/ src / txgraph.cpp
txgraph.cpp
   1  // Copyright (c) The Bitcoin Core developers
   2  // Distributed under the MIT software license, see the accompanying
   3  // file COPYING or http://www.opensource.org/licenses/mit-license.php.
   4  
   5  #include <txgraph.h>
   6  
   7  #include <cluster_linearize.h>
   8  #include <random.h>
   9  #include <util/bitset.h>
  10  #include <util/check.h>
  11  #include <util/feefrac.h>
  12  
  13  #include <compare>
  14  #include <memory>
  15  #include <set>
  16  #include <span>
  17  #include <utility>
  18  
  19  namespace {
  20  
  21  using namespace cluster_linearize;
  22  
  23  /** The maximum number of levels a TxGraph can have (0 = main, 1 = staging). */
  24  static constexpr int MAX_LEVELS{2};
  25  
  26  // Forward declare the TxGraph implementation class.
  27  class TxGraphImpl;
  28  
  29  /** Position of a DepGraphIndex within a Cluster::m_linearization. */
  30  using LinearizationIndex = uint32_t;
  31  /** Position of a Cluster within Graph::ClusterSet::m_clusters. */
  32  using ClusterSetIndex = uint32_t;
  33  
  34  /** Quality levels for cached cluster linearizations. */
  35  enum class QualityLevel
  36  {
  37      /** This cluster may have multiple disconnected components, which are all NEEDS_RELINEARIZE. */
  38      NEEDS_SPLIT,
  39      /** This cluster may have multiple disconnected components, which are all ACCEPTABLE. */
  40      NEEDS_SPLIT_ACCEPTABLE,
  41      /** This cluster has undergone changes that warrant re-linearization. */
  42      NEEDS_RELINEARIZE,
  43      /** The minimal level of linearization has been performed, but it is not known to be optimal. */
  44      ACCEPTABLE,
  45      /** The linearization is known to be optimal. */
  46      OPTIMAL,
  47      /** This cluster is not registered in any ClusterSet::m_clusters.
  48       *  This must be the last entry in QualityLevel as ClusterSet::m_clusters is sized using it. */
  49      NONE,
  50  };
  51  
  52  /** A grouping of connected transactions inside a TxGraphImpl::ClusterSet. */
  53  class Cluster
  54  {
  55      friend class TxGraphImpl;
  56      using GraphIndex = TxGraph::GraphIndex;
  57      using SetType = BitSet<MAX_CLUSTER_COUNT_LIMIT>;
  58      /** The DepGraph for this cluster, holding all feerates, and ancestors/descendants. */
  59      DepGraph<SetType> m_depgraph;
  60      /** m_mapping[i] gives the GraphIndex for the position i transaction in m_depgraph. Values for
  61       *  positions i that do not exist in m_depgraph shouldn't ever be accessed and thus don't
  62       *  matter. m_mapping.size() equals m_depgraph.PositionRange(). */
  63      std::vector<GraphIndex> m_mapping;
  64      /** The current linearization of the cluster. m_linearization.size() equals
  65       *  m_depgraph.TxCount(). This is always kept topological. */
  66      std::vector<DepGraphIndex> m_linearization;
  67      /** The quality level of m_linearization. */
  68      QualityLevel m_quality{QualityLevel::NONE};
  69      /** Which position this Cluster has in Graph::ClusterSet::m_clusters[m_quality]. */
  70      ClusterSetIndex m_setindex{ClusterSetIndex(-1)};
  71      /** Which level this Cluster is at in the graph (-1=not inserted, 0=main, 1=staging). */
  72      int m_level{-1};
  73      /** Sequence number for this Cluster (for tie-breaking comparison between equal-chunk-feerate
  74          transactions in distinct clusters). */
  75      uint64_t m_sequence;
  76  
  77  public:
  78      Cluster() noexcept = delete;
  79      /** Construct an empty Cluster. */
  80      explicit Cluster(uint64_t sequence) noexcept;
  81      /** Construct a singleton Cluster. */
  82      explicit Cluster(uint64_t sequence, TxGraphImpl& graph, const FeePerWeight& feerate, GraphIndex graph_index) noexcept;
  83  
  84      // Cannot move or copy (would invalidate Cluster* in Locator and ClusterSet). */
  85      Cluster(const Cluster&) = delete;
  86      Cluster& operator=(const Cluster&) = delete;
  87      Cluster(Cluster&&) = delete;
  88      Cluster& operator=(Cluster&&) = delete;
  89  
  90      // Generic helper functions.
  91  
  92      /** Whether the linearization of this Cluster can be exposed. */
  93      bool IsAcceptable(bool after_split = false) const noexcept
  94      {
  95          return m_quality == QualityLevel::ACCEPTABLE || m_quality == QualityLevel::OPTIMAL ||
  96                 (after_split && m_quality == QualityLevel::NEEDS_SPLIT_ACCEPTABLE);
  97      }
  98      /** Whether the linearization of this Cluster is optimal. */
  99      bool IsOptimal() const noexcept
 100      {
 101          return m_quality == QualityLevel::OPTIMAL;
 102      }
 103      /** Whether this cluster requires splitting. */
 104      bool NeedsSplitting() const noexcept
 105      {
 106          return m_quality == QualityLevel::NEEDS_SPLIT ||
 107                 m_quality == QualityLevel::NEEDS_SPLIT_ACCEPTABLE;
 108      }
 109      /** Get the number of transactions in this Cluster. */
 110      LinearizationIndex GetTxCount() const noexcept { return m_linearization.size(); }
 111      /** Given a DepGraphIndex into this Cluster, find the corresponding GraphIndex. */
 112      GraphIndex GetClusterEntry(DepGraphIndex index) const noexcept { return m_mapping[index]; }
 113      /** Only called by Graph::SwapIndexes. */
 114      void UpdateMapping(DepGraphIndex cluster_idx, GraphIndex graph_idx) noexcept { m_mapping[cluster_idx] = graph_idx; }
 115      /** Push changes to Cluster and its linearization to the TxGraphImpl Entry objects. */
 116      void Updated(TxGraphImpl& graph) noexcept;
 117      /** Create a copy of this Cluster in staging, returning a pointer to it (used by PullIn). */
 118      Cluster* CopyToStaging(TxGraphImpl& graph) const noexcept;
 119      /** Get the list of Clusters in main that conflict with this one (which is assumed to be in staging). */
 120      void GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept;
 121      /** Mark all the Entry objects belonging to this staging Cluster as missing. The Cluster must be
 122       *  deleted immediately after. */
 123      void MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept;
 124      /** Remove all transactions from a Cluster. */
 125      void Clear(TxGraphImpl& graph) noexcept;
 126      /** Change a Cluster's level from 1 (staging) to 0 (main). */
 127      void MoveToMain(TxGraphImpl& graph) noexcept;
 128  
 129      // Functions that implement the Cluster-specific side of internal TxGraphImpl mutations.
 130  
 131      /** Apply all removals from the front of to_remove that apply to this Cluster, popping them
 132       *  off. These must be at least one such entry. */
 133      void ApplyRemovals(TxGraphImpl& graph, std::span<GraphIndex>& to_remove) noexcept;
 134      /** Split this cluster (must have a NEEDS_SPLIT* quality). Returns whether to delete this
 135       *  Cluster afterwards. */
 136      [[nodiscard]] bool Split(TxGraphImpl& graph) noexcept;
 137      /** Move all transactions from cluster to *this (as separate components). */
 138      void Merge(TxGraphImpl& graph, Cluster& cluster) noexcept;
 139      /** Given a span of (parent, child) pairs that all belong to this Cluster, apply them. */
 140      void ApplyDependencies(TxGraphImpl& graph, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept;
 141      /** Improve the linearization of this Cluster. */
 142      void Relinearize(TxGraphImpl& graph, uint64_t max_iters) noexcept;
 143  
 144      // Functions that implement the Cluster-specific side of public TxGraph functions.
 145  
 146      /** Process elements from the front of args that apply to this cluster, and append Refs for the
 147       *  union of their ancestors to output. */
 148      void GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept;
 149      /** Process elements from the front of args that apply to this cluster, and append Refs for the
 150       *  union of their descendants to output. */
 151      void GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept;
 152      /** Get a vector of Refs for all elements of this Cluster, in linearization order. */
 153      std::vector<TxGraph::Ref*> GetClusterRefs(const TxGraphImpl& graph) noexcept;
 154      /** Get the individual transaction feerate of a Cluster element. */
 155      FeePerWeight GetIndividualFeerate(DepGraphIndex idx) noexcept;
 156      /** Modify the fee of a Cluster element. */
 157      void SetFee(TxGraphImpl& graph, DepGraphIndex idx, int64_t fee) noexcept;
 158  
 159      // Debugging functions.
 160  
 161      void SanityCheck(const TxGraphImpl& graph, int level) const;
 162  };
 163  
 164  
 165  /** The transaction graph, including staged changes.
 166   *
 167   * The overall design of the data structure consists of 3 interlinked representations:
 168   * - The transactions (held as a vector of TxGraphImpl::Entry inside TxGraphImpl).
 169   * - The clusters (Cluster objects in per-quality vectors inside TxGraphImpl::ClusterSet).
 170   * - The Refs (TxGraph::Ref objects, held externally by users of the TxGraph class)
 171   *
 172   * The Clusters are kept in one or two ClusterSet objects, one for the "main" graph, and one for
 173   * the proposed changes ("staging"). If a transaction occurs in both, they share the same Entry,
 174   * but there will be a separate Cluster per graph.
 175   *
 176   * Clusters and Refs contain the index of the Entry objects they refer to, and the Entry objects
 177   * refer back to the Clusters and Refs the corresponding transaction is contained in.
 178   *
 179   * While redundant, this permits moving all of them independently, without invalidating things
 180   * or costly iteration to fix up everything:
 181   * - Entry objects can be moved to fill holes left by removed transactions in the Entry vector
 182   *   (see TxGraphImpl::Compact).
 183   * - Clusters can be rewritten continuously (removals can cause them to split, new dependencies
 184   *   can cause them to be merged).
 185   * - Ref objects can be held outside the class, while permitting them to be moved around, and
 186   *   inherited from.
 187   */
 188  class TxGraphImpl final : public TxGraph
 189  {
 190      friend class Cluster;
 191  private:
 192      /** Internal RNG. */
 193      FastRandomContext m_rng;
 194      /** This TxGraphImpl's maximum cluster count limit. */
 195      const DepGraphIndex m_max_cluster_count;
 196  
 197      /** Information about one group of Clusters to be merged. */
 198      struct GroupEntry
 199      {
 200          /** Where the clusters to be merged start in m_group_clusters. */
 201          uint32_t m_cluster_offset;
 202          /** How many clusters to merge. */
 203          uint32_t m_cluster_count;
 204          /** Where the dependencies for this cluster group in m_deps_to_add start. */
 205          uint32_t m_deps_offset;
 206          /** How many dependencies to add. */
 207          uint32_t m_deps_count;
 208      };
 209  
 210      /** Information about all groups of Clusters to be merged. */
 211      struct GroupData
 212      {
 213          /** The groups of Clusters to be merged. */
 214          std::vector<GroupEntry> m_groups;
 215          /** Which clusters are to be merged. GroupEntry::m_cluster_offset indexes into this. */
 216          std::vector<Cluster*> m_group_clusters;
 217          /** Whether at least one of the groups cannot be applied because it would result in a
 218           *  Cluster that violates the cluster count limit. */
 219          bool m_group_oversized;
 220      };
 221  
 222      /** The collection of all Clusters in main or staged. */
 223      struct ClusterSet
 224      {
 225          /** The vectors of clusters, one vector per quality level. ClusterSetIndex indexes into each. */
 226          std::array<std::vector<std::unique_ptr<Cluster>>, int(QualityLevel::NONE)> m_clusters;
 227          /** Which removals have yet to be applied. */
 228          std::vector<GraphIndex> m_to_remove;
 229          /** Which dependencies are to be added ((parent,child) pairs). GroupData::m_deps_offset indexes
 230           *  into this. */
 231          std::vector<std::pair<GraphIndex, GraphIndex>> m_deps_to_add;
 232          /** Information about the merges to be performed, if known. */
 233          std::optional<GroupData> m_group_data = GroupData{};
 234          /** Which entries were removed in this ClusterSet (so they can be wiped on abort). This
 235           *  includes all entries which have an (R) removed locator at this level (staging only),
 236           *  plus optionally any transaction in m_unlinked. */
 237          std::vector<GraphIndex> m_removed;
 238          /** Total number of transactions in this graph (sum of all transaction counts in all
 239           *  Clusters, and for staging also those inherited from the main ClusterSet). */
 240          GraphIndex m_txcount{0};
 241          /** Whether this graph is oversized (if known). This roughly matches
 242           *  m_group_data->m_group_oversized, but may be known even if m_group_data is not. */
 243          std::optional<bool> m_oversized{false};
 244  
 245          ClusterSet() noexcept = default;
 246      };
 247  
 248      /** The main ClusterSet. */
 249      ClusterSet m_main_clusterset;
 250      /** The staging ClusterSet, if any. */
 251      std::optional<ClusterSet> m_staging_clusterset;
 252      /** Next sequence number to assign to created Clusters. */
 253      uint64_t m_next_sequence_counter{0};
 254  
 255      /** A Locator that describes whether, where, and in which Cluster an Entry appears.
 256       *  Every Entry has MAX_LEVELS locators, as it may appear in one Cluster per level.
 257       *
 258       *  Each level of a Locator is in one of three states:
 259       *
 260       *  - (P)resent: actually occurs in a Cluster at that level.
