Merge 6480423d79 into c5e44a0435

src/bench/cluster_linearize.cpp

@@ -229,8 +229,8 @@ void BenchLinearizeOptimally(benchmark::Bench& bench, const std::array<uint8_t,
         reader >> Using<DepGraphFormatter>(depgraph);
         uint64_t rng_seed = 0;
         bench.run([&] {
-            auto res = Linearize(depgraph, /*max_iterations=*/10000000, rng_seed++);
+            auto [_lin, optimal, _cost] = Linearize(depgraph, /*max_iterations=*/10000000, rng_seed++);
-            assert(res.second);
+            assert(optimal);
         });
     };
 

src/cluster_linearize.h

@@ -1030,19 +1030,20 @@ public:
  *                                linearize.
  * @param[in] old_linearization   An existing linearization for the cluster (which must be
  *                                topologically valid), or empty.
- * @return                        A pair of:
+ * @return                        A tuple of:
  *                                - The resulting linearization. It is guaranteed to be at least as
  *                                  good (in the feerate diagram sense) as old_linearization.
  *                                - A boolean indicating whether the result is guaranteed to be
  *                                  optimal.
+ *                                - How many optimization steps were actually performed.
  *
  * Complexity: possibly O(N * min(max_iterations + N, sqrt(2^N))) where N=depgraph.TxCount().
  */
 template<typename SetType>
-std::pair<std::vector<DepGraphIndex>, bool> Linearize(const DepGraph<SetType>& depgraph, uint64_t max_iterations, uint64_t rng_seed, std::span<const DepGraphIndex> old_linearization = {}) noexcept
+std::tuple<std::vector<DepGraphIndex>, bool, uint64_t> Linearize(const DepGraph<SetType>& depgraph, uint64_t max_iterations, uint64_t rng_seed, std::span<const DepGraphIndex> old_linearization = {}) noexcept
 {
     Assume(old_linearization.empty() || old_linearization.size() == depgraph.TxCount());
-    if (depgraph.TxCount() == 0) return {{}, true};
+    if (depgraph.TxCount() == 0) return {{}, true, 0};
 
     uint64_t iterations_left = max_iterations;
     std::vector<DepGraphIndex> linearization;
@@ -1113,7 +1114,7 @@ std::pair<std::vector<DepGraphIndex>, bool> Linearize(const DepGraph<SetType>& d
         }
     }
 
-    return {std::move(linearization), optimal};
+    return {std::move(linearization), optimal, max_iterations - iterations_left};
 }
 
 /** Improve a given linearization.

src/test/fuzz/cluster_linearize.cpp

@@ -906,7 +906,8 @@ FUZZ_TARGET(clusterlin_linearize)
 
     // Invoke Linearize().
     iter_count &= 0x7ffff;
-    auto [linearization, optimal] = Linearize(depgraph, iter_count, rng_seed, old_linearization);
+    auto [linearization, optimal, cost] = Linearize(depgraph, iter_count, rng_seed, old_linearization);
+    assert(cost <= iter_count);
     SanityCheck(depgraph, linearization);
     auto chunking = ChunkLinearization(depgraph, linearization);
 
@@ -1090,7 +1091,7 @@ FUZZ_TARGET(clusterlin_postlinearize_tree)
 
     // Try to find an even better linearization directly. This must not change the diagram for the
     // same reason.
-    auto [opt_linearization, _optimal] = Linearize(depgraph_tree, 100000, rng_seed, post_linearization);
+    auto [opt_linearization, _optimal, _cost] = Linearize(depgraph_tree, 100000, rng_seed, post_linearization);
     auto opt_chunking = ChunkLinearization(depgraph_tree, opt_linearization);
     auto cmp_opt = CompareChunks(opt_chunking, post_chunking);
     assert(cmp_opt == 0);

src/test/fuzz/txgraph.cpp

@@ -11,6 +11,7 @@
 #include <util/feefrac.h>
 
 #include <algorithm>
+#include <iterator>
 #include <map>
 #include <memory>
 #include <set>
@@ -52,9 +53,17 @@ struct SimTxGraph
     std::optional<bool> oversized;
     /** The configured maximum number of transactions per cluster. */
     DepGraphIndex max_cluster_count;
+    /** Which transactions have been modified in the graph since creation, either directly or by
+     *  being in a cluster which includes modifications. Only relevant for the staging graph. */
+    SetType modified;
+    /** The configured maximum total size of transactions per cluster. */
+    uint64_t max_cluster_size;
+    /** Whether the corresponding real graph is known to be optimally linearized. */
+    bool real_is_optimal{false};
 
     /** Construct a new SimTxGraph with the specified maximum cluster count. */
-    explicit SimTxGraph(DepGraphIndex max_cluster) : max_cluster_count(max_cluster) {}
+    explicit SimTxGraph(DepGraphIndex max_cluster, uint64_t max_size) :
+        max_cluster_count(max_cluster), max_cluster_size(max_size) {}
 
