extends RefCounted
const E = preload("res://world/owniverse/math/exact.gd")
## Quantity arithmetic without scene tree or global settings.
static var _weight_rules: Dictionary = {}
static var _output_rules: Dictionary = {}

## Parsed rules are shared read-only; the string itself is the cache key.
static func rule_weights(text: String) -> Dictionary:
    if _weight_rules.has(text): return _weight_rules[text]
    var result: Dictionary = {}
    for term in text.split("^", false):
        var pair: PackedStringArray = term.split(":")
        result[pair[0]] = E.of(pair[1])
    if _weight_rules.size() >= 128: _weight_rules.clear()
    result.make_read_only()
    _weight_rules[text] = result
    return result

static func output_terms(text: String) -> Array:
    if _output_rules.has(text): return _output_rules[text]
    var result: Array = []
    for term in text.split("&"):
        var pair: PackedStringArray = term.strip_edges().split(":")
        if pair[0].strip_edges().is_empty(): continue
        result.append({"id": pair[0].strip_edges(), "amount": E.of(pair[1]) if E.eq(pair.size(), 2) else 1})
    if _output_rules.size() >= 128: _output_rules.clear()
    result.make_read_only()
    _output_rules[text] = result
    return result

static func accumulate_whole(target: Dictionary, field: String, key: String, value: Variant) -> Variant:
    if (not target.has(field)):
        target[field] = {}
    var expected: Variant = E.add(E.maximum(0, value), target[field].get(key, 0))
    var whole: Variant = E.floor_value(expected)
    target[field][key] = E.maximum(0, E.subtract(expected, whole))
    return whole


static func allocate_whole(requests: Dictionary, limit: Variant) -> Dictionary:
    var result: Dictionary = {}
    var total: Variant = 0
    for id in requests:
        result[id] = E.floor_value(E.maximum(0, requests[id]))
        total = E.add(total, result[id])
    return _allocate_whole_with_total(requests, limit, result, total)

## Whole stocks apportioned using float shares. Both the total and every stock
## ceiling are enforced exactly after quota rounding; no quantity becomes float.
static func allocate_profiled(stocks: Dictionary, limit: Variant) -> Dictionary:
    var total: Variant = sum(stocks)
    var available: Variant = E.floor_value(E.minimum(E.maximum(0, limit), total))
    if E.eq(available, total): return stocks.duplicate()
    var ids: Array = stocks.keys()
    var shares: Array = []
    var result: Dictionary = {}
    var only: Variant = null
    var positives: int = 0
    for id in ids:
        result[id] = 0
        if E.gt(stocks[id], 0):
            only = id
            positives += 1
    if E.le(available, 0): return result
    if positives == 1:
        result[only] = available
        return result
    for id in ids: shares.append(E.ratio_approx(stocks[id], total))
    result = split_profile({}, "stock", available, ids, shares)
    var assigned: Variant = 0
    for id in ids:
        result[id] = E.minimum(stocks[id], result[id])
        assigned = E.add(assigned, result[id])
    var missing: Variant = E.subtract(available, assigned)
    for id in ids:
        if E.le(missing, 0): break
        var extra: Variant = E.minimum(missing, E.subtract(stocks[id], result[id]))
        result[id] = E.add(result[id], extra)
        missing = E.subtract(missing, extra)
    assert(E.eq(missing, 0), "Whole stock allocation must exhaust the available quota")
    return result


## Variant for callers that already maintain the sum of floored request values.
## The supplied total must equal sum(floor(max(0, requests[id]))).
static func allocate_whole_with_total(requests: Dictionary, limit: Variant, request_total: Variant) -> Dictionary:
    var result: Dictionary = {}
    for id in requests: result[id] = E.floor_value(E.maximum(0, requests[id]))
    return _allocate_whole_with_total(requests, limit, result, request_total)


static func _allocate_whole_with_total(requests: Dictionary, limit: Variant, result: Dictionary, total: Variant) -> Dictionary:
    var available: Variant = E.floor_value(E.minimum(E.maximum(0, limit), total))
    # No apportionment or rounding is necessary when every request fits.
    if E.eq(available, total): return result
    if E.le(available, 0):
        for id in result: result[id] = 0
        return result
    var only: Variant = null
    var positive_count: int = 0
    for id in result:
        if E.gt(result[id], 0):
            only = id
            positive_count += 1
    if positive_count == 1:
        result[only] = available
        return result
    var ids: Array = requests.keys()
    var rests: Array = []
    var order: Array = []
    var assigned: Variant = 0
    for i in ids.size():
        var expected: Variant = E.mul_div(result[ids[i]], available, total)
        result[ids[i]] = E.floor_value(expected)
        assigned = E.add(assigned, result[ids[i]])
        rests.append(E.subtract(expected, result[ids[i]]))
        order.append(i)
    order.sort_custom(func(a: int, b: int) -> bool: return E.gt(rests[a], rests[b]) if E.ne(rests[a], rests[b]) else str(ids[a]) < str(ids[b]))
    for i in order:
        if E.ge(assigned, available): break
        if E.lt(result[ids[i]], E.floor_value(requests[ids[i]])):
            result[ids[i]] = E.add(result[ids[i]], 1)
            assigned = E.add(assigned, 1)
    return result


