Branch data Line data Source code
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 : : #include <util/vector.h>
13 : :
14 : : #include <compare>
15 : : #include <functional>
16 : : #include <memory>
17 : : #include <ranges>
18 : : #include <set>
19 : : #include <span>
20 : : #include <unordered_set>
21 : : #include <utility>
22 : : #include <vector>
23 : :
24 : : namespace {
25 : :
26 : : using namespace cluster_linearize;
27 : :
28 : : /** The maximum number of levels a TxGraph can have (0 = main, 1 = staging). */
29 : : static constexpr int MAX_LEVELS{2};
30 : :
31 : : // Forward declare the TxGraph implementation class.
32 : : class TxGraphImpl;
33 : :
34 : : /** Position of a DepGraphIndex within a Cluster::m_linearization. */
35 : : using LinearizationIndex = uint32_t;
36 : : /** Position of a Cluster within TxGraphImpl::ClusterSet::m_clusters. */
37 : : using ClusterSetIndex = uint32_t;
38 : :
39 : : /** Quality levels for cached cluster linearizations. */
40 : : enum class QualityLevel
41 : : {
42 : : /** This is a singleton cluster consisting of a transaction that individually exceeds the
43 : : * cluster size limit. It cannot be merged with anything. */
44 : : OVERSIZED_SINGLETON,
45 : : /** This cluster may have multiple disconnected components, which are all NEEDS_FIX. */
46 : : NEEDS_SPLIT_FIX,
47 : : /** This cluster may have multiple disconnected components, which are all NEEDS_RELINEARIZE. */
48 : : NEEDS_SPLIT,
49 : : /** This cluster may be non-topological. */
50 : : NEEDS_FIX,
51 : : /** This cluster has undergone changes that warrant re-linearization. */
52 : : NEEDS_RELINEARIZE,
53 : : /** The minimal level of linearization has been performed, but it is not known to be optimal. */
54 : : ACCEPTABLE,
55 : : /** The linearization is known to be optimal. */
56 : : OPTIMAL,
57 : : /** This cluster is not registered in any ClusterSet::m_clusters.
58 : : * This must be the last entry in QualityLevel as ClusterSet::m_clusters is sized using it. */
59 : : NONE,
60 : : };
61 : :
62 : : /** Information about a transaction inside TxGraphImpl::Trim. */
63 : 64217 : struct TrimTxData
64 : : {
65 : : // Fields populated by Cluster::AppendTrimData(). These are immutable after TrimTxData
66 : : // construction.
67 : : /** Chunk feerate for this transaction. */
68 : : FeePerWeight m_chunk_feerate;
69 : : /** GraphIndex of the transaction. */
70 : : TxGraph::GraphIndex m_index;
71 : : /** Size of the transaction. */
72 : : uint32_t m_tx_size;
73 : :
74 : : // Fields only used internally by TxGraphImpl::Trim():
75 : : /** Number of unmet dependencies this transaction has. -1 if the transaction is included. */
76 : : uint32_t m_deps_left;
77 : : /** Number of dependencies that apply to this transaction as child. */
78 : : uint32_t m_parent_count;
79 : : /** Where in deps_by_child those dependencies begin. */
80 : : uint32_t m_parent_offset;
81 : : /** Number of dependencies that apply to this transaction as parent. */
82 : : uint32_t m_children_count;
83 : : /** Where in deps_by_parent those dependencies begin. */
84 : : uint32_t m_children_offset;
85 : :
86 : : // Fields only used internally by TxGraphImpl::Trim()'s union-find implementation, and only for
87 : : // transactions that are definitely included or definitely rejected.
88 : : //
89 : : // As transactions get processed, they get organized into trees which form partitions
90 : : // representing the would-be clusters up to that point. The root of each tree is a
91 : : // representative for that partition. See
92 : : // https://en.wikipedia.org/wiki/Disjoint-set_data_structure.
93 : : //
94 : : /** Pointer to another TrimTxData, towards the root of the tree. If this is a root, m_uf_parent
95 : : * is equal to this itself. */
96 : : TrimTxData* m_uf_parent;
97 : : /** If this is a root, the total number of transactions in the partition. */
98 : : uint32_t m_uf_count;
99 : : /** If this is a root, the total size of transactions in the partition. */
100 : : uint64_t m_uf_size;
101 : : };
102 : :
103 : : /** A grouping of connected transactions inside a TxGraphImpl::ClusterSet. */
104 : : class Cluster
105 : : {
106 : : friend class TxGraphImpl;
107 : : friend class BlockBuilderImpl;
108 : :
109 : : protected:
110 : : using GraphIndex = TxGraph::GraphIndex;
111 : : using SetType = BitSet<MAX_CLUSTER_COUNT_LIMIT>;
112 : : /** The quality level of m_linearization. */
113 : : QualityLevel m_quality{QualityLevel::NONE};
114 : : /** Which position this Cluster has in TxGraphImpl::ClusterSet::m_clusters[m_quality]. */
115 : : ClusterSetIndex m_setindex{ClusterSetIndex(-1)};
116 : : /** Sequence number for this Cluster (for tie-breaking comparison between equal-chunk-feerate
117 : : transactions in distinct clusters). */
118 : : uint64_t m_sequence;
119 : :
120 : 139642 : explicit Cluster(uint64_t sequence) noexcept : m_sequence(sequence) {}
121 : :
122 : : public:
123 : : // Provide virtual destructor, for safe polymorphic usage inside std::unique_ptr.
124 : 0 : virtual ~Cluster() = default;
125 : :
126 : : // Cannot move or copy (would invalidate Cluster* in Locator and ClusterSet). */
127 : : Cluster(const Cluster&) = delete;
128 : : Cluster& operator=(const Cluster&) = delete;
129 : : Cluster(Cluster&&) = delete;
130 : : Cluster& operator=(Cluster&&) = delete;
131 : :
132 : : // Generic helper functions.
133 : :
134 : : /** Whether the linearization of this Cluster can be exposed. */
135 : 330930804 : bool IsAcceptable() const noexcept
136 : : {
137 : 330930804 : return m_quality == QualityLevel::ACCEPTABLE || m_quality == QualityLevel::OPTIMAL;
138 : : }
139 : : /** Whether the linearization of this Cluster is topological. */
140 : 412977 : bool IsTopological() const noexcept
141 : : {
142 : 412977 : return m_quality != QualityLevel::NEEDS_FIX && m_quality != QualityLevel::NEEDS_SPLIT_FIX;
143 : : }
144 : : /** Whether the linearization of this Cluster is optimal. */
145 : 12080673 : bool IsOptimal() const noexcept
146 : : {
147 : 12080673 : return m_quality == QualityLevel::OPTIMAL;
148 : : }
149 : : /** Whether this cluster is oversized. Note that no changes that can cause oversizedness are
150 : : * ever applied, so the only way a materialized Cluster object can be oversized is by being
151 : : * an individually oversized transaction singleton. */
152 : 24288030 : bool IsOversized() const noexcept { return m_quality == QualityLevel::OVERSIZED_SINGLETON; }
153 : : /** Whether this cluster requires splitting. */
154 : 330540105 : bool NeedsSplitting() const noexcept
155 : : {
156 : 330540105 : return m_quality == QualityLevel::NEEDS_SPLIT || m_quality == QualityLevel::NEEDS_SPLIT_FIX;
157 : : }
158 : :
159 : : /** Get the smallest number of transactions this Cluster is intended for. */
160 : : virtual DepGraphIndex GetMinIntendedTxCount() const noexcept = 0;
161 : : /** Get the maximum number of transactions this Cluster supports. */
162 : : virtual DepGraphIndex GetMaxTxCount() const noexcept = 0;
163 : : /** Total memory usage currently for this Cluster, including all its dynamic memory, plus Cluster
164 : : * structure itself, and ClusterSet::m_clusters entry. */
165 : : virtual size_t TotalMemoryUsage() const noexcept = 0;
166 : : /** Determine the range of DepGraphIndexes used by this Cluster. */
167 : : virtual DepGraphIndex GetDepGraphIndexRange() const noexcept = 0;
168 : : /** Get the number of transactions in this Cluster. */
169 : : virtual LinearizationIndex GetTxCount() const noexcept = 0;
170 : : /** Get the total size of the transactions in this Cluster. */
171 : : virtual uint64_t GetTotalTxSize() const noexcept = 0;
172 : : /** Given a DepGraphIndex into this Cluster, find the corresponding GraphIndex. */
173 : : virtual GraphIndex GetClusterEntry(DepGraphIndex index) const noexcept = 0;
174 : : /** Append a transaction with given GraphIndex at the end of this Cluster and its
175 : : * linearization. Return the DepGraphIndex it was placed at. */
176 : : virtual DepGraphIndex AppendTransaction(GraphIndex graph_idx, FeePerWeight feerate) noexcept = 0;
177 : : /** Add dependencies to a given child in this cluster. */
178 : : virtual void AddDependencies(SetType parents, DepGraphIndex child) noexcept = 0;
179 : : /** Invoke visit1_fn for each transaction in the cluster, in linearization order, then
180 : : * visit2_fn in the same order, then wipe this Cluster. */
181 : : virtual void ExtractTransactions(const std::function<void (DepGraphIndex, GraphIndex, FeePerWeight)>& visit1_fn, const std::function<void (DepGraphIndex, GraphIndex, SetType)>& visit2_fn) noexcept = 0;
182 : : /** Figure out what level this Cluster exists at in the graph. In most cases this is known by
183 : : * the caller already (see all "int level" arguments below), but not always. */
184 : : virtual int GetLevel(const TxGraphImpl& graph) const noexcept = 0;
185 : : /** Only called by TxGraphImpl::SwapIndexes. */
186 : : virtual void UpdateMapping(DepGraphIndex cluster_idx, GraphIndex graph_idx) noexcept = 0;
187 : : /** Push changes to Cluster and its linearization to the TxGraphImpl Entry objects. Main chunk
188 : : * information is computed if the cluster is acceptable, or when rename is set. Rename is used
189 : : * when called from Compact, to recompute after GraphIndexes may have changed; in this case,
190 : : * no chunk index objects are removed or created either. */
191 : : virtual void Updated(TxGraphImpl& graph, int level, bool rename) noexcept = 0;
192 : : /** Remove all chunk index entries for this cluster (level=0 only). */
193 : : virtual void RemoveChunkData(TxGraphImpl& graph) noexcept = 0;
194 : : /** Create a copy of this Cluster in staging, returning a pointer to it (used by PullIn). */
195 : : virtual Cluster* CopyToStaging(TxGraphImpl& graph) const noexcept = 0;
196 : : /** Get the list of Clusters in main that conflict with this one (which is assumed to be in staging). */
197 : : virtual void GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept = 0;
198 : : /** Mark all the Entry objects belonging to this staging Cluster as missing. The Cluster must be
199 : : * deleted immediately after. */
200 : : virtual void MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept = 0;
201 : : /** Remove all transactions from a (non-empty) Cluster. */
202 : : virtual void Clear(TxGraphImpl& graph, int level) noexcept = 0;
203 : : /** Change a Cluster's level from 1 (staging) to 0 (main). */
204 : : virtual void MoveToMain(TxGraphImpl& graph) noexcept = 0;
205 : : /** Minimize this Cluster's memory usage. */
206 : : virtual void Compact() noexcept = 0;
207 : :
208 : : // Functions that implement the Cluster-specific side of internal TxGraphImpl mutations.
209 : :
210 : : /** Apply all removals from the front of to_remove that apply to this Cluster, popping them
211 : : * off. There must be at least one such entry. */
212 : : virtual void ApplyRemovals(TxGraphImpl& graph, int level, std::span<GraphIndex>& to_remove) noexcept = 0;
213 : : /** Split this cluster (must have a NEEDS_SPLIT* quality). Returns whether to delete this
214 : : * Cluster afterwards. */
215 : : [[nodiscard]] virtual bool Split(TxGraphImpl& graph, int level) noexcept = 0;
216 : : /** Move all transactions from cluster to *this (as separate components). */
217 : : virtual void Merge(TxGraphImpl& graph, int level, Cluster& cluster) noexcept = 0;
218 : : /** Given a span of (parent, child) pairs that all belong to this Cluster, apply them. */
219 : : virtual void ApplyDependencies(TxGraphImpl& graph, int level, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept = 0;
220 : : /** Improve the linearization of this Cluster. Returns how much work was performed and whether
221 : : * the Cluster's QualityLevel improved as a result. */
222 : : virtual std::pair<uint64_t, bool> Relinearize(TxGraphImpl& graph, int level, uint64_t max_cost) noexcept = 0;
223 : : /** For every chunk in the cluster, append its FeeFrac to ret. */
224 : : virtual void AppendChunkFeerates(std::vector<FeeFrac>& ret) const noexcept = 0;
225 : : /** Add a TrimTxData entry (filling m_chunk_feerate, m_index, m_tx_size) for every
226 : : * transaction in the Cluster to ret. Implicit dependencies between consecutive transactions
227 : : * in the linearization are added to deps. Return the Cluster's total transaction size. */
228 : : virtual uint64_t AppendTrimData(std::vector<TrimTxData>& ret, std::vector<std::pair<GraphIndex, GraphIndex>>& deps) const noexcept = 0;
229 : :
230 : : // Functions that implement the Cluster-specific side of public TxGraph functions.
231 : :
232 : : /** Process elements from the front of args that apply to this cluster, and append Refs for the
233 : : * union of their ancestors to output. */
234 : : virtual void GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept = 0;
235 : : /** Process elements from the front of args that apply to this cluster, and append Refs for the
236 : : * union of their descendants to output. */
237 : : virtual void GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept = 0;
238 : : /** Populate range with refs for the transactions in this Cluster's linearization, from
239 : : * position start_pos until start_pos+range.size()-1, inclusive. Returns whether that
240 : : * range includes the last transaction in the linearization. */
241 : : virtual bool GetClusterRefs(TxGraphImpl& graph, std::span<TxGraph::Ref*> range, LinearizationIndex start_pos) noexcept = 0;
242 : : /** Get the individual transaction feerate of a Cluster element. */
243 : : virtual FeePerWeight GetIndividualFeerate(DepGraphIndex idx) noexcept = 0;
244 : : /** Modify the fee of a Cluster element. */
245 : : virtual void SetFee(TxGraphImpl& graph, int level, DepGraphIndex idx, int64_t fee) noexcept = 0;
246 : :
247 : : // Debugging functions.
248 : :
249 : : virtual void SanityCheck(const TxGraphImpl& graph, int level) const = 0;
250 : : };
251 : :
252 : : /** An implementation of Cluster that uses a DepGraph and vectors, to support arbitrary numbers of
253 : : * transactions up to MAX_CLUSTER_COUNT_LIMIT. */
254 : : class GenericClusterImpl final : public Cluster
255 : : {
256 : : friend class TxGraphImpl;
257 : : /** The DepGraph for this cluster, holding all feerates, and ancestors/descendants. */
258 : : DepGraph<SetType> m_depgraph;
259 : : /** m_mapping[i] gives the GraphIndex for the position i transaction in m_depgraph. Values for
260 : : * positions i that do not exist in m_depgraph shouldn't ever be accessed and thus don't
261 : : * matter. m_mapping.size() equals m_depgraph.PositionRange(). */
262 : : std::vector<GraphIndex> m_mapping;
263 : : /** The current linearization of the cluster. m_linearization.size() equals
264 : : * m_depgraph.TxCount(). This is always kept topological. */
265 : : std::vector<DepGraphIndex> m_linearization;
266 : :
267 : : public:
268 : : /** The smallest number of transactions this Cluster implementation is intended for. */
269 : : static constexpr DepGraphIndex MIN_INTENDED_TX_COUNT{2};
270 : : /** The largest number of transactions this Cluster implementation supports. */
271 : : static constexpr DepGraphIndex MAX_TX_COUNT{SetType::Size()};
272 : :
273 : : GenericClusterImpl() noexcept = delete;
274 : : /** Construct an empty GenericClusterImpl. */
275 : : explicit GenericClusterImpl(uint64_t sequence) noexcept;
276 : :
277 : : size_t TotalMemoryUsage() const noexcept final;
278 : 69329 : constexpr DepGraphIndex GetMinIntendedTxCount() const noexcept final { return MIN_INTENDED_TX_COUNT; }
279 : 73862 : constexpr DepGraphIndex GetMaxTxCount() const noexcept final { return MAX_TX_COUNT; }
280 [ - + ]: 11 : DepGraphIndex GetDepGraphIndexRange() const noexcept final { return m_depgraph.PositionRange(); }
281 [ - + ]: 376883 : LinearizationIndex GetTxCount() const noexcept final { return m_linearization.size(); }
282 : : uint64_t GetTotalTxSize() const noexcept final;
283 : 404481 : GraphIndex GetClusterEntry(DepGraphIndex index) const noexcept final { return m_mapping[index]; }
284 : : DepGraphIndex AppendTransaction(GraphIndex graph_idx, FeePerWeight feerate) noexcept final;
285 : : void AddDependencies(SetType parents, DepGraphIndex child) noexcept final;
286 : : void ExtractTransactions(const std::function<void (DepGraphIndex, GraphIndex, FeePerWeight)>& visit1_fn, const std::function<void (DepGraphIndex, GraphIndex, SetType)>& visit2_fn) noexcept final;
287 : : int GetLevel(const TxGraphImpl& graph) const noexcept final;
288 : 835 : void UpdateMapping(DepGraphIndex cluster_idx, GraphIndex graph_idx) noexcept final { m_mapping[cluster_idx] = graph_idx; }
289 : : void Updated(TxGraphImpl& graph, int level, bool rename) noexcept final;
290 : : void RemoveChunkData(TxGraphImpl& graph) noexcept final;
291 : : Cluster* CopyToStaging(TxGraphImpl& graph) const noexcept final;
292 : : void GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept final;
293 : : void MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept final;
294 : : void Clear(TxGraphImpl& graph, int level) noexcept final;
295 : : void MoveToMain(TxGraphImpl& graph) noexcept final;
296 : : void Compact() noexcept final;
297 : : void ApplyRemovals(TxGraphImpl& graph, int level, std::span<GraphIndex>& to_remove) noexcept final;
298 : : [[nodiscard]] bool Split(TxGraphImpl& graph, int level) noexcept final;
299 : : void Merge(TxGraphImpl& graph, int level, Cluster& cluster) noexcept final;
300 : : void ApplyDependencies(TxGraphImpl& graph, int level, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept final;
301 : : std::pair<uint64_t, bool> Relinearize(TxGraphImpl& graph, int level, uint64_t max_cost) noexcept final;
302 : : void AppendChunkFeerates(std::vector<FeeFrac>& ret) const noexcept final;
303 : : uint64_t AppendTrimData(std::vector<TrimTxData>& ret, std::vector<std::pair<GraphIndex, GraphIndex>>& deps) const noexcept final;
304 : : void GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept final;
305 : : void GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept final;
306 : : bool GetClusterRefs(TxGraphImpl& graph, std::span<TxGraph::Ref*> range, LinearizationIndex start_pos) noexcept final;
307 : : FeePerWeight GetIndividualFeerate(DepGraphIndex idx) noexcept final;
308 : : void SetFee(TxGraphImpl& graph, int level, DepGraphIndex idx, int64_t fee) noexcept final;
309 : : void SanityCheck(const TxGraphImpl& graph, int level) const final;
310 : : };
311 : :
312 : : /** An implementation of Cluster that only supports 1 transaction. */
313 : 0 : class SingletonClusterImpl final : public Cluster
314 : : {
315 : : friend class TxGraphImpl;
316 : :
317 : : /** The feerate of the (singular) transaction in this Cluster. */
318 : : FeePerWeight m_feerate;
319 : : /** Constant to indicate that this Cluster is empty. */
320 : : static constexpr auto NO_GRAPH_INDEX = GraphIndex(-1);
321 : : /** The GraphIndex of the transaction. NO_GRAPH_INDEX if this Cluster is empty. */
322 : : GraphIndex m_graph_index = NO_GRAPH_INDEX;
323 : :
324 : : public:
325 : : /** The smallest number of transactions this Cluster implementation is intended for. */
326 : : static constexpr DepGraphIndex MIN_INTENDED_TX_COUNT{1};
327 : : /** The largest number of transactions this Cluster implementation supports. */
328 : : static constexpr DepGraphIndex MAX_TX_COUNT{1};
329 : :
330 : : SingletonClusterImpl() noexcept = delete;
331 : : /** Construct an empty SingletonClusterImpl. */
332 : 137882 : explicit SingletonClusterImpl(uint64_t sequence) noexcept : Cluster(sequence) {}
333 : :
334 : : size_t TotalMemoryUsage() const noexcept final;
335 : 12074674 : constexpr DepGraphIndex GetMinIntendedTxCount() const noexcept final { return MIN_INTENDED_TX_COUNT; }
336 : 12076106 : constexpr DepGraphIndex GetMaxTxCount() const noexcept final { return MAX_TX_COUNT; }
337 : 97463754 : LinearizationIndex GetTxCount() const noexcept final { return m_graph_index != NO_GRAPH_INDEX; }
338 : 7920 : DepGraphIndex GetDepGraphIndexRange() const noexcept final { return GetTxCount(); }
339 [ + + ]: 12214277 : uint64_t GetTotalTxSize() const noexcept final { return GetTxCount() ? m_feerate.size : 0; }
340 : 12074680 : GraphIndex GetClusterEntry(DepGraphIndex index) const noexcept final { Assume(index == 0); Assume(GetTxCount()); return m_graph_index; }
341 : : DepGraphIndex AppendTransaction(GraphIndex graph_idx, FeePerWeight feerate) noexcept final;
342 : : void AddDependencies(SetType parents, DepGraphIndex child) noexcept final;
343 : : void ExtractTransactions(const std::function<void (DepGraphIndex, GraphIndex, FeePerWeight)>& visit1_fn, const std::function<void (DepGraphIndex, GraphIndex, SetType)>& visit2_fn) noexcept final;
344 : : int GetLevel(const TxGraphImpl& graph) const noexcept final;
345 : 14594 : void UpdateMapping(DepGraphIndex cluster_idx, GraphIndex graph_idx) noexcept final { Assume(cluster_idx == 0); m_graph_index = graph_idx; }
346 : : void Updated(TxGraphImpl& graph, int level, bool rename) noexcept final;
347 : : void RemoveChunkData(TxGraphImpl& graph) noexcept final;
348 : : Cluster* CopyToStaging(TxGraphImpl& graph) const noexcept final;
349 : : void GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept final;
350 : : void MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept final;
351 : : void Clear(TxGraphImpl& graph, int level) noexcept final;
352 : : void MoveToMain(TxGraphImpl& graph) noexcept final;
353 : : void Compact() noexcept final;
354 : : void ApplyRemovals(TxGraphImpl& graph, int level, std::span<GraphIndex>& to_remove) noexcept final;
355 : : [[nodiscard]] bool Split(TxGraphImpl& graph, int level) noexcept final;
356 : : void Merge(TxGraphImpl& graph, int level, Cluster& cluster) noexcept final;
357 : : void ApplyDependencies(TxGraphImpl& graph, int level, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept final;
358 : : std::pair<uint64_t, bool> Relinearize(TxGraphImpl& graph, int level, uint64_t max_cost) noexcept final;
359 : : void AppendChunkFeerates(std::vector<FeeFrac>& ret) const noexcept final;
360 : : uint64_t AppendTrimData(std::vector<TrimTxData>& ret, std::vector<std::pair<GraphIndex, GraphIndex>>& deps) const noexcept final;
361 : : void GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept final;
362 : : void GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept final;
363 : : bool GetClusterRefs(TxGraphImpl& graph, std::span<TxGraph::Ref*> range, LinearizationIndex start_pos) noexcept final;
364 : : FeePerWeight GetIndividualFeerate(DepGraphIndex idx) noexcept final;
365 : : void SetFee(TxGraphImpl& graph, int level, DepGraphIndex idx, int64_t fee) noexcept final;
366 : : void SanityCheck(const TxGraphImpl& graph, int level) const final;
367 : : };
368 : :
369 : : /** The transaction graph, including staged changes.
370 : : *
371 : : * The overall design of the data structure consists of 3 interlinked representations:
372 : : * - The transactions (held as a vector of TxGraphImpl::Entry inside TxGraphImpl).
373 : : * - The clusters (Cluster objects in per-quality vectors inside TxGraphImpl::ClusterSet).
374 : : * - The Refs (TxGraph::Ref objects, held externally by users of the TxGraph class)
375 : : *
376 : : * The Clusters are kept in one or two ClusterSet objects, one for the "main" graph, and one for
377 : : * the proposed changes ("staging"). If a transaction occurs in both, they share the same Entry,
378 : : * but there will be a separate Cluster per graph.
379 : : *
380 : : * Clusters and Refs contain the index of the Entry objects they refer to, and the Entry objects
381 : : * refer back to the Clusters and Refs the corresponding transaction is contained in.
382 : : *
383 : : * While redundant, this permits moving all of them independently, without invalidating things
384 : : * or costly iteration to fix up everything:
385 : : * - Entry objects can be moved to fill holes left by removed transactions in the Entry vector
386 : : * (see TxGraphImpl::Compact).
387 : : * - Clusters can be rewritten continuously (removals can cause them to split, new dependencies
388 : : * can cause them to be merged).
389 : : * - Ref objects can be held outside the class, while permitting them to be moved around, and
390 : : * inherited from.
391 : : */
392 : : class TxGraphImpl final : public TxGraph
393 : : {
394 : : friend class Cluster;
395 : : friend class SingletonClusterImpl;
396 : : friend class GenericClusterImpl;
397 : : friend class BlockBuilderImpl;
398 : : private:
399 : : /** Internal RNG. */
400 : : FastRandomContext m_rng;
401 : : /** This TxGraphImpl's maximum cluster count limit. */
402 : : const DepGraphIndex m_max_cluster_count;
403 : : /** This TxGraphImpl's maximum cluster size limit. */
404 : : const uint64_t m_max_cluster_size;
405 : : /** The amount of linearization work needed per cluster to be considered acceptable. */
406 : : const uint64_t m_acceptable_cost;
407 : : /** Fallback ordering for transactions. */
408 : : const std::function<std::strong_ordering(const TxGraph::Ref&, const TxGraph::Ref&)> m_fallback_order;
409 : :
410 : : /** Information about one group of Clusters to be merged. */
411 : : struct GroupEntry
412 : : {
413 : : /** Where the clusters to be merged start in m_group_clusters. */
414 : : uint32_t m_cluster_offset;
415 : : /** How many clusters to merge. */
416 : : uint32_t m_cluster_count;
417 : : /** Where the dependencies for this cluster group in m_deps_to_add start. */
418 : : uint32_t m_deps_offset;
419 : : /** How many dependencies to add. */
420 : : uint32_t m_deps_count;
421 : : };
422 : :
423 : : /** Information about all groups of Clusters to be merged. */
424 : 100624 : struct GroupData
425 : : {
426 : : /** The groups of Clusters to be merged. */
427 : : std::vector<GroupEntry> m_groups;
428 : : /** Which clusters are to be merged. GroupEntry::m_cluster_offset indexes into this. */
429 : : std::vector<Cluster*> m_group_clusters;
430 : : };
431 : :
432 : : /** The collection of all Clusters in main or staged. */
433 : : struct ClusterSet
434 : : {
435 : : /** The vectors of clusters, one vector per quality level. ClusterSetIndex indexes into each. */
436 : : std::array<std::vector<std::unique_ptr<Cluster>>, int(QualityLevel::NONE)> m_clusters;
437 : : /** Which removals have yet to be applied. */
438 : : std::vector<GraphIndex> m_to_remove;
439 : : /** Which dependencies are to be added ((parent,child) pairs). GroupData::m_deps_offset indexes
440 : : * into this. */
441 : : std::vector<std::pair<GraphIndex, GraphIndex>> m_deps_to_add;
442 : : /** Information about the merges to be performed, if known. */
443 : : std::optional<GroupData> m_group_data = GroupData{};
444 : : /** Which entries were removed in this ClusterSet (so they can be wiped on abort). This
445 : : * includes all entries which have an (R) removed locator at this level (staging only),
446 : : * plus optionally any transaction in m_unlinked. */
447 : : std::vector<GraphIndex> m_removed;
448 : : /** Total number of transactions in this graph (sum of all transaction counts in all
449 : : * Clusters, and for staging also those inherited from the main ClusterSet). */
450 : : GraphIndex m_txcount{0};
451 : : /** Total number of individually oversized transactions in the graph. */
452 : : GraphIndex m_txcount_oversized{0};
453 : : /** Whether this graph is oversized (if known). */
454 : : std::optional<bool> m_oversized{false};
455 : : /** The combined TotalMemoryUsage of all clusters in this level (only Clusters that
456 : : * are materialized; in staging, implicit Clusters from main are not counted), */
457 : : size_t m_cluster_usage{0};
458 : :
459 : 72555 : ClusterSet() noexcept = default;
460 : : };
461 : :
462 : : /** The main ClusterSet. */
463 : : ClusterSet m_main_clusterset;
464 : : /** The staging ClusterSet, if any. */
465 : : std::optional<ClusterSet> m_staging_clusterset;
466 : : /** Next sequence number to assign to created Clusters. */
467 : : uint64_t m_next_sequence_counter{0};
468 : :
469 : : /** Information about a chunk in the main graph. */
470 : : struct ChunkData
471 : : {
472 : : /** The Entry which is the last transaction of the chunk. */
473 : : mutable GraphIndex m_graph_index;
474 : : /** How many transactions the chunk contains (-1 = singleton tail of cluster). */
475 : : LinearizationIndex m_chunk_count;
476 : :
477 : 173582 : ChunkData(GraphIndex graph_index, LinearizationIndex chunk_count) noexcept :
478 : 173582 : m_graph_index{graph_index}, m_chunk_count{chunk_count} {}
479 : : };
480 : :
481 : : /** Compare two Cluster* by their m_sequence value (while supporting nullptr). */
482 : 180733 : static std::strong_ordering CompareClusters(Cluster* a, Cluster* b) noexcept
483 : : {
484 : : // The nullptr pointer compares before everything else.