     // Permit copying and moving.
     SimTxGraph(const SimTxGraph&) noexcept = default;
@@ -74,15 +83,33 @@ struct SimTxGraph
             while (todo.Any()) {
                 auto component = graph.FindConnectedComponent(todo);
                 if (component.Count() > max_cluster_count) oversized = true;
+                uint64_t component_size{0};
+                for (auto i : component) component_size += graph.FeeRate(i).size;
+                if (component_size > max_cluster_size) oversized = true;
                 todo -= component;
             }
         }
         return *oversized;
     }
 
+    void MakeModified(DepGraphIndex index)
+    {
+        modified |= graph.GetConnectedComponent(graph.Positions(), index);
+    }
+
     /** Determine the number of (non-removed) transactions in the graph. */
     DepGraphIndex GetTransactionCount() const { return graph.TxCount(); }
 
+    /** Get the sum of all fees/sizes in the graph. */
+    FeePerWeight SumAll() const
+    {
+        FeePerWeight ret;
+        for (auto i : graph.Positions()) {
+            ret += graph.FeeRate(i);
+        }
+        return ret;
+    }
+
     /** Get the position where ref occurs in this simulated graph, or -1 if it does not. */
     Pos Find(const TxGraph::Ref* ref) const
     {
@@ -104,10 +131,14 @@ struct SimTxGraph
     {
         assert(graph.TxCount() < MAX_TRANSACTIONS);
         auto simpos = graph.AddTransaction(feerate);
+        real_is_optimal = false;
+        MakeModified(simpos);
         assert(graph.Positions()[simpos]);
         simmap[simpos] = std::make_shared<TxGraph::Ref>();
         auto ptr = simmap[simpos].get();
         simrevmap[ptr] = simpos;
+        // This may invalidate our cached oversized value.
+        if (oversized.has_value() && !*oversized) oversized = std::nullopt;
         return ptr;
     }
 
@@ -119,6 +150,8 @@ struct SimTxGraph
         auto chl_pos = Find(child);
         if (chl_pos == MISSING) return;
         graph.AddDependencies(SetType::Singleton(par_pos), chl_pos);
+        MakeModified(par_pos);
+        real_is_optimal = false;
         // This may invalidate our cached oversized value.
         if (oversized.has_value() && !*oversized) oversized = std::nullopt;
     }
@@ -128,6 +161,8 @@ struct SimTxGraph
     {
         auto pos = Find(ref);
         if (pos == MISSING) return;
+        // No need to invoke MakeModified, because this equally affects main and staging.
+        real_is_optimal = false;
         graph.FeeRate(pos).fee = fee;
     }
 
@@ -136,6 +171,8 @@ struct SimTxGraph
     {
         auto pos = Find(ref);
         if (pos == MISSING) return;
+        MakeModified(pos);
+        real_is_optimal = false;
         graph.RemoveTransactions(SetType::Singleton(pos));
         simrevmap.erase(simmap[pos].get());
         // Retain the TxGraph::Ref corresponding to this position, so the Ref destruction isn't
@@ -160,7 +197,9 @@ struct SimTxGraph
             auto remove = std::partition(removed.begin(), removed.end(), [&](auto& arg) { return arg.get() != ref; });
             removed.erase(remove, removed.end());
         } else {
+            MakeModified(pos);
             graph.RemoveTransactions(SetType::Singleton(pos));
+            real_is_optimal = false;
             simrevmap.erase(simmap[pos].get());
             simmap[pos].reset();
             // This may invalidate our cached oversized value.
@@ -238,12 +277,34 @@ FUZZ_TARGET(txgraph)
 
     // Decide the maximum number of transactions per cluster we will use in this simulation.
     auto max_count = provider.ConsumeIntegralInRange<DepGraphIndex>(1, MAX_CLUSTER_COUNT_LIMIT);
+    // And the maximum combined size of transactions per cluster.
+    auto max_size = provider.ConsumeIntegralInRange<uint64_t>(1, 0x3fffff * MAX_CLUSTER_COUNT_LIMIT);
+    // And the number of iterations to consider a cluster acceptably linearized.
+    auto acceptable_iters = provider.ConsumeIntegralInRange<uint64_t>(0, 10000);
 