 261       *
 262       *  - (M)issing:
 263       *    - In the main graph:    the transaction does not exist in main.
 264       *    - In the staging graph: the transaction's existence is the same as in main. If it doesn't
 265       *                            exist in main, (M) in staging means it does not exist there
 266       *                            either. If it does exist in main, (M) in staging means the
 267       *                            cluster it is in has not been modified in staging, and thus the
 268       *                            transaction implicitly exists in staging too (without explicit
 269       *                            Cluster object; see PullIn() to create it in staging too).
 270       *
 271       *  - (R)emoved: only possible in staging; it means the transaction exists in main, but is
 272       *               removed in staging.
 273       *
 274       * The following combinations are possible:
 275       * - (M,M): the transaction doesn't exist in either graph.
 276       * - (P,M): the transaction exists in both, but only exists explicitly in a Cluster object in
 277       *          main. Its existence in staging is inherited from main.
 278       * - (P,P): the transaction exists in both, and is materialized in both. Thus, the clusters
 279       *          and/or their linearizations may be different in main and staging.
 280       * - (M,P): the transaction is added in staging, and does not exist in main.
 281       * - (P,R): the transaction exists in main, but is removed in staging.
 282       *
 283       * When staging does not exist, only (M,M) and (P,M) are possible.
 284       */
 285      struct Locator
 286      {
 287          /** Which Cluster the Entry appears in (nullptr = missing). */
 288          Cluster* cluster{nullptr};
 289          /** Where in the Cluster it appears (if cluster == nullptr: 0 = missing, -1 = removed). */
 290          DepGraphIndex index{0};
 291  
 292          /** Mark this Locator as missing (= same as lower level, or non-existing if level 0). */
 293          void SetMissing() noexcept { cluster = nullptr; index = 0; }
 294          /** Mark this Locator as removed (not allowed in level 0). */
 295          void SetRemoved() noexcept { cluster = nullptr; index = DepGraphIndex(-1); }
 296          /** Mark this Locator as present, in the specified Cluster. */
 297          void SetPresent(Cluster* c, DepGraphIndex i) noexcept { cluster = c; index = i; }
 298          /** Check if this Locator is missing. */
 299          bool IsMissing() const noexcept { return cluster == nullptr && index == 0; }
 300          /** Check if this Locator is removed. */
 301          bool IsRemoved() const noexcept { return cluster == nullptr && index == DepGraphIndex(-1); }
 302          /** Check if this Locator is present (in some Cluster). */
 303          bool IsPresent() const noexcept { return cluster != nullptr; }
 304      };
 305  
 306      /** Internal information about each transaction in a TxGraphImpl. */
 307      struct Entry
 308      {
 309          /** Pointer to the corresponding Ref object if any, or nullptr if unlinked. */
 310          Ref* m_ref{nullptr};
 311          /** Which Cluster and position therein this Entry appears in. ([0] = main, [1] = staged). */
 312          Locator m_locator[MAX_LEVELS];
 313          /** The chunk feerate of this transaction in main (if present in m_locator[0]). */
 314          FeePerWeight m_main_chunk_feerate;
 315          /** The position this transaction has in the main linearization (if present). */
 316          LinearizationIndex m_main_lin_index;
 317      };
 318  
 319      /** The set of all transactions (in all levels combined). GraphIndex values index into this. */
 320      std::vector<Entry> m_entries;
 321  
 322      /** Set of Entries which have no linked Ref anymore. */
 323      std::vector<GraphIndex> m_unlinked;
 324  
 325      /** Compare two Cluster* by their m_sequence value (while supporting nullptr). */
 326      static std::strong_ordering CompareClusters(Cluster* a, Cluster* b) noexcept
 327      {
 328          // The nullptr pointer compares before everything else.
 329          if (a == nullptr || b == nullptr) {
 330              return (a != nullptr) <=> (b != nullptr);
 331          }
 332          // If neither pointer is nullptr, compare the Clusters' sequence numbers.
 333          Assume(a == b || a->m_sequence != b->m_sequence);
 334          return a->m_sequence <=> b->m_sequence;
 335      }
 336  
 337  public:
 338      /** Construct a new TxGraphImpl with the specified maximum cluster count. */
 339      explicit TxGraphImpl(DepGraphIndex max_cluster_count) noexcept :
 340          m_max_cluster_count(max_cluster_count)
 341      {
 342          Assume(max_cluster_count >= 1);
 343          Assume(max_cluster_count <= MAX_CLUSTER_COUNT_LIMIT);
 344      }
 345  
 346      /** Destructor. */
 347      ~TxGraphImpl() noexcept;
 348  
 349      // Cannot move or copy (would invalidate TxGraphImpl* in Ref, MiningOrder, EvictionOrder).
 350      TxGraphImpl(const TxGraphImpl&) = delete;
 351      TxGraphImpl& operator=(const TxGraphImpl&) = delete;
 352      TxGraphImpl(TxGraphImpl&&) = delete;
 353      TxGraphImpl& operator=(TxGraphImpl&&) = delete;
 354  
 355      // Simple helper functions.
 356  
 357      /** Swap the Entry referred to by a and the one referred to by b. */
 358      void SwapIndexes(GraphIndex a, GraphIndex b) noexcept;
 359      /** If idx exists in the specified level ClusterSet (explicitly, or in the level below and not
 360      *   removed), return the Cluster it is in. Otherwise, return nullptr. */
 361      Cluster* FindCluster(GraphIndex idx, int level) const noexcept;
 362      /** Extract a Cluster from its ClusterSet. */
 363      std::unique_ptr<Cluster> ExtractCluster(int level, QualityLevel quality, ClusterSetIndex setindex) noexcept;
 364      /** Delete a Cluster. */
 365      void DeleteCluster(Cluster& cluster) noexcept;
 366      /** Insert a Cluster into its ClusterSet. */
 367      ClusterSetIndex InsertCluster(int level, std::unique_ptr<Cluster>&& cluster, QualityLevel quality) noexcept;
 368      /** Change the QualityLevel of a Cluster (identified by old_quality and old_index). */
 369      void SetClusterQuality(int level, QualityLevel old_quality, ClusterSetIndex old_index, QualityLevel new_quality) noexcept;
 370      /** Get the index of the top level ClusterSet (staging if it exists, main otherwise). */
 371      int GetTopLevel() const noexcept { return m_staging_clusterset.has_value(); }
 372      /** Get the specified level (staging if it exists and main_only is not specified, main otherwise). */
 373      int GetSpecifiedLevel(bool main_only) const noexcept { return m_staging_clusterset.has_value() && !main_only; }
 374      /** Get a reference to the ClusterSet at the specified level (which must exist). */
 375      ClusterSet& GetClusterSet(int level) noexcept;
 376      const ClusterSet& GetClusterSet(int level) const noexcept;
 377      /** Make a transaction not exist at a specified level. It must currently exist there. */
 378      void ClearLocator(int level, GraphIndex index) noexcept;
 379      /** Find which Clusters in main conflict with ones in staging. */
 380      std::vector<Cluster*> GetConflicts() const noexcept;
 381  
 382      // Functions for handling Refs.
 383  
 384      /** Only called by Ref's move constructor/assignment to update Ref locations. */
 385      void UpdateRef(GraphIndex idx, Ref& new_location) noexcept final
 386      {
 387          auto& entry = m_entries[idx];
 388          Assume(entry.m_ref != nullptr);
 389          entry.m_ref = &new_location;
 390      }
 391  
 392      /** Only called by Ref::~Ref to unlink Refs, and Ref's move assignment. */
 393      void UnlinkRef(GraphIndex idx) noexcept final
 394      {
 395          auto& entry = m_entries[idx];
 396          Assume(entry.m_ref != nullptr);
 397          entry.m_ref = nullptr;
 398          // Mark the transaction as to be removed in all levels where it explicitly or implicitly
 399          // exists.
 400          bool exists_anywhere{false};
 401          bool exists{false};
 402          for (int level = 0; level <= GetTopLevel(); ++level) {
 403              if (entry.m_locator[level].IsPresent()) {
 404                  exists_anywhere = true;
 405                  exists = true;
 406              } else if (entry.m_locator[level].IsRemoved()) {
 407                  exists = false;
 408              }
 409              if (exists) {
 410                  auto& clusterset = GetClusterSet(level);
 411                  clusterset.m_to_remove.push_back(idx);
 412                  // Force recomputation of grouping data.
 413                  clusterset.m_group_data = std::nullopt;
 414                  // Do not wipe the oversized state of main if staging exists. The reason for this
 415                  // is that the alternative would mean that cluster merges may need to be applied to
 416                  // a formerly-oversized main graph while staging exists (to satisfy chunk feerate
 417                  // queries into main, for example), and such merges could conflict with pulls of
 418                  // some of their constituents into staging.
 419                  if (level == GetTopLevel() && clusterset.m_oversized == true) {
 420                      clusterset.m_oversized = std::nullopt;
 421                  }
 422              }
 423          }
 424          m_unlinked.push_back(idx);
 425          if (!exists_anywhere) Compact();
 426      }
 427  
 428      // Functions related to various normalization/application steps.
 429      /** Get rid of unlinked Entry objects in m_entries, if possible (this changes the GraphIndex
 430       *  values for remaining Entry objects, so this only does something when no to-be-applied
 431       *  operations or staged removals referring to GraphIndexes remain). */
 432      void Compact() noexcept;
 433      /** If cluster is not in staging, copy it there, and return a pointer to it.
 434      *   Staging must exist, and this modifies the locators of its
 435      *   transactions from inherited (P,M) to explicit (P,P). */
 436      Cluster* PullIn(Cluster* cluster) noexcept;
 437      /** Apply all removals queued up in m_to_remove to the relevant Clusters (which get a
 438       *  NEEDS_SPLIT* QualityLevel) up to the specified level. */
 439      void ApplyRemovals(int up_to_level) noexcept;
 440      /** Split an individual cluster. */
 441      void Split(Cluster& cluster) noexcept;
 442      /** Split all clusters that need splitting up to the specified level. */
 443      void SplitAll(int up_to_level) noexcept;
 444      /** Populate m_group_data based on m_deps_to_add in the specified level. */
 445      void GroupClusters(int level) noexcept;
 446      /** Merge the specified clusters. */
 447      void Merge(std::span<Cluster*> to_merge) noexcept;
 448      /** Apply all m_deps_to_add to the relevant Clusters in the specified level. */
 449      void ApplyDependencies(int level) noexcept;
 450      /** Make a specified Cluster have quality ACCEPTABLE or OPTIMAL. */
 451      void MakeAcceptable(Cluster& cluster) noexcept;
 452      /** Make all Clusters at the specified level have quality ACCEPTABLE or OPTIMAL. */
 453      void MakeAllAcceptable(int level) noexcept;
 454  
 455      // Implementations for the public TxGraph interface.
 456  
 457      Ref AddTransaction(const FeePerWeight& feerate) noexcept final;
 458      void RemoveTransaction(const Ref& arg) noexcept final;
 459      void AddDependency(const Ref& parent, const Ref& child) noexcept final;
 460      void SetTransactionFee(const Ref&, int64_t fee) noexcept final;
 461  
 462      void DoWork() noexcept final;
 463  
 464      void StartStaging() noexcept final;
 465      void CommitStaging() noexcept final;
 466      void AbortStaging() noexcept final;
 467      bool HaveStaging() const noexcept final { return m_staging_clusterset.has_value(); }
 468  
 469      bool Exists(const Ref& arg, bool main_only = false) noexcept final;
 470      FeePerWeight GetMainChunkFeerate(const Ref& arg) noexcept final;
 471      FeePerWeight GetIndividualFeerate(const Ref& arg) noexcept final;
 472      std::vector<Ref*> GetCluster(const Ref& arg, bool main_only = false) noexcept final;
 473      std::vector<Ref*> GetAncestors(const Ref& arg, bool main_only = false) noexcept final;
 474      std::vector<Ref*> GetDescendants(const Ref& arg, bool main_only = false) noexcept final;
 475      std::vector<Ref*> GetAncestorsUnion(std::span<const Ref* const> args, bool main_only = false) noexcept final;
 476      std::vector<Ref*> GetDescendantsUnion(std::span<const Ref* const> args, bool main_only = false) noexcept final;
 477      GraphIndex GetTransactionCount(bool main_only = false) noexcept final;
 478      bool IsOversized(bool main_only = false) noexcept final;
 479      std::strong_ordering CompareMainOrder(const Ref& a, const Ref& b) noexcept final;
 480      GraphIndex CountDistinctClusters(std::span<const Ref* const> refs, bool main_only = false) noexcept final;
 481  
 482      void SanityCheck() const final;
 483  };
 484  
 485  TxGraphImpl::ClusterSet& TxGraphImpl::GetClusterSet(int level) noexcept
 486  {
 487      if (level == 0) return m_main_clusterset;
 488      Assume(level == 1);
 489      Assume(m_staging_clusterset.has_value());
 490      return *m_staging_clusterset;
 491  }
 492  
 493  const TxGraphImpl::ClusterSet& TxGraphImpl::GetClusterSet(int level) const noexcept
 494  {
 495      if (level == 0) return m_main_clusterset;
 496      Assume(level == 1);
 497      Assume(m_staging_clusterset.has_value());
 498      return *m_staging_clusterset;
 499  }
 500  
 501  void TxGraphImpl::ClearLocator(int level, GraphIndex idx) noexcept
 502  {
 503      auto& entry = m_entries[idx];
 504      auto& clusterset = GetClusterSet(level);
 505      Assume(entry.m_locator[level].IsPresent());
 506      // Change the locator from Present to Missing or Removed.
