fix(cpp): a qualified receiver type never reaches a class that only shares its last name, and rocksdb::DB is RocksDB's own

[colbymchenry]

What

A C++ receiver declared with a namespace-qualified type should call a member of a class that qualified name can be, never one of a class that only shares its last name. Most of this landed on main while this branch waited:

• fix(cpp): a call on a declared pointer, value or reference reaches the class its type names, not a namesake #2493 / fix(cpp): a call on an auto variable or a chained call reaches the class C++ finds, not a namesake #2532: matchCppDeclaredTypeCall runs before the last-name match; isCppOutsideQualifiedType nulls absl::Status s; s.ok().
• fix(cpp): -> on an absl::StatusOr reaches the class it holds; a using-declared member of a dependent base is found #2539: -> on absl::StatusOr reaches the held type.

I re-checked on main eec9288, and the original reproduction passes there. This PR fixes the three gaps left over:

1. A namespace a macro stands for (a regression from fix(cpp): a call on an auto variable or a chained call reaches the class C++ finds, not a namesake #2532). RocksDB declares its classes in namespace ROCKSDB_NAMESPACE under #define ROCKSDB_NAMESPACE rocksdb, so the index names them ROCKSDB_NAMESPACE::DB. cppProjectScopeNames only held segments of indexed type names, so rocksdb::DB* db; db->Put(…) was judged a call on an outside type and got no edge. The same happened to ROCKSDB_NAMESPACE::Cache in a tree that declares namespace rocksdb itself. The set now adds the name a single-name macro stands for, in both directions, through cppNameMacros, the #define X y map cppNamespaceMacros already builds.
2. A lowercase outside type beside a project namesake. isCppLibraryType (fix(cpp): a receiver declared as a std or other outside type with a lowercase name calls that type's own member #2413) took a non-std lowercase type as the library's only when no project type had its name at all, so absl::flat_hash_set<int> seen; seen.insert(…) reached a project's own util::flat_hash_set::insert. It is now also the library's when every project type of that name is a namesake (cppFindsOnlyNamesakes).
3. A capitalized type whose last name finds only namesakes, on the paths fix(cpp): a call on a declared pointer, value or reference reaches the class its type names, not a namesake #2493's call-site rules leave to the last-name match. One is a declaration read outside the calling function. protobuf's _status.ok() in its RETURN_IF_ERROR macro body reached upb::Status::ok. auto bytes = ::hpb::Serialize(…); bytes.value(), typed from hpb.h's absl::StatusOr<absl::string_view>, reached hpb::StatusOr::value. The other is a namespace the project uses. Its generated DescriptorProto::ExtensionRange, which the index doesn't hold (descriptor.pb.h is over the size cap), reached Descriptor::ExtensionRange::options. Such a call now gets no edge, as isUndeclaredTypeName rules where no class shares the name. A -> on a value still goes to the name strategies, as isCppArrowThroughOutsideType does, and a template parameter's member is left alone.

cppFindsOnlyNamesakes asks whether every C/C++ type of the last name, and every owner of a method of the called name, has a qualified name that ends unlike the written one, neither being the other's tail. Method owners count because the index can hold a class's methods without the class, as with generated protobuf messages. C++ lookup only adds scopes in front of a written name, and the index drops some (a namespace a macro opens, one lost to a parse error), hence the tail rule both ways. Namespace aliases are expanded, and macro-named namespaces are read both ways. A name a project type qualifies (ns::Derived::Node), an unqualified name and an unbalanced <> are never judged, and neither is a name nothing in the project shares, which main's own rules already handle.

Validation

Tests: __tests__/cpp-outside-type-receiver.test.ts, 9 tests.

