src/wallet/coinselection.cpp - master - githubmirror/bitcoin

src/wallet/coinselection.cpp
1 067 строк51 KB
Ava Chow
Merge bitcoin/bitcoin#32150: coinselection: Optimize BnB exploration
04 июн 2026, 00:06
04 июн 2026, 00:06a5050dd
// Copyright (c) 2017-present The Bitcoin Core developers
// Distributed under the MIT software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.

#include <wallet/coinselection.h>

#include <common/system.h>
#include <consensus/amount.h>
#include <consensus/consensus.h>
#include <interfaces/chain.h>
#include <policy/feerate.h>
#include <util/check.h>
#include <util/log.h>
#include <util/moneystr.h>

#include <numeric>
#include <optional>
#include <queue>

namespace wallet {
// Common selection error across the algorithms
static util::Result<SelectionResult> ErrorMaxWeightExceeded()
{
    return util::Error{_("The inputs size exceeds the maximum weight. "
                         "Please try sending a smaller amount or manually consolidating your wallet's UTXOs")};
}

// Sort by descending (effective) value prefer lower waste on tie
struct {
    bool operator()(const OutputGroup& a, const OutputGroup& b) const
    {
        if (a.GetSelectionAmount() == b.GetSelectionAmount()) {
            // Lower waste is better when effective_values are tied
            return (a.fee - a.long_term_fee) < (b.fee - b.long_term_fee);
        }
        return a.GetSelectionAmount() > b.GetSelectionAmount();
    }
} descending;

// Sort by descending (effective) value prefer lower weight on tie
struct {
    bool operator()(const OutputGroup& a, const OutputGroup& b) const
    {
        if (a.GetSelectionAmount() == b.GetSelectionAmount()) {
            // Sort lower weight to front on tied effective_value
            return a.m_weight < b.m_weight;
        }
        return a.GetSelectionAmount() > b.GetSelectionAmount();
    }
} descending_effval_weight;

/*
 * This is the Branch and Bound Coin Selection algorithm designed by Murch. It searches for an input
 * set that can pay for the spending target and does not exceed the spending target by more than the
 * cost of creating and spending a change output. The algorithm uses a depth-first search on a binary
 * tree. In the binary tree, each node corresponds to the inclusion or the omission of a UTXO. UTXOs
 * are sorted by their effective values, tie-broken by their waste score, and the tree is explored deterministically per the inclusion
 * branch first. For each new input set candidate, the algorithm checks whether the selection is within the target range.
 * While the selection has not reached the target range, more UTXOs are included. When a selection's
 * value exceeds the target range, the complete subtree deriving from this selection prefix can be omitted.
 * At that point, the last included UTXO is deselected and the corresponding omission branch explored
 * instead starting by adding the subsequent UTXO. The search ends after the complete tree has been searched or after a limited number of tries.
 *
 * The algorithm continues to search for better solutions after one solution has been found. The best
 * solution is chosen by minimal waste score. The waste metric is defined as the cost to
 * spend the current inputs at the given fee rate minus the long term expected cost to spend the
 * inputs, plus the amount by which the selection exceeds the spending target (the "excess"):
 *
 *    excess = selected_amount - target
 *    waste = inputs × (currentFeeRate - longTermFeeRate) + excess
 *
 * Note that this means that at fee rates higher than longTermFeeRate additional inputs increase the
 * waste score, while at fee rates lower than longTermFeeRate additional inputs decrease the waste
 * score.
 *
 * The algorithm uses the following optimizations:
 * 1. Lookahead: The lookahead stores the total remaining effective value of the undecided UTXOs for
 *    every depth of the search tree. Whenever the currently selected amount plus the potential
 *    amount from the lookahead falls short of the target, we can immediately stop searching the
 *    subtree as no more input set candidates can be found in it.
 * 2. Skip clones: When two UTXOs match in weight and effective value ("are clones"), naive
 *    exploration would cause redundant work: e.g., given the UTXOs A, A', and B, where A and A' are
 *    clones, naive exploration would combine (read underscore as omission):
 *    [{}, {A}, {A, A'}, {A, A', B}, {A, _, B}, {_, A'}, {_, A', B}, {_, _, B}].
 *    In this case the input set candidates {A} and {A'} as well as {A, B} and {A', B} are
 *    equivalent. It is sufficient to explore combinations that select either both UTXOs or the
 *    first UTXO. Whenever the first UTXO is omitted, we can also skip the clone as we have already
 *    explored a set of equivalent combination as the one we could generate with the second clone.
 *    Concretely, we skip a UTXO when its predecessor is omitted and the UTXO matches the
 *    effective value and the waste of the predecessor.
 * 3. Skip similar UTXOs that are more wasteful: This search algorithm operates on the list of UTXOs
 *    sorted by effective value, tie-broken to prefer lower waste. This means that among two
 *    subsequent UTXOs with the same effective value, the second UTXO’s waste score will either be
 *    equal _or higher_ than the first UTXO’s. This allows us to apply the clone skipping idea more
 *    broadly: any combination with the second UTXO is equivalent _or worse_ than what we already
 *    combined with the first UTXO. We skip a UTXO if its predecessor is omitted and the predecessor
 *    matches in effective value.
 *
 * The Branch and Bound algorithm is described in detail in Murch's Master Thesis:
 * https://murch.one/wp-content/uploads/2016/11/erhardt2016coinselection.pdf
 *
 * @param const std::vector<OutputGroup>& utxo_pool The set of UTXO groups that we are choosing from.
 *        These UTXO groups will be sorted in descending order by effective value and the OutputGroups'
 *        values are their effective values.
 * @param const CAmount& selection_target This is the value that we want to select. It is the lower
 *        bound of the range.
 * @param const CAmount& cost_of_change This is the cost of creating and spending a change output.
 *        This plus selection_target is the upper bound of the range.
 * @param int max_selection_weight The maximum allowed weight for a selection result to be valid.
 * @returns The result of this coin selection algorithm, or std::nullopt
 */

static const size_t TOTAL_TRIES = 100000;

util::Result<SelectionResult> SelectCoinsBnB(std::vector<OutputGroup>& utxo_pool, const CAmount& selection_target, const CAmount& cost_of_change,
                                             int max_selection_weight)
{
    std::sort(utxo_pool.begin(), utxo_pool.end(), descending);
    // The sum of UTXO amounts after this UTXO index, e.g. lookahead[5] = Σ(UTXO[6+].amount)
    std::vector<CAmount> lookahead(utxo_pool.size());