 507      if (level == 0 || !entry.m_locator[level - 1].IsPresent()) {
 508          entry.m_locator[level].SetMissing();
 509      } else {
 510          entry.m_locator[level].SetRemoved();
 511          clusterset.m_removed.push_back(idx);
 512      }
 513      // Update the transaction count.
 514      --clusterset.m_txcount;
 515      // If clearing main, adjust the status of Locators of this transaction in staging, if it exists.
 516      if (level == 0 && GetTopLevel() == 1) {
 517          if (entry.m_locator[1].IsRemoved()) {
 518              entry.m_locator[1].SetMissing();
 519          } else if (!entry.m_locator[1].IsPresent()) {
 520              --m_staging_clusterset->m_txcount;
 521          }
 522      }
 523  }
 524  
 525  void Cluster::Updated(TxGraphImpl& graph) noexcept
 526  {
 527      // Update all the Locators for this Cluster's Entry objects.
 528      for (DepGraphIndex idx : m_linearization) {
 529          auto& entry = graph.m_entries[m_mapping[idx]];
 530          entry.m_locator[m_level].SetPresent(this, idx);
 531      }
 532      // If this is for the main graph (level = 0), and the Cluster's quality is ACCEPTABLE or
 533      // OPTIMAL, compute its chunking and store its information in the Entry's m_main_lin_index
 534      // and m_main_chunk_feerate. These fields are only accessed after making the entire graph
 535      // ACCEPTABLE, so it is pointless to compute these if we haven't reached that quality level
 536      // yet.
 537      if (m_level == 0 && IsAcceptable()) {
 538          LinearizationChunking chunking(m_depgraph, m_linearization);
 539          LinearizationIndex lin_idx{0};
 540          // Iterate over the chunks.
 541          for (unsigned chunk_idx = 0; chunk_idx < chunking.NumChunksLeft(); ++chunk_idx) {
 542              auto chunk = chunking.GetChunk(chunk_idx);
 543              Assume(chunk.transactions.Any());
 544              // Iterate over the transactions in the linearization, which must match those in chunk.
 545              do {
 546                  DepGraphIndex idx = m_linearization[lin_idx];
 547                  GraphIndex graph_idx = m_mapping[idx];
 548                  auto& entry = graph.m_entries[graph_idx];
 549                  entry.m_main_lin_index = lin_idx++;
 550                  entry.m_main_chunk_feerate = FeePerWeight::FromFeeFrac(chunk.feerate);
 551                  Assume(chunk.transactions[idx]);
 552                  chunk.transactions.Reset(idx);
 553              } while(chunk.transactions.Any());
 554          }
 555      }
 556  }
 557  
 558  void Cluster::GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept
 559  {
 560      Assume(m_level == 1);
 561      for (auto i : m_linearization) {
 562          auto& entry = graph.m_entries[m_mapping[i]];
 563          // For every transaction Entry in this Cluster, if it also exists in a lower-level Cluster,
 564          // then that Cluster conflicts.
 565          if (entry.m_locator[0].IsPresent()) {
 566              out.push_back(entry.m_locator[0].cluster);
 567          }
 568      }
 569  }
 570  
 571  std::vector<Cluster*> TxGraphImpl::GetConflicts() const noexcept
 572  {
 573      Assume(GetTopLevel() == 1);
 574      auto& clusterset = GetClusterSet(1);
 575      std::vector<Cluster*> ret;
 576      // All main Clusters containing transactions in m_removed (so (P,R) ones) are conflicts.
 577      for (auto i : clusterset.m_removed) {
 578          auto& entry = m_entries[i];
 579          if (entry.m_locator[0].IsPresent()) {
 580              ret.push_back(entry.m_locator[0].cluster);
 581          }
 582      }
 583      // Then go over all Clusters at this level, and find their conflicts (the (P,P) ones).
 584      for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
 585          auto& clusters = clusterset.m_clusters[quality];
 586          for (const auto& cluster : clusters) {
 587              cluster->GetConflicts(*this, ret);
 588          }
 589      }
 590      // Deduplicate the result (the same Cluster may appear multiple times).
 591      std::sort(ret.begin(), ret.end(), [](Cluster* a, Cluster* b) noexcept { return CompareClusters(a, b) < 0; });
 592      ret.erase(std::unique(ret.begin(), ret.end()), ret.end());
 593      return ret;
 594  }
 595  
 596  Cluster* Cluster::CopyToStaging(TxGraphImpl& graph) const noexcept
 597  {
 598      // Construct an empty Cluster.
 599      auto ret = std::make_unique<Cluster>(graph.m_next_sequence_counter++);
 600      auto ptr = ret.get();
 601      // Copy depgraph, mapping, and linearization/
 602      ptr->m_depgraph = m_depgraph;
 603      ptr->m_mapping = m_mapping;
 604      ptr->m_linearization = m_linearization;
 605      // Insert the new Cluster into the graph.
 606      graph.InsertCluster(1, std::move(ret), m_quality);
 607      // Update its Locators.
 608      ptr->Updated(graph);
 609      return ptr;
 610  }
 611  
 612  void Cluster::ApplyRemovals(TxGraphImpl& graph, std::span<GraphIndex>& to_remove) noexcept
 613  {
 614      // Iterate over the prefix of to_remove that applies to this cluster.
 615      Assume(!to_remove.empty());
 616      SetType todo;
 617      do {
 618          GraphIndex idx = to_remove.front();
 619          Assume(idx < graph.m_entries.size());
 620          auto& entry = graph.m_entries[idx];
 621          auto& locator = entry.m_locator[m_level];
 622          // Stop once we hit an entry that applies to another Cluster.
 623          if (locator.cluster != this) break;
 624          // - Remember it in a set of to-remove DepGraphIndexes.
 625          todo.Set(locator.index);
 626          // - Remove from m_mapping. This isn't strictly necessary as unused positions in m_mapping
 627          //   are just never accessed, but set it to -1 here to increase the ability to detect a bug
 628          //   that causes it to be accessed regardless.
 629          m_mapping[locator.index] = GraphIndex(-1);
 630          // - Remove its linearization index from the Entry (if in main).
 631          if (m_level == 0) {
 632              entry.m_main_lin_index = LinearizationIndex(-1);
 633          }
 634          // - Mark it as missing/removed in the Entry's locator.
 635          graph.ClearLocator(m_level, idx);
 636          to_remove = to_remove.subspan(1);
 637      } while(!to_remove.empty());
 638  
 639      auto quality = m_quality;
 640      Assume(todo.Any());
 641      // Wipe from the Cluster's DepGraph (this is O(n) regardless of the number of entries
 642      // removed, so we benefit from batching all the removals).
 643      m_depgraph.RemoveTransactions(todo);
 644      m_mapping.resize(m_depgraph.PositionRange());
 645  
 646      // First remove all removals at the end of the linearization.
 647      while (!m_linearization.empty() && todo[m_linearization.back()]) {
 648          todo.Reset(m_linearization.back());
 649          m_linearization.pop_back();
 650      }
 651      if (todo.None()) {
 652          // If no further removals remain, and thus all removals were at the end, we may be able
 653          // to leave the cluster at a better quality level.
 654          if (IsAcceptable(/*after_split=*/true)) {
 655              quality = QualityLevel::NEEDS_SPLIT_ACCEPTABLE;
 656          } else {
 657              quality = QualityLevel::NEEDS_SPLIT;
 658          }
 659      } else {
 660          // If more removals remain, filter those out of m_linearization.
 661          m_linearization.erase(std::remove_if(
 662              m_linearization.begin(),
 663              m_linearization.end(),
 664              [&](auto pos) { return todo[pos]; }), m_linearization.end());
 665          quality = QualityLevel::NEEDS_SPLIT;
 666      }
 667      graph.SetClusterQuality(m_level, m_quality, m_setindex, quality);
 668      Updated(graph);
 669  }
 670  
 671  void Cluster::Clear(TxGraphImpl& graph) noexcept
 672  {
 673      for (auto i : m_linearization) {
 674          graph.ClearLocator(m_level, m_mapping[i]);
 675      }
 676      m_depgraph = {};
 677      m_linearization.clear();
 678      m_mapping.clear();
 679  }
 680  
 681  void Cluster::MoveToMain(TxGraphImpl& graph) noexcept
 682  {
 683      Assume(m_level == 1);
 684      for (auto i : m_linearization) {
 685          GraphIndex idx = m_mapping[i];
 686          auto& entry = graph.m_entries[idx];
 687          entry.m_locator[1].SetMissing();
 688      }
 689      auto quality = m_quality;
 690      auto cluster = graph.ExtractCluster(1, quality, m_setindex);
 691      graph.InsertCluster(0, std::move(cluster), quality);
 692      Updated(graph);
 693  }
 694  
 695  bool Cluster::Split(TxGraphImpl& graph) noexcept
 696  {
 697      // This function can only be called when the Cluster needs splitting.
 698      Assume(NeedsSplitting());
 699      // Determine the new quality the split-off Clusters will have.
 700      QualityLevel new_quality = IsAcceptable(/*after_split=*/true) ? QualityLevel::ACCEPTABLE
 701                                                                    : QualityLevel::NEEDS_RELINEARIZE;
 702      // If we're going to produce ACCEPTABLE clusters (i.e., when in NEEDS_SPLIT_ACCEPTABLE), we
 703      // need to post-linearize to make sure the split-out versions are all connected (as
 704      // connectivity may have changed by removing part of the cluster). This could be done on each
 705      // resulting split-out cluster separately, but it is simpler to do it once up front before
 706      // splitting. This step is not necessary if the resulting clusters are NEEDS_RELINEARIZE, as
 707      // they will be post-linearized anyway in MakeAcceptable().
 708      if (new_quality == QualityLevel::ACCEPTABLE) {
 709          PostLinearize(m_depgraph, m_linearization);
 710      }
 711      /** Which positions are still left in this Cluster. */
 712      auto todo = m_depgraph.Positions();
 713      /** Mapping from transaction positions in this Cluster to the Cluster where it ends up, and
 714       *  its position therein. */
 715      std::vector<std::pair<Cluster*, DepGraphIndex>> remap(m_depgraph.PositionRange());
 716      std::vector<Cluster*> new_clusters;
 717      bool first{true};
 718      // Iterate over the connected components of this Cluster's m_depgraph.
 719      while (todo.Any()) {
 720          auto component = m_depgraph.FindConnectedComponent(todo);
 721          if (first && component == todo) {
 722              // The existing Cluster is an entire component. Leave it be, but update its quality.
 723              Assume(todo == m_depgraph.Positions());
 724              graph.SetClusterQuality(m_level, m_quality, m_setindex, new_quality);
 725              // If this made the quality ACCEPTABLE or OPTIMAL, we need to compute and cache its
 726              // chunking.
 727              Updated(graph);
 728              return false;
 729          }
 730          first = false;
 731          // Construct a new Cluster to hold the found component.
 732          auto new_cluster = std::make_unique<Cluster>(graph.m_next_sequence_counter++);
 733          new_clusters.push_back(new_cluster.get());
 734          // Remember that all the component's transactions go to this new Cluster. The positions
 735          // will be determined below, so use -1 for now.
 736          for (auto i : component) {
 737              remap[i] = {new_cluster.get(), DepGraphIndex(-1)};
 738          }
 739          graph.InsertCluster(m_level, std::move(new_cluster), new_quality);
 740          todo -= component;
 741      }
 742      // Redistribute the transactions.
 743      for (auto i : m_linearization) {
 744          /** The cluster which transaction originally in position i is moved to. */
 745          Cluster* new_cluster = remap[i].first;
 746          // Copy the transaction to the new cluster's depgraph, and remember the position.
 747          remap[i].second = new_cluster->m_depgraph.AddTransaction(m_depgraph.FeeRate(i));
 748          // Create new mapping entry.
 749          new_cluster->m_mapping.push_back(m_mapping[i]);
 750          // Create a new linearization entry. As we're only appending transactions, they equal the
 751          // DepGraphIndex.
 752          new_cluster->m_linearization.push_back(remap[i].second);
 753      }
 754      // Redistribute the dependencies.
 755      for (auto i : m_linearization) {
 756          /** The cluster transaction in position i is moved to. */
 757          Cluster* new_cluster = remap[i].first;
 758          // Copy its parents, translating positions.
 759          SetType new_parents;
 760          for (auto par : m_depgraph.GetReducedParents(i)) new_parents.Set(remap[par].second);
 761          new_cluster->m_depgraph.AddDependencies(new_parents, remap[i].second);
 762      }
 763      // Update all the Locators of moved transactions.
 764      for (Cluster* new_cluster : new_clusters) {
 765          new_cluster->Updated(graph);
 766      }
 767      // Wipe this Cluster, and return that it needs to be deleted.
 768      m_depgraph = DepGraph<SetType>{};
 769      m_mapping.clear();
 770      m_linearization.clear();
 771      return true;
 772  }
 773  
 774  void Cluster::Merge(TxGraphImpl& graph, Cluster& other) noexcept
 775  {
 776      /** Vector to store the positions in this Cluster for each position in other. */
 777      std::vector<DepGraphIndex> remap(other.m_depgraph.PositionRange());
 778      // Iterate over all transactions in the other Cluster (the one being absorbed).
 779      for (auto pos : other.m_linearization) {
 780          auto idx = other.m_mapping[pos];
 781          // Copy the transaction into this Cluster, and remember its position.