485 [ - + ]: 180733 : if (a == nullptr || b == nullptr) {
486 [ # # # # ]: 0 : return (a != nullptr) <=> (b != nullptr);
487 : : }
488 : : // If neither pointer is nullptr, compare the Clusters' sequence numbers.
489 [ + + ]: 180733 : Assume(a == b || a->m_sequence != b->m_sequence);
490 [ + + + + ]: 180733 : return a->m_sequence <=> b->m_sequence;
491 : : }
492 : :
493 : : /** Compare two entries (which must both exist within the main graph). */
494 : 161189050 : std::strong_ordering CompareMainTransactions(GraphIndex a, GraphIndex b) const noexcept
495 : : {
496 [ - + ]: 161189050 : if (a == b) return std::strong_ordering::equal;
497 [ - + + - : 161189050 : Assume(a < m_entries.size() && b < m_entries.size());
+ + ]
498 [ + + ]: 161189050 : const auto& entry_a = m_entries[a];
499 : 161189050 : const auto& entry_b = m_entries[b];
500 : : // Compare chunk feerates, and return result if it differs.
501 [ + + ]: 161189050 : auto feerate_cmp = ByRatio{entry_b.m_main_chunk_feerate} <=> ByRatio{entry_a.m_main_chunk_feerate};
502 [ + + ]: 161189050 : if (feerate_cmp != 0) return feerate_cmp;
503 : : // Compare equal-feerate chunk prefix size for comparing equal chunk feerates. This does two
504 : : // things: it distinguishes equal-feerate chunks within the same cluster (because later
505 : : // ones will always have a higher prefix size), and it may distinguish equal-feerate chunks
506 : : // from distinct clusters.
507 [ + + ]: 89787104 : if (entry_a.m_main_equal_feerate_chunk_prefix_size != entry_b.m_main_equal_feerate_chunk_prefix_size) {
508 [ + + ]: 1667031 : return entry_a.m_main_equal_feerate_chunk_prefix_size <=> entry_b.m_main_equal_feerate_chunk_prefix_size;
509 : : }
510 : : // Compare by maximum m_fallback_order element to order equal-feerate chunks in distinct
511 : : // clusters, when the equal-feerate-prefix size is also the same.
512 : 88120073 : const auto& locator_a = entry_a.m_locator[0];
513 : 88120073 : const auto& locator_b = entry_b.m_locator[0];
514 [ + + ]: 88120073 : Assume(locator_a.IsPresent() && locator_b.IsPresent());
515 [ + + ]: 88120073 : if (locator_a.cluster != locator_b.cluster) {
516 : 86982212 : auto fallback_cmp = m_fallback_order(*m_entries[entry_a.m_main_max_chunk_fallback].m_ref,
517 : 86982212 : *m_entries[entry_b.m_main_max_chunk_fallback].m_ref);
518 [ + - ]: 86982212 : if (fallback_cmp != 0) return fallback_cmp;
519 : : // This shouldn't be reachable as m_fallback_order defines a strong ordering.
520 : 0 : Assume(false);
521 : 0 : return CompareClusters(locator_a.cluster, locator_b.cluster);
522 : : }
523 : : // Within a single chunk, sort by position within cluster linearization.
524 [ + - + + ]: 1137861 : return entry_a.m_main_lin_index <=> entry_b.m_main_lin_index;
525 : : }
526 : :
527 : : /** Comparator for ChunkData objects in mining order. */
528 : : class ChunkOrder
529 : : {
530 : : const TxGraphImpl* const m_graph;
531 : : public:
532 : 1413 : explicit ChunkOrder(const TxGraphImpl* graph) : m_graph(graph) {}
533 : :
534 : 2136921 : bool operator()(const ChunkData& a, const ChunkData& b) const noexcept
535 : : {
536 : 2136921 : return m_graph->CompareMainTransactions(a.m_graph_index, b.m_graph_index) < 0;
537 : : }
538 : : };
539 : :
540 : : /** Definition for the mining index type. */
541 : : using ChunkIndex = std::set<ChunkData, ChunkOrder>;
542 : :
543 : : /** Index of ChunkData objects, indexing the last transaction in each chunk in the main
544 : : * graph. */
545 : : ChunkIndex m_main_chunkindex;
546 : : /** Number of index-observing objects in existence (BlockBuilderImpls). */
547 : : size_t m_main_chunkindex_observers{0};
548 : : /** Cache of discarded ChunkIndex node handles to reuse, avoiding additional allocation. */
549 : : std::vector<ChunkIndex::node_type> m_main_chunkindex_discarded;
550 : :
551 : : /** A Locator that describes whether, where, and in which Cluster an Entry appears.
552 : : * Every Entry has MAX_LEVELS locators, as it may appear in one Cluster per level.
553 : : *
554 : : * Each level of a Locator is in one of three states:
555 : : *
556 : : * - (P)resent: actually occurs in a Cluster at that level.
557 : : *
558 : : * - (M)issing:
559 : : * - In the main graph: the transaction does not exist in main.
560 : : * - In the staging graph: the transaction's existence is the same as in main. If it doesn't
561 : : * exist in main, (M) in staging means it does not exist there
562 : : * either. If it does exist in main, (M) in staging means the
563 : : * cluster it is in has not been modified in staging, and thus the
564 : : * transaction implicitly exists in staging too (without explicit
565 : : * Cluster object; see PullIn() to create it in staging too).
566 : : *
567 : : * - (R)emoved: only possible in staging; it means the transaction exists in main, but is
568 : : * removed in staging.
569 : : *
570 : : * The following combinations are possible:
571 : : * - (M,M): the transaction doesn't exist in either graph.
572 : : * - (P,M): the transaction exists in both, but only exists explicitly in a Cluster object in
573 : : * main. Its existence in staging is inherited from main.
574 : : * - (P,P): the transaction exists in both, and is materialized in both. Thus, the clusters
575 : : * and/or their linearizations may be different in main and staging.
576 : : * - (M,P): the transaction is added in staging, and does not exist in main.
577 : : * - (P,R): the transaction exists in main, but is removed in staging.
578 : : *
579 : : * When staging does not exist, only (M,M) and (P,M) are possible.
580 : : */
581 : : struct Locator
582 : : {
583 : : /** Which Cluster the Entry appears in (nullptr = missing). */
584 : : Cluster* cluster{nullptr};
585 : : /** Where in the Cluster it appears (if cluster == nullptr: 0 = missing, -1 = removed). */
586 : : DepGraphIndex index{0};
587 : :
588 : : /** Mark this Locator as missing (= same as lower level, or non-existing if level 0). */
589 : 2177 : void SetMissing() noexcept { cluster = nullptr; index = 0; }
590 : : /** Mark this Locator as removed (not allowed in level 0). */
591 : 2177 : void SetRemoved() noexcept { cluster = nullptr; index = DepGraphIndex(-1); }
592 : : /** Mark this Locator as present, in the specified Cluster. */
593 : 384985 : void SetPresent(Cluster* c, DepGraphIndex i) noexcept { cluster = c; index = i; }
594 : : /** Check if this Locator is missing. */
595 [ - + + + ]: 12515213 : bool IsMissing() const noexcept { return cluster == nullptr && index == 0; }
596 : : /** Check if this Locator is removed. */
597 [ + - - + : 12480764 : bool IsRemoved() const noexcept { return cluster == nullptr && index == DepGraphIndex(-1); }
- + ]
598 : : /** Check if this Locator is present (in some Cluster). */
599 [ - + + - ]: 88120084 : bool IsPresent() const noexcept { return cluster != nullptr; }
600 : : };
601 : :
602 : : /** Internal information about each transaction in a TxGraphImpl. */
603 : 136084 : struct Entry
604 : : {
605 : : /** Pointer to the corresponding Ref object if any, or nullptr if unlinked. */
606 : : Ref* m_ref{nullptr};
607 : : /** Iterator to the corresponding ChunkData, if any, and m_main_chunkindex.end() otherwise.
608 : : * This is initialized on construction of the Entry, in AddTransaction. */
609 : : ChunkIndex::iterator m_main_chunkindex_iterator;
610 : : /** Which Cluster and position therein this Entry appears in. ([0] = main, [1] = staged). */
611 : : Locator m_locator[MAX_LEVELS];
612 : : /** The chunk feerate of this transaction in main (if present in m_locator[0]). */
613 : : FeePerWeight m_main_chunk_feerate;
614 : : /** The equal-feerate chunk prefix size of this transaction in main. If the transaction is
615 : : * part of chunk C in main, then this gives the sum of the sizes of all chunks in C's
616 : : * cluster, whose feerate is equal to that of C, which do not appear after C itself in
617 : : * the cluster's linearization.
618 : : * This provides a way to sort equal-feerate chunks across clusters, in a way that agrees
619 : : * with the within-cluster chunk ordering. */
620 : : int32_t m_main_equal_feerate_chunk_prefix_size;
621 : : /** The position this transaction has in the main linearization (if present). */
622 : : LinearizationIndex m_main_lin_index;
623 : : /** Of all transactions within this transaction's chunk in main (if present there), the
624 : : * maximal one according to m_fallback_order. */
625 : : GraphIndex m_main_max_chunk_fallback = GraphIndex(-1);
626 : : };
627 : :
628 : : /** The set of all transactions (in all levels combined). GraphIndex values index into this. */
629 : : std::vector<Entry> m_entries;
630 : :
631 : : /** Set of Entries which have no linked Ref anymore. */
632 : : std::vector<GraphIndex> m_unlinked;
633 : :
634 : : public:
635 : : /** Construct a new TxGraphImpl with the specified limits and fallback order. */
636 : 1413 : explicit TxGraphImpl(
637 : : DepGraphIndex max_cluster_count,
638 : : uint64_t max_cluster_size,
639 : : uint64_t acceptable_cost,
640 : : const std::function<std::strong_ordering(const TxGraph::Ref&, const TxGraph::Ref&)>& fallback_order
641 : 1413 : ) noexcept :
642 : 1413 : m_max_cluster_count(max_cluster_count),
643 : 1413 : m_max_cluster_size(max_cluster_size),
644 : 1413 : m_acceptable_cost(acceptable_cost),
645 : 1413 : m_fallback_order(fallback_order),
646 : 2826 : m_main_chunkindex(ChunkOrder(this))
647 : : {
648 : 1413 : Assume(max_cluster_count >= 1);
649 : 1413 : Assume(max_cluster_count <= MAX_CLUSTER_COUNT_LIMIT);
650 : 1413 : }
651 : :
652 : : /** Destructor. */
653 : : ~TxGraphImpl() noexcept;
654 : :
655 : : // Cannot move or copy (would invalidate TxGraphImpl* in Ref, MiningOrder, EvictionOrder).
656 : : TxGraphImpl(const TxGraphImpl&) = delete;
657 : : TxGraphImpl& operator=(const TxGraphImpl&) = delete;
658 : : TxGraphImpl(TxGraphImpl&&) = delete;
659 : : TxGraphImpl& operator=(TxGraphImpl&&) = delete;
660 : :
661 : : // Simple helper functions.
662 : :
663 : : /** Swap the Entry referred to by a and the one referred to by b. Gather main clusters to
664 : : * update afterwards. */
665 : : void SwapIndexes(GraphIndex a, GraphIndex b, std::vector<Cluster*>& affected_main) noexcept;
666 : : /** If idx exists in the specified level ClusterSet (explicitly, or in the level below and not
667 : : * removed), return the Cluster it is in. Otherwise, return nullptr. */
668 : 719660 : Cluster* FindCluster(GraphIndex idx, int level) const noexcept { return FindClusterAndLevel(idx, level).first; }
669 : : /** Like FindCluster, but also return what level the match was found in (-1 if not found). */
670 : : std::pair<Cluster*, int> FindClusterAndLevel(GraphIndex idx, int level) const noexcept;
671 : : /** Extract a Cluster from its ClusterSet, and set its quality to QualityLevel::NONE. */
672 : : std::unique_ptr<Cluster> ExtractCluster(int level, QualityLevel quality, ClusterSetIndex setindex) noexcept;
673 : : /** Delete a Cluster. */
674 : : void DeleteCluster(Cluster& cluster, int level) noexcept;
675 : : /** Insert a Cluster into its ClusterSet. */
676 : : ClusterSetIndex InsertCluster(int level, std::unique_ptr<Cluster>&& cluster, QualityLevel quality) noexcept;
677 : : /** Change the QualityLevel of a Cluster (identified by old_quality and old_index). */
678 : : void SetClusterQuality(int level, QualityLevel old_quality, ClusterSetIndex old_index, QualityLevel new_quality) noexcept;
679 : : /** Get the index of the top level ClusterSet (staging if it exists, main otherwise). */
680 [ + + ]: 2753182 : int GetTopLevel() const noexcept { return m_staging_clusterset.has_value(); }
681 : : /** Get the specified level (staging if it exists and level is TOP, main otherwise). */
682 [ - - + + : 73781 : int GetSpecifiedLevel(Level level) const noexcept { return level == Level::TOP && m_staging_clusterset.has_value(); }
+ + - - -
- - - - -
- - + + ]
683 : : /** Get a reference to the ClusterSet at the specified level (which must exist). */
684 : : ClusterSet& GetClusterSet(int level) noexcept;
685 : : const ClusterSet& GetClusterSet(int level) const noexcept;
686 : : /** Make a transaction not exist at a specified level. It must currently exist there.
687 : : * oversized_tx indicates whether the transaction is an individually-oversized one
688 : : * (OVERSIZED_SINGLETON). */
689 : : void ClearLocator(int level, GraphIndex index, bool oversized_tx) noexcept;
690 : : /** Find which Clusters in main conflict with ones in staging. */
691 : : std::vector<Cluster*> GetConflicts() const noexcept;
692 : : /** Clear an Entry's ChunkData. */
693 : : void ClearChunkData(Entry& entry) noexcept;
694 : : /** Give an Entry a ChunkData object. */
695 : : void CreateChunkData(GraphIndex idx, LinearizationIndex chunk_count) noexcept;
696 : : /** Create an empty GenericClusterImpl object. */
697 : 1760 : std::unique_ptr<GenericClusterImpl> CreateEmptyGenericCluster() noexcept
698 : : {
699 : 1760 : return std::make_unique<GenericClusterImpl>(m_next_sequence_counter++);
700 : : }
701 : : /** Create an empty SingletonClusterImpl object. */
702 : 137882 : std::unique_ptr<SingletonClusterImpl> CreateEmptySingletonCluster() noexcept
703 : : {
704 : 137882 : return std::make_unique<SingletonClusterImpl>(m_next_sequence_counter++);
705 : : }
706 : : /** Create an empty Cluster of the appropriate implementation for the specified (maximum) tx
707 : : * count. */
708 : 137895 : std::unique_ptr<Cluster> CreateEmptyCluster(DepGraphIndex tx_count) noexcept
709 : : {
710 [ + + ]: 137895 : if (tx_count >= SingletonClusterImpl::MIN_INTENDED_TX_COUNT && tx_count <= SingletonClusterImpl::MAX_TX_COUNT) {
711 : 136467 : return CreateEmptySingletonCluster();
712 : : }
713 [ + - ]: 1428 : if (tx_count >= GenericClusterImpl::MIN_INTENDED_TX_COUNT && tx_count <= GenericClusterImpl::MAX_TX_COUNT) {
714 [ - + ]: 1428 : return CreateEmptyGenericCluster();
715 : : }
716 : 0 : AssertUnreachable();
717 : : }
718 : :
719 : : // Functions for handling Refs.
720 : :
721 : : /** Only called by Ref's move constructor/assignment to update Ref locations. */
722 : 65561 : void UpdateRef(GraphIndex idx, Ref& new_location) noexcept final
723 : : {
724 : 65561 : auto& entry = m_entries[idx];
725 : 65561 : Assume(entry.m_ref != nullptr);
726 : 65561 : entry.m_ref = &new_location;
727 : 65561 : }
728 : :
729 : : /** Only called by Ref::~Ref to unlink Refs, and Ref's move assignment. */
730 : 72073 : void UnlinkRef(GraphIndex idx) noexcept final
731 : : {
732 [ - + ]: 72073 : auto& entry = m_entries[idx];
733 [ - + ]: 72073 : Assume(entry.m_ref != nullptr);
734 [ - + + + ]: 72073 : Assume(m_main_chunkindex_observers == 0 || !entry.m_locator[0].IsPresent());
735 : : // Remove all chunk index entries for the affected cluster, to avoid any chunk indexes
736 : : // referencing unlinked/destroyed Refs.
737 [ + + ]: 72073 : if (entry.m_locator[0].IsPresent()) {
738 : 46125 : entry.m_locator[0].cluster->RemoveChunkData(*this);
739 : : }
740 : 72073 : entry.m_ref = nullptr;
741 : : // Mark the transaction as to be removed in all levels where it explicitly or implicitly
742 : : // exists.
743 : 72073 : bool exists_anywhere{false};
744 : 72073 : bool exists{false};
745 [ + + ]: 144146 : for (int level = 0; level <= GetTopLevel(); ++level) {
746 [ + + ]: 72073 : if (entry.m_locator[level].IsPresent()) {
747 : : exists_anywhere = true;
748 : : exists = true;
749 [ + - ]: 25948 : } else if (entry.m_locator[level].IsRemoved()) {
750 : : exists = false;
751 : : }
752 [ - + ]: 25948 : if (exists) {
753 : 46125 : auto& clusterset = GetClusterSet(level);
754 : 46125 : clusterset.m_to_remove.push_back(idx);
755 : : // Force recomputation of grouping data.
756 [ + + ]: 46125 : clusterset.m_group_data = std::nullopt;
757 : : // Do not wipe the oversized state of main if staging exists. The reason for this
758 : : // is that the alternative would mean that cluster merges may need to be applied to
759 : : // a formerly-oversized main graph while staging exists (to satisfy chunk feerate
760 : : // queries into main, for example), and such merges could conflict with pulls of
761 : : // some of their constituents into staging.
762 [ + - + - ]: 118198 : if (level == GetTopLevel() && clusterset.m_oversized == true) {
763 : 0 : clusterset.m_oversized = std::nullopt;
764 : : }
765 : : }
766 : : }
767 : 72073 : m_unlinked.push_back(idx);
768 [ + + ]: 72073 : if (!exists_anywhere) Compact();
769 : 72073 : }
770 : :
771 : : // Functions related to various normalization/application steps.
772 : : /** Get rid of unlinked Entry objects in m_entries, if possible (this changes the GraphIndex
773 : : * values for remaining Entry objects, so this only does something when no to-be-applied
774 : : * operations or staged removals referring to GraphIndexes remain). */
775 : : void Compact() noexcept;
776 : : /** If cluster is not in staging, copy it there, and return a pointer to it.
777 : : * Staging must exist, and this modifies the locators of its
778 : : * transactions from inherited (P,M) to explicit (P,P). */
779 : : Cluster* PullIn(Cluster* cluster, int level) noexcept;
780 : : /** Apply all removals queued up in m_to_remove to the relevant Clusters (which get a
781 : : * NEEDS_SPLIT* QualityLevel) up to the specified level. */
782 : : void ApplyRemovals(int up_to_level) noexcept;
783 : : /** Split an individual cluster. */
784 : : void Split(Cluster& cluster, int level) noexcept;
785 : : /** Split all clusters that need splitting up to the specified level. */
786 : : void SplitAll(int up_to_level) noexcept;
787 : : /** Populate m_group_data based on m_deps_to_add in the specified level. */
788 : : void GroupClusters(int level) noexcept;
789 : : /** Merge the specified clusters. */
790 : : void Merge(std::span<Cluster*> to_merge, int level) noexcept;
791 : : /** Apply all m_deps_to_add to the relevant Clusters in the specified level. */
792 : : void ApplyDependencies(int level) noexcept;
793 : : /** Make a specified Cluster have quality ACCEPTABLE or OPTIMAL. */
794 : : void MakeAcceptable(Cluster& cluster, int level) noexcept;
795 : : /** Make all Clusters at the specified level have quality ACCEPTABLE or OPTIMAL. */
796 : : void MakeAllAcceptable(int level) noexcept;
797 : :
798 : : // Implementations for the public TxGraph interface.
799 : :
800 : : void AddTransaction(Ref& arg, const FeePerWeight& feerate) noexcept final;
801 : : void RemoveTransaction(const Ref& arg) noexcept final;
802 : : void AddDependency(const Ref& parent, const Ref& child) noexcept final;
803 : : void SetTransactionFee(const Ref&, int64_t fee) noexcept final;
804 : :
805 : : bool DoWork(uint64_t max_cost) noexcept final;
806 : :
807 : : void StartStaging() noexcept final;
808 : : void CommitStaging() noexcept final;
809 : : void AbortStaging() noexcept final;
810 : 71143 : bool HaveStaging() const noexcept final { return m_staging_clusterset.has_value(); }
811 : :
812 : : bool Exists(const Ref& arg, Level level) noexcept final;
813 : : FeePerWeight GetMainChunkFeerate(const Ref& arg) noexcept final;
814 : : FeePerWeight GetIndividualFeerate(const Ref& arg) noexcept final;
815 : : std::vector<Ref*> GetCluster(const Ref& arg, Level level) noexcept final;
816 : : std::vector<Ref*> GetAncestors(const Ref& arg, Level level) noexcept final;
817 : : std::vector<Ref*> GetDescendants(const Ref& arg, Level level) noexcept final;
818 : : std::vector<Ref*> GetAncestorsUnion(std::span<const Ref* const> args, Level level) noexcept final;
819 : : std::vector<Ref*> GetDescendantsUnion(std::span<const Ref* const> args, Level level) noexcept final;
820 : : GraphIndex GetTransactionCount(Level level) noexcept final;
821 : : bool IsOversized(Level level) noexcept final;
822 : : std::strong_ordering CompareMainOrder(const Ref& a, const Ref& b) noexcept final;
823 : : GraphIndex CountDistinctClusters(std::span<const Ref* const> refs, Level level) noexcept final;
824 : : std::pair<std::vector<FeeFrac>, std::vector<FeeFrac>> GetMainStagingDiagrams() noexcept final;
825 : : std::vector<Ref*> Trim() noexcept final;
826 : :
827 : : std::unique_ptr<BlockBuilder> GetBlockBuilder() noexcept final;
828 : : std::pair<std::vector<Ref*>, FeePerWeight> GetWorstMainChunk() noexcept final;
829 : :
830 : : size_t GetMainMemoryUsage() noexcept final;
831 : :
832 : : void SanityCheck() const final;
833 : : };
834 : :
835 : 325534851 : TxGraphImpl::ClusterSet& TxGraphImpl::GetClusterSet(int level) noexcept
836 : : {
837 [ + + ]: 325534851 : if (level == 0) return m_main_clusterset;
838 : 442257 : Assume(level == 1);
839 : 442257 : Assume(m_staging_clusterset.has_value());
840 : 442257 : return *m_staging_clusterset;
841 : : }
842 : :
843 : 224028 : const TxGraphImpl::ClusterSet& TxGraphImpl::GetClusterSet(int level) const noexcept
844 : : {
845 [ + + ]: 224028 : if (level == 0) return m_main_clusterset;
846 : 48682 : Assume(level == 1);
847 : 48682 : Assume(m_staging_clusterset.has_value());
848 : 48682 : return *m_staging_clusterset;
849 : : }
850 : :
851 : : /** Implementation of the TxGraph::BlockBuilder interface. */
852 : : class BlockBuilderImpl final : public TxGraph::BlockBuilder
853 : : {
854 : : /** Which TxGraphImpl this object is doing block building for. It will have its
855 : : * m_main_chunkindex_observers incremented as long as this BlockBuilderImpl exists. */
856 : : TxGraphImpl* const m_graph;
857 : : /** Cluster sequence numbers which we're not including further transactions from. */
858 : : std::unordered_set<uint64_t> m_excluded_clusters;
859 : : /** Iterator to the current chunk in the chunk index. end() if nothing further remains. */
860 : : TxGraphImpl::ChunkIndex::const_iterator m_cur_iter;
861 : : /** Which cluster the current chunk belongs to, so we can exclude further transactions from it
862 : : * when that chunk is skipped. */
863 : : Cluster* m_cur_cluster;
864 : : /** Whether we know that m_cur_iter points to the last chunk of m_cur_cluster. */
865 : : bool m_known_end_of_cluster;
866 : :
867 : : // Move m_cur_iter / m_cur_cluster to the next acceptable chunk.
868 : : void Next() noexcept;
869 : :
870 : : public:
871 : : /** Construct a new BlockBuilderImpl to build blocks for the provided graph. */
872 : : BlockBuilderImpl(TxGraphImpl& graph) noexcept;
873 : :
874 : : // Implement the public interface.
875 : : ~BlockBuilderImpl() final;
876 : : std::optional<std::pair<std::vector<TxGraph::Ref*>, FeePerWeight>> GetCurrentChunk() noexcept final;
877 : : void Include() noexcept final;
878 : : void Skip() noexcept final;
879 : : };
880 : :
881 : 528624 : void TxGraphImpl::ClearChunkData(Entry& entry) noexcept
882 : : {
883 [ + + ]: 528624 : if (entry.m_main_chunkindex_iterator != m_main_chunkindex.end()) {
884 : 110901 : Assume(m_main_chunkindex_observers == 0);
885 : : // If the Entry has a non-empty m_main_chunkindex_iterator, extract it, and move the handle
886 : : // to the cache of discarded chunkindex entries.
887 : 110901 : m_main_chunkindex_discarded.emplace_back(m_main_chunkindex.extract(entry.m_main_chunkindex_iterator));
888 : 110901 : entry.m_main_chunkindex_iterator = m_main_chunkindex.end();
889 : : }
890 : 528624 : }
891 : :
892 : 174901 : void TxGraphImpl::CreateChunkData(GraphIndex idx, LinearizationIndex chunk_count) noexcept
893 : : {
894 [ + + ]: 174901 : auto& entry = m_entries[idx];
895 : : // Make sure to not create chunk data for unlinked entries, which would make invoking
896 : : // m_fallback_order on them impossible.
897 [ + + ]: 174901 : Assume(entry.m_ref != nullptr);
898 [ + + ]: 174901 : if (!m_main_chunkindex_discarded.empty()) {
899 : : // Reuse an discarded node handle.
900 : 1319 : auto& node = m_main_chunkindex_discarded.back().value();
901 : 1319 : node.m_graph_index = idx;
902 : 1319 : node.m_chunk_count = chunk_count;
903 : 1319 : auto insert_result = m_main_chunkindex.insert(std::move(m_main_chunkindex_discarded.back()));
904 : 1319 : Assume(insert_result.inserted);
905 : 1319 : entry.m_main_chunkindex_iterator = insert_result.position;
906 : 1319 : m_main_chunkindex_discarded.pop_back();
907 : 1319 : } else {
908 : : // Construct a new entry.