    // Calculate lookahead values, and check that there are sufficient funds
    CAmount total_available = 0;
    for (int index = static_cast<int>(utxo_pool.size()) - 1; index >= 0; --index) {
        lookahead[index] = total_available;
        // UTXOs with non-positive effective value must have been filtered
        Assume(utxo_pool[index].GetSelectionAmount() > 0);
        total_available += utxo_pool[index].GetSelectionAmount();
    }

    if (total_available < selection_target) {
        // Insufficient funds
        return util::Error();
    }


    // The current selection and the best input set found so far, stored as the utxo_pool indices of the UTXOs forming them
    std::vector<size_t> curr_selection;
    std::vector<size_t> best_selection;

    // The currently selected effective amount
    CAmount curr_amount = 0;

    // The waste score of the current selection, and the best waste score so far
    CAmount curr_selection_waste = 0;
    CAmount best_waste = MAX_MONEY;

    // The weight of the currently selected input set
    int curr_weight = 0;

    // Whether the input sets generated during this search have exceeded the maximum transaction weight at any point
    bool max_tx_weight_exceeded = false;

    // Index of the next UTXO to consider in utxo_pool
    size_t next_utxo = 0;

    auto deselect_last = [&]() {
        OutputGroup& utxo = utxo_pool[curr_selection.back()];
        curr_amount -= utxo.GetSelectionAmount();
        curr_weight -= utxo.m_weight;
        curr_selection_waste -= utxo.fee - utxo.long_term_fee;
        curr_selection.pop_back();
    };

    size_t curr_try = 0;
    SelectionResult result(selection_target, SelectionAlgorithm::BNB);
    bool is_done = false;
    // We don’t have access to the feerate here, but fee to long_term_fee is as feerate to LTFRE
    bool is_feerate_high = utxo_pool.at(0).fee > utxo_pool.at(0).long_term_fee;
    while (!is_done) {
        bool should_shift{false}, should_cut{false};
        // Select `next_utxo`
        OutputGroup& utxo = utxo_pool[next_utxo];
        curr_amount += utxo.GetSelectionAmount();
        curr_weight += utxo.m_weight;
        curr_selection_waste += utxo.fee - utxo.long_term_fee;
        curr_selection.push_back(next_utxo);
        ++next_utxo;
        ++curr_try;

        // EVALUATE current selection: check for solutions and see whether we can CUT or SHIFT before EXPLORING further
        if (curr_amount + lookahead[curr_selection.back()] < selection_target) {
            // Insufficient funds with lookahead: CUT
            should_cut = true;
        } else if (curr_weight > max_selection_weight) {
            // max_weight exceeded: SHIFT
            max_tx_weight_exceeded = true;
            should_shift = true;
        } else if (curr_amount > selection_target + cost_of_change) {
            // Overshot target range: SHIFT
            should_shift = true;
        } else if (is_feerate_high && curr_selection_waste > best_waste) {
            // At high feerates adding more inputs will increase the waste score. If the current waste is already worse
            // than the best selection’s while we have insufficient funds, it is impossible for this partial selection
            // to beat the best selection by adding more inputs: SHIFT
            // At low feerates, additional inputs lower the waste score, and using this would cause us to skip exploring
            // combinations with more inputs of lower amounts.
            should_shift = true;
        } else if (curr_amount >= selection_target) {
            // Selection is within target window: potential solution
            // Adding more UTXOs only increases fees and cannot be better: SHIFT
            should_shift = true;
            // The amount exceeding the selection_target (the "excess"), would be dropped to the fees: it is waste.
            CAmount curr_excess = curr_amount - selection_target;
            CAmount curr_waste = curr_selection_waste + curr_excess;
            if (curr_waste <= best_waste) {
                // New best solution
                best_selection = curr_selection;
                best_waste = curr_waste;
            }
        }

        if (curr_try >= TOTAL_TRIES) {
            // Solution is not guaranteed to be optimal if `curr_try` hit TOTAL_TRIES
            result.SetAlgoCompleted(false);
            break;
        }

        if (next_utxo == utxo_pool.size()) {
            // Last added UTXO was end of UTXO pool, nothing left to add on inclusion or omission branch: CUT
            should_cut = true;
        }

        if (should_cut) {
            // Neither adding to the current selection nor exploring the omission branch of the last selected UTXO can
            // find any solutions. Redirect to exploring the Omission branch of the penultimate selected UTXO (i.e.
            // set `next_utxo` to one after the penultimate selected, then deselect the last two selected UTXOs)
            deselect_last();
            should_shift = true;
        }

        while (should_shift) {
            if (curr_selection.empty()) {
                // Exhausted search space before running into attempt limit
                is_done = true;
                result.SetAlgoCompleted(true);
                break;
            }
            // Set `next_utxo` to one after last selected, then deselect last selected UTXO
            next_utxo = curr_selection.back() + 1;
            deselect_last();
            should_shift = false;

