A field-array fix survived eight days before its author reverted it

For eight days react-hook-form carried a fix that was supposed to make nested setValue calls reach every field array watching the name being written. It came back out on a Sunday night, and the commit that replaced it added no behaviour at all.

The fix landed on 9 May as part of pull request 13420, which claimed to close an issue opened in January about field values in useFieldArray failing to update after a setValue call.^[2][4] Its whole mechanism was a new helper that walked a dotted field name outward and collected every registered ancestor that looked like an array root.^[5] The loop rebuilt the prefix one segment at a time and pushed any prefix that was both a known name and immediately followed by a numeric index.^[5]

3export default (names: Set<InternalFieldName>, name: InternalFieldName) => {4  const parts = name.split('.');5  const matches: string[] = [];6  let prefix = parts[0];78  for (let i = 1; i < parts.length; prefix += '.' + parts[i++]) {9    !isNaN(+parts[i]) && names.has(prefix) && matches.push(prefix);10  }

Eight days later a user reported on the pull request that the new notification path was remounting components and stealing focus mid-keystroke.^[6] The maintainer asked for a reproduction, and then, without waiting for one, reverted the change outright about forty minutes later.^[7][1] The revert deleted the helper file in its entirety rather than narrowing the condition that produced the extra notifications.^[1]

-import type { InternalFieldName } from '../types';--export default (names: Set<InternalFieldName>, name: InternalFieldName) => {-  const parts = name.split('.');-  const matches: string[] = [];-  let prefix = parts[0];--  for (let i = 1; i < parts.length; prefix += '.' + parts[i++]) {-    !isNaN(+parts[i]) && names.has(prefix) && matches.push(prefix);-  }--  return matches;-};

Forty minutes after the revert the same maintainer landed a test-only commit adding regression coverage for exactly the key-thrashing behaviour that had just been removed.^[3] That ordering is the interesting part: the project chose to encode the symptom it had observed before it had a fix that survived contact with the symptom.^[3] A separate revert of a related field-array change had gone in six minutes before the first one, so two fixes to the same subsystem came out on the same evening.^[8]