“Know how to solve every problem that has been solved.” “What I cannot create, I do not understand.” — Richard Feynman

CSV engine, one idea at a time (modern C++)

Five self-contained C++20 programs, each one idea faster: getline baseline (130 MB/s) → mmap state machine → zero-copy string_views (proven: 0 allocations/row) → SWAR scanning → a typed GROUP-BY query at 900 MB/s, 7× the naive pipeline. RFC-4180 correct at every step, benchmark numbers in the README.

• bench.sh 27 lines · 1.1 KB • csv_test_common.hpp 103 lines · 4.2 KB • gen_data.py 89 lines · 3.2 KB • README.md 77 lines · 4.0 KB • step1_naive.cpp 98 lines · 3.8 KB • step2_mmap.cpp 119 lines · 4.8 KB • step3_zerocopy.cpp 140 lines · 5.5 KB • step4_swar.cpp 184 lines · 6.8 KB • step5_typed.cpp 281 lines · 10.2 KB