     // Construct a real graph, and a vector of simulated graphs (main, and possibly staging).
-    auto real = MakeTxGraph(max_count);
+    auto real = MakeTxGraph(max_count, max_size, acceptable_iters);
     std::vector<SimTxGraph> sims;
     sims.reserve(2);
-    sims.emplace_back(max_count);
+    sims.emplace_back(max_count, max_size);
+
+    /** Struct encapsulating information about a BlockBuilder that's currently live. */
+    struct BlockBuilderData
+    {
+        /** BlockBuilder object from real. */
+        std::unique_ptr<TxGraph::BlockBuilder> builder;
+        /** The set of transactions marked as included in *builder. */
+        SimTxGraph::SetType included;
+        /** The set of transactions marked as included or skipped in *builder. */
+        SimTxGraph::SetType done;
+        /** The last chunk feerate returned by *builder. IsEmpty() if none yet. */
+        FeePerWeight last_feerate;
+
+        BlockBuilderData(std::unique_ptr<TxGraph::BlockBuilder> builder_in) : builder(std::move(builder_in)) {}
+    };
+
+    /** Currently active block builders. */
+    std::vector<BlockBuilderData> block_builders;
 
     /** Function to pick any Ref (for either sim in sims: from sim.simmap or sim.removed, or the
      *  empty Ref). */
@@ -282,9 +343,44 @@ FUZZ_TARGET(txgraph)
         return &empty_ref;
     };
 
+    /** Function to construct the correct fee-size diagram a real graph has based on its graph
+     *  order (as reported by GetCluster(), so it works for both main and staging). */
+    auto get_diagram_fn = [&](bool main_only) -> std::vector<FeeFrac> {
+        int level = main_only ? 0 : sims.size() - 1;
+        auto& sim = sims[level];
+        // For every transaction in the graph, request its cluster, and throw them into a set.
+        std::set<std::vector<TxGraph::Ref*>> clusters;
+        for (auto i : sim.graph.Positions()) {
+            auto ref = sim.GetRef(i);
+            clusters.insert(real->GetCluster(*ref, main_only));
+        }
+        // Compute the chunkings of each (deduplicated) cluster.
+        size_t num_tx{0};
+        std::vector<FeeFrac> chunk_feerates;
+        for (const auto& cluster : clusters) {
+            num_tx += cluster.size();
+            std::vector<SimTxGraph::Pos> linearization;
+            linearization.reserve(cluster.size());
+            for (auto refptr : cluster) linearization.push_back(sim.Find(refptr));
+            for (const FeeFrac& chunk_feerate : ChunkLinearization(sim.graph, linearization)) {
+                chunk_feerates.push_back(chunk_feerate);
+            }
+        }
+        // Verify the number of transactions after deduplicating clusters. This implicitly verifies
+        // that GetCluster on each element of a cluster reports the cluster transactions in the same
+        // order.
+        assert(num_tx == sim.GetTransactionCount());
+        // Sort by feerate only, since violating topological constraints within same-feerate
+        // chunks won't affect diagram comparisons.
+        std::sort(chunk_feerates.begin(), chunk_feerates.end(), std::greater{});
+        return chunk_feerates;
+    };
+
     LIMITED_WHILE(provider.remaining_bytes() > 0, 200) {
         // Read a one-byte command.
         int command = provider.ConsumeIntegral<uint8_t>();
+        int orig_command = command;
+
         // Treat the lowest bit of a command as a flag (which selects a variant of some of the
         // operations), and the second-lowest bit as a way of selecting main vs. staging, and leave
         // the rest of the bits in command.
@@ -292,6 +388,11 @@ FUZZ_TARGET(txgraph)
         bool use_main = command & 2;
         command >>= 2;
 
