fix(cpp): a call on the member of the pair a map's iterator points at, it->second.Flush(), calls the value's method

[colbymchenry]

Summary

This is the follow-up left after #2598: calls through the key or value of a map's iterator, and through the members of a std::pair generally.

The extractor records it->second.Flush(), a range-for's kv.second.Flush() and (*it).first.Print() by their bare names. The name strategies then gave those calls to whichever class had a method of that name. Examples:

• rocksdb's 45 pair.first.c_str() calls on std::strings all went to googletest's FilePath::c_str;
• protobuf's kv.second->GenerateMembers(…) went to ImmutableEnumFieldGenerator instead of OneofGenerator.

Separately, a local set from such a member (auto& v = it->second) had no type.

What changed:

• cppPairMember: x.first / x.second (and ->) on a std::pair now type as the pair's elements in C++ expression typing. So a map's iterator, a range-for element, a declared pair, or a pair a method returns all reach their key and value. This covers the receiver forms below, auto& v = it->second, and range expressions like stats_history_.begin()->second.
• More receivers read from the source: fix(cpp): a call on a dereferenced iterator or pointer, (*it)->Visit(), calls the method of what it points at #2598's bare-name receiver reader (now cppUnspelledReceiver / matchCppUnspelledReceiverCall) also takes name->second, name.first and (*name).second.
  • A bare (*this) is still left to cppImplicitReceiver.
  • The pattern requires a dereference or a first/second member, so plain this->Foo() calls are untouched.
• Template parameters: only the calling function's own template parameter names "no particular class". A member typed by another class template's parameter goes on as before. An example is CoreLocalArray<T>::AccessElementAndIndex(), which returns a std::pair<T*, size_t>. The first A/B caught exactly that: rocksdb's p.first->GetUncompressData(idx) lost its correct link, and the guard restores it.

Validation

• New test __tests__/cpp-map-iterator-calls.test.ts (5 tests): 3 fail on main and pass here. The 2 guards pass on both: another template's parameter, and this->Flush().
• Mutations: switching off each part fails its test:
  • pair members in expression typing: 3 tests;
  • name->second receivers: 3;
  • (*name).second receivers: 1;
  • the foreign-template guard: 1.
• A/B (full index, main vs fix) on google/leveldb, fmtlib/fmt, redis (C control), facebook/rocksdb and protocolbuffers/protobuf. Every changed site was triaged, by category and line by line for every project-named method.
  • rocksdb, 130 removed: all library calls on a pair member's std::string, container, smart pointer, std::optional or weak_ptr that had been guessed onto project methods (c_str, size, data, find, get, lock, has_value, …). For example, main's filters.size() reached Slice::size at 0.9, but filters is it->second of a string-to-string map.
  • rocksdb, 55 added: Slice keys and values (entry.first.ToString()), HistogramStat, RangeDelAggregator::StripeRep, Snapshot, TrackedKeyInfo, InternalKey, and range-fors over a pair's member (ObjectLibrary::Entry).
  • rocksdb, 38 retargeted from 0.4/0.7 guesses:
    • batch_and_status.first → WriteBatch;
    • wal_and_status->first → log::Writer;
    • pair.second->Timestamp() → cassandra::ColumnBase;
    • toku range_buffer;
    • Timer::FunctionInfo;
    • an OptionTypeInfo::AreEqualByName overload, now the std::string one, matching iter->second of a string map.
  • rocksdb, 179 metadata-only: same targets, confidence up to 0.9.
  • protobuf:
    • 29 removed: library calls, plus descriptor accessors (p.second->file(), e.second->options()), which aren't indexed;
    • 1 added: ExtensionSet::Extension;
    • 14 retargeted: OneofGenerator, UnknownField;
    • 8 metadata-only.
  • leveldb: 1 metadata-only.
  • fmt, redis: byte-identical.
• Full suite, including every C/C++ test file: 7,155 passed, 0 failed.

Remaining, out of scope: element access the extractor drops (m[k].Foo(), arr_[0].Foo()), and m.at(k).Foo() chains, which would need library containers' element-returning members.

🤖 Generated with Claude Code