react-doctor/no-flush-sync

Use startTransition for non-urgent updates — flushSync forces a sync flush that skips View Transitions and concurrent rendering

• Category: Performance
• Severity: warn
• Source: oxlint-plugin-react-doctor
• Framework: global
• Enabled when: always

Validation prompt

Use this to decide whether a fired diagnostic is real or a false positive.

Confirm an ImportDeclaration whose source is exactly "react-dom" with a named specifier flushSync. The rule reports the import (one diagnostic per file) rather than every call site. False positive: integrating with a non-React imperative library that must observe a fully-committed DOM before its next line runs (e.g. before scrollIntoView).

Fix prompt

Use this once validation confirms the diagnostic is real.

Replace the flushSync wrapper with startTransition for non-urgent updates so React can batch and run View Transition snapshots. Only keep flushSync when a third-party API genuinely needs the DOM committed synchronously, and never inside a transition you want animated — flushSync silently drops the View Transition. See https://react.dev/reference/react-dom/flushSync

Related rules

More Performance rules from the rules reference:

• react-doctor/no-global-css-variable-animation: Set the variable on the nearest element instead of a parent, or use `@property` with `inherits: false` to prevent cascade. Better yet, use targeted `element.style.transform` updates
• react-doctor/no-inline-bounce-easing: Use `cubic-bezier(0.16, 1, 0.3, 1)` (ease-out-expo) for natural deceleration — objects in the real world don't bounce
• react-doctor/no-inline-prop-on-memo-component: Hoist the inline `() => ...` / `[]` / `{}` to a stable reference (useMemo, useCallback, or module scope) so the memoized child doesn't re-render every parent render
• react-doctor/no-large-animated-blur: Keep blur radius under 10px, or apply blur to a smaller element. Large blurs multiply GPU memory usage with layer size
• react-doctor/no-layout-property-animation: Use `transform: translateX()` or `scale()` instead — they run on the compositor and skip layout/paint