+        /** Use the bottom 2 bits of command to select an entry in the block_builders vector (if
+         *  any). These use the same bits as alt/use_main, so don't use those in actions below
+         *  where builder_idx is used as well. */
+        int builder_idx = block_builders.empty() ? -1 : int((orig_command & 3) % block_builders.size());
+
         // Provide convenient aliases for the top simulated graph (main, or staging if it exists),
         // one for the simulated graph selected based on use_main (for operations that can operate
         // on both graphs), and one that always refers to the main graph.
@@ -302,7 +403,7 @@ FUZZ_TARGET(txgraph)
         // Keep decrementing command for each applicable operation, until one is hit. Multiple
         // iterations may be necessary.
         while (true) {
-            if (top_sim.GetTransactionCount() < SimTxGraph::MAX_TRANSACTIONS && command-- == 0) {
+            if ((block_builders.empty() || sims.size() > 1) && top_sim.GetTransactionCount() < SimTxGraph::MAX_TRANSACTIONS && command-- == 0) {
                 // AddTransaction.
                 int64_t fee;
                 int32_t size;
@@ -314,7 +415,7 @@ FUZZ_TARGET(txgraph)
                     // Otherwise, use smaller range which consume fewer fuzz input bytes, as just
                     // these are likely sufficient to trigger all interesting code paths already.
                     fee = provider.ConsumeIntegral<uint8_t>();
-                    size = provider.ConsumeIntegral<uint8_t>() + 1;
+                    size = provider.ConsumeIntegralInRange<uint32_t>(1, 0xff);
                 }
                 FeePerWeight feerate{fee, size};
                 // Create a real TxGraph::Ref.
@@ -324,7 +425,7 @@ FUZZ_TARGET(txgraph)
                 // Move it in place.
                 *ref_loc = std::move(ref);
                 break;
-            } else if (top_sim.GetTransactionCount() + top_sim.removed.size() > 1 && command-- == 0) {
+            } else if ((block_builders.empty() || sims.size() > 1) && top_sim.GetTransactionCount() + top_sim.removed.size() > 1 && command-- == 0) {
                 // AddDependency.
                 auto par = pick_fn();
                 auto chl = pick_fn();
@@ -336,9 +437,10 @@ FUZZ_TARGET(txgraph)
                     if (top_sim.graph.Ancestors(pos_par)[pos_chl]) break;
                 }
                 top_sim.AddDependency(par, chl);
+                top_sim.real_is_optimal = false;
                 real->AddDependency(*par, *chl);
                 break;
-            } else if (top_sim.removed.size() < 100 && command-- == 0) {
+            } else if ((block_builders.empty() || sims.size() > 1) && top_sim.removed.size() < 100 && command-- == 0) {
                 // RemoveTransaction. Either all its ancestors or all its descendants are also
                 // removed (if any), to make sure TxGraph's reordering of removals and dependencies
                 // has no effect.
@@ -368,7 +470,7 @@ FUZZ_TARGET(txgraph)
                 }
                 sel_sim.removed.pop_back();
                 break;
-            } else if (command-- == 0) {
+            } else if (block_builders.empty() && command-- == 0) {
                 // ~Ref (of any transaction).
                 std::vector<TxGraph::Ref*> to_destroy;
                 to_destroy.push_back(pick_fn());
@@ -390,7 +492,7 @@ FUZZ_TARGET(txgraph)
                     }
                 }
                 break;
-            } else if (command-- == 0) {
+            } else if (block_builders.empty() && command-- == 0) {
                 // SetTransactionFee.
                 int64_t fee;
                 if (alt) {
@@ -444,6 +546,7 @@ FUZZ_TARGET(txgraph)
                     // Just do some quick checks that the reported value is in range. A full
                     // recomputation of expected chunk feerates is done at the end.
                     assert(feerate.size >= main_sim.graph.FeeRate(simpos).size);
+                    assert(feerate.size <= main_sim.SumAll().size);
                 }
                 break;
             } else if (!sel_sim.IsOversized() && command-- == 0) {
@@ -487,13 +590,17 @@ FUZZ_TARGET(txgraph)
                 assert(result.size() <= max_count);
                 // Require the result to be topologically valid and not contain duplicates.
                 auto left = sel_sim.graph.Positions();
+                uint64_t total_size{0};
                 for (auto refptr : result) {
                     auto simpos = sel_sim.Find(refptr);
+                    total_size += sel_sim.graph.FeeRate(simpos).size;
                     assert(simpos != SimTxGraph::MISSING);
                     assert(left[simpos]);
                     left.Reset(simpos);
                     assert(!sel_sim.graph.Ancestors(simpos).Overlaps(left));
                 }
+                // Check cluster size limit.
+                assert(total_size <= max_size);
                 // Require the set to be connected.
                 auto result_set = sel_sim.MakeSet(result);
                 assert(sel_sim.graph.IsConnected(result_set));
@@ -517,9 +624,10 @@ FUZZ_TARGET(txgraph)
             } else if (sims.size() < 2 && command-- == 0) {
                 // StartStaging.
                 sims.emplace_back(sims.back());
+                sims.back().modified = SimTxGraph::SetType{};
                 real->StartStaging();
                 break;
-            } else if (sims.size() > 1 && command-- == 0) {
+            } else if (block_builders.empty() && sims.size() > 1 && command-- == 0) {
                 // CommitStaging.
                 real->CommitStaging();
                 sims.erase(sims.begin());
@@ -584,7 +692,135 @@ FUZZ_TARGET(txgraph)
                 break;
             } else if (command-- == 0) {
                 // DoWork.
-                real->DoWork();
+                uint64_t iters = provider.ConsumeIntegralInRange<uint64_t>(0, alt ? 10000 : 255);
+                if (real->DoWork(iters)) {
+                    for (auto& sim : sims) sim.real_is_optimal = true;
+                }
+                break;
+            } else if (sims.size() == 2 && !sims[0].IsOversized() && !sims[1].IsOversized() && command-- == 0) {
+                // GetMainStagingDiagrams()
+                auto [main_diagram, staged_diagram] = real->GetMainStagingDiagrams();
+                auto sum_main = std::accumulate(main_diagram.begin(), main_diagram.end(), FeeFrac{});
+                auto sum_staged = std::accumulate(staged_diagram.begin(), staged_diagram.end(), FeeFrac{});
+                auto diagram_gain = sum_staged - sum_main;
+                auto real_gain = sims[1].SumAll() - sims[0].SumAll();