            // After SHIFTing to an omission branch, the `next_utxo` might have the same effective value as the
            // UTXO we just omitted. Since lower waste is our tiebreaker on UTXOs with equal effective value for sorting, if it
            // ties on the effective value, it _must_ have the same waste (i.e. be a "clone" of the prior UTXO) or a
            // higher waste.  If so, selecting `next_utxo` would produce an equivalent or worse
            // selection as one we previously evaluated. In that case, increment `next_utxo` until we find a UTXO with a
            // differing amount.
            Assume(next_utxo < utxo_pool.size());
            while (utxo_pool[next_utxo - 1].GetSelectionAmount() == utxo_pool[next_utxo].GetSelectionAmount()) {
                if (next_utxo >= utxo_pool.size() - 1) {
                    // Reached end of UTXO pool skipping clones: SHIFT instead
                    should_shift = true;
                    break;
                }
                // Skip clone: previous UTXO is equivalent and unselected
                ++next_utxo;
            }
        }
    }

    result.SetSelectionsEvaluated(curr_try);

    if (best_selection.empty()) {
        return max_tx_weight_exceeded ? ErrorMaxWeightExceeded() : util::Error();
    }

    for (const size_t& i : best_selection) {
        result.AddInput(utxo_pool.at(i));
    }

    return result;
}


/*
 * TL;DR: Coin Grinder is a DFS-based algorithm that deterministically searches for the minimum-weight input set to fund
 * the transaction. The algorithm is similar to the Branch and Bound algorithm, but will produce a transaction _with_ a
 * change output instead of a changeless transaction.
 *
 * Full description: CoinGrinder can be thought of as a graph walking algorithm. It explores a binary tree
 * representation of the powerset of the UTXO pool. Each node in the tree represents a candidate input set. The tree’s
 * root is the empty set. Each node in the tree has two children which are formed by either adding or skipping the next
 * UTXO ("inclusion/omission branch"). Each level in the tree after the root corresponds to a decision about one UTXO in
 * the UTXO pool.
 *
 * Example:
 * We represent UTXOs as _alias=[effective_value/weight]_ and indicate omitted UTXOs with an underscore. Given a UTXO
 * pool {A=[10/2], B=[7/1], C=[5/1], D=[4/2]} sorted by descending effective value, our search tree looks as follows:
 *
 *                                       _______________________ {} ________________________
 *                                      /                                                   \
 * A=[10/2]               __________ {A} _________                                __________ {_} _________
 *                       /                        \                              /                        \
 * B=[7/1]            {AB} _                      {A_} _                      {_B} _                      {__} _
 *                  /       \                   /       \                   /       \                   /       \
 * C=[5/1]     {ABC}         {AB_}         {A_C}         {A__}         {_BC}         {_B_}         {__C}         {___}
 *              / \           / \           / \           / \           / \           / \           / \           / \
 * D=[4/2] {ABCD} {ABC_} {AB_D} {AB__} {A_CD} {A_C_} {A__D} {A___} {_BCD} {_BC_} {_B_D} {_B__} {__CD} {__C_} {___D} {____}
 *
 *
 * CoinGrinder uses a depth-first search to walk this tree. It first tries inclusion branches, then omission branches. A
 * naive exploration of a tree with four UTXOs requires visiting all 31 nodes:
 *
 *     {} {A} {AB} {ABC} {ABCD} {ABC_} {AB_} {AB_D} {AB__} {A_} {A_C} {A_CD} {A_C_} {A__} {A__D} {A___} {_} {_B} {_BC}
 *     {_BCD} {_BC_} {_B_} {_B_D} {_B__} {__} {__C} {__CD} {__C} {___} {___D} {____}
 *
 * As powersets grow exponentially with the set size, walking the entire tree would quickly get computationally
 * infeasible with growing UTXO pools. Thanks to traversing the tree in a deterministic order, we can keep track of the
 * progress of the search solely on basis of the current selection (and the best selection so far). We visit as few
 * nodes as possible by recognizing and skipping any branches that can only contain solutions worse than the best
 * solution so far. This makes CoinGrinder a branch-and-bound algorithm
 * (https://en.wikipedia.org/wiki/Branch_and_bound).
 * CoinGrinder is searching for the input set with lowest weight that can fund a transaction, so for example we can only
 * ever find a _better_ candidate input set in a node that adds a UTXO, but never in a node that skips a UTXO. After
 * visiting {A} and exploring the inclusion branch {AB} and its descendants, the candidate input set in the omission
 * branch {A_} is equivalent to the parent {A} in effective value and weight. While CoinGrinder does need to visit the
 * descendants of the omission branch {A_}, it is unnecessary to evaluate the candidate input set in the omission branch
 * itself. By skipping evaluation of all nodes on an omission branch we reduce the visited nodes to 15:
 *
 *     {A} {AB} {ABC} {ABCD} {AB_D} {A_C} {A_CD} {A__D} {_B} {_BC} {_BCD} {_B_D} {__C} {__CD} {___D}
 *
 *                                       _______________________ {} ________________________
 *                                      /                                                   \
 * A=[10/2]               __________ {A} _________                                ___________\____________
 *                       /                        \                              /                        \
 * B=[7/1]            {AB} __                    __\_____                     {_B} __                    __\_____
 *                  /        \                  /        \                  /        \                  /        \
 * C=[5/1]     {ABC}          \            {A_C}          \            {_BC}          \            {__C}          \
 *              /             /             /             /             /             /             /             /
 * D=[4/2] {ABCD}        {AB_D}        {A_CD}        {A__D}        {_BCD}        {_B_D}        {__CD}        {___D}
 *
 *
 * We refer to the move from the inclusion branch {AB} via the omission branch {A_} to its inclusion-branch child {A_C}
 * as _shifting to the omission branch_ or just _SHIFT_. (The index of the ultimate element in the candidate input set
 * shifts right by one: {AB} ⇒ {A_C}.)
 * When we reach a leaf node in the last level of the tree, shifting to the omission branch is not possible. Instead we
 * go to the omission branch of the node’s last ancestor on an inclusion branch: from {ABCD}, we go to {AB_D}. From
 * {AB_D}, we go to {A_C}. We refer to this operation as a _CUT_. (The ultimate element in
 * the input set is deselected, and the penultimate element is shifted right by one: {AB_D} ⇒ {A_C}.)
 * If a candidate input set in a node has not selected sufficient funds to build the transaction, we continue directly
 * along the next inclusion branch. We call this operation _EXPLORE_. (We go from one inclusion branch to the next
 * inclusion branch: {_B} ⇒ {_BC}.)
 * Further, any prefix that already has selected sufficient effective value to fund the transaction cannot be improved
 * by adding more UTXOs. If for example the candidate input set in {AB} is a valid solution, all potential descendant
 * solutions {ABC}, {ABCD}, and {AB_D} must have a higher weight, thus instead of exploring the descendants of {AB}, we
 * can SHIFT from {AB} to {A_C}.
 *
 * Given the above UTXO set, using a target of 11, and following these initial observations, the basic implementation of
 * CoinGrinder visits the following 10 nodes:
 *
 *     Node   [eff_val/weight]  Evaluation
 *     ---------------------------------------------------------------
 *     {A}    [10/2]            Insufficient funds: EXPLORE
 *     {AB}   [17/3]            Solution: SHIFT to omission branch
 *     {A_C}  [15/3]            Better solution: SHIFT to omission branch
 *     {A__D} [14/4]            Worse solution, shift impossible due to leaf node: CUT to omission branch of {A__D},
 *                              i.e. SHIFT to omission branch of {A}
 *     {_B}   [7/1]             Insufficient funds: EXPLORE
 *     {_BC}  [12/2]            Better solution: SHIFT to omission branch
 *     {_B_D} [11/3]            Worse solution, shift impossible due to leaf node: CUT to omission branch of {_B_D},
 *                              i.e. SHIFT to omission branch of {_B}
 *     {__C}  [5/1]             Insufficient funds: EXPLORE
 *     {__CD} [9/3]             Insufficient funds, leaf node: CUT
 *     {___D} [4/2]             Insufficient funds, leaf node, cannot CUT since only one UTXO selected: done.
 *
 *                                       _______________________ {} ________________________
 *                                      /                                                   \
 * A=[10/2]               __________ {A} _________                                ___________\____________
 *                       /                        \                              /                        \
 * B=[7/1]            {AB}                       __\_____                     {_B} __                    __\_____
 *                                              /        \                  /        \                  /        \
 * C=[5/1]                                 {A_C}          \            {_BC}          \            {__C}          \
 *                                                        /                           /             /             /
 * D=[4/2]                                           {A__D}                      {_B_D}        {__CD}        {___D}
 *
 *
 * We implement this tree walk in the following algorithm:
 * 1. Add `next_utxo`
 * 2. Evaluate candidate input set
 * 3. Determine `next_utxo` by deciding whether to
 *    a) EXPLORE: Add next inclusion branch, e.g. {_B} ⇒ {_B} + `next_uxto`: C
 *    b) SHIFT: Replace last selected UTXO by next higher index, e.g. {A_C} ⇒ {A__} + `next_utxo`: D
 *    c) CUT: deselect last selected UTXO and shift to omission branch of penultimate UTXO, e.g. {AB_D} ⇒ {A_} + `next_utxo: C
 *
 * The implementation then adds further optimizations by discovering further situations in which either the inclusion
 * branch can be skipped, or both the inclusion and omission branch can be skipped after evaluating the candidate input
 * set in the node.
 *
 * @param std::vector<OutputGroup>& utxo_pool The UTXOs that we are choosing from. These UTXOs will be sorted in
 *        descending order by effective value, with lower weight preferred as a tie-breaker. (We can think of an output
 *        group with multiple as a heavier UTXO with the combined amount here.)
 * @param const CAmount& selection_target This is the minimum amount that we need for the transaction without considering change.
 * @param const CAmount& change_target The minimum budget for creating a change output, by which we increase the selection_target.
 * @param int max_selection_weight The maximum allowed weight for a selection result to be valid.
 * @returns The result of this coin selection algorithm, or std::nullopt
 */
util::Result<SelectionResult> CoinGrinder(std::vector<OutputGroup>& utxo_pool, const CAmount& selection_target, CAmount change_target, int max_selection_weight)
{
    std::sort(utxo_pool.begin(), utxo_pool.end(), descending_effval_weight);
    // The sum of UTXO amounts after this UTXO index, e.g. lookahead[5] = Σ(UTXO[6+].amount)
    std::vector<CAmount> lookahead(utxo_pool.size());
    // The minimum UTXO weight among the remaining UTXOs after this UTXO index, e.g. min_tail_weight[5] = min(UTXO[6+].weight)
    std::vector<int> min_tail_weight(utxo_pool.size());

