[ntuple] Renormalize field type names if needed in SetSchemaFromExisting

roottest/root/io/hadd/CMakeLists.txt

@@ -270,3 +270,14 @@ ROOTTEST_ADD_TEST(test_hadd_regr_20872_2
                   PASSREGEX "root://eospublic.cern.ch//eos/root-eos/h1/dstarmb.root cannot be both the target and an input!"
 )
 endif()
+
+# Verify that the hadd (or rather, the RNTupleMerger) is able to cope with RNTuples written before schema version
+# 1.0.0.1 (meaning they have the wrong type name normalization)
+configure_file(test_hadd_merge_rntuple_634_1.root . COPYONLY)
+configure_file(test_hadd_merge_rntuple_634_2.root . COPYONLY)
+configure_file(check_merge_rntuple_634.C . COPYONLY)
+ROOTTEST_ADD_TEST(test_hadd_merge_rntuple_634,
+                  PRECMD ${CMAKE_COMMAND} -E rm -f test_hadd_merge_rntuple_634_merged.root
+                  COMMAND ${ROOT_hadd_CMD} -f test_hadd_merge_rntuple_634_merged.root test_hadd_merge_rntuple_634_1.root test_hadd_merge_rntuple_634_2.root
+                  POSTCMD ${ROOT_root_CMD} -q check_merge_rntuple_634.C
+                  PASSRC 0)

roottest/root/io/hadd/check_merge_rntuple_634.C

@@ -0,0 +1,16 @@
+int check_merge_rntuple_634()
+{
+  // Verify that the given RNTuple is readable
+  auto reader = ROOT::RNTupleReader::Open("Events", "test_hadd_merge_rntuple_634_merged.root");
+  const auto &model = reader->GetModel();
+  const auto &desc = reader->GetDescriptor();
+  for (const auto &fdesc : desc.GetFieldIterable(desc.GetFieldZeroId()))  {
+    if (fdesc.GetTypeName() != ROOT::Internal::GetRenormalizedTypeName(fdesc.GetTypeName())) {
+      std::cerr << "Type name is not renormalized! " << fdesc.GetTypeName() << " vs " << 
+        ROOT::Internal::GetRenormalizedTypeName(fdesc.GetTypeName()) << "\n";
+      return 1;
+    }
+  }
+
+  return 0;
+}

tree/ntuple/inc/ROOT/RPageStorage.hxx

@@ -46,6 +46,12 @@ namespace ROOT {
 
 class RNTupleModel;
 
+namespace Experimental {
+namespace Internal {
+class RNTupleMerger;
+}
+} // namespace Experimental
+
 namespace Internal {
 
 class RPageAllocator;
@@ -445,6 +451,8 @@ public:
 */
 // clang-format on
 class RPagePersistentSink : public RPageSink {
+   friend class ROOT::Experimental::Internal::RNTupleMerger;
+
 private:
    /// Used to map the IDs of the descriptor to the physical IDs issued during header/footer serialization
    ROOT::Internal::RNTupleSerializer::RContext fSerializationContext;

tree/ntuple/src/RNTupleDescriptor.cxx

@@ -1331,6 +1331,23 @@ ROOT::RResult<void> ROOT::Internal::RNTupleDescriptorBuilder::AddClusterGroup(RC
 void ROOT::Internal::RNTupleDescriptorBuilder::SetSchemaFromExisting(const RNTupleDescriptor &descriptor)
 {
    fDescriptor = descriptor.CloneSchema();
+
+   if (descriptor.fVersionMajor == 0 && descriptor.fVersionMinor == 0 && descriptor.fVersionPatch < 1) {
+      // In case we are copying the schema from a pre-1.0.0.1 RNTuple we need to patch all field type names
+      // to use the proper normalization.
+      std::vector<ROOT::DescriptorId_t> toVisit;
+      toVisit.push_back(fDescriptor.GetFieldZeroId());
+      while (!toVisit.empty()) {
+         auto fieldId = toVisit.back();
+         toVisit.pop_back();
+         for (auto &field : fDescriptor.GetFieldIterable(fieldId)) {
+            auto prev = field.fTypeName;
+            const_cast<ROOT::RFieldDescriptor &>(field).fTypeName =
+               ROOT::Internal::GetRenormalizedTypeName(field.fTypeName);
+            toVisit.push_back(field.GetId());
+         }
+      }
+   }
 }
 
 void ROOT::Internal::RNTupleDescriptorBuilder::BeginHeaderExtension()

tree/ntuple/src/RNTupleMerger.cxx

@@ -530,7 +530,8 @@ CompareDescriptorStructure(const ROOT::RNTupleDescriptor &dst, const ROOT::RNTup
 
       // Require that fields types match
       // TODO(gparolini): allow non-identical but compatible types
-      const auto &srcTyName = field.fSrc->GetTypeName();
+      const auto &srcTyName = ROOT::Internal::GetRenormalizedTypeName(field.fSrc->GetTypeName());
+      // This is already renormalized by construction (see RNTupleDescriptorBuilder::SetSchemaFromExisting)
       const auto &dstTyName = field.fDst->GetTypeName();
       if (srcTyName != dstTyName) {
          std::stringstream ss;
@@ -1147,8 +1148,8 @@ static void AddColumnsFromField(std::vector<RColumnMergeInfo> &columns, const RO
       }
 
       // Since we disallow merging fields of different types, src and dstFieldDesc must have the same type name.
-      assert(srcFieldDesc.GetTypeName() == dstFieldDesc.GetTypeName());
-      info.fInMemoryType = ColumnInMemoryType(srcFieldDesc.GetTypeName(), info.fColumnType);
+      assert(srcDesc.GetTypeNameForComparison(srcFieldDesc) == dstFieldDesc.GetTypeName());
+      info.fInMemoryType = ColumnInMemoryType(dstFieldDesc.GetTypeName(), info.fColumnType);
       columns.emplace_back(info);
    }
 

tree/ntuple/src/RPageStorage.cxx

@@ -1109,6 +1109,8 @@ ROOT::Internal::RPagePersistentSink::InitFromDescriptor(const ROOT::RNTupleDescr
 {
    // Create new descriptor
    fDescriptorBuilder.SetSchemaFromExisting(srcDescriptor);
+   // This is needed to be able to use GetTypeNameForComparison()
+   fDescriptorBuilder.SetVersionForWriting();
    const auto &descriptor = fDescriptorBuilder.GetDescriptor();
 
    // Create column/page ranges