[ntuple] Add slow merging strategy

tree/ntuple/inc/ROOT/REntry.hxx

@@ -38,6 +38,7 @@ class RNTupleAttrSetReader;
 
 namespace Internal {
 struct RNTupleAttrEntry;
+class RNTupleMergeStrategySlow;
 }
 } // namespace Experimental
 
@@ -56,6 +57,7 @@ class REntry {
    friend class RNTupleModel;
    friend class RNTupleReader;
    friend class Experimental::RNTupleAttrSetReader;
+   friend class Experimental::Internal::RNTupleMergeStrategySlow;
    friend struct Experimental::Internal::RNTupleAttrEntry;
 
 private:

tree/ntuple/inc/ROOT/RFieldBase.hxx

@@ -64,6 +64,7 @@ void CallConnectPageSourceOnField(RFieldBase &, ROOT::Internal::RPageSource &);
 ROOT::RResult<std::unique_ptr<ROOT::RFieldBase>>
 CallFieldBaseCreate(const std::string &fieldName, const std::string &typeName, const ROOT::RCreateFieldOptions &options,
                     const ROOT::RNTupleDescriptor *desc, ROOT::DescriptorId_t fieldId);
+void CallSetArtificialOnField(RFieldBase &);
 
 } // namespace Internal
 
@@ -100,6 +101,7 @@ class RFieldBase {
    Internal::CallFieldBaseCreate(const std::string &fieldName, const std::string &typeName,
                                  const ROOT::RCreateFieldOptions &options, const ROOT::RNTupleDescriptor *desc,
                                  ROOT::DescriptorId_t fieldId);
+   friend void Internal::CallSetArtificialOnField(RFieldBase &);
 
    using ReadCallback_t = std::function<void(void *)>;
 

tree/ntuple/inc/ROOT/RNTupleMerger.hxx

@@ -27,6 +27,7 @@
 
 namespace ROOT {
 
+class REntry;
 class RNTuple;
 
 namespace Internal {
@@ -35,6 +36,10 @@ class RPageSource;
 
 namespace Experimental::Internal {
 
+class RNTupleMergeStrategySlow;
+struct RNTupleSlowMergeData;
+struct RDescriptorsComparison;
+
 enum class ENTupleMergingMode {
    /// The merger will discard all columns that aren't present in the prototype model (i.e. the model of the first
    /// source); also all subsequent RNTuples must contain at least all the columns that are present in the prototype
@@ -95,6 +100,54 @@ struct RNTupleMergeOptions {
    bool fExtraVerbose = false;
 };
 
+/// Information regarding the current slow merge process that gets passed to the RNTupleMergeActions
+struct RNTupleMergeActionContext {
+   ROOT::NTupleSize_t fSourceEntryIndex = 0;
+   ROOT::NTupleSize_t fDestinationEntryIndex = 0;
+};
+
+enum class ENTupleMergeActionResult {
+   // Don't emit this entry
+   kDrop,
+   // Emit this entry, then continue
+   kEmit,
+   // Emit this entry, then call the function again
+   kEmitMulti,
+};
+
+// clang-format off
+/**
+ * \class ROOT::Experimental::Internal::RNTupleMergeActions
+ * \ingroup NTuple
+ * \brief Container for callbacks that are invoked during the "slow merge" process.
+ *        This allows customizing how the merging happens entry-by-entry, enabling things like dropping individual
+ *        entries from the destination RNTuple, modifying their values on the fly or creating new entries artificially.
+ */
+// clang-format on
+class RNTupleMergeActions {
+public:
+   RNTupleMergeActions() = default;
+   virtual ~RNTupleMergeActions() = default;
+
+   RNTupleMergeActions(const RNTupleMergeActions &) = delete;
+   RNTupleMergeActions &operator=(const RNTupleMergeActions &) = delete;
+   RNTupleMergeActions(RNTupleMergeActions &&) = default;
+   RNTupleMergeActions &operator=(RNTupleMergeActions &&) = default;
+
+   /// Invoked every time an entry is merged from a source RNTuple to the destination RNTuple.
+   virtual ENTupleMergeActionResult OnFill(ROOT::REntry &, const RNTupleMergeActionContext &)
+   {
+      return ENTupleMergeActionResult::kEmit;
+   }
+
+   /// Invoked at the end of the merge process, right before committing the dataset.
+   /// This allows appending more entries at the end of the merge.
+   virtual ENTupleMergeActionResult BeforeCommit(ROOT::REntry &, const RNTupleMergeActionContext &)
+   {
+      return ENTupleMergeActionResult::kDrop;
+   }
+};
+
 // clang-format off
 /**
  * \class ROOT::Experimental::Internal::RNTupleMerger

tree/ntuple/src/RFieldBase.cxx

@@ -85,6 +85,10 @@ void ROOT::Internal::CallConnectPageSourceOnField(RFieldBase &field, ROOT::Inter
 {
    field.ConnectPageSource(source);
 }
+void ROOT::Internal::CallSetArtificialOnField(RFieldBase &field)
+{
+   field.SetArtificial();
+}
 
 ROOT::RResult<std::unique_ptr<ROOT::RFieldBase>>
 ROOT::Internal::CallFieldBaseCreate(const std::string &fieldName, const std::string &typeName,