fix(cpp): an auto local holding a library container's iterator calls the element's method through ->

[colbymchenry]

Summary

This is the follow-up left after #2576.

An auto local initialized with a library container's iterator had no type. rocksdb's auto it = logs_.begin(); it->IsSyncing() is an example: the auto-local reader (cppExpressionType) had nothing for a call on std::deque<LogWriterNumber>. So -> calls on such locals were matched by the method's name alone.

What changed:

• Container members: a standard-library or abseil container's begin/end/cbegin/rbegin/crbegin/find/lower_bound/upper_bound/… now types as <container>::iterator, with the right iterator kind.

• Searches: a standard search (std::find, find_if, lower_bound, upper_bound, min_element, max_element, partition_point, adjacent_find, next, prev) types as the iterator it is given first.

• From there, fix(cpp): a -> call on a declared container iterator calls the element's method #2576's cppArrowTarget reaches the container's element.

• Where the element is looked up: where the container's type is written — the auto local's declared site, or a data member's class — not at the call. So an element nested in another class (vs->obsolete_files_.begin()) finds that class's type, not a namesake visible at the call.

• Smart pointers are unchanged: their held type is still looked up at the call. A first version moved holders to the declaration site too, and the A/B caught two problems with it:

  • a pimpl's Rep, which the header only declares, wasn't found from the header, so about 80 rocksdb calls dropped from 0.9 to 0.65–0.8;
  • protobuf's index_->AddFile(…) went to SimpleDescriptorDatabase::DescriptorIndex, a namesake visible from the header.

  A new test pins this.

• Project containers keep what their own begin() returns. A project class named like a library container (util::span) isn't taken for it.

Validation

• New test __tests__/cpp-auto-iterator-arrow.test.ts (6 tests): 4 fail on main and pass here. 2 guards pass on both: the project span, and the pimpl smart pointer.
• Mutations: switching off each part fails its test:
  • the container members: 4 tests;
  • the searches: 1;
  • the std/abseil head check: 1;
  • looking elements up at the declared site: 1;
  • holders kept at the call site: 1.
• A/B (full index, main vs fix) on google/leveldb, fmtlib/fmt, redis (C control), facebook/rocksdb and protocolbuffers/protobuf. Every site was triaged:
  • rocksdb:
    • 6 added: it->LargestUserKey() on sub_compact_states.rbegin() reaches SubcompactionState; to_it->Merge(…) on a copy of pairs_.begin() reaches SeqnoTimePair; it->start() over intervals_ reaches Interval, the same target main gives . calls on that type; it->data()/it->size() over operand_list reach Slice.
    • 2 metadata-only: it->IsSyncing() and ReleaseWriter() on logs_.begin() keep their target, with confidence up from 0.7 to 0.9.
  • protobuf:
    • 3 added: it->get() over an absl::flat_hash_set<Arena::UniquePtr<int>> and over one of Arena::Ptr<int> reach Arena::UniquePtr::get and Arena::Ptr::get; it->name(*this) from std::lower_bound(by_name_flat_.begin(), …) reaches FileEntry::name.
    • 2 metadata-only: it->extendee(*this) keeps its target, with confidence up from 0.7 to 0.9.
  • leveldb, fmt, redis: byte-identical.
  • No edges removed or retargeted.
• Full suite, including every C/C++ test file: 7,146 passed, 0 failed.

Remaining, out of scope:

• (*it)->m() / (*it).m() dereferences, and auto x = *c.begin(), which would need cppPointee to dereference iterators;
• map iterators' it->second.m() chains, which need std::pair members.

🤖 Generated with Claude Code