909 : 173582 : auto emplace_result = m_main_chunkindex.emplace(idx, chunk_count);
910 : 173582 : Assume(emplace_result.second);
911 : 173582 : entry.m_main_chunkindex_iterator = emplace_result.first;
912 : : }
913 : 174901 : }
914 : :
915 : 84248 : size_t GenericClusterImpl::TotalMemoryUsage() const noexcept
916 : : {
917 : 84248 : return // Dynamic memory allocated in this Cluster.
918 [ - + - + ]: 168496 : memusage::DynamicUsage(m_mapping) + memusage::DynamicUsage(m_linearization) +
919 : : // Dynamic memory usage inside m_depgraph.
920 [ - + ]: 84248 : m_depgraph.DynamicMemoryUsage() +
921 : : // Memory usage of the allocated Cluster itself.
922 : 84248 : memusage::MallocUsage(sizeof(GenericClusterImpl)) +
923 : : // Memory usage of the ClusterSet::m_clusters entry.
924 : 84248 : sizeof(std::unique_ptr<Cluster>);
925 : : }
926 : :
927 : 12291068 : size_t SingletonClusterImpl::TotalMemoryUsage() const noexcept
928 : : {
929 : 12291068 : return // Memory usage of the allocated SingletonClusterImpl itself.
930 : 12291068 : memusage::MallocUsage(sizeof(SingletonClusterImpl)) +
931 : : // Memory usage of the ClusterSet::m_clusters entry.
932 : 12291068 : sizeof(std::unique_ptr<Cluster>);
933 : : }
934 : :
935 : 74874 : uint64_t GenericClusterImpl::GetTotalTxSize() const noexcept
936 : : {
937 : 74874 : uint64_t ret{0};
938 [ + + ]: 566314 : for (auto i : m_linearization) {
939 : 491440 : ret += m_depgraph.FeeRate(i).size;
940 : : }
941 : 74874 : return ret;
942 : : }
943 : :
944 : 17 : DepGraphIndex GenericClusterImpl::AppendTransaction(GraphIndex graph_idx, FeePerWeight feerate) noexcept
945 : : {
946 : 17 : Assume(graph_idx != GraphIndex(-1));
947 : 17 : auto ret = m_depgraph.AddTransaction(feerate);
948 : 17 : m_mapping.push_back(graph_idx);
949 : 17 : m_linearization.push_back(ret);
950 : 17 : return ret;
951 : : }
952 : :
953 : 136467 : DepGraphIndex SingletonClusterImpl::AppendTransaction(GraphIndex graph_idx, FeePerWeight feerate) noexcept
954 : : {
955 : 136467 : Assume(!GetTxCount());
956 : 136467 : m_graph_index = graph_idx;
957 : 136467 : m_feerate = feerate;
958 : 136467 : return 0;
959 : : }
960 : :
961 : 17 : void GenericClusterImpl::AddDependencies(SetType parents, DepGraphIndex child) noexcept
962 : : {
963 : 17 : m_depgraph.AddDependencies(parents, child);
964 : 17 : }
965 : :
966 : 383 : void SingletonClusterImpl::AddDependencies(SetType parents, DepGraphIndex child) noexcept
967 : : {
968 : : // Singletons cannot have any dependencies.
969 [ - + ]: 383 : Assume(child == 0);
970 [ - + ]: 383 : Assume(parents == SetType{} || parents == SetType::Fill(0));
971 : 383 : }
972 : :
973 : 11 : void GenericClusterImpl::ExtractTransactions(const std::function<void (DepGraphIndex, GraphIndex, FeePerWeight)>& visit1_fn, const std::function<void (DepGraphIndex, GraphIndex, SetType)>& visit2_fn) noexcept
974 : : {
975 [ + + ]: 47 : for (auto pos : m_linearization) {
976 : 36 : visit1_fn(pos, m_mapping[pos], FeePerWeight::FromFeeFrac(m_depgraph.FeeRate(pos)));
977 : : }
978 [ + + ]: 47 : for (auto pos : m_linearization) {
979 : 36 : visit2_fn(pos, m_mapping[pos], m_depgraph.GetReducedParents(pos));
980 : : }
981 : : // Purge this Cluster, now that everything has been moved.
982 : 11 : m_depgraph = DepGraph<SetType>{};
983 [ + - ]: 11 : m_linearization.clear();
984 [ + - ]: 11 : m_mapping.clear();
985 : 11 : }
986 : :
987 : 7920 : void SingletonClusterImpl::ExtractTransactions(const std::function<void (DepGraphIndex, GraphIndex, FeePerWeight)>& visit1_fn, const std::function<void (DepGraphIndex, GraphIndex, SetType)>& visit2_fn) noexcept
988 : : {
989 [ + - ]: 7920 : if (GetTxCount()) {
990 : 7920 : visit1_fn(0, m_graph_index, m_feerate);
991 : 7920 : visit2_fn(0, m_graph_index, SetType{});
992 : 7920 : m_graph_index = NO_GRAPH_INDEX;
993 : : }
994 : 7920 : }
995 : :
996 : 69352 : int GenericClusterImpl::GetLevel(const TxGraphImpl& graph) const noexcept
997 : : {
998 : : // GetLevel() does not work for empty Clusters.
999 [ + - ]: 69352 : if (!Assume(!m_linearization.empty())) return -1;
1000 : :
1001 : : // Pick an arbitrary Entry that occurs in this Cluster.
1002 : 69352 : const auto& entry = graph.m_entries[m_mapping[m_linearization.front()]];
1003 : : // See if there is a level whose Locator matches this Cluster, if so return that level.
1004 [ + - ]: 69374 : for (int level = 0; level < MAX_LEVELS; ++level) {
1005 [ + + ]: 69374 : if (entry.m_locator[level].cluster == this) return level;
1006 : : }
1007 : : // Given that we started with an Entry that occurs in this Cluster, one of its Locators must
1008 : : // point back to it.
1009 : 0 : AssertUnreachable();
1010 : : }
1011 : :
1012 : 12139174 : int SingletonClusterImpl::GetLevel(const TxGraphImpl& graph) const noexcept
1013 : : {
1014 : : // GetLevel() does not work for empty Clusters.
1015 [ + - ]: 12139174 : if (!Assume(GetTxCount())) return -1;
1016 : :
1017 : : // Get the Entry in this Cluster.
1018 : 12139174 : const auto& entry = graph.m_entries[m_graph_index];
1019 : : // See if there is a level whose Locator matches this Cluster, if so return that level.
1020 [ + - ]: 12139454 : for (int level = 0; level < MAX_LEVELS; ++level) {
1021 [ + + ]: 12139454 : if (entry.m_locator[level].cluster == this) return level;
1022 : : }
1023 : : // Given that we started with an Entry that occurs in this Cluster, one of its Locators must
1024 : : // point back to it.
1025 : 0 : AssertUnreachable();
1026 : : }
1027 : :
1028 : 29675 : void TxGraphImpl::ClearLocator(int level, GraphIndex idx, bool oversized_tx) noexcept
1029 : : {
1030 [ + + ]: 29675 : auto& entry = m_entries[idx];
1031 : 29675 : auto& clusterset = GetClusterSet(level);
1032 [ + + ]: 29675 : Assume(entry.m_locator[level].IsPresent());
1033 : : // Change the locator from Present to Missing or Removed.
1034 [ + + - + ]: 29675 : if (level == 0 || !entry.m_locator[level - 1].IsPresent()) {
1035 : 27498 : entry.m_locator[level].SetMissing();
1036 : : } else {
1037 : 2177 : entry.m_locator[level].SetRemoved();
1038 : 2177 : clusterset.m_removed.push_back(idx);
1039 : : }
1040 : : // Update the transaction count.
1041 : 29675 : --clusterset.m_txcount;
1042 : 29675 : clusterset.m_txcount_oversized -= oversized_tx;
1043 : : // If clearing main, adjust the status of Locators of this transaction in staging, if it exists.
1044 [ + + + + ]: 29675 : if (level == 0 && GetTopLevel() == 1) {
1045 [ + + ]: 2124 : if (entry.m_locator[1].IsRemoved()) {
1046 : 1580 : entry.m_locator[1].SetMissing();
1047 [ - + ]: 544 : } else if (!entry.m_locator[1].IsPresent()) {
1048 : 0 : --m_staging_clusterset->m_txcount;
1049 : 0 : m_staging_clusterset->m_txcount_oversized -= oversized_tx;
1050 : : }
1051 : : }
1052 [ + + ]: 29675 : if (level == 0) ClearChunkData(entry);
1053 : 29675 : }
1054 : :
1055 : 6978 : void GenericClusterImpl::RemoveChunkData(TxGraphImpl& graph) noexcept
1056 : : {
1057 [ + + ]: 176781 : for (DepGraphIndex idx : m_linearization) {
1058 : 169803 : auto& entry = graph.m_entries[m_mapping[idx]];
1059 : 169803 : graph.ClearChunkData(entry);
1060 : : }
1061 : 6978 : }
1062 : :
1063 : 39147 : void SingletonClusterImpl::RemoveChunkData(TxGraphImpl& graph) noexcept
1064 : : {
1065 [ + - ]: 39147 : if (GetTxCount() == 0) return;
1066 : 39147 : auto& entry = graph.m_entries[m_graph_index];
1067 : 39147 : graph.ClearChunkData(entry);
1068 : : }
1069 : :
1070 : 14362 : void GenericClusterImpl::Updated(TxGraphImpl& graph, int level, bool rename) noexcept
1071 : : {
1072 : : // Update all the Locators for this Cluster's Entry objects.
1073 [ + + ]: 195568 : for (DepGraphIndex idx : m_linearization) {
1074 [ + + ]: 181206 : auto& entry = graph.m_entries[m_mapping[idx]];
1075 : : // Discard any potential ChunkData prior to modifying the Cluster (as that could
1076 : : // invalidate its ordering).
1077 [ + + ]: 181206 : if (level == 0 && !rename) graph.ClearChunkData(entry);
1078 : 181206 : entry.m_locator[level].SetPresent(this, idx);
1079 : : }
1080 : : // If this is for the main graph (level = 0), and the Cluster's quality is ACCEPTABLE or
1081 : : // OPTIMAL, compute its chunking and store its information in the Entry's m_main_lin_index
1082 : : // and m_main_chunk_feerate. These fields are only accessed after making the entire graph
1083 : : // ACCEPTABLE, so it is pointless to compute these if we haven't reached that quality level
1084 : : // yet.
1085 : : // When rename=true, this is always performed for level 0, to make sure the values inside the
1086 : : // entries remain consistent with the chunk index (otherwise unrelated chunk index operations
1087 : : // could cause the index to become corrupted, by inserting elements in the wrong place).
1088 [ + + + + : 14362 : if (level == 0 && (rename || IsAcceptable())) {
+ + ]
1089 [ - + ]: 6676 : auto chunking = ChunkLinearizationInfo(m_depgraph, m_linearization);
1090 : 6676 : LinearizationIndex lin_idx{0};
1091 : : /** The sum of all chunk feerate FeeFracs with the same feerate as the current chunk,
1092 : : * up to and including the current chunk. */
1093 : 6676 : FeeFrac equal_feerate_chunk_feerate;
1094 : : // Iterate over the chunks.
1095 [ - + + + ]: 73577 : for (unsigned chunk_idx = 0; chunk_idx < chunking.size(); ++chunk_idx) {
1096 [ + + ]: 66901 : auto& chunk = chunking[chunk_idx];
1097 [ + + ]: 66901 : auto chunk_count = chunk.transactions.Count();
1098 [ + + ]: 66901 : Assume(chunk_count > 0);
1099 : : // Update equal_feerate_chunk_feerate to include this chunk, starting over when the
1100 : : // feerate changed.
1101 [ + + ]: 66901 : if (ByRatio{chunk.feerate} < ByRatio{equal_feerate_chunk_feerate}) {
1102 : 3292 : equal_feerate_chunk_feerate = chunk.feerate;
1103 : : } else {
1104 : : // Note that this is adding fees to fees, and sizes to sizes, so the overall
1105 : : // ratio remains the same; it's just accounting for the size of the added chunk.
1106 : 63609 : equal_feerate_chunk_feerate += chunk.feerate;
1107 : : }
1108 : : // Determine the m_fallback_order maximum transaction in the chunk.
1109 [ + - ]: 66901 : auto it = chunk.transactions.begin();
1110 : 66901 : GraphIndex max_element = m_mapping[*it];
1111 : 66901 : ++it;
1112 [ + + ]: 158161 : while (it != chunk.transactions.end()) {
1113 : 24359 : GraphIndex this_element = m_mapping[*it];
1114 [ + + ]: 24359 : if (graph.m_fallback_order(*graph.m_entries[this_element].m_ref, *graph.m_entries[max_element].m_ref) > 0) {
1115 : 2904 : max_element = this_element;
1116 : : }
1117 : 24359 : ++it;
1118 : : }
1119 : : // Iterate over the transactions in the linearization, which must match those in chunk.
1120 : 91260 : while (true) {
1121 [ + + ]: 91260 : DepGraphIndex idx = m_linearization[lin_idx];
1122 : 91260 : GraphIndex graph_idx = m_mapping[idx];
1123 : 91260 : auto& entry = graph.m_entries[graph_idx];
1124 : 91260 : entry.m_main_lin_index = lin_idx++;
1125 [ + + ]: 91260 : entry.m_main_chunk_feerate = FeePerWeight::FromFeeFrac(chunk.feerate);
1126 : 91260 : entry.m_main_equal_feerate_chunk_prefix_size = equal_feerate_chunk_feerate.size;
1127 : 91260 : entry.m_main_max_chunk_fallback = max_element;
1128 [ + + ]: 91260 : Assume(chunk.transactions[idx]);
1129 [ + + ]: 91260 : chunk.transactions.Reset(idx);
1130 [ + + ]: 91260 : if (chunk.transactions.None()) {
1131 : : // Last transaction in the chunk.
1132 [ + + - + : 66901 : if (chunk_count == 1 && chunk_idx + 1 == chunking.size()) {
+ + ]
1133 : : // If this is the final chunk of the cluster, and it contains just a single
1134 : : // transaction (which will always be true for the very common singleton
1135 : : // clusters), store the special value -1 as chunk count.
1136 : : chunk_count = LinearizationIndex(-1);
1137 : : }
1138 [ + + ]: 66901 : if (!rename) graph.CreateChunkData(graph_idx, chunk_count);
1139 : 66901 : break;
1140 : : }
1141 : : }
1142 : : }
1143 : 6676 : }
1144 : 14362 : }
1145 : :
1146 : 195823 : void SingletonClusterImpl::Updated(TxGraphImpl& graph, int level, bool rename) noexcept
1147 : : {
1148 : : // Don't do anything if this is empty.
1149 [ + - ]: 195823 : if (GetTxCount() == 0) return;
1150 : :
1151 [ + + ]: 195823 : auto& entry = graph.m_entries[m_graph_index];
1152 : : // Discard any potential ChunkData prior to modifying the Cluster (as that could
1153 : : // invalidate its ordering).
1154 [ + + ]: 195823 : if (level == 0 && !rename) graph.ClearChunkData(entry);
1155 : 195823 : entry.m_locator[level].SetPresent(this, 0);
1156 : : // If this is for the main graph (level = 0), compute its chunking and store its information in
1157 : : // the Entry's m_main_lin_index and m_main_chunk_feerate.
1158 [ + + + + : 195823 : if (level == 0 && (rename || IsAcceptable())) {
+ + ]
1159 : 122329 : entry.m_main_lin_index = 0;
1160 : 122329 : entry.m_main_chunk_feerate = m_feerate;
1161 : 122329 : entry.m_main_equal_feerate_chunk_prefix_size = m_feerate.size;
1162 : 122329 : entry.m_main_max_chunk_fallback = m_graph_index;
1163 : : // Always use the special LinearizationIndex(-1), indicating singleton chunk at end of
1164 : : // Cluster, here.
1165 [ + + ]: 122329 : if (!rename) graph.CreateChunkData(m_graph_index, LinearizationIndex(-1));
1166 : : }
1167 : : }
1168 : :
1169 : 135 : void GenericClusterImpl::GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept
1170 : : {
1171 [ + + ]: 1383 : for (auto i : m_linearization) {
1172 [ + + ]: 1248 : auto& entry = graph.m_entries[m_mapping[i]];
1173 : : // For every transaction Entry in this Cluster, if it also exists in a lower-level Cluster,
1174 : : // then that Cluster conflicts.
1175 [ + + ]: 1248 : if (entry.m_locator[0].IsPresent()) {
1176 : 1088 : out.push_back(entry.m_locator[0].cluster);
1177 : : }
1178 : : }
1179 : 135 : }
1180 : :
1181 : 48736 : void SingletonClusterImpl::GetConflicts(const TxGraphImpl& graph, std::vector<Cluster*>& out) const noexcept
1182 : : {
1183 : : // Empty clusters have no conflicts.
1184 [ + - ]: 48736 : if (GetTxCount() == 0) return;
1185 : :
1186 [ + + ]: 48736 : auto& entry = graph.m_entries[m_graph_index];
1187 : : // If the transaction in this Cluster also exists in a lower-level Cluster, then that Cluster
1188 : : // conflicts.
1189 [ + + ]: 48736 : if (entry.m_locator[0].IsPresent()) {
1190 : 130 : out.push_back(entry.m_locator[0].cluster);
1191 : : }
1192 : : }
1193 : :
1194 : 48682 : std::vector<Cluster*> TxGraphImpl::GetConflicts() const noexcept
1195 : : {
1196 : 48682 : Assume(GetTopLevel() == 1);
1197 : 48682 : auto& clusterset = GetClusterSet(1);
1198 : 48682 : std::vector<Cluster*> ret;
1199 : : // All main Clusters containing transactions in m_removed (so (P,R) ones) are conflicts.
1200 [ + + ]: 52428 : for (auto i : clusterset.m_removed) {
1201 [ + - ]: 3746 : auto& entry = m_entries[i];
1202 [ + - ]: 3746 : if (entry.m_locator[0].IsPresent()) {
1203 : 3746 : ret.push_back(entry.m_locator[0].cluster);
1204 : : }
1205 : : }
1206 : : // Then go over all Clusters at this level, and find their conflicts (the (P,P) ones).
1207 [ + + ]: 389456 : for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
1208 : 340774 : auto& clusters = clusterset.m_clusters[quality];
1209 [ + + ]: 389645 : for (const auto& cluster : clusters) {
1210 : 48871 : cluster->GetConflicts(*this, ret);
1211 : : }
1212 : : }
1213 : : // Deduplicate the result (the same Cluster may appear multiple times).
1214 : 48682 : std::ranges::sort(ret, [](Cluster* a, Cluster* b) noexcept { return CompareClusters(a, b) < 0; });
1215 : 48682 : ret.erase(std::ranges::unique(ret).begin(), ret.end());
1216 : 48682 : return ret;
1217 : : }
1218 : :
1219 : 332 : Cluster* GenericClusterImpl::CopyToStaging(TxGraphImpl& graph) const noexcept
1220 : : {
1221 : : // Construct an empty Cluster.
1222 : 332 : auto ret = graph.CreateEmptyGenericCluster();
1223 : 332 : auto ptr = ret.get();
1224 : : // Copy depgraph, mapping, and linearization.
1225 : 332 : ptr->m_depgraph = m_depgraph;
1226 : 332 : ptr->m_mapping = m_mapping;
1227 : 332 : ptr->m_linearization = m_linearization;
1228 : : // Insert the new Cluster into the graph.
1229 : 332 : graph.InsertCluster(/*level=*/1, std::move(ret), m_quality);
1230 : : // Update its Locators.
1231 : 332 : ptr->Updated(graph, /*level=*/1, /*rename=*/false);
1232 : : // Update memory usage.
1233 : 332 : graph.GetClusterSet(/*level=*/1).m_cluster_usage += ptr->TotalMemoryUsage();
1234 [ - + ]: 332 : return ptr;
1235 : 332 : }
1236 : :
1237 : 1415 : Cluster* SingletonClusterImpl::CopyToStaging(TxGraphImpl& graph) const noexcept
1238 : : {
1239 : : // Construct an empty Cluster.
1240 : 1415 : auto ret = graph.CreateEmptySingletonCluster();
1241 : 1415 : auto ptr = ret.get();
1242 : : // Copy data.
1243 : 1415 : ptr->m_graph_index = m_graph_index;
1244 : 1415 : ptr->m_feerate = m_feerate;
1245 : : // Insert the new Cluster into the graph.
1246 : 1415 : graph.InsertCluster(/*level=*/1, std::move(ret), m_quality);
1247 : : // Update its Locators.
1248 : 1415 : ptr->Updated(graph, /*level=*/1, /*rename=*/false);
1249 : : // Update memory usage.
1250 : 1415 : graph.GetClusterSet(/*level=*/1).m_cluster_usage += ptr->TotalMemoryUsage();
1251 : 2830 : return ptr;
1252 : 1415 : }
1253 : :
1254 : 1255 : void GenericClusterImpl::ApplyRemovals(TxGraphImpl& graph, int level, std::span<GraphIndex>& to_remove) noexcept
1255 : : {
1256 : : // Iterate over the prefix of to_remove that applies to this cluster.
1257 : 1255 : Assume(!to_remove.empty());
1258 : 1255 : SetType todo;
1259 : 1255 : graph.GetClusterSet(level).m_cluster_usage -= TotalMemoryUsage();
1260 : 6147 : do {
1261 [ - + ]: 6147 : GraphIndex idx = to_remove.front();
1262 [ - + + + ]: 6147 : Assume(idx < graph.m_entries.size());
1263 [ + + ]: 6147 : auto& entry = graph.m_entries[idx];
1264 : : auto& locator = entry.m_locator[level];
1265 : : // Stop once we hit an entry that applies to another Cluster.
1266 [ + + ]: 6147 : if (locator.cluster != this) break;
1267 : : // - Remember it in a set of to-remove DepGraphIndexes.
1268 [ + + ]: 5377 : todo.Set(locator.index);
1269 : : // - Remove from m_mapping. This isn't strictly necessary as unused positions in m_mapping
1270 : : // are just never accessed, but set it to -1 here to increase the ability to detect a bug
1271 : : // that causes it to be accessed regardless.
1272 [ + + ]: 5377 : m_mapping[locator.index] = GraphIndex(-1);
1273 : : // - Remove its linearization index from the Entry (if in main).
1274 [ + + ]: 5377 : if (level == 0) {
1275 : 4612 : entry.m_main_lin_index = LinearizationIndex(-1);
1276 : : }
1277 : : // - Mark it as missing/removed in the Entry's locator.
1278 : 5377 : graph.ClearLocator(level, idx, m_quality == QualityLevel::OVERSIZED_SINGLETON);
1279 [ + + ]: 5377 : to_remove = to_remove.subspan(1);
1280 [ + + ]: 5377 : } while(!to_remove.empty());
1281 : :
1282 : 1255 : Assume(todo.Any());
1283 : : // Wipe from the Cluster's DepGraph (this is O(n) regardless of the number of entries
1284 : : // removed, so we benefit from batching all the removals).
1285 : 1255 : m_depgraph.RemoveTransactions(todo);
1286 [ - + ]: 1255 : m_mapping.resize(m_depgraph.PositionRange());
1287 : :
1288 : : // Filter removed transactions out of m_linearization.
1289 : 1255 : std::erase_if(m_linearization, [&](auto pos) { return todo[pos]; });
1290 : :
1291 : 1255 : Compact();
1292 : 1255 : graph.GetClusterSet(level).m_cluster_usage += TotalMemoryUsage();
1293 [ + + ]: 1255 : auto new_quality = IsTopological() ? QualityLevel::NEEDS_SPLIT : QualityLevel::NEEDS_SPLIT_FIX;
1294 : 1255 : graph.SetClusterQuality(level, m_quality, m_setindex, new_quality);
1295 : 1255 : Updated(graph, /*level=*/level, /*rename=*/false);
1296 : 1255 : }
1297 : :
1298 : 22174 : void SingletonClusterImpl::ApplyRemovals(TxGraphImpl& graph, int level, std::span<GraphIndex>& to_remove) noexcept
1299 : : {
1300 : : // We can only remove the one transaction this Cluster has.
1301 : 22174 : Assume(!to_remove.empty());
1302 : 22174 : Assume(GetTxCount());
1303 : 22174 : Assume(to_remove.front() == m_graph_index);
1304 : : // Pop all copies of m_graph_index from the front of to_remove (at least one, but there may be
1305 : : // multiple).
1306 : 22174 : do {
1307 [ + + ]: 22174 : to_remove = to_remove.subspan(1);
1308 [ + + + - : 41129 : } while (!to_remove.empty() && to_remove.front() == m_graph_index);
- + ]
1309 : : // Clear this cluster.
1310 : 22174 : graph.ClearLocator(level, m_graph_index, m_quality == QualityLevel::OVERSIZED_SINGLETON);
1311 : 22174 : m_graph_index = NO_GRAPH_INDEX;
1312 : 22174 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::NEEDS_SPLIT);
1313 : : // No need to account for m_cluster_usage changes here, as SingletonClusterImpl has constant
1314 : : // memory usage.
1315 : 22174 : }
1316 : :
1317 : 230 : void GenericClusterImpl::Clear(TxGraphImpl& graph, int level) noexcept
1318 : : {
1319 [ - + ]: 230 : Assume(GetTxCount());
1320 : 230 : graph.GetClusterSet(level).m_cluster_usage -= TotalMemoryUsage();
1321 [ + + ]: 1325 : for (auto i : m_linearization) {
1322 : 1095 : graph.ClearLocator(level, m_mapping[i], m_quality == QualityLevel::OVERSIZED_SINGLETON);
1323 : : }
1324 : 230 : m_depgraph = {};
1325 [ + - ]: 230 : m_linearization.clear();
1326 [ + - ]: 230 : m_mapping.clear();
1327 : 230 : }
1328 : :
1329 : 1029 : void SingletonClusterImpl::Clear(TxGraphImpl& graph, int level) noexcept
1330 : : {
1331 : 1029 : Assume(GetTxCount());
1332 : 1029 : graph.GetClusterSet(level).m_cluster_usage -= TotalMemoryUsage();
1333 : 1029 : graph.ClearLocator(level, m_graph_index, m_quality == QualityLevel::OVERSIZED_SINGLETON);
1334 : 1029 : m_graph_index = NO_GRAPH_INDEX;
1335 : 1029 : }
1336 : :
1337 : 56 : void GenericClusterImpl::MoveToMain(TxGraphImpl& graph) noexcept
1338 : : {
1339 [ + + ]: 614 : for (auto i : m_linearization) {
1340 : 558 : GraphIndex idx = m_mapping[i];
1341 : 558 : auto& entry = graph.m_entries[idx];
1342 : 558 : entry.m_locator[1].SetMissing();
1343 : : }
1344 : 56 : auto quality = m_quality;
1345 : : // Subtract memory usage from staging and add it to main.
1346 : 56 : graph.GetClusterSet(/*level=*/1).m_cluster_usage -= TotalMemoryUsage();
1347 : 56 : graph.GetClusterSet(/*level=*/0).m_cluster_usage += TotalMemoryUsage();
1348 : : // Remove cluster itself from staging and add it to main.
1349 : 56 : auto cluster = graph.ExtractCluster(1, quality, m_setindex);
1350 : 56 : graph.InsertCluster(/*level=*/0, std::move(cluster), quality);
1351 : 56 : Updated(graph, /*level=*/0, /*rename=*/false);
1352 : 56 : }
1353 : :
1354 : 47397 : void SingletonClusterImpl::MoveToMain(TxGraphImpl& graph) noexcept
1355 : : {
1356 [ + - ]: 47397 : if (GetTxCount()) {
1357 : 47397 : auto& entry = graph.m_entries[m_graph_index];
1358 : 47397 : entry.m_locator[1].SetMissing();
1359 : : }
1360 : 47397 : auto quality = m_quality;
1361 : 47397 : graph.GetClusterSet(/*level=*/1).m_cluster_usage -= TotalMemoryUsage();
1362 : 47397 : auto cluster = graph.ExtractCluster(/*level=*/1, quality, m_setindex);
1363 : 47397 : graph.InsertCluster(/*level=*/0, std::move(cluster), quality);
1364 : 47397 : graph.GetClusterSet(/*level=*/0).m_cluster_usage += TotalMemoryUsage();
1365 : 47397 : Updated(graph, /*level=*/0, /*rename=*/false);
1366 : 47397 : }
1367 : :
1368 : 7215 : void GenericClusterImpl::Compact() noexcept
1369 : : {
1370 : 7215 : m_linearization.shrink_to_fit();
1371 : 7215 : m_mapping.shrink_to_fit();
1372 : 7215 : m_depgraph.Compact();
1373 : 7215 : }
1374 : :
1375 : 383 : void SingletonClusterImpl::Compact() noexcept
1376 : : {
1377 : : // Nothing to compact; SingletonClusterImpl is constant size.