+                // Just check that the total fee gained/lost and size gained/lost according to the
+                // diagram matches the difference in these values in the simulated graph. A more
+                // complete check of the GetMainStagingDiagrams result is performed at the end.
+                assert(diagram_gain == real_gain);
+                // Check that the feerates in each diagram are monotonically decreasing.
+                for (size_t i = 1; i < main_diagram.size(); ++i) {
+                    assert(FeeRateCompare(main_diagram[i], main_diagram[i - 1]) <= 0);
+                }
+                for (size_t i = 1; i < staged_diagram.size(); ++i) {
+                    assert(FeeRateCompare(staged_diagram[i], staged_diagram[i - 1]) <= 0);
+                }
+                break;
+            } else if (block_builders.size() < 4 && !main_sim.IsOversized() && command-- == 0) {
+                // GetBlockBuilder.
+                block_builders.emplace_back(real->GetBlockBuilder());
+                break;
+            } else if (!block_builders.empty() && command-- == 0) {
+                // ~BlockBuilder.
+                block_builders.erase(block_builders.begin() + builder_idx);
+                break;
+            } else if (!block_builders.empty() && command-- == 0) {
+                // BlockBuilder::GetCurrentChunk, followed by Include/Skip.
+                auto& builder_data = block_builders[builder_idx];
+                auto new_included = builder_data.included;
+                auto new_done = builder_data.done;
+                auto chunk = builder_data.builder->GetCurrentChunk();
+                if (chunk) {
+                    // Chunk feerates must be monotonously decreasing.
+                    if (!builder_data.last_feerate.IsEmpty()) {
+                        assert(!(chunk->second >> builder_data.last_feerate));
+                    }
+                    builder_data.last_feerate = chunk->second;
+                    // Verify the contents of GetCurrentChunk.
+                    FeePerWeight sum_feerate;
+                    for (TxGraph::Ref* ref : chunk->first) {
+                        // Each transaction in the chunk must exist in the main graph.
+                        auto simpos = main_sim.Find(ref);
+                        assert(simpos != SimTxGraph::MISSING);
+                        // Verify the claimed chunk feerate.
+                        sum_feerate += main_sim.graph.FeeRate(simpos);
+                        // Make sure no transaction is reported twice.
+                        assert(!new_done[simpos]);
+                        new_done.Set(simpos);
+                        // The concatenation of all included transactions must be topologically valid.
+                        new_included.Set(simpos);
+                        assert(main_sim.graph.Ancestors(simpos).IsSubsetOf(new_included));
+                    }
+                    assert(sum_feerate == chunk->second);
+                } else {
+                    // When we reach the end, if nothing was skipped, the entire graph should have
+                    // been reported.
+                    if (builder_data.done == builder_data.included) {
+                        assert(builder_data.done.Count() == main_sim.GetTransactionCount());
+                    }
+                }
+                // Possibly invoke GetCurrentChunk() again, which should give the same result.
+                if ((orig_command % 7) >= 5) {
+                    auto chunk2 = builder_data.builder->GetCurrentChunk();
+                    assert(chunk == chunk2);
+                }
+                // Skip or include.
+                if ((orig_command % 5) >= 3) {
+                    // Skip.
+                    builder_data.builder->Skip();
+                } else {
+                    // Include.
+                    builder_data.builder->Include();
+                    builder_data.included = new_included;
+                }
+                builder_data.done = new_done;
+                break;
+            } else if (!main_sim.IsOversized() && command-- == 0) {
+                // GetWorstMainChunk.
+                auto [worst_chunk, worst_chunk_feerate] = real->GetWorstMainChunk();
+                // Just do some sanity checks here. Consistency with GetBlockBuilder is checked
+                // below.
+                if (main_sim.GetTransactionCount() == 0) {
+                    assert(worst_chunk.empty());
+                    assert(worst_chunk_feerate.IsEmpty());
+                } else {
+                    assert(!worst_chunk.empty());
+                    SimTxGraph::SetType done;
+                    FeePerWeight sum;
+                    for (TxGraph::Ref* ref : worst_chunk) {
+                        // Each transaction in the chunk must exist in the main graph.
+                        auto simpos = main_sim.Find(ref);
+                        assert(simpos != SimTxGraph::MISSING);
+                        sum += main_sim.graph.FeeRate(simpos);
+                        // Make sure the chunk contains no duplicate transactions.
+                        assert(!done[simpos]);
+                        done.Set(simpos);
+                        // All elements are preceded by all their descendants.
+                        assert(main_sim.graph.Descendants(simpos).IsSubsetOf(done));
+                    }
+                    assert(sum == worst_chunk_feerate);
+                }
+                break;
+            } else if ((block_builders.empty() || sims.size() > 1) && command-- == 0) {
+                // Trim.
+                bool was_oversized = top_sim.IsOversized();
+                auto removed = real->Trim();
+                if (!was_oversized) {
+                    assert(removed.empty());
+                    break;
+                }
+                auto removed_set = top_sim.MakeSet(removed);
+                // The removed set must contain all its own descendants.
+                for (auto simpos : removed_set) {
+                    assert(top_sim.graph.Descendants(simpos).IsSubsetOf(removed_set));
+                }
+                // Apply all removals to the simulation, and verify the result is no longer
+                // oversized. Don't query the real graph for oversizedness; it is compared
+                // against the simulation anyway later.
+                for (auto simpos : removed_set) {
+                    top_sim.RemoveTransaction(top_sim.GetRef(simpos));
+                }
+                assert(!top_sim.IsOversized());
                 break;
             }
         }
@@ -624,6 +860,16 @@ FUZZ_TARGET(txgraph)
         }
         assert(todo.None());
 
