#pragma once

#include <bit>
#include <stdexcept>
#include <array>

#include <climits>

#ifndef _RNGXX_IMPL
#define _RNGXX_COMMON_MINIMAL
#endif
#include "rngxx/internal/common.h"
#ifndef _RNGXX_IMPL
#undef _RNGXX_COMMON_MINIMAL
#endif

#define CTOR_COPY(name) name(const name& copy)
#define CTOR_MOVE(name) name(name&& move)

// ugly hack to  get around the absolutely retarded name lookup restrictions when overriding	
#define RNG_OVERRIDE(ty, nm, rest) using Random::nm; ty nm rest override

namespace _rng__util {
	struct divp
	{
		inline consteval divp(usize n, usize o)
			: d(n / o),
			  r(n % o){}
		const usize d,r;

		inline consteval bool exact() const { return d && !r; }
		inline consteval bool none() const { return !d; }

		template<typename T>
		inline static consteval divp type(usize bytes) { return divp(bytes, sizeof(T)); }
	};
}


struct InvalidRandomSample final : public std::exception {
	inline explicit InvalidRandomSample(f64 s) : std::exception(), value(s){}
	inline CTOR_COPY(InvalidRandomSample): std::exception(copy), value(copy.value){}
	
	const f64 value;
};
struct ObjectMoved final : public std::exception{};

/// Interface for a simple random number generator
///
/// # Must override
/// f64 _sample() // A representation of a random range. The value must be between `0..=1`
///
/// # Should override
/// void next_bytes(u8* ptr, usize n); // Random bytes. The default implementation falls back to _sample() for each byte. This is very inefficient.
///
/// void next_v32(u32* ptr, usize n);  // Vectorised random bytes (4 bytes ptr iteration.) If 32-bit vectorised outputs are possible for your implementation, you should override this (falls back to next_bytes(ptr, n)) NOTE: `n` is the number of `u32`s `ptr` points to, **not** the number of bytes.
/// void next_v64(u64* ptr, usize n);  // Same as above, but for 8 byte iterations (64 bits.)
struct Random
{
	template<typename T>
	struct iterator { //TODO: Implement this in another file (has to be header because of template :/)
		friend class Random;

		//TODO: Make this work with foreach(), and STL iterator APIs somehow

		inline CTOR_COPY(iterator<T>) : rng(copy.rng){}
		inline CTOR_MOVE(iterator<T>) : rng(move.rng) {
			*const_cast<Random**>(&move.rng) = nullptr;
		}

		inline virtual ~iterator(){}

	protected:
		virtual T _sample() { if (LIKELY(rng)) return rng->next<T>(); else throw ObjectMoved(); }
		virtual inline void _init() {}
	private:
		inline explicit iterator(Random& rng) : rng(&rng){ _init(); }

		Random* const rng;
	};
	public:

	inline Random(){}
	inline virtual ~Random(){}

	inline f64 next_f64() { return sample(); }
	inline f32 next_f32() { return (f32)sample(); }

	virtual bool next_bool();

#define NDEF(t) inline virtual t next_ ## t() { return next_ ## t(_max_ ## t()); }
#define NDEFF(n) NDEF(i ## n) NDEF(u ## n)
	NDEFF(8)
	NDEFF(16)
	NDEFF(32)
	NDEFF(64)
#undef NDEFF
#undef NDEF
	virtual void next_u8(u8* a, usize n);
	virtual void next_i8(i8* a, usize n);

	virtual void next_u16(u16* a, usize n);
	virtual void next_i16(i16* a, usize n);

	virtual void next_u32(u32* a, usize n);
	virtual void next_i32(i32* a, usize n);

	virtual void next_u64(u64* a, usize n);
	virtual void next_i64(i64* a, usize n);

	virtual i8 next_i8(i8 max);
	i8 next_i8(i8 min, i8 max);

	virtual u8 next_u8(u8 max);
	u8 next_u8(u8 min, u8 max);

	virtual u16 next_u16(u16 max);
	u16 next_u16(u16 min, u16 max);

	virtual i16 next_i16(i16 max);
	i16 next_i16(i16 min, i16 max);

	virtual i32 next_i32(i32 max);
	i32 next_i32(i32 min, i32 max);

	virtual i64 next_i64(i64 max);
	i64 next_i64(i64 min, i64 max);