    // Calculate lookahead values, min_tail_weights, and check that there are sufficient funds
    CAmount total_available = 0;
    int min_group_weight = std::numeric_limits<int>::max();
    for (size_t i = 0; i < utxo_pool.size(); ++i) {
        size_t index = utxo_pool.size() - 1 - i; // Loop over every element in reverse order
        lookahead[index] = total_available;
        min_tail_weight[index] = min_group_weight;
        // UTXOs with non-positive effective value must have been filtered
        Assume(utxo_pool[index].GetSelectionAmount() > 0);
        total_available += utxo_pool[index].GetSelectionAmount();
        min_group_weight = std::min(min_group_weight, utxo_pool[index].m_weight);
    }

    const CAmount total_target = selection_target + change_target;
    if (total_available < total_target) {
        // Insufficient funds
        return util::Error();
    }

    // The current selection and the best input set found so far, stored as the utxo_pool indices of the UTXOs forming them
    std::vector<size_t> curr_selection;
    std::vector<size_t> best_selection;

    // The currently selected effective amount, and the effective amount of the best selection so far
    CAmount curr_amount = 0;
    CAmount best_selection_amount = MAX_MONEY;

    // The weight of the currently selected input set, and the weight of the best selection
    int curr_weight = 0;
    int best_selection_weight = max_selection_weight; // Tie is fine, because we prefer lower selection amount

    // Whether the input sets generated during this search have exceeded the maximum transaction weight at any point
    bool max_tx_weight_exceeded = false;

    // Index of the next UTXO to consider in utxo_pool
    size_t next_utxo = 0;

