jqueguiner / num2words2 / 29613473861

95%

main: 95%

Pull #127

github

refactor: thin num2words2 Python binder — move all logic to Rust

num2words2/ is now a pure thin binder like words2num2/: every public function is
a `return _RUST.x(...)` pass-through, with no conversion/dispatch/presentation
logic left in Python.

Moved into num2words2-core (new pyo3-free `presentation.rs`, so the crate still
publishes standalone) + the num2words2-py binder:
- `_normalize_lang` -> `presentation::resolve_lang` (locale resolution).
- `_apply_style` -> `presentation::apply_style` (style=terse/us output post-processing).
- `_normalize_cents` -> `presentation::normalize_cents` (cents kwarg handling).
- `_rust_kw_items` kwarg marshalling -> the binder (`extras_to_kwargs`).
- the whole `num2words`/`num2words_sentence` dispatch pipeline -> `#[pyfunction]`s
  (type inspection + raw-lang resolve + style, sharing the core converters).
- grouping.py: drop the pure-Python western/indian/chinese fallback; it's now a
  pass-through to `_RUST.group_digits` (which raises the same ValueError/TypeError).

__init__.py 300 -> 77 lines. num2words2-core 0.1.0 -> 0.1.1 (published).

Verified: cargo test 579; unittest OK (2780, 33 expected-failures, exit 0);
E2E 16/16; coverage source 84% / tests 98%; flake8 0; isort clean. Output parity
preserved (style/cents/lang-resolve produce identical results).

1257 of 1412 branches covered (89.02%)

Branch coverage included in aggregate %.

9 of 10 new or added lines in 3 files covered. (90.0%)

24489 of 25639 relevant lines covered (95.51%)

5.73 hits per line