[clang][reflection] Discriminate same-headed member-template reflections of a specialization in NTTP mangling

[Cfretz244]

Fixes #300.

Stacked on #287 and the deduction-guide PR (#299) — the first two commits are theirs (this change amends the same mangleReflection hash block); review the last commit only.

Problem

The NTTP discriminator for overloaded function-template reflections (#286 / #287) hashes the template parameter list plus the declaration pattern via ODRHash::AddFunctionDecl(pattern, /*SkipBody=*/true). But AddFunctionDecl silently no-ops for any declaration in "specialization context" — its decl-context walk returns on finding a ClassTemplateSpecializationDecl. A member template of an instantiated class template — the common members_of shape — therefore contributed only its template HEAD to the hash, and same-named siblings with identical heads still mangled identically. #287's own field shape (absl operator[] + its pack twin) escaped only because those heads differ.

Field shape: tl::expected<T,E>'s four value() member templates (const&/&/const&&/&&, one shared template <class U = T, enable_if_t<!is_void<U>::value>* = nullptr> head) — all four reflections got one mangled name, CodeGen silently folded the linkonce_odr dispatcher specializations of a reflection-driven binding generator, and value() never bound. Caught by a differential test suite, not a diagnostic; where the TU collides explicitly the symptom is error: definition with same mangled name ... as another definition (see #300 for the reproducer).

Fix

clang/lib/AST/ItaniumMangle.cpp, same hash block: additionally hash what AddFunctionDecl skips in specialization context —

• the pattern's function type via ODRHash::AddQualType (return type, parameter types, cv-quals). The ODR hash handles the dependent pattern types that a structural mangling of the type cannot (lifetimebound SFINAE, noexcept(...) referencing parameters — the original [clang][reflection] Discriminate same-named function-template reflections in NTTP mangling #287 constraint);
• the ref-qualifier, which even ODRHash's VisitFunctionProtoType omits — two siblings can differ in nothing else.

Test

libcxx/test/std/experimental/reflection/fn-template-nttp-mangling-spec-context.pass.cpp (new): the four-sibling value() shape (identical heads, cv/ref-qualifier + return-type differences) and a ref-qualifier-only pair, each pinned as &probe<m> NTTPs from a members_of walk over the specialization and asserted pairwise distinct at runtime.

Validation

• Base: the deduction-guide PR's branch tip (itself on [clang][reflection] Discriminate same-named function-template reflections in NTTP mangling #287 on 837da39eb88c); builds standalone.
• Without the fix: the Same-named member function templates of a class template specialization with identical template heads mangle identically as reflection NTTPs — silent linkonce_odr fold (gap in #286's fix) #300 reproducer fails at -c with the duplicate-mangled-name error (clean at -fsyntax-only; clean with the same signatures at namespace scope — the no-specialization-context control); the new test fails.
• With the fix: reproducer matrix clean; new test passes; [clang][reflection] Discriminate same-named function-template reflections in NTTP mangling #287's substitute-nested-dependent.pass.cpp and the deduction-guide test still pass on the stack tip.
• clang/test/Reflection: 16/16 with the fix on this machine — identical to base.
• Downstream soak (reflection-driven nanobind binding generator): 53-test binder suite green; the TartanLlama/expected v1.3.1 corpus run that caught this (its value() differential went from AttributeError to passing) is green end-to-end, tl::expected<int,std::string>::value() bound with all four sibling reflections mangled distinctly.

🤖 Generated with Claude Code