 782          auto new_pos = m_depgraph.AddTransaction(other.m_depgraph.FeeRate(pos));
 783          remap[pos] = new_pos;
 784          if (new_pos == m_mapping.size()) {
 785              m_mapping.push_back(idx);
 786          } else {
 787              m_mapping[new_pos] = idx;
 788          }
 789          m_linearization.push_back(new_pos);
 790          // Copy the transaction's dependencies, translating them using remap. Note that since
 791          // pos iterates over other.m_linearization, which is in topological order, all parents
 792          // of pos should already be in remap.
 793          SetType parents;
 794          for (auto par : other.m_depgraph.GetReducedParents(pos)) {
 795              parents.Set(remap[par]);
 796          }
 797          m_depgraph.AddDependencies(parents, remap[pos]);
 798          // Update the transaction's Locator. There is no need to call Updated() to update chunk
 799          // feerates, as Updated() will be invoked by Cluster::ApplyDependencies on the resulting
 800          // merged Cluster later anyway).
 801          graph.m_entries[idx].m_locator[m_level].SetPresent(this, new_pos);
 802      }
 803      // Purge the other Cluster, now that everything has been moved.
 804      other.m_depgraph = DepGraph<SetType>{};
 805      other.m_linearization.clear();
 806      other.m_mapping.clear();
 807  }
 808  
 809  void Cluster::ApplyDependencies(TxGraphImpl& graph, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept
 810  {
 811      // This function is invoked by TxGraphImpl::ApplyDependencies after merging groups of Clusters
 812      // between which dependencies are added, which simply concatenates their linearizations. Invoke
 813      // PostLinearize, which has the effect that the linearization becomes a merge-sort of the
 814      // constituent linearizations. Do this here rather than in Cluster::Merge, because this
 815      // function is only invoked once per merged Cluster, rather than once per constituent one.
 816      // This concatenation + post-linearization could be replaced with an explicit merge-sort.
 817      PostLinearize(m_depgraph, m_linearization);
 818  
 819      // Sort the list of dependencies to apply by child, so those can be applied in batch.
 820      std::sort(to_apply.begin(), to_apply.end(), [](auto& a, auto& b) { return a.second < b.second; });
 821      // Iterate over groups of to-be-added dependencies with the same child.
 822      auto it = to_apply.begin();
 823      while (it != to_apply.end()) {
 824          auto& first_child = graph.m_entries[it->second].m_locator[m_level];
 825          const auto child_idx = first_child.index;
 826          // Iterate over all to-be-added dependencies within that same child, gather the relevant
 827          // parents.
 828          SetType parents;
 829          while (it != to_apply.end()) {
 830              auto& child = graph.m_entries[it->second].m_locator[m_level];
 831              auto& parent = graph.m_entries[it->first].m_locator[m_level];
 832              Assume(child.cluster == this && parent.cluster == this);
 833              if (child.index != child_idx) break;
 834              parents.Set(parent.index);
 835              ++it;
 836          }
 837          // Push all dependencies to the underlying DepGraph. Note that this is O(N) in the size of
 838          // the cluster, regardless of the number of parents being added, so batching them together
 839          // has a performance benefit.
 840          m_depgraph.AddDependencies(parents, child_idx);
 841      }
 842  
 843      // Finally fix the linearization, as the new dependencies may have invalidated the
 844      // linearization, and post-linearize it to fix up the worst problems with it.
 845      FixLinearization(m_depgraph, m_linearization);
 846      PostLinearize(m_depgraph, m_linearization);
 847  
 848      // Finally push the changes to graph.m_entries.
 849      Updated(graph);
 850  }
 851  
 852  TxGraphImpl::~TxGraphImpl() noexcept
 853  {
 854      // If Refs outlive the TxGraphImpl they refer to, unlink them, so that their destructor does not
 855      // try to reach into a non-existing TxGraphImpl anymore.
 856      for (auto& entry : m_entries) {
 857          if (entry.m_ref != nullptr) {
 858              GetRefGraph(*entry.m_ref) = nullptr;
 859          }
 860      }
 861  }
 862  
 863  std::unique_ptr<Cluster> TxGraphImpl::ExtractCluster(int level, QualityLevel quality, ClusterSetIndex setindex) noexcept
 864  {
 865      Assume(quality != QualityLevel::NONE);
 866  
 867      auto& clusterset = GetClusterSet(level);
 868      auto& quality_clusters = clusterset.m_clusters[int(quality)];
 869      Assume(setindex < quality_clusters.size());
 870  
 871      // Extract the Cluster-owning unique_ptr.
 872      std::unique_ptr<Cluster> ret = std::move(quality_clusters[setindex]);
 873      ret->m_quality = QualityLevel::NONE;
 874      ret->m_setindex = ClusterSetIndex(-1);
 875      ret->m_level = -1;
 876  
 877      // Clean up space in quality_cluster.
 878      auto max_setindex = quality_clusters.size() - 1;
 879      if (setindex != max_setindex) {
 880          // If the cluster was not the last element of quality_clusters, move that to take its place.
 881          quality_clusters.back()->m_setindex = setindex;
 882          quality_clusters.back()->m_level = level;
 883          quality_clusters[setindex] = std::move(quality_clusters.back());
 884      }
 885      // The last element of quality_clusters is now unused; drop it.
 886      quality_clusters.pop_back();
 887  
 888      return ret;
 889  }
 890  
 891  ClusterSetIndex TxGraphImpl::InsertCluster(int level, std::unique_ptr<Cluster>&& cluster, QualityLevel quality) noexcept
 892  {
 893      // Cannot insert with quality level NONE (as that would mean not inserted).
 894      Assume(quality != QualityLevel::NONE);
 895      // The passed-in Cluster must not currently be in the TxGraphImpl.
 896      Assume(cluster->m_quality == QualityLevel::NONE);
 897  
 898      // Append it at the end of the relevant TxGraphImpl::m_cluster.
 899      auto& clusterset = GetClusterSet(level);
 900      auto& quality_clusters = clusterset.m_clusters[int(quality)];
 901      ClusterSetIndex ret = quality_clusters.size();
 902      cluster->m_quality = quality;
 903      cluster->m_setindex = ret;
 904      cluster->m_level = level;
 905      quality_clusters.push_back(std::move(cluster));
 906      return ret;
 907  }
 908  
 909  void TxGraphImpl::SetClusterQuality(int level, QualityLevel old_quality, ClusterSetIndex old_index, QualityLevel new_quality) noexcept
 910  {
 911      Assume(new_quality != QualityLevel::NONE);
 912  
 913      // Don't do anything if the quality did not change.
 914      if (old_quality == new_quality) return;
 915      // Extract the cluster from where it currently resides.
 916      auto cluster_ptr = ExtractCluster(level, old_quality, old_index);
 917      // And re-insert it where it belongs.
 918      InsertCluster(level, std::move(cluster_ptr), new_quality);
 919  }
 920  
 921  void TxGraphImpl::DeleteCluster(Cluster& cluster) noexcept
 922  {
 923      // Extract the cluster from where it currently resides.
 924      auto cluster_ptr = ExtractCluster(cluster.m_level, cluster.m_quality, cluster.m_setindex);
 925      // And throw it away.
 926      cluster_ptr.reset();
 927  }
 928  
 929  Cluster* TxGraphImpl::FindCluster(GraphIndex idx, int level) const noexcept
 930  {
 931      Assume(level >= 0 && level <= GetTopLevel());
 932      auto& entry = m_entries[idx];
 933      // Search the entry's locators from top to bottom.
 934      for (int l = level; l >= 0; --l) {
 935          // If the locator is missing, dig deeper; it may exist at a lower level and therefore be
 936          // implicitly existing at this level too.
 937          if (entry.m_locator[l].IsMissing()) continue;
 938          // If the locator has the entry marked as explicitly removed, stop.
 939          if (entry.m_locator[l].IsRemoved()) break;
 940          // Otherwise, we have found the topmost ClusterSet that contains this entry.
 941          return entry.m_locator[l].cluster;
 942      }
 943      // If no non-empty locator was found, or an explicitly removed was hit, return nothing.
 944      return nullptr;
 945  }
 946  
 947  Cluster* TxGraphImpl::PullIn(Cluster* cluster) noexcept
 948  {
 949      int to_level = GetTopLevel();
 950      Assume(to_level == 1);
 951      int level = cluster->m_level;
 952      Assume(level <= to_level);
 953      // Copy the Cluster from main to staging, if it's not already there.
 954      if (level == 0) {
 955          // Make the Cluster Acceptable before copying. This isn't strictly necessary, but doing it
 956          // now avoids doing double work later.
 957          MakeAcceptable(*cluster);
 958          cluster = cluster->CopyToStaging(*this);
 959      }
 960      return cluster;
 961  }
 962  
 963  void TxGraphImpl::ApplyRemovals(int up_to_level) noexcept
 964  {
 965      Assume(up_to_level >= 0 && up_to_level <= GetTopLevel());
 966      for (int level = 0; level <= up_to_level; ++level) {
 967          auto& clusterset = GetClusterSet(level);
 968          auto& to_remove = clusterset.m_to_remove;
 969          // Skip if there is nothing to remove in this level.
 970          if (to_remove.empty()) continue;
 971          // Pull in all Clusters that are not in staging.
 972          if (level == 1) {
 973              for (GraphIndex index : to_remove) {
 974                  auto cluster = FindCluster(index, level);
 975                  if (cluster != nullptr) PullIn(cluster);
 976              }
 977          }
 978          // Group the set of to-be-removed entries by Cluster::m_sequence.
 979          std::sort(to_remove.begin(), to_remove.end(), [&](GraphIndex a, GraphIndex b) noexcept {
 980              Cluster* cluster_a = m_entries[a].m_locator[level].cluster;
 981              Cluster* cluster_b = m_entries[b].m_locator[level].cluster;
 982              return CompareClusters(cluster_a, cluster_b) < 0;
 983          });
 984          // Process per Cluster.
 985          std::span to_remove_span{to_remove};
 986          while (!to_remove_span.empty()) {
 987              Cluster* cluster = m_entries[to_remove_span.front()].m_locator[level].cluster;
 988              if (cluster != nullptr) {
 989                  // If the first to_remove_span entry's Cluster exists, hand to_remove_span to it, so it
 990                  // can pop off whatever applies to it.
 991                  cluster->ApplyRemovals(*this, to_remove_span);
 992              } else {
 993                  // Otherwise, skip this already-removed entry. This may happen when
 994                  // RemoveTransaction was called twice on the same Ref, for example.
 995                  to_remove_span = to_remove_span.subspan(1);
 996              }
 997          }
 998          to_remove.clear();
 999      }
1000      Compact();
1001  }
1002  
1003  void TxGraphImpl::SwapIndexes(GraphIndex a, GraphIndex b) noexcept
1004  {
1005      Assume(a < m_entries.size());
1006      Assume(b < m_entries.size());
1007      // Swap the Entry objects.
1008      std::swap(m_entries[a], m_entries[b]);
1009      // Iterate over both objects.
1010      for (int i = 0; i < 2; ++i) {
1011          GraphIndex idx = i ? b : a;
1012          Entry& entry = m_entries[idx];
1013          // Update linked Ref, if any exists.
1014          if (entry.m_ref) GetRefIndex(*entry.m_ref) = idx;
1015          // Update the locators for both levels. The rest of the Entry information will not change,
1016          // so no need to invoke Cluster::Updated().
1017          for (int level = 0; level < MAX_LEVELS; ++level) {
1018              Locator& locator = entry.m_locator[level];
1019              if (locator.IsPresent()) {
1020                  locator.cluster->UpdateMapping(locator.index, idx);
1021              }
1022          }
1023      }
1024  }
1025  
1026  void TxGraphImpl::Compact() noexcept
1027  {
1028      // We cannot compact while any to-be-applied operations or staged removals remain as we'd need
1029      // to rewrite them. It is easier to delay the compaction until they have been applied.
1030      if (!m_main_clusterset.m_deps_to_add.empty()) return;
1031      if (!m_main_clusterset.m_to_remove.empty()) return;
1032      Assume(m_main_clusterset.m_removed.empty()); // non-staging m_removed is always empty
1033      if (m_staging_clusterset.has_value()) {
1034          if (!m_staging_clusterset->m_deps_to_add.empty()) return;
1035          if (!m_staging_clusterset->m_to_remove.empty()) return;
1036          if (!m_staging_clusterset->m_removed.empty()) return;
1037      }
1038  
1039      // Sort the GraphIndexes that need to be cleaned up. They are sorted in reverse, so the last
1040      // ones get processed first. This means earlier-processed GraphIndexes will not cause moving of
1041      // later-processed ones during the "swap with end of m_entries" step below (which might
1042      // invalidate them).
1043      std::sort(m_unlinked.begin(), m_unlinked.end(), std::greater{});
1044  
1045      auto last = GraphIndex(-1);
1046      for (GraphIndex idx : m_unlinked) {
1047          // m_unlinked should never contain the same GraphIndex twice (the code below would fail
1048          // if so, because GraphIndexes get invalidated by removing them).
1049          Assume(idx != last);
1050          last = idx;
1051  
1052          // Make sure the entry is unlinked.
1053          Entry& entry = m_entries[idx];
1054          Assume(entry.m_ref == nullptr);
1055          // Make sure the entry does not occur in the graph.
1056          for (int level = 0; level < MAX_LEVELS; ++level) {
1057              Assume(!entry.m_locator[level].IsPresent());
1058          }
1059  
1060          // Move the entry to the end.
1061          if (idx != m_entries.size() - 1) SwapIndexes(idx, m_entries.size() - 1);
1062          // Drop the entry for idx, now that it is at the end.
1063          m_entries.pop_back();
1064      }
1065      m_unlinked.clear();
1066  }
1067  
1068  void TxGraphImpl::Split(Cluster& cluster) noexcept
1069  {
1070      // To split a Cluster, first make sure all removals are applied (as we might need to split
1071      // again afterwards otherwise).