1378 : 383 : }
1379 : :
1380 : 410 : void GenericClusterImpl::AppendChunkFeerates(std::vector<FeeFrac>& ret) const noexcept
1381 : : {
1382 [ - + ]: 410 : auto chunk_feerates = ChunkLinearization(m_depgraph, m_linearization);
1383 [ - + - + ]: 410 : ret.reserve(ret.size() + chunk_feerates.size());
1384 : 410 : ret.insert(ret.end(), chunk_feerates.begin(), chunk_feerates.end());
1385 : 410 : }
1386 : :
1387 : 2744 : void SingletonClusterImpl::AppendChunkFeerates(std::vector<FeeFrac>& ret) const noexcept
1388 : : {
1389 [ + - ]: 2744 : if (GetTxCount()) {
1390 : 2744 : ret.push_back(m_feerate);
1391 : : }
1392 : 2744 : }
1393 : :
1394 : 0 : uint64_t GenericClusterImpl::AppendTrimData(std::vector<TrimTxData>& ret, std::vector<std::pair<GraphIndex, GraphIndex>>& deps) const noexcept
1395 : : {
1396 [ # # ]: 0 : Assume(IsAcceptable());
1397 [ # # ]: 0 : auto linchunking = ChunkLinearizationInfo(m_depgraph, m_linearization);
1398 : 0 : LinearizationIndex pos{0};
1399 : 0 : uint64_t size{0};
1400 : 0 : auto prev_index = GraphIndex(-1);
1401 : : // Iterate over the chunks of this cluster's linearization.
1402 [ # # ]: 0 : for (const auto& [chunk, chunk_feerate] : linchunking) {
1403 : : // Iterate over the transactions of that chunk, in linearization order.
1404 : 0 : auto chunk_tx_count = chunk.Count();
1405 [ # # ]: 0 : for (unsigned j = 0; j < chunk_tx_count; ++j) {
1406 : 0 : auto cluster_idx = m_linearization[pos];
1407 : : // The transaction must appear in the chunk.
1408 : 0 : Assume(chunk[cluster_idx]);
1409 : : // Construct a new element in ret.
1410 : 0 : auto& entry = ret.emplace_back();
1411 [ # # ]: 0 : entry.m_chunk_feerate = FeePerWeight::FromFeeFrac(chunk_feerate);
1412 [ # # ]: 0 : entry.m_index = m_mapping[cluster_idx];
1413 : : // If this is not the first transaction of the cluster linearization, it has an
1414 : : // implicit dependency on its predecessor.
1415 [ # # ]: 0 : if (pos != 0) deps.emplace_back(prev_index, entry.m_index);
1416 : 0 : prev_index = entry.m_index;
1417 : 0 : entry.m_tx_size = m_depgraph.FeeRate(cluster_idx).size;
1418 : 0 : size += entry.m_tx_size;
1419 : 0 : ++pos;
1420 : : }
1421 : : }
1422 : 0 : return size;
1423 : 0 : }
1424 : :
1425 : 64217 : uint64_t SingletonClusterImpl::AppendTrimData(std::vector<TrimTxData>& ret, std::vector<std::pair<GraphIndex, GraphIndex>>& deps) const noexcept
1426 : : {
1427 [ + - ]: 64217 : if (!GetTxCount()) return 0;
1428 : 64217 : auto& entry = ret.emplace_back();
1429 : 64217 : entry.m_chunk_feerate = m_feerate;
1430 : 64217 : entry.m_index = m_graph_index;
1431 : 64217 : entry.m_tx_size = m_feerate.size;
1432 : 64217 : return m_feerate.size;
1433 : : }
1434 : :
1435 : 1253 : bool GenericClusterImpl::Split(TxGraphImpl& graph, int level) noexcept
1436 : : {
1437 : : // This function can only be called when the Cluster needs splitting.
1438 [ + + ]: 1253 : Assume(NeedsSplitting());
1439 : : // Determine the new quality the split-off Clusters will have.
1440 [ + + ]: 1253 : QualityLevel new_quality = IsTopological() ? QualityLevel::NEEDS_RELINEARIZE : QualityLevel::NEEDS_FIX;
1441 : : /** Which positions are still left in this Cluster. */
1442 : 1253 : auto todo = m_depgraph.Positions();
1443 : : /** Mapping from transaction positions in this Cluster to the Cluster where it ends up, and
1444 : : * its position therein. */
1445 [ - + ]: 1253 : std::vector<std::pair<Cluster*, DepGraphIndex>> remap(m_depgraph.PositionRange());
1446 : 1253 : std::vector<Cluster*> new_clusters;
1447 : 1253 : bool first{true};
1448 : : // Iterate over the connected components of this Cluster's m_depgraph.
1449 [ + + ]: 1643 : while (todo.Any()) {
1450 : 412 : auto component = m_depgraph.FindConnectedComponent(todo);
1451 [ + + ]: 412 : auto component_size = component.Count();
1452 [ + + ]: 412 : auto split_quality = component_size <= 1 ? QualityLevel::OPTIMAL : new_quality;
1453 [ + + + + : 626 : if (first && component == todo && SetType::Fill(component_size) == component && component_size >= MIN_INTENDED_TX_COUNT) {
+ + ]
1454 : : // The existing Cluster is an entire component, without holes. Leave it be, but update
1455 : : // its quality. If there are holes, we continue, so that the Cluster is reconstructed
1456 : : // without holes, reducing memory usage. If the component's size is below the intended
1457 : : // transaction count for this Cluster implementation, continue so that it can get
1458 : : // converted.
1459 : 22 : Assume(todo == m_depgraph.Positions());
1460 : 22 : graph.SetClusterQuality(level, m_quality, m_setindex, split_quality);
1461 : : // If this made the quality ACCEPTABLE or OPTIMAL, we need to compute and cache its
1462 : : // chunking.
1463 : 22 : Updated(graph, /*level=*/level, /*rename=*/false);
1464 : 22 : return false;
1465 : : }
1466 : 390 : first = false;
1467 : : // Construct a new Cluster to hold the found component.
1468 : 390 : auto new_cluster = graph.CreateEmptyCluster(component_size);
1469 : 390 : new_clusters.push_back(new_cluster.get());
1470 : : // Remember that all the component's transactions go to this new Cluster. The positions
1471 : : // will be determined below, so use -1 for now.
1472 [ + - + + ]: 1180 : for (auto i : component) {
1473 : 400 : remap[i] = {new_cluster.get(), DepGraphIndex(-1)};
1474 : : }
1475 : 390 : graph.InsertCluster(level, std::move(new_cluster), split_quality);
1476 : 390 : todo -= component;
1477 : 390 : }
1478 : : // We have to split the Cluster up. Remove accounting for the existing one first.
1479 : 1231 : graph.GetClusterSet(level).m_cluster_usage -= TotalMemoryUsage();
1480 : : // Redistribute the transactions.
1481 [ + + ]: 1631 : for (auto i : m_linearization) {
1482 : : /** The cluster which transaction originally in position i is moved to. */
1483 : 400 : Cluster* new_cluster = remap[i].first;
1484 : : // Copy the transaction to the new cluster's depgraph, and remember the position.
1485 : 400 : remap[i].second = new_cluster->AppendTransaction(m_mapping[i], FeePerWeight::FromFeeFrac(m_depgraph.FeeRate(i)));
1486 : : }
1487 : : // Redistribute the dependencies.
1488 [ + + ]: 1631 : for (auto i : m_linearization) {
1489 : : /** The cluster transaction in position i is moved to. */
1490 : 400 : Cluster* new_cluster = remap[i].first;
1491 : : // Copy its parents, translating positions.
1492 : 400 : SetType new_parents;
1493 [ + + + + ]: 420 : for (auto par : m_depgraph.GetReducedParents(i)) new_parents.Set(remap[par].second);
1494 : 400 : new_cluster->AddDependencies(new_parents, remap[i].second);
1495 : : }
1496 : : // Update all the Locators of moved transactions, and memory usage.
1497 [ + + ]: 1621 : for (Cluster* new_cluster : new_clusters) {
1498 : 390 : new_cluster->Updated(graph, /*level=*/level, /*rename=*/false);
1499 : 390 : new_cluster->Compact();
1500 : 390 : graph.GetClusterSet(level).m_cluster_usage += new_cluster->TotalMemoryUsage();
1501 : : }
1502 : : // Wipe this Cluster, and return that it needs to be deleted.
1503 : 1231 : m_depgraph = DepGraph<SetType>{};
1504 [ + + ]: 1231 : m_mapping.clear();
1505 [ + + ]: 1453 : m_linearization.clear();
1506 : : return true;
1507 : 1253 : }
1508 : :
1509 : 22155 : bool SingletonClusterImpl::Split(TxGraphImpl& graph, int level) noexcept
1510 : : {
1511 : 22155 : Assume(NeedsSplitting());
1512 : 22155 : Assume(!GetTxCount());
1513 : 22155 : graph.GetClusterSet(level).m_cluster_usage -= TotalMemoryUsage();
1514 : 22155 : return true;
1515 : : }
1516 : :
1517 : 7931 : void GenericClusterImpl::Merge(TxGraphImpl& graph, int level, Cluster& other) noexcept
1518 : : {
1519 : : /** Vector to store the positions in this Cluster for each position in other. */
1520 : 7931 : std::vector<DepGraphIndex> remap(other.GetDepGraphIndexRange());
1521 : : // Iterate over all transactions in the other Cluster (the one being absorbed).
1522 : 7931 : other.ExtractTransactions([&](DepGraphIndex pos, GraphIndex idx, FeePerWeight feerate) noexcept {
1523 : : // Copy the transaction into this Cluster, and remember its position.
1524 : 7956 : auto new_pos = m_depgraph.AddTransaction(feerate);
1525 : : // Since this cluster must have been made hole-free before being merged into, all added
1526 : : // transactions should appear at the end.
1527 [ - + ]: 7956 : Assume(new_pos == m_mapping.size());
1528 : 7956 : remap[pos] = new_pos;
1529 : 7956 : m_mapping.push_back(idx);
1530 : 7956 : m_linearization.push_back(new_pos);
1531 : 15862 : }, [&](DepGraphIndex pos, GraphIndex idx, SetType other_parents) noexcept {
1532 : : // Copy the transaction's dependencies, translating them using remap.
1533 : 7956 : SetType parents;
1534 [ + + + + ]: 8006 : for (auto par : other_parents) {
1535 : 25 : parents.Set(remap[par]);
1536 : : }
1537 : 7956 : m_depgraph.AddDependencies(parents, remap[pos]);
1538 : : // Update the transaction's Locator. There is no need to call Updated() to update chunk
1539 : : // feerates, as Updated() will be invoked by Cluster::ApplyDependencies on the resulting
1540 : : // merged Cluster later anyway.
1541 [ + + ]: 7956 : auto& entry = graph.m_entries[idx];
1542 : : // Discard any potential ChunkData prior to modifying the Cluster (as that could
1543 : : // invalidate its ordering).
1544 [ + + ]: 7956 : if (level == 0) graph.ClearChunkData(entry);
1545 : 7956 : entry.m_locator[level].SetPresent(this, remap[pos]);
1546 : 7956 : });
1547 : 7931 : }
1548 : :
1549 : 0 : void SingletonClusterImpl::Merge(TxGraphImpl&, int, Cluster&) noexcept
1550 : : {
1551 : : // Nothing can be merged into a singleton; it should have been converted to GenericClusterImpl first.
1552 : 0 : Assume(false);
1553 : 0 : }
1554 : :
1555 : 5974 : void GenericClusterImpl::ApplyDependencies(TxGraphImpl& graph, int level, std::span<std::pair<GraphIndex, GraphIndex>> to_apply) noexcept
1556 : : {
1557 : : // Sort the list of dependencies to apply by child, so those can be applied in batch.
1558 [ - - - - : 34774 : std::ranges::sort(to_apply, [](auto& a, auto& b) { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
1559 : : // Iterate over groups of to-be-added dependencies with the same child.
1560 : 5974 : auto it = to_apply.begin();
1561 [ + + ]: 12310 : while (it != to_apply.end()) {
1562 : 6336 : auto& first_child = graph.m_entries[it->second].m_locator[level];
1563 : 6336 : const auto child_idx = first_child.index;
1564 : : // Iterate over all to-be-added dependencies within that same child, gather the relevant
1565 : : // parents.
1566 : 6336 : SetType parents;
1567 [ + + ]: 17060 : while (it != to_apply.end()) {
1568 [ + + ]: 11086 : auto& child = graph.m_entries[it->second].m_locator[level];
1569 : 11086 : auto& parent = graph.m_entries[it->first].m_locator[level];
1570 [ + + ]: 11086 : Assume(child.cluster == this && parent.cluster == this);
1571 [ + + ]: 11086 : if (child.index != child_idx) break;
1572 : 10724 : parents.Set(parent.index);
1573 : 10724 : ++it;
1574 : : }
1575 : : // Push all dependencies to the underlying DepGraph. Note that this is O(N) in the size of
1576 : : // the cluster, regardless of the number of parents being added, so batching them together
1577 : : // has a performance benefit.
1578 : 6336 : m_depgraph.AddDependencies(parents, child_idx);
1579 : : }
1580 : :
1581 : : // Finally set the cluster to NEEDS_FIX, so its linearization is fixed the next time it is
1582 : : // attempted to be made ACCEPTABLE.
1583 : 5974 : Assume(!NeedsSplitting());
1584 : 5974 : Assume(!IsOversized());
1585 : 5974 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::NEEDS_FIX);
1586 : :
1587 : : // Finally push the changes to graph.m_entries.
1588 : 5974 : Updated(graph, /*level=*/level, /*rename=*/false);
1589 : 5974 : }
1590 : :
1591 : 0 : void SingletonClusterImpl::ApplyDependencies(TxGraphImpl&, int, std::span<std::pair<GraphIndex, GraphIndex>>) noexcept
1592 : : {
1593 : : // Nothing can actually be applied.
1594 : 0 : Assume(false);
1595 : 0 : }
1596 : :
1597 : 2826 : TxGraphImpl::~TxGraphImpl() noexcept
1598 : : {
1599 : : // If Refs outlive the TxGraphImpl they refer to, unlink them, so that their destructor does not
1600 : : // try to reach into a non-existing TxGraphImpl anymore.
1601 [ + + ]: 86208 : for (auto& entry : m_entries) {
1602 [ + + ]: 84795 : if (entry.m_ref != nullptr) {
1603 : 64011 : GetRefGraph(*entry.m_ref) = nullptr;
1604 : : }
1605 : : }
1606 : 4239 : }
1607 : :
1608 : 115453 : std::unique_ptr<Cluster> TxGraphImpl::ExtractCluster(int level, QualityLevel quality, ClusterSetIndex setindex) noexcept
1609 : : {
1610 [ - + ]: 115453 : Assume(quality != QualityLevel::NONE);
1611 : :
1612 : 115453 : auto& clusterset = GetClusterSet(level);
1613 [ - + ]: 115453 : auto& quality_clusters = clusterset.m_clusters[int(quality)];
1614 [ - + - + ]: 115453 : Assume(setindex < quality_clusters.size());
1615 : :
1616 : : // Extract the Cluster-owning unique_ptr.
1617 [ - + ]: 115453 : std::unique_ptr<Cluster> ret = std::move(quality_clusters[setindex]);
1618 [ - + ]: 115453 : ret->m_quality = QualityLevel::NONE;
1619 : 115453 : ret->m_setindex = ClusterSetIndex(-1);
1620 : :
1621 : : // Clean up space in quality_cluster.
1622 [ - + ]: 115453 : auto max_setindex = quality_clusters.size() - 1;
1623 [ + + ]: 115453 : if (setindex != max_setindex) {
1624 : : // If the cluster was not the last element of quality_clusters, move that to take its place.
1625 : 21602 : quality_clusters.back()->m_setindex = setindex;
1626 : 21602 : quality_clusters[setindex] = std::move(quality_clusters.back());
1627 : : }
1628 : : // The last element of quality_clusters is now unused; drop it.
1629 : 115453 : quality_clusters.pop_back();
1630 : :
1631 : 115453 : return ret;
1632 : : }
1633 : :
1634 : 222519 : ClusterSetIndex TxGraphImpl::InsertCluster(int level, std::unique_ptr<Cluster>&& cluster, QualityLevel quality) noexcept
1635 : : {
1636 : : // Cannot insert with quality level NONE (as that would mean not inserted).
1637 [ - + ]: 222519 : Assume(quality != QualityLevel::NONE);
1638 : : // The passed-in Cluster must not currently be in the TxGraphImpl.
1639 [ - + ]: 222519 : Assume(cluster->m_quality == QualityLevel::NONE);
1640 : :
1641 : : // Append it at the end of the relevant TxGraphImpl::m_cluster.
1642 : 222519 : auto& clusterset = GetClusterSet(level);
1643 [ - + ]: 222519 : auto& quality_clusters = clusterset.m_clusters[int(quality)];
1644 [ - + ]: 222519 : ClusterSetIndex ret = quality_clusters.size();
1645 : 222519 : cluster->m_quality = quality;
1646 : 222519 : cluster->m_setindex = ret;
1647 : 222519 : quality_clusters.push_back(std::move(cluster));
1648 : 222519 : return ret;
1649 : : }
1650 : :
1651 : 35425 : void TxGraphImpl::SetClusterQuality(int level, QualityLevel old_quality, ClusterSetIndex old_index, QualityLevel new_quality) noexcept
1652 : : {
1653 [ + + ]: 35425 : Assume(new_quality != QualityLevel::NONE);
1654 : :
1655 : : // Don't do anything if the quality did not change.
1656 [ + + ]: 35425 : if (old_quality == new_quality) return;
1657 : : // Extract the cluster from where it currently resides.
1658 : 35424 : auto cluster_ptr = ExtractCluster(level, old_quality, old_index);
1659 : : // And re-insert it where it belongs.
1660 : 35424 : InsertCluster(level, std::move(cluster_ptr), new_quality);
1661 : 35424 : }
1662 : :
1663 : 32576 : void TxGraphImpl::DeleteCluster(Cluster& cluster, int level) noexcept
1664 : : {
1665 : : // Extract the cluster from where it currently resides.
1666 : 32576 : auto cluster_ptr = ExtractCluster(level, cluster.m_quality, cluster.m_setindex);
1667 : : // And throw it away.
1668 [ + - ]: 32576 : cluster_ptr.reset();
1669 : 32576 : }
1670 : :
1671 : 1375132 : std::pair<Cluster*, int> TxGraphImpl::FindClusterAndLevel(GraphIndex idx, int level) const noexcept
1672 : : {
1673 [ + - ]: 1375132 : Assume(level >= 0 && level <= GetTopLevel());
1674 : 1375132 : auto& entry = m_entries[idx];
1675 : : // Search the entry's locators from top to bottom.
1676 [ + + ]: 1411161 : for (int l = level; l >= 0; --l) {
1677 : : // If the locator is missing, dig deeper; it may exist at a lower level and therefore be
1678 : : // implicitly existing at this level too.
1679 [ + + ]: 1445870 : if (entry.m_locator[l].IsMissing()) continue;
1680 : : // If the locator has the entry marked as explicitly removed, stop.
1681 [ + + ]: 1373812 : if (entry.m_locator[l].IsRemoved()) break;
1682 : : // Otherwise, we have found the topmost ClusterSet that contains this entry.
1683 : 1373807 : return {entry.m_locator[l].cluster, l};
1684 : : }
1685 : : // If no non-empty locator was found, or an explicitly removed was hit, return nothing.
1686 : 1325 : return {nullptr, -1};
1687 : : }
1688 : :
1689 : 2482 : Cluster* TxGraphImpl::PullIn(Cluster* cluster, int level) noexcept
1690 : : {
1691 [ + + ]: 2482 : int to_level = GetTopLevel();
1692 [ + + ]: 2482 : Assume(to_level == 1);
1693 : 2482 : Assume(level <= to_level);
1694 : : // Copy the Cluster from main to staging, if it's not already there.
1695 [ + + ]: 2482 : if (level == 0) {
1696 : : // Make the Cluster Acceptable before copying. This isn't strictly necessary, but doing it
1697 : : // now avoids doing double work later.
1698 : 1747 : MakeAcceptable(*cluster, level);
1699 : 1747 : cluster = cluster->CopyToStaging(*this);
1700 : : }
1701 : 2482 : return cluster;
1702 : : }
1703 : :
1704 : 714576 : void TxGraphImpl::ApplyRemovals(int up_to_level) noexcept
1705 : : {
1706 [ + - ]: 714576 : Assume(up_to_level >= 0 && up_to_level <= GetTopLevel());
1707 [ + + ]: 1494519 : for (int level = 0; level <= up_to_level; ++level) {
1708 : 779943 : auto& clusterset = GetClusterSet(level);
1709 : 779943 : auto& to_remove = clusterset.m_to_remove;
1710 : : // Skip if there is nothing to remove in this level.
1711 [ + + ]: 779943 : if (to_remove.empty()) continue;
1712 : : // Pull in all Clusters that are not in staging.
1713 [ + + ]: 3704 : if (level == 1) {
1714 [ + + ]: 3527 : for (GraphIndex index : to_remove) {
1715 [ + - ]: 2177 : auto [cluster, cluster_level] = FindClusterAndLevel(index, level);
1716 [ + - ]: 2177 : if (cluster != nullptr) PullIn(cluster, cluster_level);
1717 : : }
1718 : : }
1719 : : // Group the set of to-be-removed entries by Cluster::m_sequence.
1720 : 3704 : std::ranges::sort(to_remove, [&](GraphIndex a, GraphIndex b) noexcept {
1721 : 155544 : Cluster* cluster_a = m_entries[a].m_locator[level].cluster;
1722 : 155544 : Cluster* cluster_b = m_entries[b].m_locator[level].cluster;
1723 : 155544 : return CompareClusters(cluster_a, cluster_b) < 0;
1724 : : });
1725 : : // Process per Cluster.
1726 [ - + ]: 3704 : std::span to_remove_span{to_remove};
1727 [ + + ]: 27133 : while (!to_remove_span.empty()) {
1728 [ + - ]: 23429 : Cluster* cluster = m_entries[to_remove_span.front()].m_locator[level].cluster;
1729 [ + - ]: 23429 : if (cluster != nullptr) {
1730 : : // If the first to_remove_span entry's Cluster exists, hand to_remove_span to it, so it
1731 : : // can pop off whatever applies to it.
1732 : 23429 : cluster->ApplyRemovals(*this, level, to_remove_span);
1733 : : } else {
1734 : : // Otherwise, skip this already-removed entry. This may happen when
1735 : : // RemoveTransaction was called twice on the same Ref, for example.
1736 : 0 : to_remove_span = to_remove_span.subspan(1);
1737 : : }
1738 : : }
1739 [ + - ]: 783647 : to_remove.clear();
1740 : : }
1741 : 714576 : Compact();
1742 : 714576 : }
1743 : :
1744 : 16234 : void TxGraphImpl::SwapIndexes(GraphIndex a, GraphIndex b, std::vector<Cluster*>& affected_main) noexcept
1745 : : {
1746 [ - + ]: 16234 : Assume(a < m_entries.size());
1747 : 16234 : Assume(b < m_entries.size());
1748 : : // Swap the Entry objects.
1749 : 16234 : std::swap(m_entries[a], m_entries[b]);
1750 : : // Iterate over both objects.
1751 [ + + ]: 48702 : for (int i = 0; i < 2; ++i) {
1752 [ + + ]: 32468 : GraphIndex idx = i ? b : a;
1753 [ + + ]: 32468 : Entry& entry = m_entries[idx];
1754 : : // Update linked Ref, if any exists.
1755 [ + + ]: 32468 : if (entry.m_ref) GetRefIndex(*entry.m_ref) = idx;
1756 : : // Update linked chunk index entries, if any exist.
1757 [ + + ]: 32468 : if (entry.m_main_chunkindex_iterator != m_main_chunkindex.end()) {
1758 : 15123 : entry.m_main_chunkindex_iterator->m_graph_index = idx;
1759 : : }
1760 : : // Update the locators for both levels.
1761 [ + + ]: 97404 : for (int level = 0; level < MAX_LEVELS; ++level) {
1762 : 64936 : Locator& locator = entry.m_locator[level];
1763 [ + + ]: 64936 : if (locator.IsPresent()) {
1764 : 15429 : locator.cluster->UpdateMapping(locator.index, idx);
1765 [ + - ]: 15429 : if (level == 0) affected_main.push_back(locator.cluster);
1766 : : }
1767 : : }
1768 : : }
1769 : 16234 : }
1770 : :
1771 : 828584 : void TxGraphImpl::Compact() noexcept
1772 : : {
1773 : : // We cannot compact while any to-be-applied operations or staged removals remain as we'd need
1774 : : // to rewrite them. It is easier to delay the compaction until they have been applied.
1775 [ + + ]: 828584 : if (!m_main_clusterset.m_deps_to_add.empty()) return;
1776 [ + + ]: 822637 : if (!m_main_clusterset.m_to_remove.empty()) return;
1777 [ + + ]: 822628 : Assume(m_main_clusterset.m_removed.empty()); // non-staging m_removed is always empty
1778 [ + + ]: 822628 : if (m_staging_clusterset.has_value()) {
1779 [ + + ]: 74215 : if (!m_staging_clusterset->m_deps_to_add.empty()) return;
1780 [ + + ]: 45652 : if (!m_staging_clusterset->m_to_remove.empty()) return;
1781 [ + + ]: 45651 : if (!m_staging_clusterset->m_removed.empty()) return;
1782 : : }
1783 : :
1784 : : // Release memory used by discarded ChunkData index entries.
1785 : 789170 : ClearShrink(m_main_chunkindex_discarded);
1786 : :
1787 : : // Sort the GraphIndexes that need to be cleaned up. They are sorted in reverse, so the last
1788 : : // ones get processed first. This means earlier-processed GraphIndexes will not cause moving of
1789 : : // later-processed ones during the "swap with end of m_entries" step below (which might
1790 : : // invalidate them).
1791 : 789170 : std::ranges::sort(m_unlinked, std::greater{});
1792 : :
1793 : 789170 : std::vector<Cluster*> affected_main;
1794 : 789170 : auto last = GraphIndex(-1);
1795 [ + + ]: 840459 : for (GraphIndex idx : m_unlinked) {
1796 : : // m_unlinked should never contain the same GraphIndex twice (the code below would fail
1797 : : // if so, because GraphIndexes get invalidated by removing them).
1798 : 51289 : Assume(idx != last);
1799 : 51289 : last = idx;
1800 : :
1801 : : // Make sure the entry is unlinked.
1802 : 51289 : Entry& entry = m_entries[idx];
1803 : 51289 : Assume(entry.m_ref == nullptr);
1804 : : // Make sure the entry does not occur in the graph.
1805 [ + + ]: 153867 : for (int level = 0; level < MAX_LEVELS; ++level) {
1806 : 102578 : Assume(!entry.m_locator[level].IsPresent());
1807 : : }
1808 : :
1809 : : // Move the entry to the end.
1810 [ - + + + ]: 51289 : if (idx != m_entries.size() - 1) SwapIndexes(idx, m_entries.size() - 1, affected_main);
1811 : : // Drop the entry for idx, now that it is at the end.
1812 : 51289 : m_entries.pop_back();
1813 : : }
1814 : :
1815 : : // Update the affected clusters, to fixup Entry::m_main_max_chunk_fallback values which may
1816 : : // have become outdated due to the compaction above.
1817 : 789170 : std::ranges::sort(affected_main);
1818 : 789170 : affected_main.erase(std::unique(affected_main.begin(), affected_main.end()), affected_main.end());
1819 [ + + ]: 800139 : for (Cluster* cluster : affected_main) {
1820 : 10969 : cluster->Updated(*this, /*level=*/0, /*rename=*/true);
1821 : : }
1822 [ + + ]: 817455 : m_unlinked.clear();
1823 : 789170 : }
1824 : :
1825 : 23408 : void TxGraphImpl::Split(Cluster& cluster, int level) noexcept
1826 : : {
1827 : : // To split a Cluster, first make sure all removals are applied (as we might need to split
1828 : : // again afterwards otherwise).