    /*
     * You can think of the current selection as a vector of booleans that has decided inclusion or exclusion of all
     * UTXOs before `next_utxo`. When we consider the next UTXO, we extend this hypothetical boolean vector either with
     * a true value if the UTXO is included or a false value if it is omitted. The equivalent state is stored more
     * compactly as the list of indices of the included UTXOs and the `next_utxo` index.
     *
     * We can never find a new solution by deselecting a UTXO, because we then revisit a previously evaluated
     * selection. Therefore, we only need to check whether we found a new solution _after adding_ a new UTXO.
     *
     * Each iteration of CoinGrinder starts by selecting the `next_utxo` and evaluating the current selection. We
     * use three state transitions to progress from the current selection to the next promising selection:
     *
     * - EXPLORE inclusion branch: We do not have sufficient funds, yet. Add `next_utxo` to the current selection, then
     *                             nominate the direct successor of the just selected UTXO as our `next_utxo` for the
     *                             following iteration.
     *
     *                             Example:
     *                                 Current Selection: {0, 5, 7}
     *                                 Evaluation: EXPLORE, next_utxo: 8
     *                                 Next Selection: {0, 5, 7, 8}
     *
     * - SHIFT to omission branch: Adding more UTXOs to the current selection cannot produce a solution that is better
     *                             than the current best, e.g. the current selection weight exceeds the max weight or
     *                             the current selection amount is equal to or greater than the target.
     *                             We designate our `next_utxo` the one after the tail of our current selection, then
     *                             deselect the tail of our current selection.
     *
     *                             Example:
     *                                 Current Selection: {0, 5, 7}
     *                                 Evaluation: SHIFT, next_utxo: 8, omit last selected: {0, 5}
     *                                 Next Selection: {0, 5, 8}
     *
     * - CUT entire subtree:       We have exhausted the inclusion branch for the penultimately selected UTXO, both the
     *                             inclusion and the omission branch of the current prefix are barren. E.g. we have
     *                             reached the end of the UTXO pool, so neither further EXPLORING nor SHIFTING can find
     *                             any solutions. We designate our `next_utxo` the one after our penultimate selected,
     *                             then deselect both the last and penultimate selected.
     *
     *                             Example:
     *                                 Current Selection: {0, 5, 7}
     *                                 Evaluation: CUT, next_utxo: 6, omit two last selected: {0}
     *                                 Next Selection: {0, 6}
     */
    auto deselect_last = [&]() {
        OutputGroup& utxo = utxo_pool[curr_selection.back()];
        curr_amount -= utxo.GetSelectionAmount();
        curr_weight -= utxo.m_weight;
        curr_selection.pop_back();
    };

    SelectionResult result(selection_target, SelectionAlgorithm::CG);
    bool is_done = false;
    size_t curr_try = 0;
    while (!is_done) {
        bool should_shift{false}, should_cut{false};
        // Select `next_utxo`
        OutputGroup& utxo = utxo_pool[next_utxo];
        curr_amount += utxo.GetSelectionAmount();
        curr_weight += utxo.m_weight;
        curr_selection.push_back(next_utxo);
        ++next_utxo;
        ++curr_try;

        // EVALUATE current selection: check for solutions and see whether we can CUT or SHIFT before EXPLORING further
        auto curr_tail = curr_selection.back();
        if (curr_amount + lookahead[curr_tail] < total_target) {
            // Insufficient funds with lookahead: CUT
            should_cut = true;
        } else if (curr_weight > best_selection_weight) {
            // best_selection_weight is initialized to max_selection_weight
            if (curr_weight > max_selection_weight) max_tx_weight_exceeded = true;
            // Worse weight than best solution. More UTXOs only increase weight:
            // CUT if last selected group had minimal weight, else SHIFT
            if (utxo_pool[curr_tail].m_weight <= min_tail_weight[curr_tail]) {
                should_cut = true;
            } else {
                should_shift  = true;
            }
        } else if (curr_amount >= total_target) {
            // Success, adding more weight cannot be better: SHIFT
            should_shift  = true;
            if (curr_weight < best_selection_weight || (curr_weight == best_selection_weight && curr_amount < best_selection_amount)) {
                // New lowest weight, or same weight with fewer funds tied up
                best_selection = curr_selection;
                best_selection_weight = curr_weight;
                best_selection_amount = curr_amount;
            }
        } else if (!best_selection.empty() && curr_weight + int64_t{min_tail_weight[curr_tail]} * ((total_target - curr_amount + utxo_pool[curr_tail].GetSelectionAmount() - 1) / utxo_pool[curr_tail].GetSelectionAmount()) > best_selection_weight) {
            // Compare minimal tail weight and last selected amount with the amount missing to gauge whether a better weight is still possible.
            if (utxo_pool[curr_tail].m_weight <= min_tail_weight[curr_tail]) {
                should_cut = true;
            } else {
                should_shift = true;
            }
        }

        if (curr_try >= TOTAL_TRIES) {
            // Solution is not guaranteed to be optimal if `curr_try` hit TOTAL_TRIES
            result.SetAlgoCompleted(false);
            break;
        }

        if (next_utxo == utxo_pool.size()) {
            // Last added UTXO was end of UTXO pool, nothing left to add on inclusion or omission branch: CUT
            should_cut = true;
        }

        if (should_cut) {
            // Neither adding to the current selection nor exploring the omission branch of the last selected UTXO can
            // find any solutions. Redirect to exploring the Omission branch of the penultimate selected UTXO (i.e.
            // set `next_utxo` to one after the penultimate selected, then deselect the last two selected UTXOs)
            deselect_last();
            should_shift  = true;
        }

        while (should_shift) {
            // Set `next_utxo` to one after last selected, then deselect last selected UTXO
            if (curr_selection.empty()) {
                // Exhausted search space before running into attempt limit
                is_done = true;
                result.SetAlgoCompleted(true);
                break;
            }
            next_utxo = curr_selection.back() + 1;
            deselect_last();
            should_shift  = false;