1072      ApplyRemovals(cluster.m_level);
1073      bool del = cluster.Split(*this);
1074      if (del) {
1075          // Cluster::Split reports whether the Cluster is to be deleted.
1076          DeleteCluster(cluster);
1077      }
1078  }
1079  
1080  void TxGraphImpl::SplitAll(int up_to_level) noexcept
1081  {
1082      Assume(up_to_level >= 0 && up_to_level <= GetTopLevel());
1083      // Before splitting all Cluster, first make sure all removals are applied.
1084      ApplyRemovals(up_to_level);
1085      for (int level = 0; level <= up_to_level; ++level) {
1086          for (auto quality : {QualityLevel::NEEDS_SPLIT, QualityLevel::NEEDS_SPLIT_ACCEPTABLE}) {
1087              auto& queue = GetClusterSet(level).m_clusters[int(quality)];
1088              while (!queue.empty()) {
1089                  Split(*queue.back().get());
1090              }
1091          }
1092      }
1093  }
1094  
1095  void TxGraphImpl::GroupClusters(int level) noexcept
1096  {
1097      auto& clusterset = GetClusterSet(level);
1098      // If the groupings have been computed already, nothing is left to be done.
1099      if (clusterset.m_group_data.has_value()) return;
1100  
1101      // Before computing which Clusters need to be merged together, first apply all removals and
1102      // split the Clusters into connected components. If we would group first, we might end up
1103      // with inefficient and/or oversized Clusters which just end up being split again anyway.
1104      SplitAll(level);
1105  
1106      /** Annotated clusters: an entry for each Cluster, together with the sequence number for the
1107       *  representative for the partition it is in (initially its own, later that of the
1108       *  to-be-merged group). */
1109      std::vector<std::pair<Cluster*, uint64_t>> an_clusters;
1110      /** Annotated dependencies: an entry for each m_deps_to_add entry (excluding ones that apply
1111       *  to removed transactions), together with the sequence number of the representative root of
1112       *  Clusters it applies to (initially that of the child Cluster, later that of the
1113       *  to-be-merged group). */
1114      std::vector<std::pair<std::pair<GraphIndex, GraphIndex>, uint64_t>> an_deps;
1115  
1116      // Construct a an_clusters entry for every parent and child in the to-be-applied dependencies,
1117      // and an an_deps entry for each dependency to be applied.
1118      an_deps.reserve(clusterset.m_deps_to_add.size());
1119      for (const auto& [par, chl] : clusterset.m_deps_to_add) {
1120          auto par_cluster = FindCluster(par, level);
1121          auto chl_cluster = FindCluster(chl, level);
1122          // Skip dependencies for which the parent or child transaction is removed.
1123          if (par_cluster == nullptr || chl_cluster == nullptr) continue;
1124          an_clusters.emplace_back(par_cluster, par_cluster->m_sequence);
1125          // Do not include a duplicate when parent and child are identical, as it'll be removed
1126          // below anyway.
1127          if (chl_cluster != par_cluster) an_clusters.emplace_back(chl_cluster, chl_cluster->m_sequence);
1128          // Add entry to an_deps, using the child sequence number.
1129          an_deps.emplace_back(std::pair{par, chl}, chl_cluster->m_sequence);
1130      }
1131      // Sort and deduplicate an_clusters, so we end up with a sorted list of all involved Clusters
1132      // to which dependencies apply.
1133      std::sort(an_clusters.begin(), an_clusters.end(), [](auto& a, auto& b) noexcept { return a.second < b.second; });
1134      an_clusters.erase(std::unique(an_clusters.begin(), an_clusters.end()), an_clusters.end());
1135      // Sort an_deps by applying the same order to the involved child cluster.
1136      std::sort(an_deps.begin(), an_deps.end(), [&](auto& a, auto& b) noexcept { return a.second < b.second; });
1137  
1138      // Run the union-find algorithm to to find partitions of the input Clusters which need to be
1139      // grouped together. See https://en.wikipedia.org/wiki/Disjoint-set_data_structure.
1140      {
1141          /** Each PartitionData entry contains information about a single input Cluster. */
1142          struct PartitionData
1143          {
1144              /** The sequence number of the cluster this holds information for. */
1145              uint64_t sequence;
1146              /** All PartitionData entries belonging to the same partition are organized in a tree.
1147               *  Each element points to its parent, or to itself if it is the root. The root is then
1148               *  a representative for the entire tree, and can be found by walking upwards from any
1149               *  element. */
1150              PartitionData* parent;
1151              /** (only if this is a root, so when parent == this) An upper bound on the height of
1152               *  tree for this partition. */
1153              unsigned rank;
1154          };
1155          /** Information about each input Cluster. Sorted by Cluster::m_sequence. */
1156          std::vector<PartitionData> partition_data;
1157  
1158          /** Given a Cluster, find its corresponding PartitionData. */
1159          auto locate_fn = [&](uint64_t sequence) noexcept -> PartitionData* {
1160              auto it = std::lower_bound(partition_data.begin(), partition_data.end(), sequence,
1161                                         [](auto& a, uint64_t seq) noexcept { return a.sequence < seq; });
1162              Assume(it != partition_data.end());
1163              Assume(it->sequence == sequence);
1164              return &*it;
1165          };
1166  
1167          /** Given a PartitionData, find the root of the tree it is in (its representative). */
1168          static constexpr auto find_root_fn = [](PartitionData* data) noexcept -> PartitionData* {
1169              while (data->parent != data) {
1170                  // Replace pointers to parents with pointers to grandparents.
1171                  // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Finding_set_representatives.
1172                  auto par = data->parent;
1173                  data->parent = par->parent;
1174                  data = par;
1175              }
1176              return data;
1177          };
1178  
1179          /** Given two PartitionDatas, union the partitions they are in, and return their
1180           *  representative. */
1181          static constexpr auto union_fn = [](PartitionData* arg1, PartitionData* arg2) noexcept {
1182              // Find the roots of the trees, and bail out if they are already equal (which would
1183              // mean they are in the same partition already).
1184              auto rep1 = find_root_fn(arg1);
1185              auto rep2 = find_root_fn(arg2);
1186              if (rep1 == rep2) return rep1;
1187              // Pick the lower-rank root to become a child of the higher-rank one.
1188              // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Union_by_rank.
1189              if (rep1->rank < rep2->rank) std::swap(rep1, rep2);
1190              rep2->parent = rep1;
1191              rep1->rank += (rep1->rank == rep2->rank);
1192              return rep1;
1193          };
1194  
1195          // Start by initializing every Cluster as its own singleton partition.
1196          partition_data.resize(an_clusters.size());
1197          for (size_t i = 0; i < an_clusters.size(); ++i) {
1198              partition_data[i].sequence = an_clusters[i].first->m_sequence;
1199              partition_data[i].parent = &partition_data[i];
1200              partition_data[i].rank = 0;
1201          }
1202  
1203          // Run through all parent/child pairs in an_deps, and union the partitions their Clusters
1204          // are in.
1205          Cluster* last_chl_cluster{nullptr};
1206          PartitionData* last_partition{nullptr};
1207          for (const auto& [dep, _] : an_deps) {
1208              auto [par, chl] = dep;
1209              auto par_cluster = FindCluster(par, level);
1210              auto chl_cluster = FindCluster(chl, level);
1211              Assume(chl_cluster != nullptr && par_cluster != nullptr);
1212              // Nothing to do if parent and child are in the same Cluster.
1213              if (par_cluster == chl_cluster) continue;
1214              Assume(par != chl);
1215              if (chl_cluster == last_chl_cluster) {
1216                  // If the child Clusters is the same as the previous iteration, union with the
1217                  // tree they were in, avoiding the need for another lookup. Note that an_deps
1218                  // is sorted by child Cluster, so batches with the same child are expected.
1219                  last_partition = union_fn(locate_fn(par_cluster->m_sequence), last_partition);
1220              } else {
1221                  last_chl_cluster = chl_cluster;
1222                  last_partition = union_fn(locate_fn(par_cluster->m_sequence), locate_fn(chl_cluster->m_sequence));
1223              }
1224          }
1225  
1226          // Update the sequence numbers in an_clusters and an_deps to be those of the partition
1227          // representative.
1228          auto deps_it = an_deps.begin();
1229          for (size_t i = 0; i < partition_data.size(); ++i) {
1230              auto& data = partition_data[i];
1231              // Find the sequence of the representative of the partition Cluster i is in, and store
1232              // it with the Cluster.
1233              auto rep_seq = find_root_fn(&data)->sequence;
1234              an_clusters[i].second = rep_seq;
1235              // Find all dependencies whose child Cluster is Cluster i, and annotate them with rep.
1236              while (deps_it != an_deps.end()) {
1237                  auto [par, chl] = deps_it->first;
1238                  auto chl_cluster = FindCluster(chl, level);
1239                  Assume(chl_cluster != nullptr);
1240                  if (chl_cluster->m_sequence > data.sequence) break;
1241                  deps_it->second = rep_seq;
1242                  ++deps_it;
1243              }
1244          }
1245      }
1246  
1247      // Sort both an_clusters and an_deps by sequence number of the representative of the
1248      // partition they are in, grouping all those applying to the same partition together.
1249      std::sort(an_deps.begin(), an_deps.end(), [](auto& a, auto& b) noexcept { return a.second < b.second; });
1250      std::sort(an_clusters.begin(), an_clusters.end(), [](auto& a, auto& b) noexcept { return a.second < b.second; });
1251  
1252      // Translate the resulting cluster groups to the m_group_data structure, and the dependencies
1253      // back to m_deps_to_add.
1254      clusterset.m_group_data = GroupData{};
1255      clusterset.m_group_data->m_group_clusters.reserve(an_clusters.size());
1256      clusterset.m_group_data->m_group_oversized = false;
1257      clusterset.m_deps_to_add.clear();
1258      clusterset.m_deps_to_add.reserve(an_deps.size());
1259      auto an_deps_it = an_deps.begin();
1260      auto an_clusters_it = an_clusters.begin();
1261      while (an_clusters_it != an_clusters.end()) {
1262          // Process all clusters/dependencies belonging to the partition with representative rep.
1263          auto rep = an_clusters_it->second;
1264          // Create and initialize a new GroupData entry for the partition.
1265          auto& new_entry = clusterset.m_group_data->m_groups.emplace_back();
1266          new_entry.m_cluster_offset = clusterset.m_group_data->m_group_clusters.size();
1267          new_entry.m_cluster_count = 0;
1268          new_entry.m_deps_offset = clusterset.m_deps_to_add.size();
1269          new_entry.m_deps_count = 0;
1270          uint32_t total_count{0};
1271          // Add all its clusters to it (copying those from an_clusters to m_group_clusters).
1272          while (an_clusters_it != an_clusters.end() && an_clusters_it->second == rep) {
1273              clusterset.m_group_data->m_group_clusters.push_back(an_clusters_it->first);
1274              total_count += an_clusters_it->first->GetTxCount();
1275              ++an_clusters_it;
1276              ++new_entry.m_cluster_count;
1277          }
1278          // Add all its dependencies to it (copying those back from an_deps to m_deps_to_add).
1279          while (an_deps_it != an_deps.end() && an_deps_it->second == rep) {
1280              clusterset.m_deps_to_add.push_back(an_deps_it->first);
1281              ++an_deps_it;
1282              ++new_entry.m_deps_count;
1283          }
1284          // Detect oversizedness.
1285          if (total_count > m_max_cluster_count) {
1286              clusterset.m_group_data->m_group_oversized = true;
1287          }
1288      }
1289      Assume(an_deps_it == an_deps.end());
1290      Assume(an_clusters_it == an_clusters.end());
1291      clusterset.m_oversized = clusterset.m_group_data->m_group_oversized;
1292      Compact();
1293  }
1294  
1295  void TxGraphImpl::Merge(std::span<Cluster*> to_merge) noexcept
1296  {
1297      Assume(!to_merge.empty());
1298      // Nothing to do if a group consists of just a single Cluster.
1299      if (to_merge.size() == 1) return;
1300  
1301      // Move the largest Cluster to the front of to_merge. As all transactions in other to-be-merged
1302      // Clusters will be moved to that one, putting the largest one first minimizes the number of
1303      // moves.
1304      size_t max_size_pos{0};
1305      DepGraphIndex max_size = to_merge[max_size_pos]->GetTxCount();
1306      for (size_t i = 1; i < to_merge.size(); ++i) {
1307          DepGraphIndex size = to_merge[i]->GetTxCount();
1308          if (size > max_size) {
1309              max_size_pos = i;
1310              max_size = size;
1311          }
1312      }
1313      if (max_size_pos != 0) std::swap(to_merge[0], to_merge[max_size_pos]);
1314  
1315      // Merge all further Clusters in the group into the first one, and delete them.
1316      for (size_t i = 1; i < to_merge.size(); ++i) {
1317          to_merge[0]->Merge(*this, *to_merge[i]);
1318          DeleteCluster(*to_merge[i]);
1319      }
1320  }
1321  
1322  void TxGraphImpl::ApplyDependencies(int level) noexcept
1323  {
1324      auto& clusterset = GetClusterSet(level);
1325      // Do not bother computing groups if we already know the result will be oversized.
1326      if (clusterset.m_oversized == true) return;
1327      // Compute the groups of to-be-merged Clusters (which also applies all removals, and splits).
1328      GroupClusters(level);
1329      Assume(clusterset.m_group_data.has_value());
1330      // Nothing to do if there are no dependencies to be added.
