Report TS1003 for JSX tag and attribute names starting with # - #65431

JSX names starting with # parse without a syntax error and emit invalid JS (React.createElement(#a, null)). <#a> only gets a checker TS2304, which noCheck suppresses; attribute and namespaced names get no error at all.

The root cause: tokenIsIdentifierOrKeyword also matches KindPrivateIdentifier. As a result, ScanJsxIdentifier turned #a into an ordinary identifier, and the parser created an Identifier with text #a.

<#a></#a>;    // now TS1003: Identifier expected.
<div #b />;   // now TS1003
<div #c:d />; // now TS1003
<a:#e />;     // now TS1003

<a.#g /> already errored, but only because parseRightSideOfDot checks for private identifiers before identifier names.

Changes

• Scanner: ScanJsxIdentifier keeps KindPrivateIdentifier instead of converting it to an identifier. It still appends JSX identifier parts, so #a-b stays one token.

• Parser: the new parseJsxIdentifierName replaces the four JSX name call sites (tag name, attribute name, and both sides of a namespaced name). On a private identifier it:

  • reports TS1003 on the token;
  • consumes the token;
  • returns a missing identifier.

  This follows the <a.#g /> path. The missing name means no TS2304 follows, and emit no longer contains #.

• astnav: the skipped token isn't covered by any node. When getTokenAtPosition rescanned the gap, the regular scanner split #a-b into #a, -, b. The trailing identifier then hit the did not expect … in its trivia panic, e.g. on hover inside <div #n-o={1} />.

  • scanNavigationToken now rescans such a private identifier with ScanJsxIdentifier, matching how the parser tokenized it.
  • This applies inside JSX opening, closing and self-closing elements, attributes and namespaced names, and in the leftmost part of a dotted tag name (<#a-b.c />).
  • It is skipped for <a.#b-c />, where the parser read #b with the regular scanner.

Notes for review

• Emit for these error cases now contains an empty name where the # name was (React.createElement(, null), { "": true }).
• Alternatives rejected:
  • Narrowing tokenIsIdentifierOrKeyword globally: class members like #foo depend on it.
  • Keeping the #a Identifier: it cascades into checker errors and still emits invalid JS.
  • Not extending #a-b with dashes: -b then produces cascading parse errors.

Tests

• jsxNamesStartingWithHash.tsx covers the issue cases plus namespaced and hyphenated variants, for jsx: preserve, react.
• TestGetTokenAtPositionJsxNameStartingWithHash in tsc/internal/astnav/tokens_test.go panicked before the astnav change.

• Fixes JSX tag and attribute names starting with # are accepted without a syntax error #65396