1829 : 23408 : ApplyRemovals(level);
1830 : 23408 : bool del = cluster.Split(*this, level);
1831 [ + + ]: 23408 : if (del) {
1832 : : // Cluster::Split reports whether the Cluster is to be deleted.
1833 : 23386 : DeleteCluster(cluster, level);
1834 : : }
1835 : 23408 : }
1836 : :
1837 : 635976 : void TxGraphImpl::SplitAll(int up_to_level) noexcept
1838 : : {
1839 [ + - ]: 635976 : Assume(up_to_level >= 0 && up_to_level <= GetTopLevel());
1840 : : // Before splitting all Cluster, first make sure all removals are applied.
1841 : 635976 : ApplyRemovals(up_to_level);
1842 [ + + ]: 1280718 : for (int level = 0; level <= up_to_level; ++level) {
1843 [ + + ]: 1934226 : for (auto quality : {QualityLevel::NEEDS_SPLIT_FIX, QualityLevel::NEEDS_SPLIT}) {
1844 : 1289484 : auto& queue = GetClusterSet(level).m_clusters[int(quality)];
1845 [ + + ]: 1312892 : while (!queue.empty()) {
1846 : 23408 : Split(*queue.back().get(), level);
1847 : : }
1848 : : }
1849 : : }
1850 : 635976 : }
1851 : :
1852 : 160838919 : void TxGraphImpl::GroupClusters(int level) noexcept
1853 : : {
1854 : 160838919 : auto& clusterset = GetClusterSet(level);
1855 : : // If the groupings have been computed already, nothing is left to be done.
1856 [ + + ]: 160838919 : if (clusterset.m_group_data.has_value()) return;
1857 : :
1858 : : // Before computing which Clusters need to be merged together, first apply all removals and
1859 : : // split the Clusters into connected components. If we would group first, we might end up
1860 : : // with inefficient and/or oversized Clusters which just end up being split again anyway.
1861 : 11151 : SplitAll(level);
1862 : :
1863 : : /** Annotated clusters: an entry for each Cluster, together with the sequence number for the
1864 : : * representative for the partition it is in (initially its own, later that of the
1865 : : * to-be-merged group). */
1866 : 11151 : std::vector<std::pair<Cluster*, uint64_t>> an_clusters;
1867 : : /** Annotated dependencies: an entry for each m_deps_to_add entry (excluding ones that apply
1868 : : * to removed transactions), together with the sequence number of the representative root of
1869 : : * Clusters it applies to (initially that of the child Cluster, later that of the
1870 : : * to-be-merged group). */
1871 : 11151 : std::vector<std::pair<std::pair<GraphIndex, GraphIndex>, uint64_t>> an_deps;
1872 : :
1873 : : // Construct an an_clusters entry for every oversized cluster, including ones from levels below,
1874 : : // as they may be inherited in this one.
1875 [ + + ]: 31068 : for (int level_iter = 0; level_iter <= level; ++level_iter) {
1876 [ + + ]: 19923 : for (auto& cluster : GetClusterSet(level_iter).m_clusters[int(QualityLevel::OVERSIZED_SINGLETON)]) {
1877 : 6 : auto graph_idx = cluster->GetClusterEntry(0);
1878 : 6 : auto cur_cluster = FindCluster(graph_idx, level);
1879 [ - + ]: 6 : if (cur_cluster == nullptr) continue;
1880 : 6 : an_clusters.emplace_back(cur_cluster, cur_cluster->m_sequence);
1881 : : }
1882 : : }
1883 : :
1884 : : // Construct a an_clusters entry for every parent and child in the to-be-applied dependencies,
1885 : : // and an an_deps entry for each dependency to be applied.
1886 [ - + ]: 11151 : an_deps.reserve(clusterset.m_deps_to_add.size());
1887 [ + + + + ]: 151829 : for (const auto& [par, chl] : clusterset.m_deps_to_add) {
1888 : 140678 : auto par_cluster = FindCluster(par, level);
1889 : 140678 : auto chl_cluster = FindCluster(chl, level);
1890 : : // Skip dependencies for which the parent or child transaction is removed.
1891 [ + + - + ]: 140678 : if (par_cluster == nullptr || chl_cluster == nullptr) continue;
1892 : 140673 : an_clusters.emplace_back(par_cluster, par_cluster->m_sequence);
1893 : : // Do not include a duplicate when parent and child are identical, as it'll be removed
1894 : : // below anyway.
1895 [ + + ]: 140673 : if (chl_cluster != par_cluster) an_clusters.emplace_back(chl_cluster, chl_cluster->m_sequence);
1896 : : // Add entry to an_deps, using the child sequence number.
1897 : 140673 : an_deps.emplace_back(std::pair{par, chl}, chl_cluster->m_sequence);
1898 : : }
1899 : : // Sort and deduplicate an_clusters, so we end up with a sorted list of all involved Clusters
1900 : : // to which dependencies apply, or which are oversized.
1901 [ + + + + : 4905001 : std::ranges::sort(an_clusters, [](auto& a, auto& b) noexcept { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
1902 : 11151 : an_clusters.erase(std::ranges::unique(an_clusters).begin(), an_clusters.end());
1903 : : // Sort an_deps by applying the same order to the involved child cluster.
1904 [ - - - - : 2352630 : std::ranges::sort(an_deps, [&](auto& a, auto& b) noexcept { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
1905 : :
1906 : : // Run the union-find algorithm to find partitions of the input Clusters which need to be
1907 : : // grouped together. See https://en.wikipedia.org/wiki/Disjoint-set_data_structure.
1908 : 11151 : {
1909 : : /** Each PartitionData entry contains information about a single input Cluster. */
1910 : 11151 : struct PartitionData
1911 : : {
1912 : : /** The sequence number of the cluster this holds information for. */
1913 : : uint64_t sequence;
1914 : : /** All PartitionData entries belonging to the same partition are organized in a tree.
1915 : : * Each element points to its parent, or to itself if it is the root. The root is then
1916 : : * a representative for the entire tree, and can be found by walking upwards from any
1917 : : * element. */
1918 : : PartitionData* parent;
1919 : : /** (only if this is a root, so when parent == this) An upper bound on the height of
1920 : : * tree for this partition. */
1921 : : unsigned rank;
1922 : : };
1923 : : /** Information about each input Cluster. Sorted by Cluster::m_sequence. */
1924 : 11151 : std::vector<PartitionData> partition_data;
1925 : :
1926 : : /** Given a Cluster, find its corresponding PartitionData. */
1927 : 284169 : auto locate_fn = [&](uint64_t sequence) noexcept -> PartitionData* {
1928 : 273018 : auto it = std::lower_bound(partition_data.begin(), partition_data.end(), sequence,
1929 [ + + ]: 4090800 : [](auto& a, uint64_t seq) noexcept { return a.sequence < seq; });
1930 : 273018 : Assume(it != partition_data.end());
1931 : 273018 : Assume(it->sequence == sequence);
1932 : 273018 : return &*it;
1933 : 11151 : };
1934 : :
1935 : : /** Given a PartitionData, find the root of the tree it is in (its representative). */
1936 : 295428 : static constexpr auto find_root_fn = [](PartitionData* data) noexcept -> PartitionData* {
1937 [ + + + + ]: 485674 : while (data->parent != data) {
1938 : : // Replace pointers to parents with pointers to grandparents.
1939 : : // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Finding_set_representatives.
1940 : 323397 : auto par = data->parent;
1941 : 323397 : data->parent = par->parent;
1942 : 323397 : data = par;
1943 : : }
1944 : 284277 : return data;
1945 : : };
1946 : :
1947 : : /** Given two PartitionDatas, union the partitions they are in, and return their
1948 : : * representative. */
1949 : 150280 : static constexpr auto union_fn = [](PartitionData* arg1, PartitionData* arg2) noexcept {
1950 : : // Find the roots of the trees, and bail out if they are already equal (which would
1951 : : // mean they are in the same partition already).
1952 [ + + ]: 400258 : auto rep1 = find_root_fn(arg1);
1953 : 139129 : auto rep2 = find_root_fn(arg2);
1954 [ + + ]: 139129 : if (rep1 == rep2) return rep1;
1955 : : // Pick the lower-rank root to become a child of the higher-rank one.
1956 : : // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Union_by_rank.
1957 [ + + ]: 136393 : if (rep1->rank < rep2->rank) std::swap(rep1, rep2);
1958 : 136393 : rep2->parent = rep1;
1959 : 136393 : rep1->rank += (rep1->rank == rep2->rank);
1960 : 136393 : return rep1;
1961 : : };
1962 : :
1963 : : // Start by initializing every Cluster as its own singleton partition.
1964 [ - + ]: 11151 : partition_data.resize(an_clusters.size());
1965 [ - + + + ]: 156299 : for (size_t i = 0; i < an_clusters.size(); ++i) {
1966 : 145148 : partition_data[i].sequence = an_clusters[i].first->m_sequence;
1967 : 145148 : partition_data[i].parent = &partition_data[i];
1968 : 145148 : partition_data[i].rank = 0;
1969 : : }
1970 : :
1971 : : // Run through all parent/child pairs in an_deps, and union the partitions their Clusters
1972 : : // are in.
1973 : 11151 : Cluster* last_chl_cluster{nullptr};
1974 : 11151 : PartitionData* last_partition{nullptr};
1975 [ + + + + ]: 151824 : for (const auto& [dep, _] : an_deps) {
1976 [ + + ]: 140673 : auto [par, chl] = dep;
1977 : 140673 : auto par_cluster = FindCluster(par, level);
1978 : 140673 : auto chl_cluster = FindCluster(chl, level);
1979 [ + + ]: 140673 : Assume(chl_cluster != nullptr && par_cluster != nullptr);
1980 : : // Nothing to do if parent and child are in the same Cluster.
1981 [ + + ]: 140673 : if (par_cluster == chl_cluster) continue;
1982 [ + + ]: 139129 : Assume(par != chl);
1983 [ + + ]: 139129 : if (chl_cluster == last_chl_cluster) {
1984 : : // If the child Clusters is the same as the previous iteration, union with the
1985 : : // tree they were in, avoiding the need for another lookup. Note that an_deps
1986 : : // is sorted by child Cluster, so batches with the same child are expected.
1987 : 5240 : last_partition = union_fn(locate_fn(par_cluster->m_sequence), last_partition);
1988 : : } else {
1989 : 133889 : last_chl_cluster = chl_cluster;
1990 : 133889 : last_partition = union_fn(locate_fn(par_cluster->m_sequence), locate_fn(chl_cluster->m_sequence));
1991 : : }
1992 : : }
1993 : :
1994 : : // Update the sequence numbers in an_clusters and an_deps to be those of the partition
1995 : : // representative.
1996 : 11151 : auto deps_it = an_deps.begin();
1997 [ - + + + ]: 156299 : for (size_t i = 0; i < partition_data.size(); ++i) {
1998 : 145148 : auto& data = partition_data[i];
1999 : : // Find the sequence of the representative of the partition Cluster i is in, and store
2000 : : // it with the Cluster.
2001 : 145148 : auto rep_seq = find_root_fn(&data)->sequence;
2002 : 145148 : an_clusters[i].second = rep_seq;
2003 : : // Find all dependencies whose child Cluster is Cluster i, and annotate them with rep.
2004 [ + + ]: 285821 : while (deps_it != an_deps.end()) {
2005 [ + + ]: 277708 : auto [par, chl] = deps_it->first;
2006 : 277708 : auto chl_cluster = FindCluster(chl, level);
2007 [ + + ]: 277708 : Assume(chl_cluster != nullptr);
2008 [ + + ]: 277708 : if (chl_cluster->m_sequence > data.sequence) break;
2009 : 140673 : deps_it->second = rep_seq;
2010 : 140673 : ++deps_it;
2011 : : }
2012 : : }
2013 : 11151 : }
2014 : :
2015 : : // Sort both an_clusters and an_deps by sequence number of the representative of the
2016 : : // partition they are in, grouping all those applying to the same partition together.
2017 [ - - - - : 1806349 : std::ranges::sort(an_deps, [](auto& a, auto& b) noexcept { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
2018 [ + + + + : 1799856 : std::ranges::sort(an_clusters, [](auto& a, auto& b) noexcept { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
2019 : :
2020 : : // Translate the resulting cluster groups to the m_group_data structure, and the dependencies
2021 : : // back to m_deps_to_add.
2022 : 11151 : clusterset.m_group_data = GroupData{};
2023 [ - + ]: 11151 : clusterset.m_group_data->m_group_clusters.reserve(an_clusters.size());
2024 [ + + ]: 11151 : clusterset.m_deps_to_add.clear();
2025 [ - + ]: 11151 : clusterset.m_deps_to_add.reserve(an_deps.size());
2026 : 11151 : clusterset.m_oversized = false;
2027 : 11151 : auto an_deps_it = an_deps.begin();
2028 : 11151 : auto an_clusters_it = an_clusters.begin();
2029 [ + + ]: 19906 : while (an_clusters_it != an_clusters.end()) {
2030 : : // Process all clusters/dependencies belonging to the partition with representative rep.
2031 : 8755 : auto rep = an_clusters_it->second;
2032 : : // Create and initialize a new GroupData entry for the partition.
2033 : 8755 : auto& new_entry = clusterset.m_group_data->m_groups.emplace_back();
2034 [ - + ]: 8755 : new_entry.m_cluster_offset = clusterset.m_group_data->m_group_clusters.size();
2035 : 8755 : new_entry.m_cluster_count = 0;
2036 [ - + ]: 8755 : new_entry.m_deps_offset = clusterset.m_deps_to_add.size();
2037 : 8755 : new_entry.m_deps_count = 0;
2038 : 8755 : uint32_t total_count{0};
2039 : 8755 : uint64_t total_size{0};
2040 : : // Add all its clusters to it (copying those from an_clusters to m_group_clusters).
2041 [ + + + + ]: 153903 : while (an_clusters_it != an_clusters.end() && an_clusters_it->second == rep) {
2042 : 145148 : clusterset.m_group_data->m_group_clusters.push_back(an_clusters_it->first);
2043 : 145148 : total_count += an_clusters_it->first->GetTxCount();
2044 : 145148 : total_size += an_clusters_it->first->GetTotalTxSize();
2045 : 145148 : ++an_clusters_it;
2046 : 145148 : ++new_entry.m_cluster_count;
2047 : : }
2048 : : // Add all its dependencies to it (copying those back from an_deps to m_deps_to_add).
2049 [ + + + + ]: 149428 : while (an_deps_it != an_deps.end() && an_deps_it->second == rep) {
2050 : 140673 : clusterset.m_deps_to_add.push_back(an_deps_it->first);
2051 : 140673 : ++an_deps_it;
2052 : 140673 : ++new_entry.m_deps_count;
2053 : : }
2054 : : // Detect oversizedness.
2055 [ + + + + ]: 8755 : if (total_count > m_max_cluster_count || total_size > m_max_cluster_size) {
2056 : 146 : clusterset.m_oversized = true;
2057 : : }
2058 : : }
2059 : 11151 : Assume(an_deps_it == an_deps.end());
2060 : 11151 : Assume(an_clusters_it == an_clusters.end());
2061 : 11151 : Compact();
2062 : 11151 : }
2063 : :
2064 : 5974 : void TxGraphImpl::Merge(std::span<Cluster*> to_merge, int level) noexcept
2065 : : {
2066 [ + + ]: 5974 : Assume(!to_merge.empty());
2067 : : // Nothing to do if a group consists of just a single Cluster.
2068 [ + + ]: 5974 : if (to_merge.size() == 1) return;
2069 : :
2070 : : // Move the largest Cluster to the front of to_merge. As all transactions in other to-be-merged
2071 : : // Clusters will be moved to that one, putting the largest one first minimizes the number of
2072 : : // moves.
2073 : 5953 : size_t max_size_pos{0};
2074 : 5953 : DepGraphIndex max_size = to_merge[max_size_pos]->GetTxCount();
2075 : 5953 : GetClusterSet(level).m_cluster_usage -= to_merge[max_size_pos]->TotalMemoryUsage();
2076 : 5953 : DepGraphIndex total_size = max_size;
2077 [ + + ]: 12463 : for (size_t i = 1; i < to_merge.size(); ++i) {
2078 : 6510 : GetClusterSet(level).m_cluster_usage -= to_merge[i]->TotalMemoryUsage();
2079 : 6510 : DepGraphIndex size = to_merge[i]->GetTxCount();
2080 : 6510 : total_size += size;
2081 [ + + ]: 6510 : if (size > max_size) {
2082 : 66 : max_size_pos = i;
2083 : 66 : max_size = size;
2084 : : }
2085 : : }
2086 [ + + ]: 5953 : if (max_size_pos != 0) std::swap(to_merge[0], to_merge[max_size_pos]);
2087 : :
2088 : 5953 : size_t start_idx = 1;
2089 : 5953 : Cluster* into_cluster = to_merge[0];
2090 [ + + ]: 5953 : if (total_size > into_cluster->GetMaxTxCount()) {
2091 : : // The into_merge cluster is too small to fit all transactions being merged. Construct a
2092 : : // a new Cluster using an implementation that matches the total size, and merge everything
2093 : : // in there.
2094 : 1421 : auto new_cluster = CreateEmptyCluster(total_size);
2095 : 1421 : into_cluster = new_cluster.get();
2096 : 1421 : InsertCluster(level, std::move(new_cluster), QualityLevel::OPTIMAL);
2097 : 1421 : start_idx = 0;
2098 : 1421 : }
2099 : :
2100 : : // Merge all further Clusters in the group into the result (first one, or new one), and delete
2101 : : // them.
2102 [ + + ]: 13884 : for (size_t i = start_idx; i < to_merge.size(); ++i) {
2103 : 7931 : into_cluster->Merge(*this, level, *to_merge[i]);
2104 : 7931 : DeleteCluster(*to_merge[i], level);
2105 : : }
2106 : 5953 : into_cluster->Compact();
2107 : 5953 : GetClusterSet(level).m_cluster_usage += into_cluster->TotalMemoryUsage();
2108 : : }
2109 : :
2110 : 160831374 : void TxGraphImpl::ApplyDependencies(int level) noexcept
2111 : : {
2112 : 160831374 : auto& clusterset = GetClusterSet(level);
2113 : : // Do not bother computing groups if we already know the result will be oversized.
2114 [ + + ]: 160831374 : if (clusterset.m_oversized == true) return;
2115 : : // Compute the groups of to-be-merged Clusters (which also applies all removals, and splits).
2116 : 160831374 : GroupClusters(level);
2117 [ + + ]: 160831374 : Assume(clusterset.m_group_data.has_value());
2118 : : // Nothing to do if there are no dependencies to be added.
2119 [ + + ]: 160831374 : if (clusterset.m_deps_to_add.empty()) return;
2120 : : // Dependencies cannot be applied if it would result in oversized clusters.
2121 [ + - ]: 5767 : if (clusterset.m_oversized == true) return;
2122 : :
2123 : : // For each group of to-be-merged Clusters.
2124 [ + + ]: 11741 : for (const auto& group_entry : clusterset.m_group_data->m_groups) {
2125 [ - + ]: 5974 : auto cluster_span = std::span{clusterset.m_group_data->m_group_clusters}
2126 [ + + ]: 5974 : .subspan(group_entry.m_cluster_offset, group_entry.m_cluster_count);
2127 : : // Pull in all the Clusters that contain dependencies.
2128 [ + + ]: 5974 : if (level == 1) {
2129 [ + + ]: 384 : for (Cluster*& cluster : cluster_span) {
2130 : 305 : cluster = PullIn(cluster, cluster->GetLevel(*this));
2131 : : }
2132 : : }
2133 : : // Invoke Merge() to merge them into a single Cluster.
2134 : 5974 : Merge(cluster_span, level);
2135 : : // Actually apply all to-be-added dependencies (all parents and children from this grouping
2136 : : // belong to the same Cluster at this point because of the merging above).
2137 [ - + ]: 5974 : auto deps_span = std::span{clusterset.m_deps_to_add}
2138 : 5974 : .subspan(group_entry.m_deps_offset, group_entry.m_deps_count);
2139 : 5974 : Assume(!deps_span.empty());
2140 : 5974 : const auto& loc = m_entries[deps_span[0].second].m_locator[level];
2141 : 5974 : Assume(loc.IsPresent());
2142 : 5974 : loc.cluster->ApplyDependencies(*this, level, deps_span);
2143 : : }
2144 : :
2145 : : // Wipe the list of to-be-added dependencies now that they are applied.
2146 [ + - ]: 5767 : clusterset.m_deps_to_add.clear();
2147 : 5767 : Compact();
2148 : : // Also no further Cluster mergings are needed (note that we clear, but don't set to
2149 : : // std::nullopt, as that would imply the groupings are unknown).
2150 : 11534 : clusterset.m_group_data = GroupData{};
2151 : : }
2152 : :
2153 : 5988 : std::pair<uint64_t, bool> GenericClusterImpl::Relinearize(TxGraphImpl& graph, int level, uint64_t max_cost) noexcept
2154 : : {
2155 : : // We can only relinearize Clusters that do not need splitting.
2156 [ - + ]: 5988 : Assume(!NeedsSplitting());
2157 : : // No work is required for Clusters which are already optimally linearized.
2158 [ - + ]: 5988 : if (IsOptimal()) return {0, false};
2159 : : // Invoke the actual linearization algorithm (passing in the existing one).
2160 : 5988 : uint64_t rng_seed = graph.m_rng.rand64();
2161 : 95054 : const auto fallback_order = [&](DepGraphIndex a, DepGraphIndex b) noexcept {
2162 : 89066 : const auto ref_a = graph.m_entries[m_mapping[a]].m_ref;
2163 : 89066 : const auto ref_b = graph.m_entries[m_mapping[b]].m_ref;
2164 : 89066 : return graph.m_fallback_order(*ref_a, *ref_b);
2165 : 5988 : };
2166 [ - + ]: 5988 : auto [linearization, optimal, cost] = Linearize(
2167 : 5988 : /*depgraph=*/m_depgraph,
2168 : : /*max_cost=*/max_cost,
2169 : : /*rng_seed=*/rng_seed,
2170 : : /*fallback_order=*/fallback_order,
2171 : 11976 : /*old_linearization=*/m_linearization,
2172 [ - + - + ]: 5988 : /*is_topological=*/IsTopological());
2173 : : // Postlinearize to improve the linearization (if optimal, only the sub-chunk order).
2174 : : // This also guarantees that all chunks are connected (even when non-optimal).
2175 [ - + ]: 5988 : PostLinearize(m_depgraph, linearization);
2176 : : // Update the linearization.
2177 : 5988 : m_linearization = std::move(linearization);
2178 : : // Update the Cluster's quality.
2179 : 5988 : bool improved = false;
2180 [ + - ]: 5988 : if (optimal) {
2181 : 5988 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::OPTIMAL);
2182 : 5988 : improved = true;
2183 [ # # # # ]: 0 : } else if (max_cost >= graph.m_acceptable_cost && !IsAcceptable()) {
2184 : 0 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::ACCEPTABLE);
2185 : 0 : improved = true;
2186 [ # # ]: 0 : } else if (!IsTopological()) {
2187 : 0 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::NEEDS_RELINEARIZE);
2188 : 0 : improved = true;
2189 : : }
2190 : : // Update the Entry objects.
2191 : 5988 : Updated(graph, /*level=*/level, /*rename=*/false);
2192 : 5988 : return {cost, improved};
2193 : 5988 : }
2194 : :
2195 : 0 : std::pair<uint64_t, bool> SingletonClusterImpl::Relinearize(TxGraphImpl& graph, int level, uint64_t max_cost) noexcept
2196 : : {
2197 : : // All singletons are optimal, oversized, or need splitting. Each of these precludes
2198 : : // Relinearize from being called.
2199 : 0 : AssertUnreachable();
2200 : : }
2201 : :
2202 : 318326749 : void TxGraphImpl::MakeAcceptable(Cluster& cluster, int level) noexcept
2203 : : {
2204 : : // Relinearize the Cluster if needed.
2205 [ + - + + : 318326749 : if (!cluster.NeedsSplitting() && !cluster.IsAcceptable() && !cluster.IsOversized()) {
+ + ]
2206 : 92 : cluster.Relinearize(*this, level, m_acceptable_cost);
2207 : : }
2208 : 318326749 : }
2209 : :
2210 : 217161 : void TxGraphImpl::MakeAllAcceptable(int level) noexcept
2211 : : {
2212 : 217161 : ApplyDependencies(level);
2213 : 217161 : auto& clusterset = GetClusterSet(level);
2214 [ + - ]: 217161 : if (clusterset.m_oversized == true) return;
2215 [ + + ]: 651483 : for (auto quality : {QualityLevel::NEEDS_FIX, QualityLevel::NEEDS_RELINEARIZE}) {
2216 : 434322 : auto& queue = clusterset.m_clusters[int(quality)];
2217 [ + + ]: 434405 : while (!queue.empty()) {
2218 : 83 : MakeAcceptable(*queue.back().get(), level);
2219 : : }
2220 : : }
2221 : : }
2222 : :
2223 : 1760 : GenericClusterImpl::GenericClusterImpl(uint64_t sequence) noexcept : Cluster{sequence} {}
2224 : :
2225 : 136084 : void TxGraphImpl::AddTransaction(Ref& arg, const FeePerWeight& feerate) noexcept
2226 : : {
2227 [ - + - + ]: 136084 : Assume(m_main_chunkindex_observers == 0 || GetTopLevel() != 0);
2228 : 136084 : Assume(feerate.size > 0);
2229 : 136084 : Assume(GetRefGraph(arg) == nullptr);
2230 : : // Construct a new Entry, and link it with the Ref.
2231 [ - + ]: 136084 : auto idx = m_entries.size();
2232 : 136084 : m_entries.emplace_back();
2233 : 136084 : auto& entry = m_entries.back();
2234 : 136084 : entry.m_main_chunkindex_iterator = m_main_chunkindex.end();
2235 : 136084 : entry.m_ref = &arg;
2236 : 136084 : GetRefGraph(arg) = this;
2237 : 136084 : GetRefIndex(arg) = idx;
2238 : : // Construct a new singleton Cluster (which is necessarily optimally linearized).
2239 : 136084 : bool oversized = uint64_t(feerate.size) > m_max_cluster_size;
2240 : 136084 : auto cluster = CreateEmptyCluster(1);
2241 : 136084 : cluster->AppendTransaction(idx, feerate);
2242 [ + + ]: 136084 : auto cluster_ptr = cluster.get();
2243 [ + + ]: 136084 : int level = GetTopLevel();
2244 : 136084 : auto& clusterset = GetClusterSet(level);
2245 [ + + ]: 272159 : InsertCluster(level, std::move(cluster), oversized ? QualityLevel::OVERSIZED_SINGLETON : QualityLevel::OPTIMAL);
2246 : 136084 : cluster_ptr->Updated(*this, /*level=*/level, /*rename=*/false);
2247 : 136084 : clusterset.m_cluster_usage += cluster_ptr->TotalMemoryUsage();
2248 : 136084 : ++clusterset.m_txcount;
2249 : : // Deal with individually oversized transactions.
2250 [ + + ]: 136084 : if (oversized) {
2251 : 9 : ++clusterset.m_txcount_oversized;
2252 : 9 : clusterset.m_oversized = true;
2253 [ + + ]: 9 : clusterset.m_group_data = std::nullopt;
2254 : : }
2255 : 136084 : }
2256 : :
2257 : 2188 : void TxGraphImpl::RemoveTransaction(const Ref& arg) noexcept
2258 : : {
2259 : : // Don't do anything if the Ref is empty (which may be indicative of the transaction already
2260 : : // having been removed).
2261 [ + - ]: 2188 : if (GetRefGraph(arg) == nullptr) return;
2262 [ - + ]: 2188 : Assume(GetRefGraph(arg) == this);
2263 [ - + + - ]: 2188 : Assume(m_main_chunkindex_observers == 0 || GetTopLevel() != 0);
2264 : : // Find the Cluster the transaction is in, and stop if it isn't in any.
2265 [ + - ]: 2188 : int level = GetTopLevel();
2266 [ + - ]: 2188 : auto cluster = FindCluster(GetRefIndex(arg), level);
2267 [ + - ]: 2188 : if (cluster == nullptr) return;
2268 : : // Remember that the transaction is to be removed.
2269 : 2188 : auto& clusterset = GetClusterSet(level);
2270 : 2188 : clusterset.m_to_remove.push_back(GetRefIndex(arg));
2271 : : // Wipe m_group_data (as it will need to be recomputed).