            // After SHIFTing to an omission branch, the `next_utxo` might have the same effective value as the UTXO we
            // just omitted. Since lower weight is our tiebreaker on UTXOs with equal effective value for sorting, if it
            // ties on the effective value, it _must_ have the same weight (i.e. be a "clone" of the prior UTXO) or a
            // higher weight. If so, selecting `next_utxo` would produce an equivalent or worse selection as one we
            // previously evaluated. In that case, increment `next_utxo` until we find a UTXO with a differing amount.
            while (utxo_pool[next_utxo - 1].GetSelectionAmount() == utxo_pool[next_utxo].GetSelectionAmount()) {
                if (next_utxo >= utxo_pool.size() - 1) {
                    // Reached end of UTXO pool skipping clones: SHIFT instead
                    should_shift = true;
                    break;
                }
                // Skip clone: previous UTXO is equivalent and unselected
                ++next_utxo;
            }
        }
    }

    result.SetSelectionsEvaluated(curr_try);

    if (best_selection.empty()) {
        return max_tx_weight_exceeded ? ErrorMaxWeightExceeded() : util::Error();
    }

    for (const size_t& i : best_selection) {
        result.AddInput(utxo_pool[i]);
    }

    return result;
}

class MinOutputGroupComparator
{
public:
    int operator() (const OutputGroup& group1, const OutputGroup& group2) const
    {
        return descending_effval_weight(group1, group2);
    }
};

util::Result<SelectionResult> SelectCoinsSRD(const std::vector<OutputGroup>& utxo_pool, CAmount target_value, CAmount change_fee, FastRandomContext& rng,
                                             int max_selection_weight)
{
    SelectionResult result(target_value, SelectionAlgorithm::SRD);
    std::priority_queue<OutputGroup, std::vector<OutputGroup>, MinOutputGroupComparator> heap;

    // Include change for SRD as we want to avoid making really small change if the selection just
    // barely meets the target. Just use the lower bound change target instead of the randomly
    // generated one, since SRD will result in a random change amount anyway; avoid making the
    // target needlessly large.
    target_value += CHANGE_LOWER + change_fee;

    std::vector<size_t> indexes;
    indexes.resize(utxo_pool.size());
    std::iota(indexes.begin(), indexes.end(), 0);
    std::shuffle(indexes.begin(), indexes.end(), rng);

    CAmount selected_eff_value = 0;
    int weight = 0;
    bool max_tx_weight_exceeded = false;
    for (const size_t i : indexes) {
        const OutputGroup& group = utxo_pool.at(i);
        Assume(group.GetSelectionAmount() > 0);

        // Add group to selection
        heap.push(group);
        selected_eff_value += group.GetSelectionAmount();
        weight += group.m_weight;

        // If the selection weight exceeds the maximum allowed size, remove the least valuable inputs until we
        // are below max weight.
        if (weight > max_selection_weight) {
            max_tx_weight_exceeded = true; // mark it in case we don't find any useful result.
            do {
                const OutputGroup& to_remove_group = heap.top();
                selected_eff_value -= to_remove_group.GetSelectionAmount();
                weight -= to_remove_group.m_weight;
                heap.pop();
            } while (!heap.empty() && weight > max_selection_weight);
        }

        // Now check if we are above the target
        if (selected_eff_value >= target_value) {
            // Result found, add it.
            while (!heap.empty()) {
                result.AddInput(heap.top());
                heap.pop();
            }
            return result;
        }
    }
    return max_tx_weight_exceeded ? ErrorMaxWeightExceeded() : util::Error();
}

/** Find a subset of the OutputGroups that is at least as large as, but as close as possible to, the
 * target amount; solve subset sum.
 * @param[in]   groups          OutputGroups to choose from, sorted by value in descending order.
 * @param[in]   nTotalLower     Total (effective) value of the UTXOs in groups.
 * @param[in]   nTargetValue    Subset sum target, not including change.
 * @param[out]  vfBest          Boolean vector representing the subset chosen that is closest to
 *                              nTargetValue, with indices corresponding to groups. If the ith
 *                              entry is true, that means the ith group in groups was selected.
 * @param[out]  nBest           Total amount of subset chosen that is closest to nTargetValue.
 * @param[in]   max_selection_weight  The maximum allowed weight for a selection result to be valid.
 * @param[in]   iterations      Maximum number of tries.
 */
static void ApproximateBestSubset(FastRandomContext& insecure_rand, const std::vector<OutputGroup>& groups,
                                  const CAmount& nTotalLower, const CAmount& nTargetValue,
                                  std::vector<char>& vfBest, CAmount& nBest, int max_selection_weight, int iterations = 1000)
{
    std::vector<char> vfIncluded;

    // Worst case "best" approximation is just all of the groups.
    vfBest.assign(groups.size(), true);
    nBest = nTotalLower;

    for (int nRep = 0; nRep < iterations && nBest != nTargetValue; nRep++)
    {
        vfIncluded.assign(groups.size(), false);
        CAmount nTotal = 0;
        int selected_coins_weight{0};
        bool fReachedTarget = false;
        for (int nPass = 0; nPass < 2 && !fReachedTarget; nPass++)
        {
            for (unsigned int i = 0; i < groups.size(); i++)
            {
                //The solver here uses a randomized algorithm,
                //the randomness serves no real security purpose but is just
                //needed to prevent degenerate behavior and it is important
                //that the rng is fast. We do not use a constant random sequence,
                //because there may be some privacy improvement by making
                //the selection random.
                if (nPass == 0 ? insecure_rand.randbool() : !vfIncluded[i])
                {
                    nTotal += groups[i].GetSelectionAmount();
                    selected_coins_weight += groups[i].m_weight;
                    vfIncluded[i] = true;
                    if (nTotal >= nTargetValue && selected_coins_weight <= max_selection_weight) {
                        fReachedTarget = true;
                        // If the total is between nTargetValue and nBest, it's our new best
                        // approximation.
                        if (nTotal < nBest)
                        {
                            nBest = nTotal;
                            vfBest = vfIncluded;
                        }
                        nTotal -= groups[i].GetSelectionAmount();
                        selected_coins_weight -= groups[i].m_weight;
                        vfIncluded[i] = false;
                    }
                }
            }
        }
    }
}

util::Result<SelectionResult> KnapsackSolver(std::vector<OutputGroup>& groups, const CAmount& nTargetValue,
                                             CAmount change_target, FastRandomContext& rng, int max_selection_weight)
{
    SelectionResult result(nTargetValue, SelectionAlgorithm::KNAPSACK);

    bool max_weight_exceeded{false};
    // List of values less than target
    std::optional<OutputGroup> lowest_larger;
    // Groups with selection amount smaller than the target and any change we might produce.
    // Don't include groups larger than this, because they will only cause us to overshoot.
    std::vector<OutputGroup> applicable_groups;
    CAmount nTotalLower = 0;