+        // If the real graph claims to be optimal (the last DoWork() call returned true), verify
+        // that calling Linearize on it does not improve it further.
+        if (sims[0].real_is_optimal) {
+            auto real_diagram = ChunkLinearization(sims[0].graph, vec1);
+            auto [sim_lin, _optimal, _cost] = Linearize(sims[0].graph, 300000, rng.rand64(), vec1);
+            auto sim_diagram = ChunkLinearization(sims[0].graph, sim_lin);
+            auto cmp = CompareChunks(real_diagram, sim_diagram);
+            assert(cmp == 0);
+        }
+
         // For every transaction in the total ordering, find a random one before it and after it,
         // and compare their chunk feerates, which must be consistent with the ordering.
         for (size_t pos = 0; pos < vec1.size(); ++pos) {
@@ -639,6 +885,94 @@ FUZZ_TARGET(txgraph)
                 assert(FeeRateCompare(after_feerate, pos_feerate) <= 0);
             }
         }
+
+        // The same order should be obtained through a BlockBuilder as implied by CompareMainOrder,
+        // if nothing is skipped.
+        auto builder = real->GetBlockBuilder();
+        std::vector<SimTxGraph::Pos> vec_builder;
+        std::vector<TxGraph::Ref*> last_chunk;
+        FeePerWeight last_chunk_feerate;
+        while (auto chunk = builder->GetCurrentChunk()) {
+            FeePerWeight sum;
+            for (TxGraph::Ref* ref : chunk->first) {
+                // The reported chunk feerate must match the chunk feerate obtained by asking
+                // it for each of the chunk's transactions individually.
+                assert(real->GetMainChunkFeerate(*ref) == chunk->second);
+                // Verify the chunk feerate matches the sum of the reported individual feerates.
+                sum += real->GetIndividualFeerate(*ref);
+                // Chunks must contain transactions that exist in the graph.
+                auto simpos = sims[0].Find(ref);
+                assert(simpos != SimTxGraph::MISSING);
+                vec_builder.push_back(simpos);
+            }
+            assert(sum == chunk->second);
+            last_chunk = std::move(chunk->first);
+            last_chunk_feerate = chunk->second;
+            builder->Include();
+        }
+        assert(vec_builder == vec1);
+
+        // The last chunk returned by the BlockBuilder must match GetWorstMainChunk, in reverse.
+        std::reverse(last_chunk.begin(), last_chunk.end());
+        auto [worst_chunk, worst_chunk_feerate] = real->GetWorstMainChunk();
+        assert(last_chunk == worst_chunk);
+        assert(last_chunk_feerate == worst_chunk_feerate);
+
+        // Check that the implied ordering gives rise to a combined diagram that matches the
+        // diagram constructed from the individual cluster linearization chunkings.
+        auto main_real_diagram = get_diagram_fn(/*main_only=*/true);
+        auto main_implied_diagram = ChunkLinearization(sims[0].graph, vec1);
+        assert(CompareChunks(main_real_diagram, main_implied_diagram) == 0);
+
+        if (sims.size() >= 2 && !sims[1].IsOversized()) {
+            // When the staging graph is not oversized as well, call GetMainStagingDiagrams, and
+            // fully verify the result.
+            auto [main_cmp_diagram, stage_cmp_diagram] = real->GetMainStagingDiagrams();
+            // Check that the feerates in each diagram are monotonically decreasing.
+            for (size_t i = 1; i < main_cmp_diagram.size(); ++i) {
+                assert(FeeRateCompare(main_cmp_diagram[i], main_cmp_diagram[i - 1]) <= 0);
+            }
+            for (size_t i = 1; i < stage_cmp_diagram.size(); ++i) {
+                assert(FeeRateCompare(stage_cmp_diagram[i], stage_cmp_diagram[i - 1]) <= 0);
+            }
+            // Treat the diagrams as sets of chunk feerates, and sort them in the same way so that
+            // std::set_difference can be used on them below. The exact ordering does not matter
+            // here, but it has to be consistent with the one used in main_diagram and
+            // stage_diagram).
+            std::sort(main_cmp_diagram.begin(), main_cmp_diagram.end(), std::greater{});
+            std::sort(stage_cmp_diagram.begin(), stage_cmp_diagram.end(), std::greater{});
+            // Find the chunks that appear in main_diagram but are missing from main_cmp_diagram.
+            // This is allowed, because GetMainStagingDiagrams omits clusters in main unaffected
+            // by staging.
+            std::vector<FeeFrac> missing_main_cmp;
+            std::set_difference(main_real_diagram.begin(), main_real_diagram.end(),
+                                main_cmp_diagram.begin(), main_cmp_diagram.end(),
+                                std::inserter(missing_main_cmp, missing_main_cmp.end()),
+                                std::greater{});
+            assert(main_cmp_diagram.size() + missing_main_cmp.size() == main_real_diagram.size());
+            // Do the same for chunks in stage_diagram missing from stage_cmp_diagram.
+            auto stage_real_diagram = get_diagram_fn(false);
+            std::vector<FeeFrac> missing_stage_cmp;
+            std::set_difference(stage_real_diagram.begin(), stage_real_diagram.end(),
+                                stage_cmp_diagram.begin(), stage_cmp_diagram.end(),
+                                std::inserter(missing_stage_cmp, missing_stage_cmp.end()),
+                                std::greater{});
+            assert(stage_cmp_diagram.size() + missing_stage_cmp.size() == stage_real_diagram.size());
+            // The missing chunks must be equal across main & staging (otherwise they couldn't have
+            // been omitted).
+            assert(missing_main_cmp == missing_stage_cmp);
+
+            // The missing part must include at least all transactions in staging which have not been
+            // modified, or been in a cluster together with modified transactions, since they were
+            // copied from main. Note that due to the reordering of removals w.r.t. dependency
+            // additions, it is possible that the real implementation found more unaffected things.
+            FeeFrac missing_real;
+            for (const auto& feerate : missing_main_cmp) missing_real += feerate;
+            FeeFrac missing_expected = sims[1].graph.FeeRate(sims[1].graph.Positions() - sims[1].modified);
+            // Note that missing_real.fee < missing_expected.fee is possible to due the presence of
+            // negative-fee transactions.
+            assert(missing_real.size >= missing_expected.size);
+        }
     }
 