2272 [ + + ]: 2188 : clusterset.m_group_data.reset();
2273 [ + - ]: 2188 : if (clusterset.m_oversized == true) clusterset.m_oversized = std::nullopt;
2274 : : }
2275 : :
2276 : 77678 : void TxGraphImpl::AddDependency(const Ref& parent, const Ref& child) noexcept
2277 : : {
2278 : : // Don't do anything if either Ref is empty (which may be indicative of it having already been
2279 : : // removed).
2280 [ + - + - ]: 77678 : if (GetRefGraph(parent) == nullptr || GetRefGraph(child) == nullptr) return;
2281 [ - + ]: 77678 : Assume(GetRefGraph(parent) == this && GetRefGraph(child) == this);
2282 [ - + + - ]: 77678 : Assume(m_main_chunkindex_observers == 0 || GetTopLevel() != 0);
2283 : : // Don't do anything if this is a dependency on self.
2284 [ + - ]: 77678 : if (GetRefIndex(parent) == GetRefIndex(child)) return;
2285 : : // Find the Cluster the parent and child transaction are in, and stop if either appears to be
2286 : : // already removed.
2287 [ + - ]: 77678 : int level = GetTopLevel();
2288 : 77678 : auto par_cluster = FindCluster(GetRefIndex(parent), level);
2289 [ + - ]: 77678 : if (par_cluster == nullptr) return;
2290 : 77678 : auto chl_cluster = FindCluster(GetRefIndex(child), level);
2291 [ + - ]: 77678 : if (chl_cluster == nullptr) return;
2292 : : // Remember that this dependency is to be applied.
2293 : 77678 : auto& clusterset = GetClusterSet(level);
2294 : 77678 : clusterset.m_deps_to_add.emplace_back(GetRefIndex(parent), GetRefIndex(child));
2295 : : // Wipe m_group_data (as it will need to be recomputed).
2296 [ + + ]: 77678 : clusterset.m_group_data.reset();
2297 [ + + ]: 85223 : if (clusterset.m_oversized == false) clusterset.m_oversized = std::nullopt;
2298 : : }
2299 : :
2300 : 302 : bool TxGraphImpl::Exists(const Ref& arg, Level level_select) noexcept
2301 : : {
2302 [ + - ]: 302 : if (GetRefGraph(arg) == nullptr) return false;
2303 [ + - ]: 302 : Assume(GetRefGraph(arg) == this);
2304 [ + - ]: 302 : size_t level = GetSpecifiedLevel(level_select);
2305 : : // Make sure the transaction isn't scheduled for removal.
2306 : 302 : ApplyRemovals(level);
2307 : 302 : auto cluster = FindCluster(GetRefIndex(arg), level);
2308 : 302 : return cluster != nullptr;
2309 : : }
2310 : :
2311 : 45945 : void GenericClusterImpl::GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
2312 : : {
2313 : : /** The union of all ancestors to be returned. */
2314 : 45945 : SetType ancestors_union;
2315 : : // Process elements from the front of args, as long as they apply.
2316 [ + + ]: 91890 : while (!args.empty()) {
2317 [ + - ]: 45945 : if (args.front().first != this) break;
2318 : 45945 : ancestors_union |= m_depgraph.Ancestors(args.front().second);
2319 : 45945 : args = args.subspan(1);
2320 : : }
2321 [ + - ]: 45945 : Assume(ancestors_union.Any());
2322 : : // Translate all ancestors (in arbitrary order) to Refs (if they have any), and return them.
2323 [ + - + + ]: 357082 : for (auto idx : ancestors_union) {
2324 : 265192 : const auto& entry = graph.m_entries[m_mapping[idx]];
2325 : 265192 : Assume(entry.m_ref != nullptr);
2326 : 265192 : output.push_back(entry.m_ref);
2327 : : }
2328 : 45945 : }
2329 : :
2330 : 121502 : void SingletonClusterImpl::GetAncestorRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
2331 : : {
2332 : 121502 : Assume(GetTxCount());
2333 [ + + ]: 243004 : while (!args.empty()) {
2334 [ + - ]: 121502 : if (args.front().first != this) break;
2335 : 121502 : args = args.subspan(1);
2336 : : }
2337 : 121502 : const auto& entry = graph.m_entries[m_graph_index];
2338 : 121502 : Assume(entry.m_ref != nullptr);
2339 : 121502 : output.push_back(entry.m_ref);
2340 : 121502 : }
2341 : :
2342 : 46741 : void GenericClusterImpl::GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
2343 : : {
2344 : : /** The union of all descendants to be returned. */
2345 : 46741 : SetType descendants_union;
2346 : : // Process elements from the front of args, as long as they apply.
2347 [ + + ]: 93482 : while (!args.empty()) {
2348 [ + + ]: 46743 : if (args.front().first != this) break;
2349 : 46741 : descendants_union |= m_depgraph.Descendants(args.front().second);
2350 : 46741 : args = args.subspan(1);
2351 : : }
2352 [ + - ]: 46741 : Assume(descendants_union.Any());
2353 : : // Translate all descendants (in arbitrary order) to Refs (if they have any), and return them.
2354 [ + - + + ]: 497984 : for (auto idx : descendants_union) {
2355 : 404502 : const auto& entry = graph.m_entries[m_mapping[idx]];
2356 : 404502 : Assume(entry.m_ref != nullptr);
2357 : 404502 : output.push_back(entry.m_ref);
2358 : : }
2359 : 46741 : }
2360 : :
2361 : 42805 : void SingletonClusterImpl::GetDescendantRefs(const TxGraphImpl& graph, std::span<std::pair<Cluster*, DepGraphIndex>>& args, std::vector<TxGraph::Ref*>& output) noexcept
2362 : : {
2363 : : // In a singleton cluster, the ancestors or descendants are always just the entire cluster.
2364 : 42805 : GetAncestorRefs(graph, args, output);
2365 : 42805 : }
2366 : :
2367 : 224440 : bool GenericClusterImpl::GetClusterRefs(TxGraphImpl& graph, std::span<TxGraph::Ref*> range, LinearizationIndex start_pos) noexcept
2368 : : {
2369 : : // Translate the transactions in the Cluster (in linearization order, starting at start_pos in
2370 : : // the linearization) to Refs, and fill them in range.
2371 [ + + ]: 665042 : for (auto& ref : range) {
2372 [ - + ]: 440602 : Assume(start_pos < m_linearization.size());
2373 : 440602 : const auto& entry = graph.m_entries[m_mapping[m_linearization[start_pos++]]];
2374 : 440602 : Assume(entry.m_ref != nullptr);
2375 : 440602 : ref = entry.m_ref;
2376 : : }
2377 : : // Return whether start_pos has advanced to the end of the Cluster.
2378 [ - + ]: 224440 : return start_pos == m_linearization.size();
2379 : : }
2380 : :
2381 : 76409 : bool SingletonClusterImpl::GetClusterRefs(TxGraphImpl& graph, std::span<TxGraph::Ref*> range, LinearizationIndex start_pos) noexcept
2382 : : {
2383 : 76409 : Assume(!range.empty());
2384 : 76409 : Assume(GetTxCount());
2385 : 76409 : Assume(start_pos == 0);
2386 : 76409 : const auto& entry = graph.m_entries[m_graph_index];
2387 : 76409 : Assume(entry.m_ref != nullptr);
2388 : 76409 : range[0] = entry.m_ref;
2389 : 76409 : return true;
2390 : : }
2391 : :
2392 : 9 : FeePerWeight GenericClusterImpl::GetIndividualFeerate(DepGraphIndex idx) noexcept
2393 : : {
2394 : 9 : return FeePerWeight::FromFeeFrac(m_depgraph.FeeRate(idx));
2395 : : }
2396 : :
2397 : 2 : FeePerWeight SingletonClusterImpl::GetIndividualFeerate(DepGraphIndex idx) noexcept
2398 : : {
2399 : 2 : Assume(GetTxCount());
2400 : 2 : Assume(idx == 0);
2401 : 2 : return m_feerate;
2402 : : }
2403 : :
2404 : 24 : void GenericClusterImpl::MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept
2405 : : {
2406 : : // Mark all transactions of a Cluster missing, needed when aborting staging, so that the
2407 : : // corresponding Locators don't retain references into aborted Clusters.
2408 [ + + ]: 157 : for (auto ci : m_linearization) {
2409 : 133 : GraphIndex idx = m_mapping[ci];
2410 : 133 : auto& entry = graph.m_entries[idx];
2411 : 133 : entry.m_locator[1].SetMissing();
2412 : : }
2413 : 24 : }
2414 : :
2415 : 24355 : void SingletonClusterImpl::MakeStagingTransactionsMissing(TxGraphImpl& graph) noexcept
2416 : : {
2417 [ + - ]: 24355 : if (GetTxCount()) {
2418 : 24355 : auto& entry = graph.m_entries[m_graph_index];
2419 : 24355 : entry.m_locator[1].SetMissing();
2420 : : }
2421 : 24355 : }
2422 : :
2423 : 125954 : std::vector<TxGraph::Ref*> TxGraphImpl::GetAncestors(const Ref& arg, Level level_select) noexcept
2424 : : {
2425 : : // Return the empty vector if the Ref is empty.
2426 [ - + ]: 125954 : if (GetRefGraph(arg) == nullptr) return {};
2427 [ - + ]: 125954 : Assume(GetRefGraph(arg) == this);
2428 : : // Apply all removals and dependencies, as the result might be incorrect otherwise.
2429 [ - + ]: 125954 : size_t level = GetSpecifiedLevel(level_select);
2430 : 125954 : ApplyDependencies(level);
2431 : : // Ancestry cannot be known if unapplied dependencies remain.
2432 [ + + ]: 125954 : Assume(GetClusterSet(level).m_deps_to_add.empty());
2433 : : // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
2434 [ + + ]: 125954 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(arg), level);
2435 [ + + ]: 125954 : if (cluster == nullptr) return {};
2436 : : // Dispatch to the Cluster.
2437 : 124642 : std::pair<Cluster*, DepGraphIndex> match = {cluster, m_entries[GetRefIndex(arg)].m_locator[cluster_level].index};
2438 : 124642 : auto matches = std::span(&match, 1);
2439 : 124642 : std::vector<TxGraph::Ref*> ret;
2440 : 124642 : cluster->GetAncestorRefs(*this, matches, ret);
2441 : 124642 : return ret;
2442 : 124642 : }
2443 : :
2444 : 89462 : std::vector<TxGraph::Ref*> TxGraphImpl::GetDescendants(const Ref& arg, Level level_select) noexcept
2445 : : {
2446 : : // Return the empty vector if the Ref is empty.
2447 [ - + ]: 89462 : if (GetRefGraph(arg) == nullptr) return {};
2448 [ - + ]: 89462 : Assume(GetRefGraph(arg) == this);
2449 : : // Apply all removals and dependencies, as the result might be incorrect otherwise.
2450 [ - + ]: 89462 : size_t level = GetSpecifiedLevel(level_select);
2451 : 89462 : ApplyDependencies(level);
2452 : : // Ancestry cannot be known if unapplied dependencies remain.
2453 [ - + ]: 89462 : Assume(GetClusterSet(level).m_deps_to_add.empty());
2454 : : // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
2455 [ - + ]: 89462 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(arg), level);
2456 [ - + ]: 89462 : if (cluster == nullptr) return {};
2457 : : // Dispatch to the Cluster.
2458 : 89462 : std::pair<Cluster*, DepGraphIndex> match = {cluster, m_entries[GetRefIndex(arg)].m_locator[cluster_level].index};
2459 : 89462 : auto matches = std::span(&match, 1);
2460 : 89462 : std::vector<TxGraph::Ref*> ret;
2461 : 89462 : cluster->GetDescendantRefs(*this, matches, ret);
2462 : 89462 : return ret;
2463 : 89462 : }
2464 : :
2465 : 0 : std::vector<TxGraph::Ref*> TxGraphImpl::GetAncestorsUnion(std::span<const Ref* const> args, Level level_select) noexcept
2466 : : {
2467 : : // Apply all dependencies, as the result might be incorrect otherwise.
2468 [ # # ]: 0 : size_t level = GetSpecifiedLevel(level_select);
2469 : 0 : ApplyDependencies(level);
2470 : : // Ancestry cannot be known if unapplied dependencies remain.
2471 : 0 : Assume(GetClusterSet(level).m_deps_to_add.empty());
2472 : :
2473 : : // Translate args to matches.
2474 : 0 : std::vector<std::pair<Cluster*, DepGraphIndex>> matches;
2475 : 0 : matches.reserve(args.size());
2476 [ # # ]: 0 : for (auto arg : args) {
2477 [ # # ]: 0 : Assume(arg);
2478 : : // Skip empty Refs.
2479 [ # # ]: 0 : if (GetRefGraph(*arg) == nullptr) continue;
2480 [ # # ]: 0 : Assume(GetRefGraph(*arg) == this);
2481 : : // Find the Cluster the argument is in, and skip if none is found.
2482 [ # # ]: 0 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(*arg), level);
2483 [ # # ]: 0 : if (cluster == nullptr) continue;
2484 : : // Append to matches.
2485 : 0 : matches.emplace_back(cluster, m_entries[GetRefIndex(*arg)].m_locator[cluster_level].index);
2486 : : }
2487 : : // Group by Cluster.
2488 : 0 : std::ranges::sort(matches, [](auto& a, auto& b) noexcept { return CompareClusters(a.first, b.first) < 0; });
2489 : : // Dispatch to the Clusters.
2490 [ # # ]: 0 : std::span match_span(matches);
2491 : 0 : std::vector<TxGraph::Ref*> ret;
2492 [ # # ]: 0 : while (!match_span.empty()) {
2493 : 0 : match_span.front().first->GetAncestorRefs(*this, match_span, ret);
2494 : : }
2495 : 0 : return ret;
2496 : 0 : }
2497 : :
2498 : 17308 : std::vector<TxGraph::Ref*> TxGraphImpl::GetDescendantsUnion(std::span<const Ref* const> args, Level level_select) noexcept
2499 : : {
2500 : : // Apply all dependencies, as the result might be incorrect otherwise.
2501 [ - + ]: 17308 : size_t level = GetSpecifiedLevel(level_select);
2502 : 17308 : ApplyDependencies(level);
2503 : : // Ancestry cannot be known if unapplied dependencies remain.
2504 : 17308 : Assume(GetClusterSet(level).m_deps_to_add.empty());
2505 : :
2506 : : // Translate args to matches.
2507 : 17308 : std::vector<std::pair<Cluster*, DepGraphIndex>> matches;
2508 : 17308 : matches.reserve(args.size());
2509 [ + + ]: 17392 : for (auto arg : args) {
2510 [ - + ]: 84 : Assume(arg);
2511 : : // Skip empty Refs.
2512 [ - + ]: 84 : if (GetRefGraph(*arg) == nullptr) continue;
2513 [ - + ]: 84 : Assume(GetRefGraph(*arg) == this);
2514 : : // Find the Cluster the argument is in, and skip if none is found.
2515 [ - + ]: 84 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(*arg), level);
2516 [ - + ]: 84 : if (cluster == nullptr) continue;
2517 : : // Append to matches.
2518 : 84 : matches.emplace_back(cluster, m_entries[GetRefIndex(*arg)].m_locator[cluster_level].index);
2519 : : }
2520 : : // Group by Cluster.
2521 : 17308 : std::ranges::sort(matches, [](auto& a, auto& b) noexcept { return CompareClusters(a.first, b.first) < 0; });
2522 : : // Dispatch to the Clusters.
2523 [ - + ]: 17308 : std::span match_span(matches);
2524 : 17308 : std::vector<TxGraph::Ref*> ret;
2525 [ + + ]: 17392 : while (!match_span.empty()) {
2526 : 84 : match_span.front().first->GetDescendantRefs(*this, match_span, ret);
2527 : : }
2528 : 17308 : return ret;
2529 : 17308 : }
2530 : :
2531 : 96556 : std::vector<TxGraph::Ref*> TxGraphImpl::GetCluster(const Ref& arg, Level level_select) noexcept
2532 : : {
2533 : : // Return the empty vector if the Ref is empty (which may be indicative of the transaction
2534 : : // having been removed already.
2535 [ - + ]: 96556 : if (GetRefGraph(arg) == nullptr) return {};
2536 [ - + ]: 96556 : Assume(GetRefGraph(arg) == this);
2537 : : // Apply all removals and dependencies, as the result might be incorrect otherwise.
2538 [ - + ]: 96556 : size_t level = GetSpecifiedLevel(level_select);
2539 : 96556 : ApplyDependencies(level);
2540 : : // Cluster linearization cannot be known if unapplied dependencies remain.
2541 [ - + ]: 96556 : Assume(GetClusterSet(level).m_deps_to_add.empty());
2542 : : // Find the Cluster the argument is in, and return the empty vector if it isn't in any.
2543 [ - + ]: 96556 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(arg), level);
2544 [ - + ]: 96556 : if (cluster == nullptr) return {};
2545 : : // Make sure the Cluster has an acceptable quality level, and then dispatch to it.
2546 : 96556 : MakeAcceptable(*cluster, cluster_level);
2547 [ - + ]: 96556 : std::vector<TxGraph::Ref*> ret(cluster->GetTxCount());
2548 [ - + ]: 96556 : cluster->GetClusterRefs(*this, ret, 0);
2549 : 96556 : return ret;
2550 : 96556 : }
2551 : :
2552 : 23 : TxGraph::GraphIndex TxGraphImpl::GetTransactionCount(Level level_select) noexcept
2553 : : {
2554 [ + + ]: 23 : size_t level = GetSpecifiedLevel(level_select);
2555 : 23 : ApplyRemovals(level);
2556 : 23 : return GetClusterSet(level).m_txcount;
2557 : : }
2558 : :
2559 : 11 : FeePerWeight TxGraphImpl::GetIndividualFeerate(const Ref& arg) noexcept
2560 : : {
2561 : : // Return the empty FeePerWeight if the passed Ref is empty.
2562 [ - + ]: 11 : if (GetRefGraph(arg) == nullptr) return {};
2563 : 11 : Assume(GetRefGraph(arg) == this);
2564 : : // Find the cluster the argument is in (the level does not matter as individual feerates will
2565 : : // be identical if it occurs in both), and return the empty FeePerWeight if it isn't in any.
2566 : 11 : Cluster* cluster{nullptr};
2567 : 11 : int level;
2568 [ + - ]: 11 : for (level = 0; level <= GetTopLevel(); ++level) {
2569 : : // Apply removals, so that we can correctly report FeePerWeight{} for non-existing
2570 : : // transactions.
2571 : 11 : ApplyRemovals(level);
2572 [ - + ]: 11 : if (m_entries[GetRefIndex(arg)].m_locator[level].IsPresent()) {
2573 : : cluster = m_entries[GetRefIndex(arg)].m_locator[level].cluster;
2574 : : break;
2575 : : }
2576 : : }
2577 [ - + ]: 11 : if (cluster == nullptr) return {};
2578 : : // Dispatch to the Cluster.
2579 : 11 : return cluster->GetIndividualFeerate(m_entries[GetRefIndex(arg)].m_locator[level].index);
2580 : : }
2581 : :
2582 : 59888 : FeePerWeight TxGraphImpl::GetMainChunkFeerate(const Ref& arg) noexcept
2583 : : {
2584 : : // Return the empty FeePerWeight if the passed Ref is empty.
2585 [ - + ]: 59888 : if (GetRefGraph(arg) == nullptr) return {};
2586 : 59888 : Assume(GetRefGraph(arg) == this);
2587 : : // Apply all removals and dependencies, as the result might be inaccurate otherwise.
2588 : 59888 : ApplyDependencies(/*level=*/0);
2589 : : // Chunk feerates cannot be accurately known if unapplied dependencies remain.
2590 [ - + ]: 59888 : Assume(m_main_clusterset.m_deps_to_add.empty());
2591 : : // Find the cluster the argument is in, and return the empty FeePerWeight if it isn't in any.
2592 [ - + ]: 59888 : auto [cluster, cluster_level] = FindClusterAndLevel(GetRefIndex(arg), /*level=*/0);
2593 [ - + ]: 59888 : if (cluster == nullptr) return {};
2594 : : // Make sure the Cluster has an acceptable quality level, and then return the transaction's
2595 : : // chunk feerate.
2596 : 59888 : MakeAcceptable(*cluster, cluster_level);
2597 : 59888 : const auto& entry = m_entries[GetRefIndex(arg)];
2598 : 59888 : return entry.m_main_chunk_feerate;
2599 : : }
2600 : :
2601 : 250189 : bool TxGraphImpl::IsOversized(Level level_select) noexcept
2602 : : {
2603 [ + + ]: 250189 : size_t level = GetSpecifiedLevel(level_select);
2604 : 250189 : auto& clusterset = GetClusterSet(level);
2605 [ + + ]: 250189 : if (clusterset.m_oversized.has_value()) {
2606 : : // Return cached value if known.
2607 : 242676 : return *clusterset.m_oversized;
2608 : : }
2609 : 7513 : ApplyRemovals(level);
2610 [ - + ]: 7513 : if (clusterset.m_txcount_oversized > 0) {
2611 : 0 : clusterset.m_oversized = true;
2612 : : } else {
2613 : : // Find which Clusters will need to be merged together, as that is where the oversize
2614 : : // property is assessed.
2615 : 7513 : GroupClusters(level);
2616 : : }
2617 : 7513 : Assume(clusterset.m_oversized.has_value());
2618 : 7513 : return *clusterset.m_oversized;
2619 : : }
2620 : :
2621 : 71142 : void TxGraphImpl::StartStaging() noexcept
2622 : : {
2623 : : // Staging cannot already exist.
2624 : 71142 : Assume(!m_staging_clusterset.has_value());
2625 : : // Apply all remaining dependencies in main before creating a staging graph. Once staging
2626 : : // exists, we cannot merge Clusters anymore (because of interference with Clusters being
2627 : : // pulled into staging), so to make sure all inspectors are available (if not oversized), do
2628 : : // all merging work now. Call SplitAll() first, so that even if ApplyDependencies does not do
2629 : : // any thing due to knowing the result is oversized, splitting is still performed.
2630 : 71142 : SplitAll(0);
2631 : 71142 : ApplyDependencies(0);
2632 : : // Construct the staging ClusterSet.
2633 : 71142 : m_staging_clusterset.emplace();
2634 : : // Copy statistics, precomputed data, and to-be-applied dependencies (only if oversized) to
2635 : : // the new graph. To-be-applied removals will always be empty at this point.
2636 : 71142 : m_staging_clusterset->m_txcount = m_main_clusterset.m_txcount;
2637 : 71142 : m_staging_clusterset->m_txcount_oversized = m_main_clusterset.m_txcount_oversized;
2638 : 71142 : m_staging_clusterset->m_deps_to_add = m_main_clusterset.m_deps_to_add;
2639 : 71142 : m_staging_clusterset->m_group_data = m_main_clusterset.m_group_data;
2640 : 71142 : m_staging_clusterset->m_oversized = m_main_clusterset.m_oversized;
2641 : 71142 : Assume(m_staging_clusterset->m_oversized.has_value());
2642 : 71142 : m_staging_clusterset->m_cluster_usage = 0;
2643 : 71142 : }
2644 : :
2645 : 23799 : void TxGraphImpl::AbortStaging() noexcept
2646 : : {
2647 : : // Staging must exist.
2648 : 23799 : Assume(m_staging_clusterset.has_value());
2649 : : // Mark all removed transactions as Missing (so the staging locator for these transactions
2650 : : // can be reused if another staging is created).
2651 [ + + ]: 24396 : for (auto idx : m_staging_clusterset->m_removed) {
2652 : 597 : m_entries[idx].m_locator[1].SetMissing();
2653 : : }
2654 : : // Do the same with the non-removed transactions in staging Clusters.
2655 [ + + ]: 190392 : for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
2656 [ + + ]: 190972 : for (auto& cluster : m_staging_clusterset->m_clusters[quality]) {
2657 : 24379 : cluster->MakeStagingTransactionsMissing(*this);
2658 : : }
2659 : : }
2660 : : // Destroy the staging ClusterSet.
2661 : 23799 : m_staging_clusterset.reset();
2662 : 23799 : Compact();
2663 [ - + ]: 23799 : if (!m_main_clusterset.m_group_data.has_value()) {
2664 : : // In case m_oversized in main was kept after a Ref destruction while staging exists, we
2665 : : // need to re-evaluate m_oversized now.
2666 [ # # # # ]: 0 : if (m_main_clusterset.m_to_remove.empty() && m_main_clusterset.m_txcount_oversized > 0) {
2667 : : // It is possible that a Ref destruction caused a removal in main while staging existed.
2668 : : // In this case, m_txcount_oversized may be inaccurate.
2669 : 0 : m_main_clusterset.m_oversized = true;
2670 : : } else {
2671 [ # # ]: 0 : m_main_clusterset.m_oversized = std::nullopt;
2672 : : }
2673 : : }
2674 : 23799 : }
2675 : :
2676 : 47343 : void TxGraphImpl::CommitStaging() noexcept
2677 : : {
2678 : : // Staging must exist.
2679 : 47343 : Assume(m_staging_clusterset.has_value());
2680 : 47343 : Assume(m_main_chunkindex_observers == 0);
2681 : : // Get rid of removed transactions in staging before moving to main, so they do not need to be
2682 : : // added to the chunk index there. Doing so is impossible if they were unlinked, and thus have
2683 : : // no Ref anymore to pass to the fallback comparator.
2684 : 47343 : ApplyRemovals(/*up_to_level=*/1);
2685 : : // Delete all conflicting Clusters in main, to make place for moving the staging ones
2686 : : // there. All of these have been copied to staging in PullIn().
2687 : 47343 : auto conflicts = GetConflicts();
2688 [ + + ]: 48602 : for (Cluster* conflict : conflicts) {
2689 : 1259 : conflict->Clear(*this, /*level=*/0);
2690 : 1259 : DeleteCluster(*conflict, /*level=*/0);
2691 : : }
2692 : : // Mark the removed transactions as Missing (so the staging locator for these transactions
2693 : : // can be reused if another staging is created).
2694 [ + + ]: 48923 : for (auto idx : m_staging_clusterset->m_removed) {
2695 : 1580 : m_entries[idx].m_locator[1].SetMissing();
2696 : : }
2697 : : // Then move all Clusters in staging to main.
2698 [ + + ]: 378744 : for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
2699 : 331401 : auto& stage_sets = m_staging_clusterset->m_clusters[quality];
2700 [ + + ]: 378854 : while (!stage_sets.empty()) {
2701 : 47453 : stage_sets.back()->MoveToMain(*this);
2702 : : }
2703 : : }
2704 : : // Move all statistics, precomputed data, and to-be-applied removals and dependencies.
2705 : 47343 : m_main_clusterset.m_deps_to_add = std::move(m_staging_clusterset->m_deps_to_add);
2706 : 47343 : m_main_clusterset.m_to_remove = std::move(m_staging_clusterset->m_to_remove);
2707 : 47343 : m_main_clusterset.m_group_data = std::move(m_staging_clusterset->m_group_data);
2708 : 47343 : m_main_clusterset.m_oversized = m_staging_clusterset->m_oversized;
2709 : 47343 : m_main_clusterset.m_txcount = m_staging_clusterset->m_txcount;
2710 : 47343 : m_main_clusterset.m_txcount_oversized = m_staging_clusterset->m_txcount_oversized;
2711 : : // Delete the old staging graph, after all its information was moved to main.
2712 : 47343 : m_staging_clusterset.reset();
2713 : 47343 : Compact();
2714 : 47343 : }
2715 : :
2716 : 16 : void GenericClusterImpl::SetFee(TxGraphImpl& graph, int level, DepGraphIndex idx, int64_t fee) noexcept
2717 : : {
2718 : : // Make sure the specified DepGraphIndex exists in this Cluster.
2719 [ + - ]: 16 : Assume(m_depgraph.Positions()[idx]);
2720 : : // Bail out if the fee isn't actually being changed.
2721 [ + - ]: 16 : if (m_depgraph.FeeRate(idx).fee == fee) return;
2722 : : // Update the fee, remember that relinearization will be necessary, and update the Entries
2723 : : // in the same Cluster.
2724 [ + - ]: 16 : m_depgraph.FeeRate(idx).fee = fee;
2725 [ + - ]: 16 : if (m_quality == QualityLevel::OVERSIZED_SINGLETON) {
2726 : : // Nothing to do.