    std::shuffle(groups.begin(), groups.end(), rng);

    for (const OutputGroup& group : groups) {
        if (group.m_weight > max_selection_weight) {
            max_weight_exceeded = true;
            continue;
        }
        if (group.GetSelectionAmount() == nTargetValue) {
            result.AddInput(group);
            return result;
        } else if (group.GetSelectionAmount() < nTargetValue + change_target) {
            applicable_groups.push_back(group);
            nTotalLower += group.GetSelectionAmount();
        } else if (!lowest_larger || group.GetSelectionAmount() < lowest_larger->GetSelectionAmount()) {
            lowest_larger = group;
        }
    }

    if (nTotalLower == nTargetValue) {
        for (const auto& group : applicable_groups) {
            result.AddInput(group);
        }
        if (result.GetWeight() <= max_selection_weight) return result;
        else max_weight_exceeded = true;

        // Try something else
        result.Clear();
    }

    if (nTotalLower < nTargetValue) {
        if (!lowest_larger) {
            if (max_weight_exceeded) return ErrorMaxWeightExceeded();
            return util::Error();
        }
        result.AddInput(*lowest_larger);
        return result;
    }

    // Solve subset sum by stochastic approximation
    std::sort(applicable_groups.begin(), applicable_groups.end(), descending);
    std::vector<char> vfBest;
    CAmount nBest;

    ApproximateBestSubset(rng, applicable_groups, nTotalLower, nTargetValue, vfBest, nBest, max_selection_weight);
    if (nBest != nTargetValue && nTotalLower >= nTargetValue + change_target) {
        ApproximateBestSubset(rng, applicable_groups, nTotalLower, nTargetValue + change_target, vfBest, nBest, max_selection_weight);
    }

    // If we have a bigger coin and (either the stochastic approximation didn't find a good solution,
    //                                   or the next bigger coin is closer), return the bigger coin
    if (lowest_larger &&
        ((nBest != nTargetValue && nBest < nTargetValue + change_target) || lowest_larger->GetSelectionAmount() <= nBest)) {
        result.AddInput(*lowest_larger);
    } else {
        for (unsigned int i = 0; i < applicable_groups.size(); i++) {
            if (vfBest[i]) {
                result.AddInput(applicable_groups[i]);
            }
        }

        // If the result exceeds the maximum allowed size, return closest UTXO above the target
        if (result.GetWeight() > max_selection_weight) {
            // No coin above target, nothing to do.
            if (!lowest_larger) return ErrorMaxWeightExceeded();

            // Return closest UTXO above target
            result.Clear();
            result.AddInput(*lowest_larger);
        }

        if (util::log::ShouldDebugLog(BCLog::SELECTCOINS)) {
            std::string log_message{"Coin selection best subset: "};
            for (unsigned int i = 0; i < applicable_groups.size(); i++) {
                if (vfBest[i]) {
                    log_message += strprintf("%s ", FormatMoney(applicable_groups[i].m_value));
                }
            }
            LogDebug(BCLog::SELECTCOINS, "%stotal %s\n", log_message, FormatMoney(nBest));
        }
    }
    Assume(result.GetWeight() <= max_selection_weight);
    return result;
}

/******************************************************************************

 OutputGroup

 ******************************************************************************/

void OutputGroup::Insert(const std::shared_ptr<COutput>& output, size_t ancestors, size_t cluster_count) {
    m_outputs.push_back(output);
    auto& coin = *m_outputs.back();

    fee += coin.GetFee();

    coin.long_term_fee = coin.input_bytes < 0 ? 0 : m_long_term_feerate.GetFee(coin.input_bytes);
    long_term_fee += coin.long_term_fee;

    effective_value += coin.GetEffectiveValue();

    m_from_me &= coin.from_me;
    m_value += coin.txout.nValue;
    m_depth = std::min(m_depth, coin.depth);
    // ancestors here express the number of ancestors the new coin will end up having, which is
    // the sum, rather than the max; this will overestimate in the cases where multiple inputs
    // have common ancestors
    m_ancestors += ancestors;
    // This is the maximum cluster count among all outputs. If these outputs are from distinct clusters but spent in the
    // same transaction, their clusters will be merged, potentially exceeding the mempool's max cluster count.
    m_max_cluster_count = std::max(m_max_cluster_count, cluster_count);

    if (output->input_bytes > 0) {
        m_weight += output->input_bytes * WITNESS_SCALE_FACTOR;
    }
}

bool OutputGroup::EligibleForSpending(const CoinEligibilityFilter& eligibility_filter) const
{
    return m_depth >= (m_from_me ? eligibility_filter.conf_mine : eligibility_filter.conf_theirs)
        && m_ancestors <= eligibility_filter.max_ancestors
        && m_max_cluster_count <= eligibility_filter.max_cluster_count;
}

CAmount OutputGroup::GetSelectionAmount() const
{
    return m_subtract_fee_outputs ? m_value : effective_value;
}

void OutputGroupTypeMap::Push(const OutputGroup& group, OutputType type, bool insert_positive, bool insert_mixed)
{
    if (group.m_outputs.empty()) return;

    Groups& groups = groups_by_type[type];
    if (insert_positive && group.GetSelectionAmount() > 0) {
        groups.positive_group.emplace_back(group);
        all_groups.positive_group.emplace_back(group);
    }
    if (insert_mixed) {
        groups.mixed_group.emplace_back(group);
        all_groups.mixed_group.emplace_back(group);
    }
}

CAmount GenerateChangeTarget(const CAmount payment_value, const CAmount change_fee, FastRandomContext& rng)
{
    if (payment_value <= CHANGE_LOWER / 2) {
        return change_fee + CHANGE_LOWER;
    } else {
        // random value between 50ksat and min (payment_value * 2, 1milsat)
        const auto upper_bound = std::min(payment_value * 2, CHANGE_UPPER);
        return change_fee + rng.randrange(upper_bound - CHANGE_LOWER) + CHANGE_LOWER;
    }
}

void SelectionResult::SetBumpFeeDiscount(const CAmount discount)
{
    // Overlapping ancestry can only lower the fees, not increase them
    assert (discount >= 0);
    bump_fee_group_discount = discount;
}

void SelectionResult::RecalculateWaste(const CAmount min_viable_change, const CAmount change_cost, const CAmount change_fee)
{
    // This function should not be called with empty inputs as that would mean the selection failed
    assert(!m_selected_inputs.empty());