     assert(real->HaveStaging() == (sims.size() > 1));
@@ -680,13 +1014,17 @@ FUZZ_TARGET(txgraph)
                     // linearization).
                     std::vector<DepGraphIndex> simlin;
                     SimTxGraph::SetType done;
+                    uint64_t total_size{0};
                     for (TxGraph::Ref* ptr : cluster) {
                         auto simpos = sim.Find(ptr);
                         assert(sim.graph.Descendants(simpos).IsSubsetOf(component - done));
                         done.Set(simpos);
                         assert(sim.graph.Ancestors(simpos).IsSubsetOf(done));
                         simlin.push_back(simpos);
+                        total_size += sim.graph.FeeRate(simpos).size;
                     }
+                    // Check cluster size.
+                    assert(total_size <= max_size);
                     // Construct a chunking object for the simulated graph, using the reported cluster
                     // linearization as ordering, and compare it against the reported chunk feerates.
                     if (sims.size() == 1 || main_only) {
@@ -714,6 +1052,8 @@ FUZZ_TARGET(txgraph)
     // Sanity check again (because invoking inspectors may modify internal unobservable state).
     real->SanityCheck();
 
+    // Kill the block builders.
+    block_builders.clear();
     // Kill the TxGraph object.
     real.reset();
     // Kill the simulated graphs, with all remaining Refs in it. If any, this verifies that Refs

src/txgraph.h

@@ -3,9 +3,11 @@
 // file COPYING or http://www.opensource.org/licenses/mit-license.php.
 
 #include <compare>
-#include <stdint.h>
 #include <memory>
+#include <optional>
+#include <stdint.h>
 #include <vector>
+#include <utility>
 
 #include <util/feefrac.h>
 
@@ -61,10 +63,10 @@ public:
     /** Virtual destructor, so inheriting is safe. */
     virtual ~TxGraph() = default;
     /** Construct a new transaction with the specified feerate, and return a Ref to it.
-     *  If a staging graph exists, the new transaction is only created there. In all
+     *  If a staging graph exists, the new transaction is only created there. In all further
-     *  further calls, only Refs created by AddTransaction() are allowed to be passed to this
+     *  calls, only Refs created by AddTransaction() are allowed to be passed to this TxGraph
-     *  TxGraph object (or empty Ref objects). Ref objects may outlive the TxGraph they were
+     *  object (or empty Ref objects). Ref objects may outlive the TxGraph they were created
-     *  created for. */
+     *  for. */
     [[nodiscard]] virtual Ref AddTransaction(const FeePerWeight& feerate) noexcept = 0;
     /** Remove the specified transaction. If a staging graph exists, the removal only happens
      *  there. This is a no-op if the transaction was already removed.
@@ -92,9 +94,10 @@ public:
     virtual void SetTransactionFee(const Ref& arg, int64_t fee) noexcept = 0;
 