1331      if (clusterset.m_deps_to_add.empty()) return;
1332      // Dependencies cannot be applied if it would result in oversized clusters.
1333      if (clusterset.m_group_data->m_group_oversized) return;
1334  
1335      // For each group of to-be-merged Clusters.
1336      for (const auto& group_entry : clusterset.m_group_data->m_groups) {
1337          auto cluster_span = std::span{clusterset.m_group_data->m_group_clusters}
1338                                  .subspan(group_entry.m_cluster_offset, group_entry.m_cluster_count);
1339          // Pull in all the Clusters that contain dependencies.
1340          if (level == 1) {
1341              for (Cluster*& cluster : cluster_span) {
1342                  cluster = PullIn(cluster);
1343              }
1344          }
1345          // Invoke Merge() to merge them into a single Cluster.
1346          Merge(cluster_span);
1347          // Actually apply all to-be-added dependencies (all parents and children from this grouping
1348          // belong to the same Cluster at this point because of the merging above).
1349          auto deps_span = std::span{clusterset.m_deps_to_add}
1350                               .subspan(group_entry.m_deps_offset, group_entry.m_deps_count);
1351          Assume(!deps_span.empty());
1352          const auto& loc = m_entries[deps_span[0].second].m_locator[level];
1353          Assume(loc.IsPresent());
1354          loc.cluster->ApplyDependencies(*this, deps_span);
1355      }
1356  
1357      // Wipe the list of to-be-added dependencies now that they are applied.
1358      clusterset.m_deps_to_add.clear();
1359      Compact();
1360      // Also no further Cluster mergings are needed (note that we clear, but don't set to
1361      // std::nullopt, as that would imply the groupings are unknown).
1362      clusterset.m_group_data = GroupData{};
1363  }
1364  
1365  void Cluster::Relinearize(TxGraphImpl& graph, uint64_t max_iters) noexcept
1366  {
1367      // We can only relinearize Clusters that do not need splitting.
1368      Assume(!NeedsSplitting());
1369      // No work is required for Clusters which are already optimally linearized.
1370      if (IsOptimal()) return;
1371      // Invoke the actual linearization algorithm (passing in the existing one).
1372      uint64_t rng_seed = graph.m_rng.rand64();
1373      auto [linearization, optimal] = Linearize(m_depgraph, max_iters, rng_seed, m_linearization);
1374      // Postlinearize if the result isn't optimal already. This guarantees (among other things)
1375      // that the chunks of the resulting linearization are all connected.
1376      if (!optimal) PostLinearize(m_depgraph, linearization);
1377      // Update the linearization.
1378      m_linearization = std::move(linearization);
1379      // Update the Cluster's quality.
1380      auto new_quality = optimal ? QualityLevel::OPTIMAL : QualityLevel::ACCEPTABLE;
1381      graph.SetClusterQuality(m_level, m_quality, m_setindex, new_quality);
1382      // Update the Entry objects.
1383      Updated(graph);
1384  }
1385  
1386  void TxGraphImpl::MakeAcceptable(Cluster& cluster) noexcept
1387  {
1388      // Relinearize the Cluster if needed.
1389      if (!cluster.NeedsSplitting() && !cluster.IsAcceptable()) {
1390          cluster.Relinearize(*this, 10000);
1391      }
1392  }
1393  
1394  void TxGraphImpl::MakeAllAcceptable(int level) noexcept
1395  {
1396      ApplyDependencies(level);
1397      auto& clusterset = GetClusterSet(level);
1398      if (clusterset.m_oversized == true) return;
1399      auto& queue = clusterset.m_clusters[int(QualityLevel::NEEDS_RELINEARIZE)];
1400      while (!queue.empty()) {
1401          MakeAcceptable(*queue.back().get());
1402      }
1403  }
1404  
1405  Cluster::Cluster(uint64_t sequence) noexcept : m_sequence{sequence} {}
1406  
1407  Cluster::Cluster(uint64_t sequence, TxGraphImpl& graph, const FeePerWeight& feerate, GraphIndex graph_index) noexcept :
1408      m_sequence{sequence}
1409  {
1410      // Create a new transaction in the DepGraph, and remember its position in m_mapping.
1411      auto cluster_idx = m_depgraph.AddTransaction(feerate);
1412      m_mapping.push_back(graph_index);
1413      m_linearization.push_back(cluster_idx);
1414  }
1415  
1416  TxGraph::Ref TxGraphImpl::AddTransaction(const FeePerWeight& feerate) noexcept
1417  {
1418      // Construct a new Ref.
1419      Ref ret;
1420      // Construct a new Entry, and link it with the Ref.
1421      auto idx = m_entries.size();
1422      m_entries.emplace_back();
1423      auto& entry = m_entries.back();
1424      entry.m_ref = &ret;
1425      GetRefGraph(ret) = this;
1426      GetRefIndex(ret) = idx;
1427      // Construct a new singleton Cluster (which is necessarily optimally linearized).
1428      auto cluster = std::make_unique<Cluster>(m_next_sequence_counter++, *this, feerate, idx);
1429      auto cluster_ptr = cluster.get();
1430      int level = GetTopLevel();
1431      auto& clusterset = GetClusterSet(level);
1432      InsertCluster(level, std::move(cluster), QualityLevel::OPTIMAL);
1433      cluster_ptr->Updated(*this);
1434      ++clusterset.m_txcount;
1435      // Return the Ref.
1436      return ret;
1437  }
1438  
1439  void TxGraphImpl::RemoveTransaction(const Ref& arg) noexcept
1440  {
1441      // Don't do anything if the Ref is empty (which may be indicative of the transaction already
1442      // having been removed).
1443      if (GetRefGraph(arg) == nullptr) return;
1444      Assume(GetRefGraph(arg) == this);
1445      // Find the Cluster the transaction is in, and stop if it isn't in any.
1446      int level = GetTopLevel();
1447      auto cluster = FindCluster(GetRefIndex(arg), level);
1448      if (cluster == nullptr) return;
1449      // Remember that the transaction is to be removed.
1450      auto& clusterset = GetClusterSet(level);
1451      clusterset.m_to_remove.push_back(GetRefIndex(arg));
1452      // Wipe m_group_data (as it will need to be recomputed).
1453      clusterset.m_group_data.reset();
1454      if (clusterset.m_oversized == true) clusterset.m_oversized = std::nullopt;
1455  }
1456  
1457  void TxGraphImpl::AddDependency(const Ref& parent, const Ref& child) noexcept
1458  {
1459      // Don't do anything if either Ref is empty (which may be indicative of it having already been
1460      // removed).
1461      if (GetRefGraph(parent) == nullptr || GetRefGraph(child) == nullptr) return;
1462      Assume(GetRefGraph(parent) == this && GetRefGraph(child) == this);
1463      // Don't do anything if this is a dependency on self.
1464      if (GetRefIndex(parent) == GetRefIndex(child)) return;
1465      // Find the Cluster the parent and child transaction are in, and stop if either appears to be
1466      // already removed.
1467      int level = GetTopLevel();
1468      auto par_cluster = FindCluster(GetRefIndex(parent), level);
1469      if (par_cluster == nullptr) return;
1470      auto chl_cluster = FindCluster(GetRefIndex(child), level);
1471      if (chl_cluster == nullptr) return;
1472      // Remember that this dependency is to be applied.
1473      auto& clusterset = GetClusterSet(level);
1474      clusterset.m_deps_to_add.emplace_back(GetRefIndex(parent), GetRefIndex(child));
1475      // Wipe m_group_data (as it will need to be recomputed).
1476      clusterset.m_group_data.reset();
1477      if (clusterset.m_oversized == false) clusterset.m_oversized = std::nullopt;
1478  }
1479  
1480  bool TxGraphImpl::Exists(const Ref& arg, bool main_only) noexcept
1481  {
1482      if (GetRefGraph(arg) == nullptr) return false;
1483      Assume(GetRefGraph(arg) == this);
1484      size_t level = GetSpecifiedLevel(main_only);
1485      // Make sure the transaction isn't scheduled for removal.
1486      ApplyRemovals(level);
1487      auto cluster = FindCluster(GetRefIndex(arg), level);
1488      return cluster != nullptr;
1489  }
1490  
1491  void Cluster::GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
1492  {
1493      /** The union of all ancestors to be returned. */
1494      SetType ancestors_union;
1495      // Process elements from the front of args, as long as they apply.
1496      while (!args.empty()) {
1497          if (args.front().first != this) break;
1498          ancestors_union |= m_depgraph.Ancestors(args.front().second);
1499          args = args.subspan(1);
1500      }
1501      Assume(ancestors_union.Any());
1502      // Translate all ancestors (in arbitrary order) to Refs (if they have any), and return them.
1503      for (auto idx : ancestors_union) {
1504          const auto& entry = graph.m_entries[m_mapping[idx]];
1505          Assume(entry.m_ref != nullptr);
1506          output.push_back(entry.m_ref);
1507      }
1508  }
1509  
1510  void Cluster::GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
1511  {
1512      /** The union of all descendants to be returned. */
1513      SetType descendants_union;
1514      // Process elements from the front of args, as long as they apply.
1515      while (!args.empty()) {
1516          if (args.front().first != this) break;
1517          descendants_union |= m_depgraph.Descendants(args.front().second);
1518          args = args.subspan(1);
1519      }
1520      Assume(descendants_union.Any());
1521      // Translate all descendants (in arbitrary order) to Refs (if they have any), and return them.
1522      for (auto idx : descendants_union) {
1523          const auto& entry = graph.m_entries[m_mapping[idx]];
1524          Assume(entry.m_ref != nullptr);
1525          output.push_back(entry.m_ref);
1526      }
1527  }
1528  
1529  std::vector<TxGraph::Ref*> Cluster::GetClusterRefs(const TxGraphImpl& graph) noexcept
1530  {
1531      std::vector<TxGraph::Ref*> ret;
1532      ret.reserve(m_linearization.size());
1533      // Translate all transactions in the Cluster (in linearization order) to Refs.
1534      for (auto idx : m_linearization) {
1535          const auto& entry = graph.m_entries[m_mapping[idx]];
1536          Assume(entry.m_ref != nullptr);
1537          ret.push_back(entry.m_ref);
1538      }
1539      return ret;
1540  }
1541  
1542  FeePerWeight Cluster::GetIndividualFeerate(DepGraphIndex idx) noexcept
1543  {
1544      return FeePerWeight::FromFeeFrac(m_depgraph.FeeRate(idx));
1545  }
1546  
1547  void Cluster::MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept
1548  {
1549      Assume(m_level == 1);
1550      // Mark all transactions of a Cluster missing, needed when aborting staging, so that the
1551      // corresponding Locators don't retain references into aborted Clusters.
1552      for (auto ci : m_linearization) {
1553          GraphIndex idx = m_mapping[ci];
1554          auto& entry = graph.m_entries[idx];
1555          entry.m_locator[1].SetMissing();
1556      }
1557  }
1558  
1559  std::vector<TxGraph::Ref*> TxGraphImpl::GetAncestors(const Ref& arg, bool main_only) noexcept
1560  {
1561      // Return the empty vector if the Ref is empty.
1562      if (GetRefGraph(arg) == nullptr) return {};
1563      Assume(GetRefGraph(arg) == this);
1564      // Apply all removals and dependencies, as the result might be incorrect otherwise.
1565      size_t level = GetSpecifiedLevel(main_only);
1566      ApplyDependencies(level);
1567      // Ancestry cannot be known if unapplied dependencies remain.
1568      Assume(GetClusterSet(level).m_deps_to_add.empty());
1569      // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
1570      auto cluster = FindCluster(GetRefIndex(arg), level);
1571      if (cluster == nullptr) return {};
1572      // Dispatch to the Cluster.
1573      std::pair<Cluster*, DepGraphIndex> match = {cluster, m_entries[GetRefIndex(arg)].m_locator[cluster->m_level].index};
1574      auto matches = std::span(&match, 1);
1575      std::vector<TxGraph::Ref*> ret;
1576      cluster->GetAncestorRefs(*this, matches, ret);
1577      return ret;
1578  }
1579  
1580  std::vector<TxGraph::Ref*> TxGraphImpl::GetDescendants(const Ref& arg, bool main_only) noexcept
1581  {
1582      // Return the empty vector if the Ref is empty.
1583      if (GetRefGraph(arg) == nullptr) return {};
1584      Assume(GetRefGraph(arg) == this);
1585      // Apply all removals and dependencies, as the result might be incorrect otherwise.
1586      size_t level = GetSpecifiedLevel(main_only);
1587      ApplyDependencies(level);
1588      // Ancestry cannot be known if unapplied dependencies remain.
1589      Assume(GetClusterSet(level).m_deps_to_add.empty());
1590      // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
1591      auto cluster = FindCluster(GetRefIndex(arg), level);
1592      if (cluster == nullptr) return {};
1593      // Dispatch to the Cluster.
1594      std::pair<Cluster*, DepGraphIndex> match = {cluster, m_entries[GetRefIndex(arg)].m_locator[cluster->m_level].index};
1595      auto matches = std::span(&match, 1);
1596      std::vector<TxGraph::Ref*> ret;
1597      cluster->GetDescendantRefs(*this, matches, ret);
1598      return ret;
1599  }
1600  
1601  std::vector<TxGraph::Ref*> TxGraphImpl::GetAncestorsUnion(std::span<const Ref* const> args, bool main_only) noexcept
1602  {
1603      // Apply all dependencies, as the result might be incorrect otherwise.
1604      size_t level = GetSpecifiedLevel(main_only);
1605      ApplyDependencies(level);
1606      // Ancestry cannot be known if unapplied dependencies remain.
