cluster_linearize.h

Go to the documentation of this file.

// Copyright (c) The Bitcoin Core developers
// Distributed under the MIT software license, see the accompanying
// file COPYING or http://www.opensource.org/licenses/mit-license.php.

#ifndef BITCOIN_TEST_UTIL_CLUSTER_LINEARIZE_H
#define BITCOIN_TEST_UTIL_CLUSTER_LINEARIZE_H

#include <cluster_linearize.h>
#include <serialize.h>
#include <span.h>
#include <streams.h>
#include <util/bitset.h>
#include <util/feefrac.h>

#include <cstdint>
#include <numeric>
#include <utility>
#include <vector>

namespace {

using namespace cluster_linearize;

using TestBitSet = BitSet<32>;

struct DepGraphFormatter
{
    [[maybe_unused]] static uint64_t SignedToUnsigned(int64_t x) noexcept
    {
        if (x < 0) {
            return 2 * uint64_t(-(x + 1)) + 1;
        } else {
            return 2 * uint64_t(x);
        }
    }

    [[maybe_unused]] static int64_t UnsignedToSigned(uint64_t x) noexcept
    {
        if (x & 1) {
            return -int64_t(x / 2) - 1;
        } else {
            return int64_t(x / 2);
        }
    }

    template <typename Stream, typename SetType>
    static void Ser(Stream& s, const DepGraph<SetType>& depgraph)
    {
        std::vector<DepGraphIndex> topo_order;
        topo_order.reserve(depgraph.TxCount());
        for (auto i : depgraph.Positions()) topo_order.push_back(i);
        std::sort(topo_order.begin(), topo_order.end(), [&](DepGraphIndex a, DepGraphIndex b) {
            auto anc_a = depgraph.Ancestors(a).Count(), anc_b = depgraph.Ancestors(b).Count();
            if (anc_a != anc_b) return anc_a < anc_b;
            return a < b;
        });

        SetType done;

        // Loop over the transactions in topological order.
        for (DepGraphIndex topo_idx = 0; topo_idx < topo_order.size(); ++topo_idx) {
            DepGraphIndex idx = topo_order[topo_idx];
            // Write size, which must be larger than 0.
            s << VARINT_MODE(depgraph.FeeRate(idx).size, VarIntMode::NONNEGATIVE_SIGNED);
            // Write fee, encoded as an unsigned varint (odd=negative, even=non-negative).
            s << VARINT(SignedToUnsigned(depgraph.FeeRate(idx).fee));
            // Write dependency information.
            SetType written_parents;
            uint64_t diff = 0; 
            for (DepGraphIndex dep_dist = 0; dep_dist < topo_idx; ++dep_dist) {
                DepGraphIndex dep_idx = topo_order[topo_idx - 1 - dep_dist];
                // Ignore transactions which are already known to be ancestors.
                if (depgraph.Descendants(dep_idx).Overlaps(written_parents)) continue;
                if (depgraph.Ancestors(idx)[dep_idx]) {
                    // When an actual parent is encountered, encode how many non-parents were skipped
                    // before it.
                    s << VARINT(diff);
                    diff = 0;
                    written_parents.Set(dep_idx);
                } else {
                    // When a non-parent is encountered, increment the skip counter.
                    ++diff;
                }
            }
            // Write position information.
            auto add_holes = SetType::Fill(idx) - done - depgraph.Positions();
            if (add_holes.None()) {
                // The new transaction is to be inserted N positions back from the end of the
                // cluster. Emit N to indicate that that many insertion choices are skipped.
                auto skips = (done - SetType::Fill(idx)).Count();
                s << VARINT(diff + skips);
            } else {
                // The new transaction is to be appended at the end of the cluster, after N holes.
                // Emit current_cluster_size + N, to indicate all insertion choices are skipped,
                // plus N possibilities for the number of holes.
                s << VARINT(diff + done.Count() + add_holes.Count());
                done |= add_holes;
            }
            done.Set(idx);
        }

        // Output a final 0 to denote the end of the graph.
        s << uint8_t{0};
    }

    template <typename Stream, typename SetType>
    void Unser(Stream& s, DepGraph<SetType>& depgraph)
    {
        DepGraph<SetType> topo_depgraph;
        std::vector<DepGraphIndex> reordering;
        DepGraphIndex total_size{0};