2727 [ + + ]: 16 : } else if (IsAcceptable()) {
2728 : 12 : graph.SetClusterQuality(level, m_quality, m_setindex, QualityLevel::NEEDS_RELINEARIZE);
2729 : : }
2730 : 16 : Updated(graph, /*level=*/level, /*rename=*/false);
2731 : : }
2732 : :
2733 : 287 : void SingletonClusterImpl::SetFee(TxGraphImpl& graph, int level, DepGraphIndex idx, int64_t fee) noexcept
2734 : : {
2735 : 287 : Assume(GetTxCount());
2736 : 287 : Assume(idx == 0);
2737 : 287 : m_feerate.fee = fee;
2738 : 287 : Updated(graph, /*level=*/level, /*rename=*/false);
2739 : 287 : }
2740 : :
2741 : 303 : void TxGraphImpl::SetTransactionFee(const Ref& ref, int64_t fee) noexcept
2742 : : {
2743 : : // Don't do anything if the passed Ref is empty.
2744 [ + - ]: 303 : if (GetRefGraph(ref) == nullptr) return;
2745 : 303 : Assume(GetRefGraph(ref) == this);
2746 : 303 : Assume(m_main_chunkindex_observers == 0);
2747 : : // Find the entry, its locator, and inform its Cluster about the new feerate, if any.
2748 : 303 : auto& entry = m_entries[GetRefIndex(ref)];
2749 [ + + ]: 909 : for (int level = 0; level < MAX_LEVELS; ++level) {
2750 : 606 : auto& locator = entry.m_locator[level];
2751 [ + + ]: 606 : if (locator.IsPresent()) {
2752 : 303 : locator.cluster->SetFee(*this, level, locator.index, fee);
2753 : : }
2754 : : }
2755 : : }
2756 : :
2757 : 159052129 : std::strong_ordering TxGraphImpl::CompareMainOrder(const Ref& a, const Ref& b) noexcept
2758 : : {
2759 : : // The references must not be empty.
2760 : 159052129 : Assume(GetRefGraph(a) == this);
2761 : 159052129 : Assume(GetRefGraph(b) == this);
2762 : : // Apply dependencies in main.
2763 : 159052129 : ApplyDependencies(0);
2764 : 159052129 : Assume(m_main_clusterset.m_deps_to_add.empty());
2765 : : // Make both involved Clusters acceptable, so chunk feerates are relevant.
2766 : 159052129 : const auto& entry_a = m_entries[GetRefIndex(a)];
2767 : 159052129 : const auto& entry_b = m_entries[GetRefIndex(b)];
2768 : 159052129 : const auto& locator_a = entry_a.m_locator[0];
2769 : 159052129 : const auto& locator_b = entry_b.m_locator[0];
2770 : 159052129 : Assume(locator_a.IsPresent());
2771 : 159052129 : Assume(locator_b.IsPresent());
2772 : 159052129 : MakeAcceptable(*locator_a.cluster, /*level=*/0);
2773 : 159052129 : MakeAcceptable(*locator_b.cluster, /*level=*/0);
2774 : : // Invoke comparison logic.
2775 : 159052129 : return CompareMainTransactions(GetRefIndex(a), GetRefIndex(b));
2776 : : }
2777 : :
2778 : 1395 : TxGraph::GraphIndex TxGraphImpl::CountDistinctClusters(std::span<const Ref* const> refs, Level level_select) noexcept
2779 : : {
2780 [ - + ]: 1395 : size_t level = GetSpecifiedLevel(level_select);
2781 : 1395 : ApplyDependencies(level);
2782 : 1395 : auto& clusterset = GetClusterSet(level);
2783 : 1395 : Assume(clusterset.m_deps_to_add.empty());
2784 : : // Build a vector of Clusters that the specified Refs occur in.
2785 : 1395 : std::vector<Cluster*> clusters;
2786 : 1395 : clusters.reserve(refs.size());
2787 [ + + ]: 4144 : for (const Ref* ref : refs) {
2788 [ - + ]: 2749 : Assume(ref);
2789 [ - + ]: 2749 : if (GetRefGraph(*ref) == nullptr) continue;
2790 [ + - ]: 2749 : Assume(GetRefGraph(*ref) == this);
2791 [ + - ]: 2749 : auto cluster = FindCluster(GetRefIndex(*ref), level);
2792 [ + - ]: 2749 : if (cluster != nullptr) clusters.push_back(cluster);
2793 : : }
2794 : : // Count the number of distinct elements in clusters.
2795 : 1395 : std::ranges::sort(clusters, [](Cluster* a, Cluster* b) noexcept { return CompareClusters(a, b) < 0; });
2796 : 1395 : Cluster* last{nullptr};
2797 : 1395 : GraphIndex ret{0};
2798 [ + + ]: 4144 : for (Cluster* cluster : clusters) {
2799 : 2749 : ret += (cluster != last);
2800 : 2749 : last = cluster;
2801 : : }
2802 : 1395 : return ret;
2803 : 1395 : }
2804 : :
2805 : 1339 : std::pair<std::vector<FeeFrac>, std::vector<FeeFrac>> TxGraphImpl::GetMainStagingDiagrams() noexcept
2806 : : {
2807 : 1339 : Assume(m_staging_clusterset.has_value());
2808 : 1339 : MakeAllAcceptable(0);
2809 : 1339 : Assume(m_main_clusterset.m_deps_to_add.empty()); // can only fail if main is oversized
2810 : 1339 : MakeAllAcceptable(1);
2811 : 1339 : Assume(m_staging_clusterset->m_deps_to_add.empty()); // can only fail if staging is oversized
2812 : : // For all Clusters in main which conflict with Clusters in staging (i.e., all that are removed
2813 : : // by, or replaced in, staging), gather their chunk feerates.
2814 : 1339 : auto main_clusters = GetConflicts();
2815 : 1339 : std::vector<FeeFrac> main_feerates, staging_feerates;
2816 [ + + ]: 3075 : for (Cluster* cluster : main_clusters) {
2817 : 1736 : cluster->AppendChunkFeerates(main_feerates);
2818 : : }
2819 : : // Do the same for the Clusters in staging themselves.
2820 [ + + ]: 10712 : for (int quality = 0; quality < int(QualityLevel::NONE); ++quality) {
2821 [ + + ]: 10791 : for (const auto& cluster : m_staging_clusterset->m_clusters[quality]) {
2822 : 1418 : cluster->AppendChunkFeerates(staging_feerates);
2823 : : }
2824 : : }
2825 : : // Sort both by decreasing feerate to obtain diagrams, and return them.
2826 : 1339 : std::ranges::sort(main_feerates, std::greater<ByRatioNegSize<FeeFrac>>{});
2827 : 1339 : std::ranges::sort(staging_feerates, std::greater<ByRatioNegSize<FeeFrac>>{});
2828 : 1339 : return std::make_pair(std::move(main_feerates), std::move(staging_feerates));
2829 : 1339 : }
2830 : :
2831 : 69330 : void GenericClusterImpl::SanityCheck(const TxGraphImpl& graph, int level) const
2832 : : {
2833 : : // There must be an m_mapping for each m_depgraph position (including holes).
2834 [ - + - + : 69330 : assert(m_depgraph.PositionRange() == m_mapping.size());
- + ]
2835 : : // The linearization for this Cluster must contain every transaction once.
2836 [ - + - + ]: 69330 : assert(m_depgraph.TxCount() == m_linearization.size());
2837 : : // Unless a split is to be applied, the cluster cannot have any holes.
2838 [ + + ]: 69330 : if (!NeedsSplitting()) {
2839 [ - + ]: 69329 : assert(m_depgraph.Positions() == SetType::Fill(m_depgraph.TxCount()));
2840 : : }
2841 : :
2842 : : // Compute the chunking of m_linearization.
2843 : 69330 : auto linchunking = ChunkLinearizationInfo(m_depgraph, m_linearization);
2844 : 69330 : unsigned chunk_num{0};
2845 : :
2846 : : // Verify m_linearization.
2847 : 69330 : SetType m_done;
2848 : 69330 : LinearizationIndex linindex{0};
2849 : 69330 : DepGraphIndex chunk_pos{0}; //!< position within the current chunk
2850 [ - + ]: 69330 : assert(m_depgraph.IsAcyclic());
2851 [ - + ]: 69330 : if (m_linearization.empty()) return;
2852 : 69330 : FeeFrac equal_feerate_prefix = linchunking[chunk_num].feerate;
2853 [ + + ]: 473811 : for (auto lin_pos : m_linearization) {
2854 [ - + - + ]: 404481 : assert(lin_pos < m_mapping.size());
2855 [ + - ]: 404481 : const auto& entry = graph.m_entries[m_mapping[lin_pos]];
2856 : : // Check that the linearization is topological.
2857 [ + - ]: 404481 : m_done.Set(lin_pos);
2858 [ + - ]: 404481 : if (IsTopological()) {
2859 [ - + ]: 404481 : assert(m_done.IsSupersetOf(m_depgraph.Ancestors(lin_pos)));
2860 : : }
2861 : : // Check that the Entry has a locator pointing back to this Cluster & position within it.
2862 [ - + ]: 404481 : assert(entry.m_locator[level].cluster == this);
2863 [ - + ]: 404481 : assert(entry.m_locator[level].index == lin_pos);
2864 : : // For main-level entries, check linearization position and chunk feerate.
2865 [ + - + + ]: 404481 : if (level == 0 && IsAcceptable()) {
2866 [ - + ]: 404480 : assert(entry.m_main_lin_index == linindex);
2867 : 404480 : ++linindex;
2868 [ + + ]: 404480 : if (!linchunking[chunk_num].transactions[lin_pos]) {
2869 : : // First transaction of a new chunk.
2870 : 189418 : ++chunk_num;
2871 [ - + - + ]: 189418 : assert(chunk_num < linchunking.size());
2872 : 189418 : chunk_pos = 0;
2873 [ + + ]: 189418 : if (ByRatio{linchunking[chunk_num].feerate} < ByRatio{equal_feerate_prefix}) {
2874 : 8751 : equal_feerate_prefix = linchunking[chunk_num].feerate;
2875 : : } else {
2876 [ - + ]: 180667 : assert(ByRatio{linchunking[chunk_num].feerate} == ByRatio{equal_feerate_prefix});
2877 : 180667 : equal_feerate_prefix += linchunking[chunk_num].feerate;
2878 : : }
2879 : : }
2880 [ + - ]: 404480 : assert(entry.m_main_chunk_feerate == linchunking[chunk_num].feerate);
2881 [ - + ]: 404480 : assert(entry.m_main_equal_feerate_chunk_prefix_size == equal_feerate_prefix.size);
2882 : : // Verify that an entry in the chunk index exists for every chunk-ending transaction.
2883 : 404480 : ++chunk_pos;
2884 [ + - ]: 404480 : if (graph.m_main_clusterset.m_to_remove.empty()) {
2885 [ - + ]: 404480 : bool is_chunk_end = (chunk_pos == linchunking[chunk_num].transactions.Count());
2886 [ - + ]: 404480 : assert((entry.m_main_chunkindex_iterator != graph.m_main_chunkindex.end()) == is_chunk_end);
2887 [ + + ]: 404480 : if (is_chunk_end) {
2888 [ + + ]: 258747 : auto& chunk_data = *entry.m_main_chunkindex_iterator;
2889 [ + + + + ]: 258747 : if (m_done == m_depgraph.Positions() && chunk_pos == 1) {
2890 [ - + ]: 58788 : assert(chunk_data.m_chunk_count == LinearizationIndex(-1));
2891 : : } else {
2892 [ - + ]: 199959 : assert(chunk_data.m_chunk_count == chunk_pos);
2893 : : }
2894 : : }
2895 : : }
2896 : : // If this Cluster has an acceptable quality level, its chunks must be connected.
2897 [ - + ]: 404480 : assert(m_depgraph.IsConnected(linchunking[chunk_num].transactions));
2898 : : }
2899 : : }
2900 : : // Verify that each element of m_depgraph occurred in m_linearization.
2901 [ - + ]: 69330 : assert(m_done == m_depgraph.Positions());
2902 : 69330 : }
2903 : :
2904 : 12074685 : void SingletonClusterImpl::SanityCheck(const TxGraphImpl& graph, int level) const
2905 : : {
2906 : : // All singletons are optimal, oversized, or need splitting.
2907 [ + + + + : 12074685 : Assume(IsOptimal() || IsOversized() || NeedsSplitting());
+ + ]
2908 [ + + ]: 12074685 : if (GetTxCount()) {
2909 [ - + ]: 12074674 : const auto& entry = graph.m_entries[m_graph_index];
2910 : : // Check that the Entry has a locator pointing back to this Cluster & position within it.
2911 [ - + ]: 12074674 : assert(entry.m_locator[level].cluster == this);
2912 [ - + ]: 12074674 : assert(entry.m_locator[level].index == 0);
2913 : : // For main-level entries, check linearization position and chunk feerate.
2914 [ + - + + ]: 12074674 : if (level == 0 && IsAcceptable()) {
2915 [ - + ]: 12074662 : assert(entry.m_main_lin_index == 0);
2916 [ + - ]: 12074662 : assert(entry.m_main_chunk_feerate == m_feerate);
2917 [ - + ]: 12074662 : assert(entry.m_main_equal_feerate_chunk_prefix_size == m_feerate.size);
2918 [ + + ]: 12074662 : if (graph.m_main_clusterset.m_to_remove.empty()) {
2919 [ - + ]: 12074368 : assert(entry.m_main_chunkindex_iterator != graph.m_main_chunkindex.end());
2920 [ - + ]: 12074368 : auto& chunk_data = *entry.m_main_chunkindex_iterator;
2921 [ - + ]: 12074368 : assert(chunk_data.m_chunk_count == LinearizationIndex(-1));
2922 : : }
2923 : : }
2924 : : }
2925 : 12074685 : }
2926 : :
2927 : 175346 : void TxGraphImpl::SanityCheck() const
2928 : : {
2929 : : /** Which GraphIndexes ought to occur in m_unlinked, based on m_entries. */
2930 : 175346 : std::set<GraphIndex> expected_unlinked;
2931 : : /** Which Clusters ought to occur in ClusterSet::m_clusters, based on m_entries. */
2932 [ + + ]: 876730 : std::set<const Cluster*> expected_clusters[MAX_LEVELS];
2933 : : /** Which GraphIndexes ought to occur in ClusterSet::m_removed, based on m_entries. */
2934 [ + + ]: 876730 : std::set<GraphIndex> expected_removed[MAX_LEVELS];
2935 : : /** Which Cluster::m_sequence values have been encountered. */
2936 : 175346 : std::set<uint64_t> sequences;
2937 : : /** Which Clusters are live in ClusterSet::m_clusters up to the level being checked. */
2938 : 175346 : std::set<const Cluster*> live_clusters;
2939 : : /** Which GraphIndexes ought to occur in m_main_chunkindex, based on m_entries. */
2940 : 175346 : std::set<GraphIndex> expected_chunkindex;
2941 : : /** Whether compaction is possible in the current state. */
2942 : 175346 : bool compact_possible{true};
2943 : :
2944 : : // Go over all Entry objects in m_entries.
2945 [ - + + + ]: 12654513 : for (GraphIndex idx = 0; idx < m_entries.size(); ++idx) {
2946 [ - + ]: 12479167 : const auto& entry = m_entries[idx];
2947 [ - + ]: 12479167 : if (entry.m_ref == nullptr) {
2948 : : // Unlinked Entry must have indexes appear in m_unlinked.
2949 [ # # ]: 0 : expected_unlinked.insert(idx);
2950 : : } else {
2951 : : // Every non-unlinked Entry must have a Ref that points back to it.
2952 [ - + ]: 12479167 : assert(GetRefGraph(*entry.m_ref) == this);
2953 [ - + ]: 12479167 : assert(GetRefIndex(*entry.m_ref) == idx);
2954 : : }
2955 [ + + ]: 12479167 : if (entry.m_main_chunkindex_iterator != m_main_chunkindex.end()) {
2956 : : // Remember which entries we see a chunkindex entry for.
2957 [ - + ]: 12333409 : assert(entry.m_locator[0].IsPresent());
2958 [ + - ]: 12333409 : expected_chunkindex.insert(idx);
2959 : : }
2960 : : // Verify the Entry m_locators.
2961 : : bool was_present{false}, was_removed{false};
2962 [ + + ]: 37437501 : for (int level = 0; level < MAX_LEVELS; ++level) {
2963 : 24958334 : const auto& locator = entry.m_locator[level];
2964 : : // Every Locator must be in exactly one of these 3 states.
2965 [ + + + + : 74875002 : assert(locator.IsMissing() + locator.IsRemoved() + locator.IsPresent() == 1);
- + ]
2966 [ + + ]: 24958334 : if (locator.IsPresent()) {
2967 : : // Once removed, a transaction cannot be revived.
2968 [ - + ]: 12479155 : assert(!was_removed);
2969 : : // Verify that the Cluster agrees with where the Locator claims the transaction is.
2970 [ - + ]: 12479155 : assert(locator.cluster->GetClusterEntry(locator.index) == idx);
2971 : : // Remember that we expect said Cluster to appear in the ClusterSet::m_clusters.
2972 [ + - ]: 12479155 : expected_clusters[level].insert(locator.cluster);
2973 : : was_present = true;
2974 [ - + ]: 24958334 : } else if (locator.IsRemoved()) {
2975 : : // Level 0 (main) cannot have IsRemoved locators (IsMissing there means non-existing).
2976 [ # # ]: 0 : assert(level > 0);
2977 : : // A Locator can only be IsRemoved if it was IsPresent before, and only once.
2978 [ # # ]: 0 : assert(was_present && !was_removed);
2979 : : // Remember that we expect this GraphIndex to occur in the ClusterSet::m_removed.
2980 [ # # ]: 0 : expected_removed[level].insert(idx);
2981 : : was_removed = true;
2982 : : }
2983 : : }
2984 : : }
2985 : :
2986 : : // For all levels (0 = main, 1 = staged)...
2987 [ + + ]: 350692 : for (int level = 0; level <= GetTopLevel(); ++level) {
2988 : 175346 : assert(level < MAX_LEVELS);
2989 : 175346 : auto& clusterset = GetClusterSet(level);
2990 : 175346 : std::set<const Cluster*> actual_clusters;
2991 : 175346 : size_t recomputed_cluster_usage{0};
2992 : :
2993 : : // For all quality levels...
2994 [ + + ]: 1402768 : for (int qual = 0; qual < int(QualityLevel::NONE); ++qual) {
2995 : 1227422 : QualityLevel quality{qual};
2996 : 1227422 : const auto& quality_clusters = clusterset.m_clusters[qual];
2997 : : // ... for all clusters in them ...
2998 [ - + + + ]: 13371437 : for (ClusterSetIndex setindex = 0; setindex < quality_clusters.size(); ++setindex) {
2999 [ + - ]: 12144015 : const auto& cluster = *quality_clusters[setindex];
3000 [ + - ]: 12144015 : live_clusters.insert(&cluster);
3001 : : // The number of transactions in a Cluster cannot exceed m_max_cluster_count.
3002 [ - + ]: 12144015 : assert(cluster.GetTxCount() <= m_max_cluster_count);
3003 : : // The level must match the Cluster's own idea of what level it is in (but GetLevel
3004 : : // can only be called for non-empty Clusters).
3005 [ + + - + ]: 12144015 : assert(cluster.GetTxCount() == 0 || level == cluster.GetLevel(*this));
3006 : : // The sum of their sizes cannot exceed m_max_cluster_size, unless it is an
3007 : : // individually oversized transaction singleton. Note that groups of to-be-merged
3008 : : // clusters which would exceed this limit are marked oversized, which means they
3009 : : // are never applied.
3010 [ + + - + ]: 12144015 : assert(cluster.IsOversized() || cluster.GetTotalTxSize() <= m_max_cluster_size);
3011 : : // OVERSIZED clusters are singletons.
3012 [ + + - + ]: 12144015 : assert(!cluster.IsOversized() || cluster.GetTxCount() == 1);
3013 : : // Transaction counts cannot exceed the Cluster implementation's maximum
3014 : : // supported transactions count.
3015 [ - + ]: 12144015 : assert(cluster.GetTxCount() <= cluster.GetMaxTxCount());
3016 : : // Unless a Split is yet to be applied, the number of transactions must not be
3017 : : // below the Cluster implementation's intended transaction count.
3018 [ + + ]: 12144015 : if (!cluster.NeedsSplitting()) {
3019 [ - + ]: 12144003 : assert(cluster.GetTxCount() >= cluster.GetMinIntendedTxCount());
3020 : : }
3021 : :
3022 : : // Check the sequence number.
3023 [ - + ]: 12144015 : assert(cluster.m_sequence < m_next_sequence_counter);
3024 [ - + ]: 12144015 : assert(!sequences.contains(cluster.m_sequence));
3025 [ + - ]: 12144015 : sequences.insert(cluster.m_sequence);
3026 : : // Remember we saw this Cluster (only if it is non-empty; empty Clusters aren't
3027 : : // expected to be referenced by the Entry vector).
3028 [ + + ]: 12144015 : if (cluster.GetTxCount() != 0) {
3029 [ + - ]: 12144004 : actual_clusters.insert(&cluster);
3030 : : }
3031 : : // Sanity check the cluster, according to the Cluster's internal rules.
3032 [ + - ]: 12144015 : cluster.SanityCheck(*this, level);
3033 : : // Check that the cluster's quality and setindex matches its position in the quality list.
3034 [ - + ]: 12144015 : assert(cluster.m_quality == quality);
3035 [ - + ]: 12144015 : assert(cluster.m_setindex == setindex);
3036 : : // Count memory usage.
3037 : 12144015 : recomputed_cluster_usage += cluster.TotalMemoryUsage();
3038 : : }
3039 : : }
3040 : :
3041 : : // Verify memory usage.
3042 [ - + ]: 175346 : assert(clusterset.m_cluster_usage == recomputed_cluster_usage);
3043 : :
3044 : : // Verify that all to-be-removed transactions have valid identifiers.
3045 [ + + ]: 175354 : for (GraphIndex idx : clusterset.m_to_remove) {
3046 [ - + - + ]: 8 : assert(idx < m_entries.size());
3047 : : // We cannot assert that all m_to_remove transactions are still present: ~Ref on a
3048 : : // (P,M) transaction (present in main, inherited in staging) will cause an m_to_remove
3049 : : // addition in both main and staging, but a subsequence ApplyRemovals in main will
3050 : : // cause it to disappear from staging too, leaving the m_to_remove in place.
3051 : : }
3052 : :
3053 : : // Verify that all to-be-added dependencies have valid identifiers.
3054 [ - + + + ]: 366762 : for (auto [par_idx, chl_idx] : clusterset.m_deps_to_add) {
3055 [ - + ]: 191416 : assert(par_idx != chl_idx);
3056 [ - + - + ]: 191416 : assert(par_idx < m_entries.size());
3057 [ - + ]: 191416 : assert(chl_idx < m_entries.size());
3058 : : }
3059 : :
3060 : : // Verify that the actually encountered clusters match the ones occurring in Entry vector.
3061 [ - + ]: 175346 : assert(actual_clusters == expected_clusters[level]);
3062 : :
3063 : : // Validate m_group_data cluster pointers refer to live clusters at this
3064 : : // level or below (staging group data may reference not-yet-pulled-in
3065 : : // main clusters).
3066 [ + + ]: 175346 : if (clusterset.m_group_data.has_value()) {
3067 [ - + ]: 175336 : for (const Cluster* cl : clusterset.m_group_data->m_group_clusters) {
3068 [ # # ]: 0 : assert(live_clusters.contains(cl));
3069 [ # # ]: 0 : assert(cl->GetTxCount() > 0);
3070 [ # # ]: 0 : assert(FindCluster(cl->GetClusterEntry(0), level) == cl);
3071 : : }
3072 : : }
3073 : : // Verify that the contents of m_removed matches what was expected based on the Entry vector.
3074 [ + - ]: 175346 : std::set<GraphIndex> actual_removed(clusterset.m_removed.begin(), clusterset.m_removed.end());
3075 [ - + ]: 175346 : for (auto i : expected_unlinked) {
3076 : : // If a transaction exists in both main and staging, and is removed from staging (adding
3077 : : // it to m_removed there), and consequently destroyed (wiping the locator completely),
3078 : : // it can remain in m_removed despite not having an IsRemoved() locator. Exclude those
3079 : : // transactions from the comparison here.
3080 : 0 : actual_removed.erase(i);
3081 : 0 : expected_removed[level].erase(i);
3082 : : }
3083 : :
3084 [ - + ]: 175346 : assert(actual_removed == expected_removed[level]);
3085 : :
3086 : : // If any GraphIndex entries remain in this ClusterSet, compact is not possible.
3087 [ + + ]: 175346 : if (!clusterset.m_deps_to_add.empty()) compact_possible = false;
3088 [ + + ]: 175346 : if (!clusterset.m_to_remove.empty()) compact_possible = false;
3089 [ - + ]: 175346 : if (!clusterset.m_removed.empty()) compact_possible = false;
3090 : :
3091 : : // For non-top levels, m_oversized must be known (as it cannot change until the level
3092 : : // on top is gone).
3093 [ - + - - ]: 175346 : if (level < GetTopLevel()) assert(clusterset.m_oversized.has_value());
3094 : 175346 : }
3095 : :
3096 : : // Verify that the contents of m_unlinked matches what was expected based on the Entry vector.
3097 [ + - ]: 175346 : std::set<GraphIndex> actual_unlinked(m_unlinked.begin(), m_unlinked.end());
3098 [ - + ]: 175346 : assert(actual_unlinked == expected_unlinked);
3099 : :
3100 : : // If compaction was possible, it should have been performed already, and m_unlinked must be
3101 : : // empty (to prevent memory leaks due to an ever-growing m_entries vector).
3102 [ + + ]: 175346 : if (compact_possible) {
3103 [ - + ]: 175338 : assert(actual_unlinked.empty());
3104 : : }
3105 : :
3106 : : // Finally, check the chunk index.
3107 : 175346 : std::set<GraphIndex> actual_chunkindex;
3108 : 175346 : FeeFrac last_chunk_feerate;
3109 [ + + ]: 12508755 : for (const auto& chunk : m_main_chunkindex) {
3110 : 12333409 : GraphIndex idx = chunk.m_graph_index;
3111 [ + - ]: 12333409 : actual_chunkindex.insert(idx);
3112 [ + + ]: 12333409 : auto chunk_feerate = m_entries[idx].m_main_chunk_feerate;
3113 [ + + ]: 12333409 : if (!last_chunk_feerate.IsEmpty()) {
3114 [ - + ]: 12296061 : assert(ByRatio{last_chunk_feerate} >= ByRatio{FeeFrac{chunk_feerate}});
3115 : : }
3116 : 12333409 : last_chunk_feerate = chunk_feerate;
3117 : : }
3118 [ - + ]: 175346 : assert(actual_chunkindex == expected_chunkindex);
3119 [ + + + + : 1227422 : }
- - - - ]
3120 : :
3121 : 182232 : bool TxGraphImpl::DoWork(uint64_t max_cost) noexcept
3122 : : {
3123 : 182232 : uint64_t cost_done{0};
3124 : : // First linearize everything in NEEDS_RELINEARIZE to an acceptable level. If more budget
3125 : : // remains after that, try to make everything optimal.
3126 [ + + ]: 728928 : for (QualityLevel quality : {QualityLevel::NEEDS_FIX, QualityLevel::NEEDS_RELINEARIZE, QualityLevel::ACCEPTABLE}) {
3127 : : // First linearize staging, if it exists, then main.
3128 [ + + ]: 1093392 : for (int level = GetTopLevel(); level >= 0; --level) {
3129 : : // Do not modify main if it has any observers.
3130 [ + - - + ]: 546696 : if (level == 0 && m_main_chunkindex_observers != 0) continue;
3131 : 546696 : ApplyDependencies(level);
3132 : 546696 : auto& clusterset = GetClusterSet(level);
3133 : : // Do not modify oversized levels.
3134 [ + - ]: 546696 : if (clusterset.m_oversized == true) continue;
3135 : 546696 : auto& queue = clusterset.m_clusters[int(quality)];
3136 [ + + ]: 552592 : while (!queue.empty()) {
3137 [ + - ]: 5896 : if (cost_done >= max_cost) return false;
3138 : : // Randomize the order in which we process, so that if the first cluster somehow
3139 : : // needs more work than what max_cost allows, we don't keep spending it on the same
3140 : : // one.
