[willeastcott]

getType() returns null for a class member whose name isn't an identifier: a private field (#secret), or a computed (['computed']), quoted ('quoted') or numeric (42) name. Two callers used the result without checking it:

1. getAttributes() read type.name for every non-static property of an ESM script, so any script class with such a member threw TypeError: Cannot read properties of null (reading 'name').
2. AttributeParser.getNodeAsAttribute() destructured the result, so an @attribute tag on such a member, or such a member inside an @interface type, threw TypeError: Cannot destructure property 'type' of 'getType(...)' as it is null. getAttributes() calls parseAttributes() first, so a script with an @interface attribute or a tagged private field hits this one first.

Both threw out of the parser, so the whole file returned no attributes and no diagnostics. The PlayCanvas Editor's code editor runs getAttributes() in a web worker for attribute autofill, so a script with a plain private field crashed that worker on every parse, and these crashes show up in the editor's error reports.

Changes

• getAttributes() skips a member when getType() returns null.
• getNodeAsAttribute() returns early in the same case. Both of its callers already treat an empty result as "not an attribute".
• Add member-names.valid.js regression coverage: one script with a private field, a private field tagged @attribute, computed, quoted and numeric names, an @interface attribute whose type has a private field, and two valid attributes. It checks that parseAttributes() and getAttributes() both succeed with no errors, that only the identifier-named attributes come back, and that the interface's schema leaves out its private field.

Members with these names can't be attributes, so they're skipped silently, which matches how other non-attribute members are treated. An @attribute tag on a private field is ignored rather than reported. That could become an error in a follow-up if it's worth flagging.

Testing

• npm test: the 5 new tests pass. Before the fix, both before hooks threw the destructuring error; with only the getNodeAsAttribute() change, the getAttributes() test threw the type.name error, so both guards are needed. All other tests are unchanged. On my Windows checkout misc.valid.js "a" fails the same way with and without this change, because core.autocrlf adds a trailing \r to its multi-line description.
• npm run lint and tsc -p tsconfig.build.json --noEmit are clean.

Manual smoke test

1. In an ESM script, add #secret = 1; alongside /** @attribute */ speed = 1;, then parse and open attribute autofill in the code editor.
   • Expected: speed is still offered and nothing throws.
2. Tag the private field: /** @attribute @type {number} */ #secret = 1;.
   • Expected: no exception; #secret isn't listed as an attribute.

🤖 Generated with Claude Code