        // Read transactions in topological order.
        while (true) {
            FeeFrac new_feerate; 
            SetType new_ancestors; 
            uint64_t diff{0}; 
            bool read_error{false};
            try {
                // Read size. Size 0 signifies the end of the DepGraph.
                int32_t size;
                s >> VARINT_MODE(size, VarIntMode::NONNEGATIVE_SIGNED);
                size &= 0x3FFFFF; // Enough for size up to 4M.
                static_assert(0x3FFFFF >= 4000000);
                if (size == 0 || topo_depgraph.TxCount() == SetType::Size()) break;
                // Read fee, encoded as an unsigned varint (odd=negative, even=non-negative).
                uint64_t coded_fee;
                s >> VARINT(coded_fee);
                coded_fee &= 0xFFFFFFFFFFFFF; // Enough for fee between -21M...21M BTC.
                static_assert(0xFFFFFFFFFFFFF > uint64_t{2} * 21000000 * 100000000);
                new_feerate = {UnsignedToSigned(coded_fee), size};
                // Read dependency information.
                auto topo_idx = reordering.size();
                s >> VARINT(diff);
                for (DepGraphIndex dep_dist = 0; dep_dist < topo_idx; ++dep_dist) {
                    DepGraphIndex dep_topo_idx = topo_idx - 1 - dep_dist;
                    // Ignore transactions which are already known ancestors of topo_idx.
                    if (new_ancestors[dep_topo_idx]) continue;
                    if (diff == 0) {
                        // When the skip counter has reached 0, add an actual dependency.
                        new_ancestors |= topo_depgraph.Ancestors(dep_topo_idx);
                        // And read the number of skips after it.
                        s >> VARINT(diff);
                    } else {
                        // Otherwise, dep_topo_idx is not a parent. Decrement and continue.
                        --diff;
                    }
                }
            } catch (const std::ios_base::failure&) {
                // Continue even if a read error was encountered.
                read_error = true;
            }
            // Construct a new transaction whenever we made it past the new_feerate construction.
            if (new_feerate.IsEmpty()) break;
            assert(reordering.size() < SetType::Size());
            auto topo_idx = topo_depgraph.AddTransaction(new_feerate);
            topo_depgraph.AddDependencies(new_ancestors, topo_idx);
            if (total_size < SetType::Size()) {
                // Normal case.
                diff %= SetType::Size();
                if (diff <= total_size) {
                    // Insert the new transaction at distance diff back from the end.
                    for (auto& pos : reordering) {
                        pos += (pos >= total_size - diff);
                    }
                    reordering.push_back(total_size++ - diff);
                } else {
                    // Append diff - total_size holes at the end, plus the new transaction.
                    total_size = diff;
                    reordering.push_back(total_size++);
                }
            } else {
                // In case total_size == SetType::Size, it is not possible to insert the new
                // transaction without exceeding SetType's size. Instead, interpret diff as an
                // index into the holes, and overwrite a position there. This branch is never used
                // when deserializing the output of the serializer, but gives meaning to otherwise
                // invalid input.
                diff %= (SetType::Size() - reordering.size());
                SetType holes = SetType::Fill(SetType::Size());
                for (auto pos : reordering) holes.Reset(pos);
                for (auto pos : holes) {
                    if (diff == 0) {
                        reordering.push_back(pos);
                        break;
                    }
                    --diff;
                }
            }
            // Stop if a read error was encountered during deserialization.
            if (read_error) break;
        }

        // Construct the original cluster order depgraph.
        depgraph = DepGraph(topo_depgraph, reordering, total_size);
    }
};

template<typename SetType>
void SanityCheck(const DepGraph<SetType>& depgraph)
{
    // Verify Positions and PositionRange consistency.
    DepGraphIndex num_positions{0};
    DepGraphIndex position_range{0};
    for (DepGraphIndex i : depgraph.Positions()) {
        ++num_positions;
        position_range = i + 1;
    }
    assert(num_positions == depgraph.TxCount());
    assert(position_range == depgraph.PositionRange());
    assert(position_range >= num_positions);
    assert(position_range <= SetType::Size());
    // Consistency check between ancestors internally.
    for (DepGraphIndex i : depgraph.Positions()) {
        // Transactions include themselves as ancestors.
        assert(depgraph.Ancestors(i)[i]);
        // If a is an ancestor of b, then b's ancestors must include all of a's ancestors.
        for (auto a : depgraph.Ancestors(i)) {
            assert(depgraph.Ancestors(i).IsSupersetOf(depgraph.Ancestors(a)));
        }
    }
    // Consistency check between ancestors and descendants.
    for (DepGraphIndex i : depgraph.Positions()) {
        for (DepGraphIndex j : depgraph.Positions()) {
            assert(depgraph.Ancestors(i)[j] == depgraph.Descendants(j)[i]);
        }
        // No transaction is a parent or child of itself.
        auto parents = depgraph.GetReducedParents(i);
        auto children = depgraph.GetReducedChildren(i);
        assert(!parents[i]);
        assert(!children[i]);
        // Parents of a transaction do not have ancestors inside those parents (except itself).
        // Note that even the transaction itself may be missing (if it is part of a cycle).
        for (auto parent : parents) {
            assert((depgraph.Ancestors(parent) & parents).IsSubsetOf(SetType::Singleton(parent)));
        }
        // Similar for children and descendants.
        for (auto child : children) {
            assert((depgraph.Descendants(child) & children).IsSubsetOf(SetType::Singleton(child)));
        }
    }
    if (depgraph.IsAcyclic()) {
        // If DepGraph is acyclic, serialize + deserialize must roundtrip.
        std::vector<unsigned char> ser;
        VectorWriter writer(ser, 0);
        writer << Using<DepGraphFormatter>(depgraph);
        SpanReader reader(ser);
        DepGraph<SetType> decoded_depgraph;
        reader >> Using<DepGraphFormatter>(decoded_depgraph);
        assert(depgraph == decoded_depgraph);
        assert(reader.empty());
        // It must also deserialize correctly without the terminal 0 byte (as the deserializer
        // will upon EOF still return what it read so far).
        assert(ser.size() >= 1 && ser.back() == 0);
        ser.pop_back();
        reader = SpanReader{ser};
        decoded_depgraph = {};
        reader >> Using<DepGraphFormatter>(decoded_depgraph);
        assert(depgraph == decoded_depgraph);
        assert(reader.empty());

