[gouyt13]

Summary

Make naming and style consistent across the library and let the tools enforce it. Three commits, best reviewed separately:

1. style: brace every control statement body: enables clang-format's InsertBraces and reformats first-party sources. Formatting only (about 1270 bodies were already braced, about 50 were not).
2. refactor: unify naming and style across the library: one convention per concept in core, SIMD, FastScan, HNSW, IVF, SymphonyQG, clustering, utils and the bindings, following the role names in docs/development.md:
   • role names such as internal_id, assignments/cluster_id, num_threads (requested) vs num_workers (resolved), batch_size/num_points, lookup_table/accumulated_values, *_bytes/*_offset;
   • HNSW's hnswlib-style private names, the empty-graph sentinel (kPidMax), and fixed-width link counts;
   • parameter names agree between declarations, definitions and backends; PascalCase template parameters; anonymous namespaces for file-local helpers; internal linkage for headers compiled under several ISA flags;
   • error messages read "<Component> <text>";
   • test suites follow <Module><Area>Test, with the CI filters and tests/README.md updated.
3. chore(tidy): enforce identifier naming and declaration parameter names: extends readability-identifier-naming to every identifier kind, lists the public spellings kept for compatibility, makes naming findings blocking, and enables Strict parameter-name checking. docs/development.md gains a library-wide conventions section.

Compatibility

• No public API, Python keyword, default, saved-index format or numerical change. Public names that break the conventions are kept and listed in .clang-tidy.
• Python-visible message text changes: HNSW messages drop the colon after HNSW; QuantizedGraph messages start with QuantizedGraph; an unknown metric reports metric must be 'l2' or 'ip'; SymqgIndex.save before build reports ... must be built or loaded before save; two IVF messages name IVF. The untrained k-means error is now std::runtime_error (Python RuntimeError before and after).
• Detail-header changes (no stability promise): VisitedListPool → VisitedSetPool with acquire()/release(), rabitqkmeans::detail::RaBitQAssigner → RaBitQAssignment, pipnn_impl folded into symqg::detail, the unused hnsw::detail::search_knn adapter removed.

Testing

Linux x86-64 (Xeon Gold 6418H), GCC 14.3 and Clang 23, RABITQ_ENABLE_NATIVE_OPTIMIZATION=OFF.

• CTest (Release with samples and benchmarks, --no-tests=error): 443 passed, 1 skipped (OpenMP placement). The first commit alone: 427 passed.
• Test inventory: 389 GoogleTest cases, the same set as before apart from renames; every CI --gtest_filter and -R pattern still selects tests.
• Equivalence (checked during the change by independent reviewers against main):
  • HNSW and IVF saved files byte-identical across 12 and 24 save/load cases;
  • 2,400 HNSW search results identical;
  • core quantization and estimation outputs byte-identical over dims 32–1000 and ex_bits 0–8;
  • core, SIMD and FastScan object code identical apart from names.
• Static analysis: check-format.sh, full check-tidy.sh with clang-tidy 23 (bindings on; 0 findings under the new rules), ShellCheck, git diff --check.
• Python (package rebuilt and installed): check-python.sh; pytest tests/python 812 passed, 2 skipped; tests/test_pyright.py and tests/test_mypy.py pass.
• Not run locally: NEON (compiled for aarch64 with Clang in review, not executed), Windows/MSVC, and AVX2-only execution; CI covers them.