C++ API Overview

Headers under include/zedda/ — namespace zedda, C++17, no external runtime deps.

The Zedda core engine is C++17. All public C++ symbols live in namespace zedda, and headers are under include/zedda/.

The C++ API is not part of the public Python surface — it is the engine that powers the Python bindings via nanobind. You would only use these headers directly if you were embedding Zedda into another C++ project.

Header map

Header Key types
column_accumulator.hpp ColumnType, ColumnAccumulator (Welford)
correlation_engine.hpp ColumnPairAccumulator, CorrelationResult, pair_idx, pair_count
profile_result.hpp ColumnProfile, DatasetProfile
hyperloglog.hpp HyperLogLog (14-bit, 16 KB)
mmap_reader.hpp MmapFile (RAII mmap)
simd_scanner.hpp has_avx2(), has_avx512f(), select_best_scanner()
stream_reader.hpp CsvStreamReader, StreamReaderConfig, ChunkResult
profile_builder.hpp ProfileBuilder (orchestrator)
arrow_profiler.hpp ArrowProfiler, ArrowSchema, ArrowArray
parsing_utils.hpp fast_atod, fast_is_null, fast_detect_type, fast_parse_bool

Vendored third-party (not part of the public API)

These live under include/zedda/ but are third-party libraries, not Zedda code:

Library License Path
fast_float v8.0.0 Apache 2.0 / MIT include/zedda/fast_float/
BS::thread_pool MIT include/zedda/BS_thread_pool.hpp

See License & Third-Party Notices for full attribution.

C++ standard and compiler requirements

  • C++17 (required, no extensions)
  • CMake ≥ 3.21
  • GCC, Clang, or MSVC with C++17 support
  • POSIX threads (Linux / macOS) or Win32 threads (Windows)

Compiler flags applied by CMakeLists.txt:

  • GCC/Clang: -O3 -Wall -Wextra -D_FORTIFY_SOURCE=2 -fstack-protector-strong -fPIC
  • MSVC: /O2 /W3 /wd4996 /wd4834 /D_CRT_SECURE_NO_WARNINGS
  • LTO enabled in Release builds via check_ipo_supported() (15–25% expected speedup, PERF-2)
  • simd_scanner.cpp gets extra flags: GCC/Clang -mavx2 -mavx512f -mavx512bw -mpopcnt, MSVC /arch:AVX2

The Python binding

The C++ extension is loaded as fasteda_core (legacy name retained for backwards compatibility; will be renamed zedda_core in v0.6.0 per audit finding P-M26).

The binding (src/bindings/bindings.cpp) exposes:

  • ColumnProfile, CorrelationResult, DatasetProfile — read-only / read-write attribute access
  • ArrowProfiler(file_name, total_rows) — consume_batch(schema_ptr, array_ptr) + finalize()
  • profile(path, show_progress=True, is_sampled=False, sample_size=1000000, correlate=False) — the GIL-releasing entry point used by the Python scan() / profile()

You normally never need to import fasteda_core directly — use the Python API.

Next steps