3141 [ - + ]: 5896 : auto pos = m_rng.randrange<size_t>(queue.size());
3142 : 5896 : auto cost_now = max_cost - cost_done;
3143 [ + - ]: 5896 : if (quality == QualityLevel::NEEDS_FIX || quality == QualityLevel::NEEDS_RELINEARIZE) {
3144 : : // If we're working with clusters that need relinearization still, only perform
3145 : : // up to m_acceptable_cost work. If they become ACCEPTABLE, and we still
3146 : : // have budget after all other clusters are ACCEPTABLE too, we'll spend the
3147 : : // remaining budget on trying to make them OPTIMAL.
3148 [ - + ]: 5896 : cost_now = std::min(cost_now, m_acceptable_cost);
3149 : : }
3150 [ + - ]: 5896 : auto [cost, improved] = queue[pos].get()->Relinearize(*this, level, cost_now);
3151 : 5896 : cost_done += cost;
3152 : : // If no improvement was made to the Cluster, it means we've essentially run out of
3153 : : // budget. Even though it may be the case that cost_done < max_cost still, the
3154 : : // linearizer decided there wasn't enough budget left to attempt anything with.
3155 : : // To avoid an infinite loop that keeps trying clusters with minuscule budgets,
3156 : : // stop here too.
3157 [ + - ]: 5896 : if (!improved) return false;
3158 : : }
3159 : : }
3160 : : }
3161 : : // All possible work has been performed, so we can return true. Note that this does *not* mean
3162 : : // that all clusters are optimally linearized now. It may be that there is nothing to do left
3163 : : // because all non-optimal clusters are in oversized and/or observer-bearing levels.
3164 : : return true;
3165 : : }
3166 : :
3167 : 12235756 : void BlockBuilderImpl::Next() noexcept
3168 : : {
3169 : : // Don't do anything if we're already done.
3170 [ + - ]: 12235756 : if (m_cur_iter == m_graph->m_main_chunkindex.end()) return;
3171 : 12235769 : while (true) {
3172 : : // Advance the pointer, and stop if we reach the end.
3173 [ + + ]: 12235769 : ++m_cur_iter;
3174 : 12235769 : m_cur_cluster = nullptr;
3175 [ + + ]: 12235769 : if (m_cur_iter == m_graph->m_main_chunkindex.end()) break;
3176 : : // Find the cluster pointed to by m_cur_iter.
3177 : 12197007 : const auto& chunk_data = *m_cur_iter;
3178 : 12197007 : const auto& chunk_end_entry = m_graph->m_entries[chunk_data.m_graph_index];
3179 : 12197007 : m_cur_cluster = chunk_end_entry.m_locator[0].cluster;
3180 : 12197007 : m_known_end_of_cluster = false;
3181 : : // If we previously skipped a chunk from this cluster we cannot include more from it.
3182 [ + + ]: 12197007 : if (!m_excluded_clusters.contains(m_cur_cluster->m_sequence)) break;
3183 : : }
3184 : : }
3185 : :
3186 : 12450165 : std::optional<std::pair<std::vector<TxGraph::Ref*>, FeePerWeight>> BlockBuilderImpl::GetCurrentChunk() noexcept
3187 : : {
3188 : 12450165 : std::optional<std::pair<std::vector<TxGraph::Ref*>, FeePerWeight>> ret;
3189 : : // Populate the return value if we are not done.
3190 [ + + ]: 12450165 : if (m_cur_iter != m_graph->m_main_chunkindex.end()) {
3191 : 12235802 : ret.emplace();
3192 [ + + ]: 12235802 : const auto& chunk_data = *m_cur_iter;
3193 [ + + ]: 12235802 : const auto& chunk_end_entry = m_graph->m_entries[chunk_data.m_graph_index];
3194 [ + + ]: 12235802 : if (chunk_data.m_chunk_count == LinearizationIndex(-1)) {
3195 : : // Special case in case just a single transaction remains, avoiding the need to
3196 : : // dispatch to and dereference Cluster.
3197 : 12031515 : ret->first.resize(1);
3198 : 12031515 : Assume(chunk_end_entry.m_ref != nullptr);
3199 : 12031515 : ret->first[0] = chunk_end_entry.m_ref;
3200 : 12031515 : m_known_end_of_cluster = true;
3201 : : } else {
3202 : 204287 : Assume(m_cur_cluster);
3203 : 204287 : ret->first.resize(chunk_data.m_chunk_count);
3204 : 204287 : auto start_pos = chunk_end_entry.m_main_lin_index + 1 - chunk_data.m_chunk_count;
3205 [ - + ]: 204287 : m_known_end_of_cluster = m_cur_cluster->GetClusterRefs(*m_graph, ret->first, start_pos);
3206 : : // If the chunk size was 1 and at end of cluster, then the special case above should
3207 : : // have been used.
3208 : 204287 : Assume(!m_known_end_of_cluster || chunk_data.m_chunk_count > 1);
3209 : : }
3210 : 12235802 : ret->second = chunk_end_entry.m_main_chunk_feerate;
3211 : : }
3212 : 12450165 : return ret;
3213 : : }
3214 : :
3215 : 214430 : BlockBuilderImpl::BlockBuilderImpl(TxGraphImpl& graph) noexcept : m_graph(&graph)
3216 : : {
3217 : : // Make sure all clusters in main are up to date, and acceptable.
3218 : 214430 : m_graph->MakeAllAcceptable(0);
3219 : : // There cannot remain any inapplicable dependencies (only possible if main is oversized).
3220 [ + + ]: 214430 : Assume(m_graph->m_main_clusterset.m_deps_to_add.empty());
3221 : : // Remember that this object is observing the graph's index, so that we can detect concurrent
3222 : : // modifications.
3223 : 214430 : ++m_graph->m_main_chunkindex_observers;
3224 : : // Find the first chunk.
3225 [ + + ]: 214430 : m_cur_iter = m_graph->m_main_chunkindex.begin();
3226 : 214430 : m_cur_cluster = nullptr;
3227 [ + + ]: 214430 : if (m_cur_iter != m_graph->m_main_chunkindex.end()) {
3228 : : // Find the cluster pointed to by m_cur_iter.
3229 : 38829 : const auto& chunk_data = *m_cur_iter;
3230 : 38829 : const auto& chunk_end_entry = m_graph->m_entries[chunk_data.m_graph_index];
3231 : 38829 : m_cur_cluster = chunk_end_entry.m_locator[0].cluster;
3232 : : }
3233 : 214430 : }
3234 : :
3235 : 428860 : BlockBuilderImpl::~BlockBuilderImpl()
3236 : : {
3237 : 214430 : Assume(m_graph->m_main_chunkindex_observers > 0);
3238 : : // Permit modifications to the main graph again after destroying the BlockBuilderImpl.
3239 : 214430 : --m_graph->m_main_chunkindex_observers;
3240 : 428860 : }
3241 : :
3242 : 12191864 : void BlockBuilderImpl::Include() noexcept
3243 : : {
3244 : : // The actual inclusion of the chunk is done by the calling code. All we have to do is switch
3245 : : // to the next chunk.
3246 : 12191864 : Next();
3247 : 12191864 : }
3248 : :
3249 : 43892 : void BlockBuilderImpl::Skip() noexcept
3250 : : {
3251 : : // When skipping a chunk we need to not include anything more of the cluster, as that could make
3252 : : // the result topologically invalid. However, don't do this if the chunk is known to be the last
3253 : : // chunk of the cluster. This may significantly reduce the size of m_excluded_clusters,
3254 : : // especially when many singleton clusters are ignored.
3255 [ + - + + ]: 43892 : if (m_cur_cluster != nullptr && !m_known_end_of_cluster) {
3256 : 1 : m_excluded_clusters.insert(m_cur_cluster->m_sequence);
3257 : : }
3258 : 43892 : Next();
3259 : 43892 : }
3260 : :
3261 : 214430 : std::unique_ptr<TxGraph::BlockBuilder> TxGraphImpl::GetBlockBuilder() noexcept
3262 : : {
3263 [ - + ]: 214430 : return std::make_unique<BlockBuilderImpl>(*this);
3264 : : }
3265 : :
3266 : 53 : std::pair<std::vector<TxGraph::Ref*>, FeePerWeight> TxGraphImpl::GetWorstMainChunk() noexcept
3267 : : {
3268 : 53 : std::pair<std::vector<Ref*>, FeePerWeight> ret;
3269 : : // Make sure all clusters in main are up to date, and acceptable.
3270 : 53 : MakeAllAcceptable(0);
3271 [ + - ]: 53 : Assume(m_main_clusterset.m_deps_to_add.empty());
3272 : : // If the graph is not empty, populate ret.
3273 [ + - ]: 53 : if (!m_main_chunkindex.empty()) {
3274 [ + + ]: 53 : const auto& chunk_data = *m_main_chunkindex.rbegin();
3275 [ + + ]: 53 : const auto& chunk_end_entry = m_entries[chunk_data.m_graph_index];
3276 : 53 : Cluster* cluster = chunk_end_entry.m_locator[0].cluster;
3277 [ + + ]: 53 : if (chunk_data.m_chunk_count == LinearizationIndex(-1) || chunk_data.m_chunk_count == 1) {
3278 : : // Special case for singletons.
3279 : 47 : ret.first.resize(1);
3280 : 47 : Assume(chunk_end_entry.m_ref != nullptr);
3281 : 47 : ret.first[0] = chunk_end_entry.m_ref;
3282 : : } else {
3283 : 6 : ret.first.resize(chunk_data.m_chunk_count);
3284 : 6 : auto start_pos = chunk_end_entry.m_main_lin_index + 1 - chunk_data.m_chunk_count;
3285 [ - + ]: 6 : cluster->GetClusterRefs(*this, ret.first, start_pos);
3286 : 6 : std::reverse(ret.first.begin(), ret.first.end());
3287 : : }
3288 : 53 : ret.second = chunk_end_entry.m_main_chunk_feerate;
3289 : : }
3290 : 53 : return ret;
3291 : : }
3292 : :
3293 : 2917 : std::vector<TxGraph::Ref*> TxGraphImpl::Trim() noexcept
3294 : : {
3295 [ + + ]: 2917 : int level = GetTopLevel();
3296 [ + + ]: 2917 : Assume(m_main_chunkindex_observers == 0 || level != 0);
3297 : 2917 : std::vector<TxGraph::Ref*> ret;
3298 : :
3299 : : // Compute the groups of to-be-merged Clusters (which also applies all removals, and splits).
3300 : 2917 : auto& clusterset = GetClusterSet(level);
3301 [ + + ]: 2917 : if (clusterset.m_oversized == false) return ret;
3302 : 32 : GroupClusters(level);
3303 [ + - ]: 32 : Assume(clusterset.m_group_data.has_value());
3304 : : // Nothing to do if not oversized.
3305 : 32 : Assume(clusterset.m_oversized.has_value());
3306 [ + - ]: 32 : if (clusterset.m_oversized == false) return ret;
3307 : :
3308 : : // In this function, would-be clusters (as precomputed in m_group_data by GroupClusters) are
3309 : : // trimmed by removing transactions in them such that the resulting clusters satisfy the size
3310 : : // and count limits.
3311 : : //
3312 : : // It works by defining for each would-be cluster a rudimentary linearization: at every point
3313 : : // the highest-chunk-feerate remaining transaction is picked among those with no unmet
3314 : : // dependencies. "Dependency" here means either a to-be-added dependency (m_deps_to_add), or
3315 : : // an implicit dependency added between any two consecutive transaction in their current
3316 : : // cluster linearization. So it can be seen as a "merge sort" of the chunks of the clusters,
3317 : : // but respecting the dependencies being added.
3318 : : //
3319 : : // This rudimentary linearization is computed lazily, by putting all eligible (no unmet
3320 : : // dependencies) transactions in a heap, and popping the highest-feerate one from it. Along the
3321 : : // way, the counts and sizes of the would-be clusters up to that point are tracked (by
3322 : : // partitioning the involved transactions using a union-find structure). Any transaction whose
3323 : : // addition would cause a violation is removed, along with all their descendants.
3324 : : //
3325 : : // A next invocation of GroupClusters (after applying the removals) will compute the new
3326 : : // resulting clusters, and none of them will violate the limits.
3327 : :
3328 : : /** All dependencies (both to be added ones, and implicit ones between consecutive transactions
3329 : : * in existing cluster linearizations), sorted by parent. */
3330 : 4 : std::vector<std::pair<GraphIndex, GraphIndex>> deps_by_parent;
3331 : : /** Same, but sorted by child. */
3332 : 4 : std::vector<std::pair<GraphIndex, GraphIndex>> deps_by_child;
3333 : : /** Information about all transactions involved in a Cluster group to be trimmed, sorted by
3334 : : * GraphIndex. It contains entries both for transactions that have already been included,
3335 : : * and ones that have not yet been. */
3336 : 4 : std::vector<TrimTxData> trim_data;
3337 : : /** Iterators into trim_data, treated as a max heap according to cmp_fn below. Each entry is
3338 : : * a transaction that has not yet been included yet, but all its ancestors have. */
3339 : 4 : std::vector<std::vector<TrimTxData>::iterator> trim_heap;
3340 : : /** The list of representatives of the partitions a given transaction depends on. */
3341 : 4 : std::vector<TrimTxData*> current_deps;
3342 : :
3343 : : /** Function to define the ordering of trim_heap. */
3344 : 1084529 : static constexpr auto cmp_fn = [](auto a, auto b) noexcept {
3345 : : // Sort by increasing chunk feerate, and then by decreasing size.
3346 : : // We do not need to sort by cluster or within clusters, because due to the implicit
3347 : : // dependency between consecutive linearization elements, no two transactions from the
3348 : : // same Cluster will ever simultaneously be in the heap.
3349 : 1084525 : return ByRatioNegSize{a->m_chunk_feerate} < ByRatioNegSize{b->m_chunk_feerate};
3350 : : };
3351 : :
3352 : : /** Given a TrimTxData entry, find the representative of the partition it is in. */
3353 : 127308 : static constexpr auto find_fn = [](TrimTxData* arg) noexcept {
3354 [ + + - + ]: 189351 : while (arg != arg->m_uf_parent) {
3355 : : // Replace pointer to parent with pointer to grandparent (path splitting).
3356 : : // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Finding_set_representatives.
3357 : 62047 : auto par = arg->m_uf_parent;
3358 : 62047 : arg->m_uf_parent = par->m_uf_parent;
3359 : 62047 : arg = par;
3360 : : }
3361 : 127304 : return arg;
3362 : : };
3363 : :
3364 : : /** Given two TrimTxData entries, union the partitions they are in, and return the
3365 : : * representative. */
3366 : 63100 : static constexpr auto union_fn = [](TrimTxData* arg1, TrimTxData* arg2) noexcept {
3367 : : // Replace arg1 and arg2 by their representatives.
3368 [ - + ]: 126192 : auto rep1 = find_fn(arg1);
3369 : 63096 : auto rep2 = find_fn(arg2);
3370 : : // Bail out if both representatives are the same, because that means arg1 and arg2 are in
3371 : : // the same partition already.
3372 [ + - ]: 63096 : if (rep1 == rep2) return rep1;
3373 : : // Pick the lower-count root to become a child of the higher-count one.
3374 : : // See https://en.wikipedia.org/wiki/Disjoint-set_data_structure#Union_by_size.
3375 [ + + ]: 63096 : if (rep1->m_uf_count < rep2->m_uf_count) std::swap(rep1, rep2);
3376 : 63096 : rep2->m_uf_parent = rep1;
3377 : : // Add the statistics of arg2 (which is no longer a representative) to those of arg1 (which
3378 : : // is now the representative for both).
3379 : 63096 : rep1->m_uf_size += rep2->m_uf_size;
3380 : 63096 : rep1->m_uf_count += rep2->m_uf_count;
3381 : 63096 : return rep1;
3382 : : };
3383 : :
3384 : : /** Get iterator to TrimTxData entry for a given index. */
3385 : 128420 : auto locate_fn = [&](GraphIndex index) noexcept {
3386 : 128416 : auto it = std::lower_bound(trim_data.begin(), trim_data.end(), index, [](TrimTxData& elem, GraphIndex idx) noexcept {
3387 [ + + ]: 2048000 : return elem.m_index < idx;
3388 : : });
3389 [ + - ]: 128416 : Assume(it != trim_data.end() && it->m_index == index);
3390 : 128416 : return it;
3391 : 4 : };
3392 : :
3393 : : // For each group of to-be-merged Clusters.
3394 [ + + ]: 13 : for (const auto& group_data : clusterset.m_group_data->m_groups) {
3395 [ + + ]: 9 : trim_data.clear();
3396 [ - + ]: 9 : trim_heap.clear();
3397 [ - + ]: 9 : deps_by_child.clear();
3398 [ - + ]: 9 : deps_by_parent.clear();
3399 : :
3400 : : // Gather trim data and implicit dependency data from all involved Clusters.
3401 [ - + ]: 9 : auto cluster_span = std::span{clusterset.m_group_data->m_group_clusters}
3402 : 9 : .subspan(group_data.m_cluster_offset, group_data.m_cluster_count);
3403 : 9 : uint64_t size{0};
3404 [ + + ]: 64226 : for (Cluster* cluster : cluster_span) {
3405 : 64217 : MakeAcceptable(*cluster, cluster->GetLevel(*this));
3406 : 64217 : size += cluster->AppendTrimData(trim_data, deps_by_child);
3407 : : }
3408 : : // If this group of Clusters does not violate any limits, continue to the next group.
3409 [ - + + + : 9 : if (trim_data.size() <= m_max_cluster_count && size <= m_max_cluster_size) continue;
+ - ]
3410 : : // Sort the trim data by GraphIndex. In what follows, we will treat this sorted vector as
3411 : : // a map from GraphIndex to TrimTxData via locate_fn, and its ordering will not change
3412 : : // anymore.
3413 [ + + + + : 1315115 : std::ranges::sort(trim_data, [](auto& a, auto& b) noexcept { return a.m_index < b.m_index; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
3414 : :
3415 : : // Add the explicitly added dependencies to deps_by_child.
3416 : 9 : deps_by_child.insert(deps_by_child.end(),
3417 : 9 : clusterset.m_deps_to_add.begin() + group_data.m_deps_offset,
3418 : 9 : clusterset.m_deps_to_add.begin() + group_data.m_deps_offset + group_data.m_deps_count);
3419 : :
3420 : : // Sort deps_by_child by child transaction GraphIndex. The order will not be changed
3421 : : // anymore after this.
3422 [ + + + + : 1453049 : std::ranges::sort(deps_by_child, [](auto& a, auto& b) noexcept { return a.second < b.second; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
3423 : : // Fill m_parents_count and m_parents_offset in trim_data, as well as m_deps_left, and
3424 : : // initially populate trim_heap. Because of the sort above, all dependencies involving the
3425 : : // same child are grouped together, so a single linear scan suffices.
3426 : 9 : auto deps_it = deps_by_child.begin();
3427 [ + + ]: 64226 : for (auto trim_it = trim_data.begin(); trim_it != trim_data.end(); ++trim_it) {
3428 : 64217 : trim_it->m_parent_offset = deps_it - deps_by_child.begin();
3429 : 64217 : trim_it->m_deps_left = 0;
3430 [ + + + + ]: 128425 : while (deps_it != deps_by_child.end() && deps_it->second == trim_it->m_index) {
3431 : 64208 : ++trim_it->m_deps_left;
3432 : 64208 : ++deps_it;
3433 : : }
3434 : 64217 : trim_it->m_parent_count = trim_it->m_deps_left;
3435 : : // If this transaction has no unmet dependencies, and is not oversized, add it to the
3436 : : // heap (just append for now, the heapification happens below).
3437 [ + + + + ]: 64217 : if (trim_it->m_deps_left == 0 && trim_it->m_tx_size <= m_max_cluster_size) {
3438 : 1150 : trim_heap.push_back(trim_it);
3439 : : }
3440 : : }
3441 : 9 : Assume(deps_it == deps_by_child.end());
3442 : :
3443 : : // Construct deps_by_parent, sorted by parent transaction GraphIndex. The order will not be
3444 : : // changed anymore after this.
3445 : 9 : deps_by_parent = deps_by_child;
3446 [ + + + + : 1748590 : std::ranges::sort(deps_by_parent, [](auto& a, auto& b) noexcept { return a.first < b.first; });
+ + + + +
+ + + + +
+ + + + +
+ - - +
+ ]
3447 : : // Fill m_children_offset and m_children_count in trim_data. Because of the sort above, all
3448 : : // dependencies involving the same parent are grouped together, so a single linear scan
3449 : : // suffices.
3450 : 9 : deps_it = deps_by_parent.begin();
3451 [ + + ]: 64226 : for (auto& trim_entry : trim_data) {
3452 : 64217 : trim_entry.m_children_count = 0;
3453 : 64217 : trim_entry.m_children_offset = deps_it - deps_by_parent.begin();
3454 [ + + + + ]: 128425 : while (deps_it != deps_by_parent.end() && deps_it->first == trim_entry.m_index) {
3455 : 64208 : ++trim_entry.m_children_count;
3456 : 64208 : ++deps_it;
3457 : : }
3458 : : }
3459 : 9 : Assume(deps_it == deps_by_parent.end());
3460 : :
3461 : : // Build a heap of all transactions with 0 unmet dependencies.
3462 : 9 : std::make_heap(trim_heap.begin(), trim_heap.end(), cmp_fn);
3463 : :
3464 : : // Iterate over to-be-included transactions, and convert them to included transactions, or
3465 : : // skip them if adding them would violate resource limits of the would-be cluster.
3466 : : //
3467 : : // It is possible that the heap empties without ever hitting either cluster limit, in case
3468 : : // the implied graph (to be added dependencies plus implicit dependency between each
3469 : : // original transaction and its predecessor in the linearization it came from) contains
3470 : : // cycles. Such cycles will be removed entirely, because each of the transactions in the
3471 : : // cycle permanently have unmet dependencies. However, this cannot occur in real scenarios
3472 : : // where Trim() is called to deal with reorganizations that would violate cluster limits,
3473 : : // as all added dependencies are in the same direction (from old mempool transactions to
3474 : : // new from-block transactions); cycles require dependencies in both directions to be
3475 : : // added.
3476 [ + + ]: 64229 : while (!trim_heap.empty()) {
3477 : : // Move the best remaining transaction to the end of trim_heap.
3478 : 64211 : std::pop_heap(trim_heap.begin(), trim_heap.end(), cmp_fn);
3479 : : // Pop it, and find its TrimTxData.
3480 [ + + ]: 64211 : auto& entry = *trim_heap.back();
3481 [ + + ]: 64211 : trim_heap.pop_back();
3482 : :
3483 : : // Initialize it as a singleton partition.
3484 : 64211 : entry.m_uf_parent = &entry;
3485 : 64211 : entry.m_uf_count = 1;
3486 : 64211 : entry.m_uf_size = entry.m_tx_size;
3487 : :
3488 : : // Find the distinct transaction partitions this entry depends on.
3489 [ + + ]: 64211 : current_deps.clear();
3490 [ - + + + ]: 128419 : for (auto& [par, chl] : std::span{deps_by_child}.subspan(entry.m_parent_offset, entry.m_parent_count)) {
3491 : 64208 : Assume(chl == entry.m_index);
3492 : 128416 : current_deps.push_back(find_fn(&*locate_fn(par)));
3493 : : }
3494 : 64211 : std::ranges::sort(current_deps);
3495 : 64211 : current_deps.erase(std::ranges::unique(current_deps).begin(), current_deps.end());
3496 : :
3497 : : // Compute resource counts.
3498 : 64211 : uint32_t new_count = 1;
3499 : 64211 : uint64_t new_size = entry.m_tx_size;
3500 [ + + ]: 128419 : for (TrimTxData* ptr : current_deps) {
3501 : 64208 : new_count += ptr->m_uf_count;
3502 : 64208 : new_size += ptr->m_uf_size;
3503 : : }
3504 : : // Skip the entry if this would violate any limit.
3505 [ + + - + ]: 64211 : if (new_count > m_max_cluster_count || new_size > m_max_cluster_size) continue;
3506 : :
3507 : : // Union the partitions this transaction and all its dependencies are in together.
3508 : 64198 : auto rep = &entry;
3509 [ + + ]: 127294 : for (TrimTxData* ptr : current_deps) rep = union_fn(ptr, rep);
3510 : : // Mark the entry as included (so the loop below will not remove the transaction).
3511 : 64198 : entry.m_deps_left = uint32_t(-1);
3512 : : // Mark each to-be-added dependency involving this transaction as parent satisfied.
3513 [ - + + + ]: 128406 : for (auto& [par, chl] : std::span{deps_by_parent}.subspan(entry.m_children_offset, entry.m_children_count)) {
3514 : 64208 : Assume(par == entry.m_index);
3515 : 64208 : auto chl_it = locate_fn(chl);
3516 : : // Reduce the number of unmet dependencies of chl_it, and if that brings the number
3517 : : // to zero, add it to the heap of includable transactions.
3518 [ + + ]: 64208 : Assume(chl_it->m_deps_left > 0);
3519 [ + + ]: 64208 : if (--chl_it->m_deps_left == 0) {
3520 : 63061 : trim_heap.push_back(chl_it);
3521 : 63061 : std::push_heap(trim_heap.begin(), trim_heap.end(), cmp_fn);
3522 : : }
3523 : : }
3524 : : }
3525 : :
3526 : : // Remove all the transactions that were not processed above. Because nothing gets
3527 : : // processed until/unless all its dependencies are met, this automatically guarantees
3528 : : // that if a transaction is removed, all its descendants, or would-be descendants, are
3529 : : // removed as well.
3530 [ + + ]: 64226 : for (const auto& trim_entry : trim_data) {
3531 [ + + ]: 64217 : if (trim_entry.m_deps_left != uint32_t(-1)) {
3532 : 19 : ret.push_back(m_entries[trim_entry.m_index].m_ref);
3533 : 19 : clusterset.m_to_remove.push_back(trim_entry.m_index);
3534 : : }
3535 : : }
3536 : : }
3537 [ + - ]: 4 : clusterset.m_group_data.reset();
3538 : 4 : clusterset.m_oversized = false;
3539 : 4 : Assume(!ret.empty());
3540 : 4 : return ret;
3541 : 4 : }
3542 : :
3543 : 553683 : size_t TxGraphImpl::GetMainMemoryUsage() noexcept
3544 : : {
3545 : : // Make sure splits/merges are applied, as memory usage may not be representative otherwise.
3546 : 553683 : SplitAll(/*up_to_level=*/0);
3547 : 553683 : ApplyDependencies(/*level=*/0);
3548 : : // Compute memory usage
3549 : 553683 : size_t usage = /* From clusters */
3550 : 553683 : m_main_clusterset.m_cluster_usage +
3551 : : /* From Entry objects. */
3552 : 553683 : sizeof(Entry) * m_main_clusterset.m_txcount +
3553 : : /* From the chunk index. */
3554 : 553683 : memusage::DynamicUsage(m_main_chunkindex);
3555 : 553683 : return usage;
3556 : : }
3557 : :
3558 : : } // namespace
3559 : :
3560 : 410992 : TxGraph::Ref::~Ref()
3561 : : {
3562 [ + + ]: 410992 : if (m_graph) {
3563 : : // Inform the TxGraph about the Ref being destroyed.
3564 : 72073 : m_graph->UnlinkRef(m_index);
3565 : 72073 : m_graph = nullptr;
3566 : : }
3567 : 410992 : }
3568 : :
3569 : 74161 : TxGraph::Ref::Ref(Ref&& other) noexcept
3570 : : {
3571 : : // Inform the TxGraph of other that its Ref is being moved.
3572 [ + + ]: 74161 : if (other.m_graph) other.m_graph->UpdateRef(other.m_index, *this);
3573 : : // Actually move the contents.
3574 : 74161 : std::swap(m_graph, other.m_graph);
3575 : 74161 : std::swap(m_index, other.m_index);
3576 : 74161 : }
3577 : :
3578 : 1413 : std::unique_ptr<TxGraph> MakeTxGraph(
3579 : : unsigned max_cluster_count,
3580 : : uint64_t max_cluster_size,
3581 : : uint64_t acceptable_cost,
3582 : : const std::function<std::strong_ordering(const TxGraph::Ref&, const TxGraph::Ref&)>& fallback_order) noexcept
3583 : : {
3584 [ - + ]: 1413 : return std::make_unique<TxGraphImpl>(
3585 : : /*max_cluster_count=*/max_cluster_count,
3586 : : /*max_cluster_size=*/max_cluster_size,
3587 : : /*acceptable_cost=*/acceptable_cost,
3588 [ - + ]: 1413 : /*fallback_order=*/fallback_order);
3589 : : }
|