreading fire heavily on modern React codebases without pointing at real defects. Proposing they be disabled by default in the React preset.

react-perf/jsx-no-new-object-as-prop / jsx-no-new-function-as-prop

These rules predate the React Compiler. Their whole purpose is to stop inline objects and functions from breaking referential equality and causing re-renders — which is exactly what the compiler now memoises automatically.

Worse, they push in the opposite direction to the compiler: satisfying them means hand-rolling useMemo/useCallback, and manual memoisation can make the compiler bail on the component entirely rather than optimise it. So following the rule can measurably reduce optimisation.

They also collide with animation libraries, where prop objects are the API:

<m.div
    animate={{ opacity: 1, rotate: 0, scale: 1 }}
    initial={{ opacity: 0, rotate: -180, scale: 0.95 }}
    transition={{ damping: 20, stiffness: 200, type: "spring" }}
/>

Every one of those three props is a violation. In one app in this org these two rules alone produced 211 of 1027 problems (~21%), essentially all of them Motion props, with no genuine performance finding among them.

react/jsx-props-no-spreading

{...props} is the standard composition pattern for Radix / shadcn-style primitives — a wrapper forwards the underlying element's full prop surface:

const Tabs = ({ className, ref, ...properties }: TabsProperties) => (
    <TabsPrimitive.Root className={cn("flex flex-col gap-2", className)} ref={ref} {...properties} />
);

Complying means enumerating every prop of every primitive by hand and re-doing it whenever the upstream library changes. Same app: 49 violations, all in UI primitives.

Suggested change

Default these off in the React preset, with a comment explaining why, and let consumers opt back in:

// The React Compiler memoises component output, so these flag library prop
// objects (Motion animate/initial/transition) with no benefit — and satisfying
// them via manual memoisation can stop the compiler optimising the component.
"react-perf/jsx-no-new-object-as-prop": "off",
"react-perf/jsx-no-new-function-as-prop": "off",
"react-perf/jsx-no-new-array-as-prop": "off",
"react-perf/jsx-no-jsx-as-prop": "off",

// Radix/shadcn-style primitives compose by forwarding the full prop surface.
"react/jsx-props-no-spreading": "off",

Worth also considering whether react-x/no-unstable-context-value and react/jsx-no-constructed-context-values belong in the same group for compiler-enabled projects.

Happy to open a PR if the direction looks right.

[prisis]

Verified both claims against main and opened #1120.

• react/jsx-props-no-spreading was error in the main React rules block, disabled only for Storybook files.
• react-perf/jsx-no-new-object-as-prop, jsx-no-new-array-as-prop and jsx-no-new-function-as-prop came in at error via the plugin's spread recommended config.

All five are now off by default, jsx-no-jsx-as-prop included. The reasoning about the compiler bailing on manually-memoised components is the part that decided it — a rule that can reduce optimisation while claiming to improve it does not belong on by default.

One implementation note in case it is ever refactored: the disables have to sit after the recommended spread in the same rules object. My first attempt placed them earlier and the spread silently re-enabled all three; the probe still reported every violation.

On the two context rules

I left react-x/no-unstable-context-value and react/jsx-no-constructed-context-values enabled, rather than grouping them with the rest.

They are a different case. The compiler does not memoise across a context provider boundary, so an unstable provider value still re-renders every consumer of that context regardless of what the compiler does inside the component. Those two flag a real effect; the react-perf rules flag a referential-equality technicality the compiler now handles. Happy to revisit with a concrete false positive.

Note this ships as a breaking change, so it will land on the next major.