static func production_step(costs: Array):
    var step = 1
    for cost in costs:
        if E.le(cost, 0): continue
        var terms: Array = E.parts(cost)
        var denominator = E.make(E.Big.divmod(terms[1], E.Big.gcd(terms[0], terms[1]))[0])
        step = E.multiply(E.divide(step, whole_gcd(step, denominator)), denominator)
    return step


static func whole_gcd(a: Variant, b: Variant) -> Variant:
    while E.gt(b, 0):
        var remainder: Variant = E.remainder(a, b)
        a = b
        b = remainder
    return a


static func normalize_distribution(values: Dictionary) -> Dictionary:
    var normalized_distribution: Dictionary = {}
    var distribution_total: Variant = 0
    for id in values:
        if E.gt(values[id], 0):
            distribution_total = E.add(distribution_total, values[id])
            normalized_distribution[id] = values[id]
    for id in normalized_distribution:
        normalized_distribution[id] = E.divide(normalized_distribution[id], distribution_total)
    return normalized_distribution


static func sum_array(values: Array) -> Variant:
    var result: Variant = 0
    for value in values: result = E.add(result, value)
    return result

static func sum(pop: Dictionary) -> Variant:
    var result: Variant = 0
    for n in pop.values(): result = E.add(result, n)
    return result


static var _split_keys: Dictionary = {}

## Split a whole population using normalized float shares. Quantizing the six
## shares to 15 decimal places matches their stored float precision, while the
## integer denominator avoids a second expensive exact normalization pass.
static func split_profile(remainders: Dictionary, key: String, count: Variant, ids: Array, shares: Array) -> Dictionary:
    var result: Dictionary = {}
    if E.le(count, 0): return result
    assert(E.eq(count, E.floor_value(count)) and ids.size() == shares.size() and not ids.is_empty())
    const SCALE: int = 1000000000000000
    var weights: Array = []
    var assigned_weight: int = 0
    for i in ids.size() - 1:
        var weight: int = clampi(roundi(clampf(float(shares[i]), 0.0, 1.0) * SCALE), 0, SCALE)
        weights.append(weight)
        assigned_weight += weight
    var last_weight: int = SCALE - assigned_weight
    if last_weight < 0:
        var excess: int = -last_weight
        for i in range(weights.size() - 1, -1, -1):
            var reduction: int = mini(excess, weights[i])
            weights[i] -= reduction
            excess -= reduction
            if excess == 0: break
        last_weight = 0
    weights.append(last_weight)
    var cached: Dictionary = _split_keys.get(key, {})
    if cached.is_empty() or cached.ids != ids:
        var keys: Array = []
        for id in ids: keys.append(key + ":" + str(id))
        if _split_keys.size() >= 512: _split_keys.clear()
        cached = {"ids": ids.duplicate(), "keys": keys}
        _split_keys[key] = cached
    var quotas: Array = E.whole_quotas(count, weights, SCALE)
    var rests: Array = []
    var order: Array = []
    var total: Variant = 0
    for i in ids.size():
        var fraction: float = quotas[i][1] + E.approx(remainders.get(cached.keys[i], 0))
        var adjustment: int = floori(fraction)
        result[ids[i]] = E.maximum(0, E.add(quotas[i][0], adjustment))
        total = E.add(total, result[ids[i]])
        rests.append(fraction - adjustment if E.ge(quotas[i][0], -adjustment) else fraction)
        order.append(i)
    order.sort_custom(func(a: int, b: int) -> bool: return rests[a] > rests[b] if rests[a] != rests[b] else str(ids[a]) < str(ids[b]))
    if E.lt(total, count):
        for i in order:
            if E.ge(total, count): break
            result[ids[i]] = E.add(result[ids[i]], 1)
            rests[i] -= 1.0
            total = E.add(total, 1)
    elif E.gt(total, count):
        order.reverse()
        for i in order:
            if E.le(total, count): break
            if E.le(result[ids[i]], 0): continue
            result[ids[i]] = E.subtract(result[ids[i]], 1)
            rests[i] += 1.0
            total = E.subtract(total, 1)
    for i in ids.size():
        remainders[cached.keys[i]] = clampf(rests[i], -1.0, 1.0)
    return result

static func split(remainders: Dictionary, key: String, count: Variant, weights: Dictionary) -> Dictionary:
    if E.le(count, 0): return {}
    return split_arrays(remainders, key, count, weights.keys(), weights.values())