• Red on main: 4 fail, the four bug cases: rocksdb::DB::Put and ROCKSDB_NAMESPACE::Cache getting no edge, absl::flat_hash_set reaching util::flat_hash_set::insert, and DescriptorProto::ExtensionRange reaching Descriptor::ExtensionRange::options. The 5 guard tests pass on main too.
• Green: all 41 C/C++ test files pass, 392 tests. Full suite: 7,117 passed; the 2 failures were load-bound (daemon-pid-reuse EBUSY on teardown, function-ref Caller/impact graph misses methods passed as first-class references (callbacks), e.g. executor.submit(obj.method, …) #1820 hitting its 60 s in-file timeout), and both files pass on their own (36/36).

Mutations: each guard switched off fails exactly its own test.

Mutation	Test that fails
macro mapping, forward / reverse	the two rocksdb tests
lowercase rule	absl::flat_hash_set reproduction
gate off	DescriptorProto::ExtensionRange reproduction
owners.length > 0	outside type with no namesake stays unguessed (grpc::ClientReader)
method owners	a class whose methods alone are indexed (DescriptorProto::Clear beside an upb::DescriptorProto)
template-member exclusion	typename PolicyT::Payload
tail rule / alias expansion	fmt's fmt::detail::buffer / u::flat_hash_set through namespace u = util;
qualifier guard	ns::Derived::Node inherited from Base
-> on a value → name strategies	folly::Optional<Widget> holder

The one guard-less mutation, the gate extended to lowercase names, changes nothing: those stay with isCppLibraryType.

Site-keyed A/B (main eec9288 vs this branch, kit %TEMP%\cgstd, diff-sites.mjs):

repo	removed	added	retargeted
google/leveldb	0	0	0 (byte-identical)
fmtlib/fmt	0	0	0 (byte-identical)
redis (C control)	0	0	0 (byte-identical)
facebook/rocksdb	0	0	0 (byte-identical)
protocolbuffers/protobuf	7	0	0

All 7 protobuf removals were wrong edges:

• hpb_generator/tests/test_generated.cc:488, :495: bytes.value() on auto bytes = ::hpb::Serialize(…), an absl::StatusOr<absl::string_view>, reached hpb::StatusOr::value.
• src/google/protobuf/compiler/importer_unittest.cc:567, 569, 579, 581: range.options() on const DescriptorProto::ExtensionRange& range, the generated message, reached the descriptor class's Descriptor::ExtensionRange::options.
• src/google/protobuf/stubs/status_macros.h:33: _status.ok() on const absl::Status _status in RETURN_IF_ERROR reached upb::Status::ok.

A logging probe on main listed every qualified-type site main still sends down the last-name path. Besides these 7, the only owner-mismatch sites left are right ones:

• inherited methods, like DefaultFieldComparator → SimpleFieldComparator::set_treat_nan_as_equal;
• typename PolicyT::Payload template members, kept on purpose;
• hpb/arena.h's backend::Arena → hpb::Arena (3 sites), an alias question for the found-class path rather than an outside type.

The rocksdb and lowercase gaps don't occur in these five repos: rocksdb itself writes ROCKSDB_NAMESPACE:: (8,602 uses against a handful of rocksdb:: calls). The fixtures pin them.

Overlap / convergence

• No second outside rule. fix(cpp): a call on an auto variable or a chained call reaches the class C++ finds, not a namesake #2532's isCppOutsideQualifiedType keeps its namespace-head test and gains the macro mapping through cppProjectScopeNames, which the held-type and chain paths also read. fix(cpp): a receiver declared as a std or other outside type with a lowercase name calls that type's own member #2413's isCppLibraryType and the new gate share cppFindsOnlyNamesakes.
• An earlier version of this branch, before fix(cpp): a call on a declared pointer, value or reference reaches the class its type names, not a namesake #2493 / fix(cpp): a call on an auto variable or a chained call reaches the class C++ finds, not a namesake #2532 / fix(cpp): -> on an absl::StatusOr reaches the class it holds; a using-declared member of a dependent base is found #2539 landed, had its own outside verdict before the typed match. It was dropped for the above.

CHANGELOG: #2413's and #2532's [Unreleased] bullets were extended rather than given new ones, since neither is released yet (latest release 1.6.2), as #2530 did.

🤖 Generated with Claude Code