	virtual u32 next_u32(u32 max);
	u32 next_u32(u32 min, u32 max);

	virtual u64 next_u64(u64 max);
	u64 next_u64(u64 min, u64 max);

	virtual inline f32 next_f32(f32 max) { return next_f32() * max; }
		inline f32 next_f32(f32 min, f32 max) { return min + next_f32(max-min); }

	virtual inline f64 next_f64(f64 max) { return next_f64() * max; }
		inline f64 next_f64(f64 min, f64 max) { return min + next_f64(max-min); }

	virtual void next_bytes(u8* bytes, usize n);

	template<usize N>
	inline void next_bytes(u8 (&a)[N]) { 
		_next_bytes<N>(a);
	}
	template<usize N>
	inline void next_bytes(std::array<u8, N>& ar)
	{
		// XXX: this doesn't seem to work (aliasing issues?)
		_next_bytes<N>(&ar[0]);
	}

	template<typename T> 
	inline T next() {
		std::array<u8, sizeof(T)> arr;
		
		next_bytes(arr);

		return std::bit_cast<T>(arr);
	}
	template<typename T>
	inline T next(T min, T max);
	template<typename T>
	inline T next(T max);

	template<typename T>
	inline iterator<T> iter() { return iterator(*this); }  //TODO: An iterator that yields `next<T>()` forever.

	template<typename T>
	constexpr inline T max_for() const;
protected:
	//TODO: Should we have _min_* functions too? or just continue to use 0 as the lower bound for next_*(..max)? I think use 0...
#define MAX(T, M) constexpr inline virtual T _max_ ## T() const { return M; }
#define MAXX(n) MAX(u ## n, UINT ## n ## _MAX) MAX(i ## n, INT ## n ## _MAX)
	MAXX(8)
	MAXX(16)
	MAXX(32)
	MAXX(64)
#undef MAXX
#undef MAX
protected:
	//constexpr inline virtual i8 _max() const { return 100; } // use limits.h stuff instead.

	// Vectorised versions of `next_bytes()`. These will fall back to that if they are not overriden, but if the implementation has a more efficient way of generating 4/8 bytes of random data it should override these.
	//
	// These should produce entirely random (not bounded or weighted) results, not confined to the integer types they take.
	virtual void next_v64(u64* p, usize n);
	virtual void next_v32(u32* p, usize n);

	// Main sample function. Must return between 0..=1
	// If nothing else is overrided, this value is used for everything else.
	// It is recommended to override `next_bytes()` too however.
	virtual f64 _sample() = 0;

	f64 sample();

	private:
	template<usize N>
	inline void _next_bytes(u8* a)
	{
		using namespace _rng__util;
		constexpr const auto rem64 = divp::type<u64>(N);
		u8* ptr = a;
		if constexpr(rem64.d) {
			next_v64(reinterpret_cast<u64*>(ptr), rem64.d); ptr+= rem64.d * sizeof(u64);
		}
		constexpr const auto rem32 = divp::type<u32>(rem64.r);
		if constexpr(rem32.d) {
			next_v32(reinterpret_cast<u32*>(ptr), rem32.d); ptr+=rem32.d * sizeof(u32);
		}
		if constexpr(rem32.r) {
			next_bytes(ptr, rem32.r);
		}
	}
};
#define DEFTT(n) DEFT(i ## n) DEFT(u ## n)
#define DEFT(T) template<> inline T Random::max_for< T >() const { return _max_ ## T(); }
#define DEF \
	DEFTT(8) \
	DEFTT(16) \
	DEFTT(32) \
	DEFTT(64) \
	//DEFTT(128)
DEF
#undef DEFT
#define DEFT(T) template<> inline T Random::next< T >() { return next_ ## T(); } \
		template<> inline T Random::next< T >(T min, T max) { return next_ ## T(min, max); } \
		template<> inline T Random::next< T >(T max) { return next_ ## T(max); }
template<> inline bool Random::next<bool>() { return next_bool(); }
template<> inline f64 Random::next<f64>() { return next_f64(); }
template<> inline f32 Random::next<f32>() { return next_f32(); }
DEF
#undef DEF
#undef DEFTT
#undef DEFT

#undef CTOR_COPY
#undef CTOR_MOVE