        // In acyclic graphs, the union of parents with parents of parents etc. yields the
        // full ancestor set (and similar for children and descendants).
        std::vector<SetType> parents(depgraph.PositionRange()), children(depgraph.PositionRange());
        for (DepGraphIndex i : depgraph.Positions()) {
            parents[i] = depgraph.GetReducedParents(i);
            children[i] = depgraph.GetReducedChildren(i);
        }
        for (auto i : depgraph.Positions()) {
            // Initialize the set of ancestors with just the current transaction itself.
            SetType ancestors = SetType::Singleton(i);
            // Iteratively add parents of all transactions in the ancestor set to itself.
            while (true) {
                const auto old_ancestors = ancestors;
                for (auto j : ancestors) ancestors |= parents[j];
                // Stop when no more changes are being made.
                if (old_ancestors == ancestors) break;
            }
            assert(ancestors == depgraph.Ancestors(i));

            // Initialize the set of descendants with just the current transaction itself.
            SetType descendants = SetType::Singleton(i);
            // Iteratively add children of all transactions in the descendant set to itself.
            while (true) {
                const auto old_descendants = descendants;
                for (auto j : descendants) descendants |= children[j];
                // Stop when no more changes are being made.
                if (old_descendants == descendants) break;
            }
            assert(descendants == depgraph.Descendants(i));
        }
    }
}

template<typename SetType>
void SanityCheck(const DepGraph<SetType>& depgraph, std::span<const DepGraphIndex> linearization)
{
    // Check completeness.
    assert(linearization.size() == depgraph.TxCount());
    SetType done;
    for (auto i : linearization) {
        // Check transaction position is in range.
        assert(depgraph.Positions()[i]);
        // Check topology and lack of duplicates.
        assert((depgraph.Ancestors(i) - done) == SetType::Singleton(i));
        done.Set(i);
    }
}

inline uint64_t MaxOptimalLinearizationCost(DepGraphIndex cluster_count)
{
    // These are the largest numbers seen returned as cost by Linearize(), in a large randomized
    // trial. There exist almost certainly far worse cases, but they are unlikely to be
    // encountered in randomized tests. The purpose of these numbers is guaranteeing that for
    // *some* reasonable cost bound, optimal linearizations are always found.
    static constexpr uint64_t COSTS[65] = {
        0,
        0, 545, 928, 1633, 2647, 4065, 5598, 8258,
        9505, 11471, 14137, 19553, 20460, 26191, 28397, 32599,
        41631, 47419, 56329, 57767, 72196, 63652, 95366, 96537,
        115653, 125407, 131734, 145090, 156349, 164665, 194224, 203953,
        207710, 225878, 239971, 252284, 256534, 222142, 251332, 357098,
        325788, 295867, 410053, 497483, 533892, 576572, 577845, 572400,
        592536, 455082, 609249, 659130, 714091, 544507, 718788, 562378,
        601926, 1025081, 732725, 708896, 738224, 900445, 1092519, 1139946
    };
    assert(cluster_count < std::size(COSTS));
    // Multiply the table number by two, to account for the fact that they are not absolutes.
    return COSTS[cluster_count] * 2;
}

} // namespace

#endif // BITCOIN_TEST_UTIL_CLUSTER_LINEARIZE_H