[ASDAlexander77]

Three failures of a @value type that may be missing (T | null, T | undefined), all on main before #760, plus printing of interface values. Found while working on #760.

@value T | null (98e92da)

• rc: return d < 0 ? null : { at: d, owner: new Owner(d) } in a function returning Mark | null read freed memory. The branch released the new Owner before the retain after the ?: ran. consumeConditionalResults' producerOf now sees through the cast from a tuple to the @value type it is laid out as. So yieldsLiteralOfBranchTemporary recognizes the literal, and the branch retains first. 2,000,000 calls peak at 4.7 MB, so nothing leaks.
• ??: find(-1) ?? find(3) with find(): Hit | null was typed Hit | (Hit | null) and did not compile. An optional member of a union contributed its whole element type as one member. processUnionTypeItem now processes the element like any other member, so a union inside it contributes its members.
• Printing:
  • an undefined T | undefined printed null, or a zeroed T; it now prints undefined;
  • a null T | null stored without a tag (@value T | null without a tag when T holds an interface or a class; null tested by descriptor kind #760) printed T's fields; it now prints null. castNullableToString now accepts a @value base.

The new test value/value_missing_ops.ts runs under gc, rc and none, for a T holding a class (no tag) and a T of numbers only (tagged). Own rejects it: it can't yet tell which branch of a ?: owns a literal it merges.

Printing an interface (7fa8c15)

Printing an interface value, or putting one into a string, crashed at run time, whatever object was behind it: a class instance, an object literal, or an interface held by a @value. Per your ruling, an interface now prints its fields, as an object literal does ({ n: 3, name: 'b' }), with fields of the interfaces it extends first.

• An interface that declares toString calls it, as a class does.
• A null interface prints null.
• A field whose interface type is already being printed shows [Object], because interface Chain { next: Chain } would otherwise expand forever.

castTupleToString and the new castInterfaceToString share one formatter, fieldsToString. The new test tests/00interface_print runs under every model.

A C string result from a .d.ts an imported module loaded first (2da1465, TSLANG_ISSUES 21)

The string a C function returns is copied (#575), except when the .d.ts declaring it was first reached through a module this file imports. declaredFunctions kept that earlier pass's module, and isForeignCall required the current one, so the raw char* was used as is. Stored strerror results then all showed the buffer's last text, and under -mm=rc the reference count was written into C's memory and the release freed it. Now a top-level declare function with no body counts as C whichever module loaded its .d.ts. The new test is in c-string-import/, under gc, rc, none and own, compiled and JIT. Its .d.ts lives outside tests/ because tests/*.d.ts are build output and git ignores them.

Testing

• ctest 6560/6560 (Windows Release).
• Fuzz seeds 83000-85599 clean.
• DefaultLib compiles under own, rc, gc and none.

🤖 Generated with Claude Code