1607      Assume(GetClusterSet(level).m_deps_to_add.empty());
1608  
1609      // Translate args to matches.
1610      std::vector<std::pair<Cluster*, DepGraphIndex>> matches;
1611      matches.reserve(args.size());
1612      for (auto arg : args) {
1613          Assume(arg);
1614          // Skip empty Refs.
1615          if (GetRefGraph(*arg) == nullptr) continue;
1616          Assume(GetRefGraph(*arg) == this);
1617          // Find the Cluster the argument is in, and skip if none is found.
1618          auto cluster = FindCluster(GetRefIndex(*arg), level);
1619          if (cluster == nullptr) continue;
1620          // Append to matches.
1621          matches.emplace_back(cluster, m_entries[GetRefIndex(*arg)].m_locator[cluster->m_level].index);
1622      }
1623      // Group by Cluster.
1624      std::sort(matches.begin(), matches.end(), [](auto& a, auto& b) noexcept { return CompareClusters(a.first, b.first) < 0; });
1625      // Dispatch to the Clusters.
1626      std::span match_span(matches);
1627      std::vector<TxGraph::Ref*> ret;
1628      while (!match_span.empty()) {
1629          match_span.front().first->GetAncestorRefs(*this, match_span, ret);
1630      }
1631      return ret;
1632  }
1633  
1634  std::vector<TxGraph::Ref*> TxGraphImpl::GetDescendantsUnion(std::span<const Ref* const> args, bool main_only) noexcept
1635  {
1636      // Apply all dependencies, as the result might be incorrect otherwise.
1637      size_t level = GetSpecifiedLevel(main_only);
1638      ApplyDependencies(level);
1639      // Ancestry cannot be known if unapplied dependencies remain.
1640      Assume(GetClusterSet(level).m_deps_to_add.empty());
1641  
1642      // Translate args to matches.
1643      std::vector<std::pair<Cluster*, DepGraphIndex>> matches;
1644      matches.reserve(args.size());
1645      for (auto arg : args) {
1646          Assume(arg);
1647          // Skip empty Refs.
1648          if (GetRefGraph(*arg) == nullptr) continue;
1649          Assume(GetRefGraph(*arg) == this);
1650          // Find the Cluster the argument is in, and skip if none is found.
1651          auto cluster = FindCluster(GetRefIndex(*arg), level);
1652          if (cluster == nullptr) continue;
1653          // Append to matches.
1654          matches.emplace_back(cluster, m_entries[GetRefIndex(*arg)].m_locator[cluster->m_level].index);
1655      }
1656      // Group by Cluster.
1657      std::sort(matches.begin(), matches.end(), [](auto& a, auto& b) noexcept { return CompareClusters(a.first, b.first) < 0; });
1658      // Dispatch to the Clusters.
1659      std::span match_span(matches);
1660      std::vector<TxGraph::Ref*> ret;
1661      while (!match_span.empty()) {
1662          match_span.front().first->GetDescendantRefs(*this, match_span, ret);
1663      }
1664      return ret;
1665  }
1666  
1667  std::vector<TxGraph::Ref*> TxGraphImpl::GetCluster(const Ref& arg, bool main_only) noexcept
1668  {
1669      // Return the empty vector if the Ref is empty (which may be indicative of the transaction
1670      // having been removed already.
1671      if (GetRefGraph(arg) == nullptr) return {};
1672      Assume(GetRefGraph(arg) == this);
1673      // Apply all removals and dependencies, as the result might be incorrect otherwise.
1674      size_t level = GetSpecifiedLevel(main_only);
1675      ApplyDependencies(level);
1676      // Cluster linearization cannot be known if unapplied dependencies remain.
1677      Assume(GetClusterSet(level).m_deps_to_add.empty());
1678      // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
1679      auto cluster = FindCluster(GetRefIndex(arg), level);
1680      if (cluster == nullptr) return {};
1681      // Make sure the Cluster has an acceptable quality level, and then dispatch to it.
1682      MakeAcceptable(*cluster);
1683      return cluster->GetClusterRefs(*this);
1684  }
1685  
1686  TxGraph::GraphIndex TxGraphImpl::GetTransactionCount(bool main_only) noexcept
1687  {
1688      size_t level = GetSpecifiedLevel(main_only);
1689      ApplyRemovals(level);
1690      return GetClusterSet(level).m_txcount;
1691  }
1692  
1693  FeePerWeight TxGraphImpl::GetIndividualFeerate(const Ref& arg) noexcept
1694  {
1695      // Return the empty FeePerWeight if the passed Ref is empty.
1696      if (GetRefGraph(arg) == nullptr) return {};
1697      Assume(GetRefGraph(arg) == this);
1698      // Find the cluster the argument is in (the level does not matter as individual feerates will
1699      // be identical if it occurs in both), and return the empty FeePerWeight if it isn't in any.
1700      Cluster* cluster{nullptr};
1701      for (int level = 0; level <= GetTopLevel(); ++level) {
1702          // Apply removals, so that we can correctly report FeePerWeight{} for non-existing
1703          // transactions.
1704          ApplyRemovals(level);
1705          if (m_entries[GetRefIndex(arg)].m_locator[level].IsPresent()) {
1706              cluster = m_entries[GetRefIndex(arg)].m_locator[level].cluster;
1707              break;
1708          }
1709      }
1710      if (cluster == nullptr) return {};
1711      // Dispatch to the Cluster.
1712      return cluster->GetIndividualFeerate(m_entries[GetRefIndex(arg)].m_locator[cluster->m_level].index);
1713  }
1714  
1715  FeePerWeight TxGraphImpl::GetMainChunkFeerate(const Ref& arg) noexcept
1716  {
1717      // Return the empty FeePerWeight if the passed Ref is empty.
1718      if (GetRefGraph(arg) == nullptr) return {};
1719      Assume(GetRefGraph(arg) == this);
1720      // Apply all removals and dependencies, as the result might be inaccurate otherwise.
1721      ApplyDependencies(/*level=*/0);
1722      // Chunk feerates cannot be accurately known if unapplied dependencies remain.
1723      Assume(m_main_clusterset.m_deps_to_add.empty());
1724      // Find the cluster the argument is in, and return the empty FeePerWeight if it isn't in any.
1725      auto cluster = FindCluster(GetRefIndex(arg), 0);
1726      if (cluster == nullptr) return {};
1727      // Make sure the Cluster has an acceptable quality level, and then return the transaction's
1728      // chunk feerate.
1729      MakeAcceptable(*cluster);
1730      const auto& entry = m_entries[GetRefIndex(arg)];
1731      return entry.m_main_chunk_feerate;
1732  }
1733  
1734  bool TxGraphImpl::IsOversized(bool main_only) noexcept
1735  {
1736      size_t level = GetSpecifiedLevel(main_only);
1737      auto& clusterset = GetClusterSet(level);
1738      if (clusterset.m_oversized.has_value()) {
1739          // Return cached value if known.
1740          return *clusterset.m_oversized;
1741      }
1742      // Find which Clusters will need to be merged together, as that is where the oversize
1743      // property is assessed.
1744      GroupClusters(level);
1745      Assume(clusterset.m_group_data.has_value());
1746      clusterset.m_oversized = clusterset.m_group_data->m_group_oversized;
1747      return *clusterset.m_oversized;
1748  }
1749  
1750  void TxGraphImpl::StartStaging() noexcept
1751  {
1752      // Staging cannot already exist.
1753      Assume(!m_staging_clusterset.has_value());
1754      // Apply all remaining dependencies in main before creating a staging graph. Once staging
1755      // exists, we cannot merge Clusters anymore (because of interference with Clusters being
1756      // pulled into staging), so to make sure all inspectors are available (if not oversized), do
1757      // all merging work now. Call SplitAll() first, so that even if ApplyDependencies does not do
1758      // any thing due to knowing the result is oversized, splitting is still performed.
1759      SplitAll(0);
1760      ApplyDependencies(0);
1761      // Construct the staging ClusterSet.
1762      m_staging_clusterset.emplace();
1763      // Copy statistics, precomputed data, and to-be-applied dependencies (only if oversized) to
1764      // the new graph. To-be-applied removals will always be empty at this point.
1765      m_staging_clusterset->m_txcount = m_main_clusterset.m_txcount;
1766      m_staging_clusterset->m_deps_to_add = m_main_clusterset.m_deps_to_add;
1767      m_staging_clusterset->m_group_data = m_main_clusterset.m_group_data;
1768      m_staging_clusterset->m_oversized = m_main_clusterset.m_oversized;
1769      Assume(m_staging_clusterset->m_oversized.has_value());
1770  }
1771  
1772  void TxGraphImpl::AbortStaging() noexcept
1773  {
1774      // Staging must exist.
1775      Assume(m_staging_clusterset.has_value());
1776      // Mark all removed transactions as Missing (so the staging locator for these transactions
1777      // can be reused if another staging is created).
1778      for (auto idx : m_staging_clusterset->m_removed) {
1779          m_entries[idx].m_locator[1].SetMissing();
1780      }
1781      // Do the same with the non-removed transactions in staging Clusters.
1782      for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
1783          for (auto& cluster : m_staging_clusterset->m_clusters[quality]) {
1784              cluster->MakeStagingTransactionsMissing(*this);
1785          }
1786      }
1787      // Destroy the staging ClusterSet.
1788      m_staging_clusterset.reset();
1789      Compact();
1790      if (!m_main_clusterset.m_group_data.has_value()) {
1791          // In case m_oversized in main was kept after a Ref destruction while staging exists, we
1792          // need to re-evaluate m_oversized now.
1793          m_main_clusterset.m_oversized = std::nullopt;
1794      }
1795  }
1796  
1797  void TxGraphImpl::CommitStaging() noexcept
1798  {
1799      // Staging must exist.
1800      Assume(m_staging_clusterset.has_value());
1801      // Delete all conflicting Clusters in main, to make place for moving the staging ones
1802      // there. All of these have been copied to staging in PullIn().
1803      auto conflicts = GetConflicts();
1804      for (Cluster* conflict : conflicts) {
1805          conflict->Clear(*this);
1806          DeleteCluster(*conflict);
1807      }
1808      // Mark the removed transactions as Missing (so the staging locator for these transactions
1809      // can be reused if another staging is created).
1810      for (auto idx : m_staging_clusterset->m_removed) {
1811          m_entries[idx].m_locator[1].SetMissing();
1812      }
1813      // Then move all Clusters in staging to main.
1814      for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
1815          auto& stage_sets = m_staging_clusterset->m_clusters[quality];
1816          while (!stage_sets.empty()) {
1817              stage_sets.back()->MoveToMain(*this);
1818          }
1819      }
1820      // Move all statistics, precomputed data, and to-be-applied removals and dependencies.
1821      m_main_clusterset.m_deps_to_add = std::move(m_staging_clusterset->m_deps_to_add);
1822      m_main_clusterset.m_to_remove = std::move(m_staging_clusterset->m_to_remove);
1823      m_main_clusterset.m_group_data = std::move(m_staging_clusterset->m_group_data);
1824      m_main_clusterset.m_oversized = std::move(m_staging_clusterset->m_oversized);
1825      m_main_clusterset.m_txcount = std::move(m_staging_clusterset->m_txcount);
1826      // Delete the old staging graph, after all its information was moved to main.
1827      m_staging_clusterset.reset();
1828      Compact();
1829  }
1830  
1831  void Cluster::SetFee(TxGraphImpl& graph, DepGraphIndex idx, int64_t fee) noexcept
1832  {
1833      // Make sure the specified DepGraphIndex exists in this Cluster.
1834      Assume(m_depgraph.Positions()[idx]);
1835      // Bail out if the fee isn't actually being changed.
1836      if (m_depgraph.FeeRate(idx).fee == fee) return;
1837      // Update the fee, remember that relinearization will be necessary, and update the Entries
1838      // in the same Cluster.
1839      m_depgraph.FeeRate(idx).fee = fee;
1840      if (!NeedsSplitting()) {
1841          graph.SetClusterQuality(m_level, m_quality, m_setindex, QualityLevel::NEEDS_RELINEARIZE);
1842      } else {
1843          graph.SetClusterQuality(m_level, m_quality, m_setindex, QualityLevel::NEEDS_SPLIT);
1844      }
1845      Updated(graph);
1846  }
1847  
1848  void TxGraphImpl::SetTransactionFee(const Ref& ref, int64_t fee) noexcept
1849  {
1850      // Don't do anything if the passed Ref is empty.
1851      if (GetRefGraph(ref) == nullptr) return;
1852      Assume(GetRefGraph(ref) == this);
1853      // Find the entry, its locator, and inform its Cluster about the new feerate, if any.
1854      auto& entry = m_entries[GetRefIndex(ref)];
1855      for (int level = 0; level < MAX_LEVELS; ++level) {
1856          auto& locator = entry.m_locator[level];
1857          if (locator.IsPresent()) {
1858              locator.cluster->SetFee(*this, locator.index, fee);
1859          }
1860      }
1861  }
1862  
1863  std::strong_ordering TxGraphImpl::CompareMainOrder(const Ref& a, const Ref& b) noexcept
1864  {
1865      // The references must not be empty.
1866      Assume(GetRefGraph(a) == this);
1867      Assume(GetRefGraph(b) == this);
1868      // Apply dependencies in main.
1869      ApplyDependencies(0);
1870      Assume(m_main_clusterset.m_deps_to_add.empty());
1871      // Make both involved Clusters acceptable, so chunk feerates are relevant.