    // Always consider the cost of spending an input now vs in the future.
    CAmount waste = 0;
    for (const auto& coin_ptr : m_selected_inputs) {
        const COutput& coin = *coin_ptr;
        waste += coin.GetFee() - coin.long_term_fee;
    }
    // Bump fee of whole selection may diverge from sum of individual bump fees
    waste -= bump_fee_group_discount;

    if (GetChange(min_viable_change, change_fee)) {
        // if we have a minimum viable amount after deducting fees, account for
        // cost of creating and spending change
        waste += change_cost;
    } else {
        // When we are not making change (GetChange(…) == 0), consider the excess we are throwing away to fees
        CAmount selected_effective_value = m_use_effective ? GetSelectedEffectiveValue() : GetSelectedValue();
        assert(selected_effective_value >= m_target);
        waste += selected_effective_value - m_target;
    }

    m_waste = waste;
}

void SelectionResult::SetAlgoCompleted(bool algo_completed)
{
    m_algo_completed = algo_completed;
}

bool SelectionResult::GetAlgoCompleted() const
{
    return m_algo_completed;
}

void SelectionResult::SetSelectionsEvaluated(size_t attempts)
{
    m_selections_evaluated = attempts;
}

size_t SelectionResult::GetSelectionsEvaluated() const
{
    return m_selections_evaluated;
}

CAmount SelectionResult::GetWaste() const
{
    return *Assert(m_waste);
}

CAmount SelectionResult::GetSelectedValue() const
{
    return std::accumulate(m_selected_inputs.cbegin(), m_selected_inputs.cend(), CAmount{0}, [](CAmount sum, const auto& coin) { return sum + coin->txout.nValue; });
}

CAmount SelectionResult::GetSelectedEffectiveValue() const
{
    return std::accumulate(m_selected_inputs.cbegin(), m_selected_inputs.cend(), CAmount{0}, [](CAmount sum, const auto& coin) { return sum + coin->GetEffectiveValue(); }) + bump_fee_group_discount;
}

CAmount SelectionResult::GetTotalBumpFees() const
{
    return std::accumulate(m_selected_inputs.cbegin(), m_selected_inputs.cend(), CAmount{0}, [](CAmount sum, const auto& coin) { return sum + coin->ancestor_bump_fees; }) - bump_fee_group_discount;
}

void SelectionResult::Clear()
{
    m_selected_inputs.clear();
    m_waste.reset();
    m_weight = 0;
}

void SelectionResult::AddInput(const OutputGroup& group)
{
    // As it can fail, combine inputs first
    InsertInputs(group.m_outputs);
    m_use_effective = !group.m_subtract_fee_outputs;

    m_weight += group.m_weight;
}

void SelectionResult::AddInputs(const OutputSet& inputs, bool subtract_fee_outputs)
{
    // As it can fail, combine inputs first
    InsertInputs(inputs);
    m_use_effective = !subtract_fee_outputs;

    m_weight += std::accumulate(inputs.cbegin(), inputs.cend(), 0, [](int sum, const auto& coin) {
        return sum + std::max(coin->input_bytes, 0) * WITNESS_SCALE_FACTOR;
    });
}

void SelectionResult::Merge(const SelectionResult& other)
{
    // As it can fail, combine inputs first
    InsertInputs(other.m_selected_inputs);

    m_target += other.m_target;
    m_use_effective |= other.m_use_effective;
    if (m_algo == SelectionAlgorithm::MANUAL) {
        m_algo = other.m_algo;
    }

    m_weight += other.m_weight;
}

const OutputSet& SelectionResult::GetInputSet() const
{
    return m_selected_inputs;
}

std::vector<std::shared_ptr<COutput>> SelectionResult::GetShuffledInputVector() const
{
    std::vector<std::shared_ptr<COutput>> coins(m_selected_inputs.begin(), m_selected_inputs.end());
    std::shuffle(coins.begin(), coins.end(), FastRandomContext());
    return coins;
}

bool SelectionResult::operator<(SelectionResult other) const
{
    Assert(m_waste.has_value());
    Assert(other.m_waste.has_value());
    // As this operator is only used in std::min_element, we want the result that has more inputs when waste are equal.
    return *m_waste < *other.m_waste || (*m_waste == *other.m_waste && m_selected_inputs.size() > other.m_selected_inputs.size());
}

std::string COutput::ToString() const
{
    return strprintf("COutput(%s, %d, %d) [%s]", outpoint.hash.ToString(), outpoint.n, depth, FormatMoney(txout.nValue));
}

std::string GetAlgorithmName(const SelectionAlgorithm algo)
{
    switch (algo)
    {
    case SelectionAlgorithm::BNB: return "bnb";
    case SelectionAlgorithm::KNAPSACK: return "knapsack";
    case SelectionAlgorithm::SRD: return "srd";
    case SelectionAlgorithm::CG: return "cg";
    case SelectionAlgorithm::MANUAL: return "manual";
    } // no default case, so the compiler can warn about missing cases
    assert(false);
}

CAmount SelectionResult::GetChange(const CAmount min_viable_change, const CAmount change_fee) const
{
    // change = SUM(inputs) - SUM(outputs) - fees
    // 1) With SFFO we don't pay any fees
    // 2) Otherwise we pay all the fees:
    //  - input fees are covered by GetSelectedEffectiveValue()
    //  - non_input_fee is included in m_target
    //  - change_fee
    const CAmount change = m_use_effective
                           ? GetSelectedEffectiveValue() - m_target - change_fee
                           : GetSelectedValue() - m_target;

    if (change < min_viable_change) {
        return 0;
    }

    return change;
}

} // namespace wallet