     /** TxGraph is internally lazy, and will not compute many things until they are needed.
-     *  Calling DoWork will compute everything now, so that future operations are fast. This can be
+     *  Calling DoWork will perform some work now (controlled by iters) so that future operations
-     *  invoked while oversized. */
+     *  are fast, if there is any. Returns whether all work is done. This can be invoked while
-    virtual void DoWork() noexcept = 0;
+     *  oversized. */
+    virtual bool DoWork(uint64_t iters) noexcept = 0;
 
     /** Create a staging graph (which cannot exist already). This acts as if a full copy of
      *  the transaction graph is made, upon which further modifications are made. This copy can
@@ -162,6 +165,44 @@ public:
      *  main clusters are counted. Refs that do not exist in the queried graph are ignored. Refs
      *  can not be null. The queried graph must not be oversized. */
     virtual GraphIndex CountDistinctClusters(std::span<const Ref* const>, bool main_only = false) noexcept = 0;
+    /** For both main and staging (which must both exist and not be oversized), return the combined
+     *  respective feerate diagrams, including chunks from all clusters, but excluding clusters
+     *  that appear identically in both. Use FeeFrac rather than FeePerWeight so CompareChunks is
+     *  usable without type-conversion. */
+    virtual std::pair<std::vector<FeeFrac>, std::vector<FeeFrac>> GetMainStagingDiagrams() noexcept = 0;
+    /** Remove transactions (including their own descendants) according to a fast but best-effort
+     *  strategy such that the TxGraph's cluster and size limits are respected. Applies to staging
+     *  if it exists, and to main otherwise. Returns the list of all removed transactions in
+     *  unspecified order. This has no effect unless the relevant graph is oversized. */
+    virtual std::vector<Ref*> Trim() noexcept = 0;
+
+    /** Interface returned by GetBlockBuilder. */
+    class BlockBuilder
+    {
+    protected:
+        /** Make constructor non-public (use TxGraph::GetBlockBuilder()). */
+        BlockBuilder() noexcept = default;
+    public:
+        /** Support safe inheritance. */
+        virtual ~BlockBuilder() = default;
+        /** Get the chunk that is currently suggested to be included, plus its feerate, if any. */
+        virtual std::optional<std::pair<std::vector<Ref*>, FeePerWeight>> GetCurrentChunk() noexcept = 0;
+        /** Mark the current chunk as included, and progress to the next one. */
+        virtual void Include() noexcept = 0;
+        /** Mark the current chunk as skipped, and progress to the next one. Further chunks from
+         *  the same cluster as the current one will not be reported anymore. */
+        virtual void Skip() noexcept = 0;
+    };
+
+    /** Construct a block builder, drawing chunks in order, from the main graph, which cannot be
+     *  oversized. While the returned object exists, no mutators on the main graph are allowed.
+     *  The BlockBuilder object must not outlive the TxGraph it was created with. */
+    virtual std::unique_ptr<BlockBuilder> GetBlockBuilder() noexcept = 0;
+    /** Get the last chunk in the main graph, i.e., the last chunk that would be returned by a
+     *  BlockBuilder created now, together with its feerate. The chunk is returned in
+     *  reverse-topological order, so every element is preceded by all its descendants. The main
+     *  graph must not be oversized. If the graph is empty, {{}, FeePerWeight{}} is returned. */
+    virtual std::pair<std::vector<Ref*>, FeePerWeight> GetWorstMainChunk() noexcept = 0;
 
     /** Perform an internal consistency check on this object. */
     virtual void SanityCheck() const = 0;
@@ -205,8 +246,10 @@ public:
     };
 };
 
-/** Construct a new TxGraph with the specified limit on transactions within a cluster. That
+/** Construct a new TxGraph with the specified limit on transactions within a cluster, and the
- *  number cannot exceed MAX_CLUSTER_COUNT_LIMIT. */
+ *  specified limit on the sum of transaction sizes within a cluster. max_cluster_count cannot
-std::unique_ptr<TxGraph> MakeTxGraph(unsigned max_cluster_count) noexcept;
+ *  exceed MAX_CLUSTER_COUNT_LIMIT. acceptable_iters controls how many linearization optimization
+ *  steps will be performed before it is considered to be of acceptable quality. */
+std::unique_ptr<TxGraph> MakeTxGraph(unsigned max_cluster_count, uint64_t max_cluster_size, uint64_t acceptable_iters) noexcept;
 
 #endif // BITCOIN_TXGRAPH_H