1872      const auto& entry_a = m_entries[GetRefIndex(a)];
1873      const auto& entry_b = m_entries[GetRefIndex(b)];
1874      const auto& locator_a = entry_a.m_locator[0];
1875      const auto& locator_b = entry_b.m_locator[0];
1876      Assume(locator_a.IsPresent());
1877      Assume(locator_b.IsPresent());
1878      MakeAcceptable(*locator_a.cluster);
1879      MakeAcceptable(*locator_b.cluster);
1880      // Compare chunk feerates, and return result if it differs.
1881      auto feerate_cmp = FeeRateCompare(entry_b.m_main_chunk_feerate, entry_a.m_main_chunk_feerate);
1882      if (feerate_cmp < 0) return std::strong_ordering::less;
1883      if (feerate_cmp > 0) return std::strong_ordering::greater;
1884      // Compare Cluster* as tie-break for equal chunk feerates.
1885      if (locator_a.cluster != locator_b.cluster) {
1886          return CompareClusters(locator_a.cluster, locator_b.cluster);
1887      }
1888      // As final tie-break, compare position within cluster linearization.
1889      return entry_a.m_main_lin_index <=> entry_b.m_main_lin_index;
1890  }
1891  
1892  TxGraph::GraphIndex TxGraphImpl::CountDistinctClusters(std::span<const Ref* const> refs, bool main_only) noexcept
1893  {
1894      size_t level = GetSpecifiedLevel(main_only);
1895      ApplyDependencies(level);
1896      auto& clusterset = GetClusterSet(level);
1897      Assume(clusterset.m_deps_to_add.empty());
1898      // Build a vector of Clusters that the specified Refs occur in.
1899      std::vector<Cluster*> clusters;
1900      clusters.reserve(refs.size());
1901      for (const Ref* ref : refs) {
1902          Assume(ref);
1903          if (GetRefGraph(*ref) == nullptr) continue;
1904          Assume(GetRefGraph(*ref) == this);
1905          auto cluster = FindCluster(GetRefIndex(*ref), level);
1906          if (cluster != nullptr) clusters.push_back(cluster);
1907      }
1908      // Count the number of distinct elements in clusters.
1909      std::sort(clusters.begin(), clusters.end(), [](Cluster* a, Cluster* b) noexcept { return CompareClusters(a, b) < 0; });
1910      Cluster* last{nullptr};
1911      GraphIndex ret{0};
1912      for (Cluster* cluster : clusters) {
1913          ret += (cluster != last);
1914          last = cluster;
1915      }
1916      return ret;
1917  }
1918  
1919  void Cluster::SanityCheck(const TxGraphImpl& graph, int level) const
1920  {
1921      // There must be an m_mapping for each m_depgraph position (including holes).
1922      assert(m_depgraph.PositionRange() == m_mapping.size());
1923      // The linearization for this Cluster must contain every transaction once.
1924      assert(m_depgraph.TxCount() == m_linearization.size());
1925      // The number of transactions in a Cluster cannot exceed m_max_cluster_count.
1926      assert(m_linearization.size() <= graph.m_max_cluster_count);
1927      // The level must match the level the Cluster occurs in.
1928      assert(m_level == level);
1929      // m_quality and m_setindex are checked in TxGraphImpl::SanityCheck.
1930  
1931      // Compute the chunking of m_linearization.
1932      LinearizationChunking linchunking(m_depgraph, m_linearization);
1933  
1934      // Verify m_linearization.
1935      SetType m_done;
1936      LinearizationIndex linindex{0};
1937      assert(m_depgraph.IsAcyclic());
1938      for (auto lin_pos : m_linearization) {
1939          assert(lin_pos < m_mapping.size());
1940          const auto& entry = graph.m_entries[m_mapping[lin_pos]];
1941          // Check that the linearization is topological.
1942          m_done.Set(lin_pos);
1943          assert(m_done.IsSupersetOf(m_depgraph.Ancestors(lin_pos)));
1944          // Check that the Entry has a locator pointing back to this Cluster & position within it.
1945          assert(entry.m_locator[level].cluster == this);
1946          assert(entry.m_locator[level].index == lin_pos);
1947          // For main-level entries, check linearization position and chunk feerate.
1948          if (level == 0 && IsAcceptable()) {
1949              assert(entry.m_main_lin_index == linindex);
1950              ++linindex;
1951              if (!linchunking.GetChunk(0).transactions[lin_pos]) {
1952                  linchunking.MarkDone(linchunking.GetChunk(0).transactions);
1953              }
1954              assert(entry.m_main_chunk_feerate == linchunking.GetChunk(0).feerate);
1955              // If this Cluster has an acceptable quality level, its chunks must be connected.
1956              assert(m_depgraph.IsConnected(linchunking.GetChunk(0).transactions));
1957          }
1958      }
1959      // Verify that each element of m_depgraph occurred in m_linearization.
1960      assert(m_done == m_depgraph.Positions());
1961  }
1962  
1963  void TxGraphImpl::SanityCheck() const
1964  {
1965      /** Which GraphIndexes ought to occur in m_unlinked, based on m_entries. */
1966      std::set<GraphIndex> expected_unlinked;
1967      /** Which Clusters ought to occur in ClusterSet::m_clusters, based on m_entries. */
1968      std::set<const Cluster*> expected_clusters[MAX_LEVELS];
1969      /** Which GraphIndexes ought to occur in ClusterSet::m_removed, based on m_entries. */
1970      std::set<GraphIndex> expected_removed[MAX_LEVELS];
1971      /** Which Cluster::m_sequence values have been encountered. */
1972      std::set<uint64_t> sequences;
1973      /** Whether compaction is possible in the current state. */
1974      bool compact_possible{true};
1975  
1976      // Go over all Entry objects in m_entries.
1977      for (GraphIndex idx = 0; idx < m_entries.size(); ++idx) {
1978          const auto& entry = m_entries[idx];
1979          if (entry.m_ref == nullptr) {
1980              // Unlinked Entry must have indexes appear in m_unlinked.
1981              expected_unlinked.insert(idx);
1982          } else {
1983              // Every non-unlinked Entry must have a Ref that points back to it.
1984              assert(GetRefGraph(*entry.m_ref) == this);
1985              assert(GetRefIndex(*entry.m_ref) == idx);
1986          }
1987          // Verify the Entry m_locators.
1988          bool was_present{false}, was_removed{false};
1989          for (int level = 0; level < MAX_LEVELS; ++level) {
1990              const auto& locator = entry.m_locator[level];
1991              // Every Locator must be in exactly one of these 3 states.
1992              assert(locator.IsMissing() + locator.IsRemoved() + locator.IsPresent() == 1);
1993              if (locator.IsPresent()) {
1994                  // Once removed, a transaction cannot be revived.
1995                  assert(!was_removed);
1996                  // Verify that the Cluster agrees with where the Locator claims the transaction is.
1997                  assert(locator.cluster->GetClusterEntry(locator.index) == idx);
1998                  // Remember that we expect said Cluster to appear in the ClusterSet::m_clusters.
1999                  expected_clusters[level].insert(locator.cluster);
2000                  was_present = true;
2001              } else if (locator.IsRemoved()) {
2002                  // Level 0 (main) cannot have IsRemoved locators (IsMissing there means non-existing).
2003                  assert(level > 0);
2004                  // A Locator can only be IsRemoved if it was IsPresent before, and only once.
2005                  assert(was_present && !was_removed);
2006                  // Remember that we expect this GraphIndex to occur in the ClusterSet::m_removed.
2007                  expected_removed[level].insert(idx);
2008                  was_removed = true;
2009              }
2010          }
2011      }
2012  
2013      // For all levels (0 = main, 1 = staged)...
2014      for (int level = 0; level <= GetTopLevel(); ++level) {
2015          assert(level < MAX_LEVELS);
2016          auto& clusterset = GetClusterSet(level);
2017          std::set<const Cluster*> actual_clusters;
2018  
2019          // For all quality levels...
2020          for (int qual = 0; qual < int(QualityLevel::NONE); ++qual) {
2021              QualityLevel quality{qual};
2022              const auto& quality_clusters = clusterset.m_clusters[qual];
2023              // ... for all clusters in them ...
2024              for (ClusterSetIndex setindex = 0; setindex < quality_clusters.size(); ++setindex) {
2025                  const auto& cluster = *quality_clusters[setindex];
2026                  // Check the sequence number.
2027                  assert(cluster.m_sequence < m_next_sequence_counter);
2028                  assert(sequences.count(cluster.m_sequence) == 0);
2029                  sequences.insert(cluster.m_sequence);
2030                  // Remember we saw this Cluster (only if it is non-empty; empty Clusters aren't
2031                  // expected to be referenced by the Entry vector).
2032                  if (cluster.GetTxCount() != 0) {
2033                      actual_clusters.insert(&cluster);
2034                  }
2035                  // Sanity check the cluster, according to the Cluster's internal rules.
2036                  cluster.SanityCheck(*this, level);
2037                  // Check that the cluster's quality and setindex matches its position in the quality list.
2038                  assert(cluster.m_quality == quality);
2039                  assert(cluster.m_setindex == setindex);
2040              }
2041          }
2042  
2043          // Verify that all to-be-removed transactions have valid identifiers.
2044          for (GraphIndex idx : clusterset.m_to_remove) {
2045              assert(idx < m_entries.size());
2046              // We cannot assert that all m_to_remove transactions are still present: ~Ref on a
2047              // (P,M) transaction (present in main, inherited in staging) will cause an m_to_remove
2048              // addition in both main and staging, but a subsequence ApplyRemovals in main will
2049              // cause it to disappear from staging too, leaving the m_to_remove in place.
2050          }
2051  
2052          // Verify that all to-be-added dependencies have valid identifiers.
2053          for (auto [par_idx, chl_idx] : clusterset.m_deps_to_add) {
2054              assert(par_idx != chl_idx);
2055              assert(par_idx < m_entries.size());
2056              assert(chl_idx < m_entries.size());
2057          }
2058  
2059          // Verify that the actually encountered clusters match the ones occurring in Entry vector.
2060          assert(actual_clusters == expected_clusters[level]);
2061  
2062          // Verify that the contents of m_removed matches what was expected based on the Entry vector.
2063          std::set<GraphIndex> actual_removed(clusterset.m_removed.begin(), clusterset.m_removed.end());
2064          for (auto i : expected_unlinked) {
2065              // If a transaction exists in both main and staging, and is removed from staging (adding
2066              // it to m_removed there), and consequently destroyed (wiping the locator completely),
2067              // it can remain in m_removed despite not having an IsRemoved() locator. Exclude those
2068              // transactions from the comparison here.
2069              actual_removed.erase(i);
2070              expected_removed[level].erase(i);
2071          }
2072  
2073          assert(actual_removed == expected_removed[level]);
2074  
2075          // If any GraphIndex entries remain in this ClusterSet, compact is not possible.
2076          if (!clusterset.m_deps_to_add.empty()) compact_possible = false;
2077          if (!clusterset.m_to_remove.empty()) compact_possible = false;
2078          if (!clusterset.m_removed.empty()) compact_possible = false;
2079  
2080          // If m_group_data exists, its m_group_oversized must match m_oversized.
2081          if (clusterset.m_group_data.has_value()) {
2082              assert(clusterset.m_oversized == clusterset.m_group_data->m_group_oversized);
2083          }
2084  
2085          // For non-top levels, m_oversized must be known (as it cannot change until the level
2086          // on top is gone).
2087          if (level < GetTopLevel()) assert(clusterset.m_oversized.has_value());
2088      }
2089  
2090      // Verify that the contents of m_unlinked matches what was expected based on the Entry vector.
2091      std::set<GraphIndex> actual_unlinked(m_unlinked.begin(), m_unlinked.end());
2092      assert(actual_unlinked == expected_unlinked);
2093  
2094      // If compaction was possible, it should have been performed already, and m_unlinked must be
2095      // empty (to prevent memory leaks due to an ever-growing m_entries vector).
2096      if (compact_possible) {
2097          assert(actual_unlinked.empty());
2098      }
2099  }
2100  
2101  void TxGraphImpl::DoWork() noexcept
2102  {
2103      for (int level = 0; level <= GetTopLevel(); ++level) {
2104          MakeAllAcceptable(level);
2105      }
2106  }
2107  
2108  } // namespace
2109  
2110  TxGraph::Ref::~Ref()
2111  {
2112      if (m_graph) {
2113          // Inform the TxGraph about the Ref being destroyed.
2114          m_graph->UnlinkRef(m_index);
2115          m_graph = nullptr;
2116      }
2117  }
2118  
2119  TxGraph::Ref& TxGraph::Ref::operator=(Ref&& other) noexcept
2120  {
2121      // Unlink the current graph, if any.
2122      if (m_graph) m_graph->UnlinkRef(m_index);
2123      // Inform the other's graph about the move, if any.
2124      if (other.m_graph) other.m_graph->UpdateRef(other.m_index, *this);
2125      // Actually update the contents.
2126      m_graph = other.m_graph;
2127      m_index = other.m_index;
2128      other.m_graph = nullptr;
2129      other.m_index = GraphIndex(-1);
2130      return *this;
2131  }
2132  
2133  TxGraph::Ref::Ref(Ref&& other) noexcept
2134  {
2135      // Inform the TxGraph of other that its Ref is being moved.
2136      if (other.m_graph) other.m_graph->UpdateRef(other.m_index, *this);
2137      // Actually move the contents.
2138      std::swap(m_graph, other.m_graph);
2139      std::swap(m_index, other.m_index);
2140  }
2141  
2142  std::unique_ptr<TxGraph> MakeTxGraph(unsigned max_cluster_count) noexcept
2143  {
2144      return std::make_unique<TxGraphImpl>(max_cluster_count);
2145  }