[cpruijsen]

regex() now accepts \} and \/ instead of reporting them as unnecessary escapes. Fixes #1690.

With the u or v flag, an unescaped } is a SyntaxError (Lone quantifier brackets), so \} is the only way to match a literal }. Before this change, regex("\\{\\}", "u") failed to typecheck with Escape preceding } is unnecessary and should be removed.

parseEscapedChar (ark/regex/escape.ts) rejects any escaped char that is not in Control. Control in ark/regex/state.ts held every ECMAScript syntax character except }, so this adds }. Escapes inside a char class use the same function, so [\}] is fixed too.

It also adds /. \/ failed for the same reason and is a valid escape in every mode, including u and v. The issue names only }, so if you want / out, remove | "/".

The new escaped braces (with the "u" case from the issue) and escaped forward slash tests in ark/regex/__tests__/regex.test.ts fail on the base commit and pass with the change.

Not in this PR: regex("}", "u") still typechecks, but new RegExp("}", "u") throws. Catching that needs a flag-aware check for unescaped literals.

[pullfrog]

✅ No new issues found.

Reviewed changes

• \} accepted by Control — ark/regex/state.ts:95 adds "}" to the Control union, so parseEscapedChar (ark/regex/escape.ts:55) now returns } instead of writeUnnecessaryEscapeMessage. Correct: } is an ECMAScript SyntaxCharacter, and under u/v the escape is the only way to match it literally. The same function backs the char-class path (ark/regex/charset.ts:67), so [\}] is fixed too.
• \/ accepted by Control — ark/regex/state.ts:98 adds "/". Valid as an identity escape in every mode (verified new RegExp("a\\/c", "u") and "v" both parse), so this is safe; see the note below on whether it belongs in the same set as the syntax characters.
• New tests — escaped braces (including the u-flag case from the issue) and escaped forward slash in ark/regex/__tests__/regex.test.ts. I confirmed both fail on the base commit with Escape preceding } / is unnecessary and should be removed. and pass with the change, and that the existing escape/charset suite still passes (57 passing).

ℹ️ "/" is a deliberate scope expansion beyond issue #1690

The issue names only }, and / is not a regex syntax character — in a string-form RegExp an unescaped / is already literal, so \/ is never required, unlike every other member of Control. It is nonetheless a valid identity escape in all modes and eases porting regex literals (where \/ is mandatory) into regex(). This is a judgment call the author already flagged; keeping it is reasonable, but it should be an intentional exception to the "unnecessary escape" rule.

Technical details

# `"/"` in `Control` changes behavior without addressing a required escape

## Affected sites
- `ark/regex/state.ts:98` — `"/"` added to the `Control` union.

## Required outcome
- Decide whether `regex("a\\/b")` should typecheck. If yes, the exception is fine as-is. If no (to stay consistent with rejecting escapes that are never necessary in string form), remove `| "/"` and drop the `escaped forward slash` test.

## Notes
- Verified: `new RegExp("a\\/c", "")`, `"u"`, and `"v"` all parse — `\/` is valid everywhere.
- Contrast with `"-"` (already in `Control`): `new RegExp("a\\-c", "u")` throws `Invalid escape`, so `Control` already contains at least one escape that is only valid inside a char class. Fully flag-aware validation would need to distinguish those, which this PR (correctly) does not attempt.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[ssalbdivad]

thanks @cpruijsen! merged with one tweak: dropped \/, since an unescaped / is already literal in a string pattern under every flag, so that escape is never needed (unlike } with u/v)