static func split_arrays(remainders: Dictionary, key: String, count: Variant, ids: Array, weights: Array, population_factor: int = 0, skip_conserved_identity: bool = false) -> Dictionary:
    var result: Dictionary = {}
    if E.le(count, 0): return result
    var sum_weights: Variant = 0
    var positive_index: int = -1
    var positive_count: int = 0
    for i in weights.size():
        sum_weights = E.add(sum_weights, weights[i])
        if E.gt(weights[i], 0):
            positive_index = i
            positive_count += 1
    if E.le(sum_weights, 0): return result
    var cached: Dictionary = _split_keys.get(key, {})
    if cached.is_empty() or cached.ids != ids:
        var keys: Array = []
        for id in ids: keys.append(key + ":" + str(id))
        if _split_keys.size() >= 512: _split_keys.clear()
        cached = {"ids": ids.duplicate(), "keys": keys}
        _split_keys[key] = cached
    var remainder_keys: Array = cached.keys
    # A single branch is exact only when no saved rounding error redistributes it.
    var single: bool = positive_count == 1 and E.eq(count, E.floor_value(count))
    if single:
        for i in ids.size():
            if E.lt(weights[i], 0) or E.ne(remainders.get(remainder_keys[i], 0), 0):
                single = false
                break
    if single:
        for i in ids.size():
            result[ids[i]] = count if i == positive_index else 0
            remainders[remainder_keys[i]] = 0
        return result
    var expected: Array = []
    if population_factor > 0 and E.eq(sum_weights, E.multiply(count, population_factor)):
        # Ordinary drift: count cancels exactly from count*weight/(factor*count).
        expected = E.population_quotas(count, weights, population_factor)
    else:
        expected = E.mul_div_many(count, weights, sum_weights)
    var positives: Array = []
    var positive_sum: Variant = 0
    var whole_quotas: bool = true
    for i in ids.size():
        expected[i] = E.add(expected[i], remainders.get(remainder_keys[i], 0))
        var positive: Variant = E.maximum(0, expected[i])
        whole_quotas = whole_quotas and E.eq(positive, E.floor_value(positive))
        positives.append(positive)
        positive_sum = E.add(positive_sum, positive)
    # Fractional identity ratios still round-trip through parts/make in the
    # original algorithm. Retain them to preserve ties and saved errors exactly.
    var normalized: Array = positives
    if E.gt(positive_sum, 0) and (E.ne(positive_sum, count) or (not whole_quotas and not (skip_conserved_identity and E.eq(positive_sum, count)))):
        normalized = E.mul_div_many(count, positives, positive_sum)
    var order: Array = []
    var rests: Array = []
    var assigned: Variant = 0
    for i in ids.size():
        result[ids[i]] = E.floor_value(normalized[i])
        assigned = E.add(assigned, result[ids[i]])
        rests.append(E.subtract(normalized[i], result[ids[i]]))
        order.append(i)
    order.sort_custom(func(a: int, b: int) -> bool: return E.gt(rests[a], rests[b]) if E.ne(rests[a], rests[b]) else str(ids[a]) < str(ids[b]))
    for i in order:
        if E.ge(assigned, count): break
        result[ids[i]] = E.add(result[ids[i]], 1)
        assigned = E.add(assigned, 1)
    for i in ids.size():
        remainders[remainder_keys[i]] = E.clamp_value(E.subtract(expected[i], result[ids[i]]), E.neg(1), 1)
    return result


static func accumulate_rate(remainders: Dictionary, key: String, count: Variant, probability: Variant) -> Variant:
    # Reward rates are floats while populations are exact whole values. Keep
    # the product as quotient plus fractional carry instead of constructing an
    # intermediate Exact Value and flooring it again.
    if probability is float and is_finite(probability) and probability >= 0.0 and (count is int or (count is E.Value and count.fraction == 0.0)) and E.ge(count, 0):
        var terms: Array = E.parts(probability)
        if terms[2] > 0:
            var population: Array = E._blocks(count)[0]
            var whole_product: Array = E.Big.multiply(population, terms[0])
            var qr: Array = E.Big.divmod(whole_product, terms[1])
            var fractional: float = E._fraction_ratio(qr[1], terms[1])
            var carry: float = E.approx(remainders.get(key, 0))
            if carry >= 0.0 and carry < 1.0:
                if qr[1] == [0]:
                    remainders[key] = carry
                    return E._compose(qr[0])
                fractional += carry
                if fractional > 0.5 and absf(fractional - roundf(fractional)) <= 1e-12:
                    fractional = roundf(fractional)
                var extra: int = floori(fractional)
                remainders[key] = fractional - extra
                return E.add(E._compose(qr[0]), extra)
    var product: Variant = E.multiply(count, E.maximum(0, probability))
    var carry: float = E.approx(remainders.get(key, 0))
    if carry >= 0.0 and carry < 1.0 and E.ge(product, 0):
        if product is int:
            remainders[key] = carry
            return product
        if product is E.Value:
            var fraction: float = product.fraction + carry
            if product.fraction > 0.0 and carry > 0.0 and fraction > 0.5 and absf(fraction - roundf(fraction)) <= 1e-12:
                fraction = roundf(fraction)
            var extra: int = floori(fraction)
            remainders[key] = fraction - extra
            return E.add(E.floor_value(product), extra)
    var expected: Variant = E.add(product, remainders.get(key, 0))
    var actual: Variant = E.floor_value(expected)
    remainders[key] = E.maximum(0, E.subtract(expected, actual))
    return actual