libexopt/include/leven.h

#ifndef _LEVEN_H
#define _LEVEN_H

#include "exopt.h"

#ifdef __cplusplus
//#include "util.hh"

#include <string_view>
#include <concepts>
#include <vector>
#include <map>
#include <array>
#include <span>
#include <tuple>

extern "C" {
#endif

size_t _EO(leven_diff)(const char* _EO(restrict), const char* _EO(restrict), size_t) 
	_EO(internal) 
	_EO(readonly)
	__attribute__((nonnull(1, 2)))
;

//TODO: C interface for `string_ord` and `sim_map`?

#ifdef __cplusplus
}

namespace exopt { namespace util [[gnu::visibility("internal")]] {
	template<size_t N =0> // NOTE: Do not change this manually //
	constexpr auto leven_diff(std::string_view s1, std::string_view s2) noexcept {
		const size_t 
			m(s1.size()),
			n(s2.size());

		if(__builtin_expect(!m, false)) return n;
		if(__builtin_expect(!n, false)) return m;

		auto&& work_on_d = [&](auto&& costs) {
			for(size_t i=0;i <= n; i++) costs[i] = i;
			
			size_t i{0};
			for(auto const& c1: s1) {
				costs[0] = i + 1;
				size_t	corner	{ i },
					j	{ 0 };
				
				for(auto const& c2: s2) {
					size_t upper { costs[j+1] };
					if( c1 == c2 ) costs[j+1] = corner;
					else {
						using std::min;
						size_t t{min(upper, corner)};//upper<corner ? upper : corner
						costs[j+1] = min(costs[j], t) + 1;
						
					}
					corner = upper;
					j += 1;
				}
				i += 1;
			}
			return costs[n];
		};

		if constexpr(N > 0) {
			return work_on_d(std::array<size_t, N+1>{});
		} else if consteval {
			return work_on_d(std::vector<size_t>(n + 1));
		} else {
			thread_local static std::vector<size_t> d;
			const size_t n1 = n + 1;
			if(__builtin_expect(d.size() < n1, false)) d.resize(n1);
			// We don't need to clear the buffer, it will be reinitialised by `work_on_d()`.
			return work_on_d(std::span<size_t>{d.begin(), n1});
		}
	}
	template<std::convertible_to<std::string_view> S1, std::convertible_to<std::string_view> S2>
	constexpr decltype(auto) leven_diff(const S1& sa, const S2& sb) noexcept {
		using str = std::string_view;
		return leven_diff(str{sa}, str{sb});
	}

	//TODO: Add overloads for when one bound is known statically but the other is not. Would this be useful?
	template<size_t M, size_t N>
	constexpr decltype(auto) leven_diff(const char (&sa)[M], const char (&sb)[N]) noexcept {
		if constexpr(!N) return M;
		else if constexpr(!M) return N;

		//else if constexpr(M < N) return N - M; // These lines are not correct, just fixing the sizes does not make the edit valid
		//else if constexpr(N < M) return M - N; // ^
		else return leven_diff<N>(std::string_view{sa}, std::string_view{sb});
	}

	/// String that is ordered by Levenshtein distance.
	///
	/// Used as sorted map key for finding suggestions for possible typos. All valid strings are added as `string_ord`, and the invalid string lookup looks for its closest neighbour(s), those are presented to the user in a "did you mean ..." format with the lowest distance first and so on.
	template<std::convertible_to<std::string_view> S>
	struct string_ord {
		using string_type = std::remove_reference_t<S>;

		constexpr string_ord(string_type&& str) noexcept
			: m_string(std::move(str)) {}

		constexpr string_ord(string_ord const&) = default;
		constexpr string_ord(string_ord &&) = default;
		constexpr string_ord& operator=(string_ord const&) = default;
		constexpr string_ord& operator=(string_ord &&) = default;
		constexpr ~string_ord() = default;

		constexpr auto difference_from(std::convertible_to<std::string_view> auto const& str) const noexcept {
			return leven_diff(m_string, std::forward<decltype(str)>(str));
		}
		constexpr auto difference_to(std::convertible_to<std::string_view> auto const& str) const noexcept {
			return leven_diff(std::forward<decltype(str)>(str), m_string);
		}
		constexpr friend auto operator<=>(const string_ord& a, const string_ord& b) noexcept {
			auto ab = leven_diff(a.m_string, b.m_string);
			auto ba = leven_diff(b.m_string, a.m_string);
			return ab <=> ba;
		}

		constexpr operator std::string_view() const noexcept { return m_string; }

		constexpr string_type& string() noexcept { return m_string; }
		constexpr string_type const& string() const noexcept { return m_string; }

		constexpr string_type& operator*() & noexcept { return m_string; }
		constexpr string_type const& operator*() const& noexcept { return m_string; }

		constexpr string_type&& operator*() && noexcept { return std::move(m_string); }
		constexpr string_type const&& operator*() const&& noexcept { return std::move(m_string); }

	private:
		string_type m_string;
	};

	/// Used to store all valid command names, so when an invalid one is found, the closest matching one(s) can be suggested to the user in a "did you mean ..." format with the *lowest difference* neighbour(s) to the invalid string first.
	template<typename T, std::convertible_to<std::string_view> S = std::string_view>
	using sim_map = std::map<string_ord<S>, T>;

} // ::util // 
	}


#endif

#endif /* _LEVEN_H */