[ntuple] Initial minimal support for SoA I/O

core/clingutils/res/TClingUtils.h

@@ -203,6 +203,7 @@ class AnnotatedRecordDecl {
    bool fRequestOnlyTClass;
    int  fRequestedVersionNumber;
    int  fRequestedRNTupleSerializationMode;
+   std::string fRequestedRNTupleSoARecord;
    // clang-format on
 
 public:
@@ -228,6 +229,7 @@ class AnnotatedRecordDecl {
                        bool rRequestOnlyTClass,
                        int rRequestedVersionNumber,
                        int rRequestedRNTupleSerializationMode,
+                       const std::string &rRequestedRNTupleSoARecord,
                        const cling::Interpreter &interpret,
                        const TNormalizedCtxt &normCtxt);
 
@@ -240,6 +242,7 @@ class AnnotatedRecordDecl {
                        bool rRequestOnlyTClass,
                        int rRequestedVersionNumber,
                        int rRequestedRNTupleSerializationMode,
+                       const std::string &rRequestedRNTupleSoARecord,
                        const cling::Interpreter &interpret,
                        const TNormalizedCtxt &normCtxt);
 
@@ -253,6 +256,7 @@ class AnnotatedRecordDecl {
                        bool rRequestOnlyTClass,
                        int rRequestedVersionNumber,
                        int rRequestedRNTupleSerializationMode,
+                       const std::string &rRequestedRNTupleSoARecord,
                        const cling::Interpreter &interpret,
                        const TNormalizedCtxt &normCtxt);
 
@@ -267,6 +271,7 @@ class AnnotatedRecordDecl {
                        bool rRequestOnlyTClass,
                        int rRequestedVersionNumber,
                        int rRequestedRNTupleSerializationMode,
+                       const std::string &rRequestedRNTupleSoARecord,
                        const cling::Interpreter &interpret,
                        const TNormalizedCtxt &normCtxt);
    // clang-format on
@@ -292,6 +297,7 @@ class AnnotatedRecordDecl {
    bool RequestOnlyTClass() const { return fRequestOnlyTClass; }
    int  RequestedVersionNumber() const { return fRequestedVersionNumber; }
    int  RequestedRNTupleSerializationMode() const { return fRequestedRNTupleSerializationMode; }
+   const std::string &RequestedRNTupleSoARecord() const { return fRequestedRNTupleSoARecord; }
    // clang-format on
    int  RootFlag() const {
       // Return the request (streamerInfo, has_version, etc.) combined in a single

core/clingutils/src/RStl.cxx

@@ -101,6 +101,7 @@ void ROOT::Internal::RStl::GenerateTClassFor(const clang::QualType &type, const
                                                       false,
                                                       -1,
                                                       0,
+                                                      "",
                                                       interp,
                                                       normCtxt));
    // clang-format on
@@ -163,6 +164,7 @@ void ROOT::Internal::RStl::GenerateTClassFor(const char *requestedName, const cl
                                                       false,
                                                       -1,
                                                       0,
+                                                      "",
                                                       interp,
                                                       normCtxt));
    // clang-format on

core/clingutils/src/TClingUtils.cxx

@@ -386,6 +386,7 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
                                          bool rRequestOnlyTClass,
                                          int rRequestedVersionNumber,
                                          int rRequestedRNTupleSerializationMode,
+                                         const std::string &rRequestedRNTupleSoARecord,
                                          const cling::Interpreter &interpreter,
                                          const TNormalizedCtxt &normCtxt)
    : fRuleIndex(index),
@@ -395,7 +396,8 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
      fRequestNoInputOperator(rRequestNoInputOperator),
      fRequestOnlyTClass(rRequestOnlyTClass),
      fRequestedVersionNumber(rRequestedVersionNumber),
-     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode)
+     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode),
+     fRequestedRNTupleSoARecord(rRequestedRNTupleSoARecord)
 // clang-format on
 {
    TMetaUtils::GetNormalizedName(fNormalizedName, decl->getASTContext().getTypeDeclType(decl), interpreter,normCtxt);
@@ -417,6 +419,7 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
                                          bool rRequestOnlyTClass,
                                          int rRequestVersionNumber,
                                          int rRequestedRNTupleSerializationMode,
+                                         const std::string &rRequestedRNTupleSoARecord,
                                          const cling::Interpreter &interpreter,
                                          const TNormalizedCtxt &normCtxt)
    : fRuleIndex(index),
@@ -427,7 +430,8 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
      fRequestNoInputOperator(rRequestNoInputOperator),
      fRequestOnlyTClass(rRequestOnlyTClass),
      fRequestedVersionNumber(rRequestVersionNumber),
-     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode)
+     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode),
+     fRequestedRNTupleSoARecord(rRequestedRNTupleSoARecord)
 // clang-format on
 {
    // For comparison purposes.
@@ -456,6 +460,7 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
                                          bool rRequestOnlyTClass,
                                          int rRequestVersionNumber,
                                          int rRequestedRNTupleSerializationMode,
+                                         const std::string &rRequestedRNTupleSoARecord,
                                          const cling::Interpreter &interpreter,
                                          const TNormalizedCtxt &normCtxt)
    : fRuleIndex(index),
@@ -466,7 +471,8 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
      fRequestNoInputOperator(rRequestNoInputOperator),
      fRequestOnlyTClass(rRequestOnlyTClass),
      fRequestedVersionNumber(rRequestVersionNumber),
-     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode)
+     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode),
+     fRequestedRNTupleSoARecord(rRequestedRNTupleSoARecord)
 // clang-format on
 {
    // For comparison purposes.
@@ -490,6 +496,7 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
                                          bool rRequestOnlyTClass,
                                          int rRequestVersionNumber,
                                          int rRequestedRNTupleSerializationMode,
+                                         const std::string &rRequestedRNTupleSoARecord,
                                          const cling::Interpreter &interpreter,
                                          const TNormalizedCtxt &normCtxt)
    : fRuleIndex(index),
@@ -500,7 +507,8 @@ AnnotatedRecordDecl::AnnotatedRecordDecl(long index,
      fRequestNoInputOperator(rRequestNoInputOperator),
      fRequestOnlyTClass(rRequestOnlyTClass),
      fRequestedVersionNumber(rRequestVersionNumber),
-     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode)
+     fRequestedRNTupleSerializationMode(rRequestedRNTupleSerializationMode),
+     fRequestedRNTupleSoARecord(rRequestedRNTupleSoARecord)
 // clang-format on
 {
    // const clang::ClassTemplateSpecializationDecl *tmplt_specialization = llvm::dyn_cast<clang::ClassTemplateSpecializationDecl> (decl);
@@ -2063,6 +2071,14 @@ void ROOT::TMetaUtils::WriteClassInit(std::ostream& finalString,
 
    }
 
+   //---------------------------------------------------------------------------
+   // Register underlying SoA record for RNTuple SoA layouts
+   /////////////////////////////////////////////////////////////////////////////
+
+   if (!cl.RequestedRNTupleSoARecord().empty()) {
+      finalString << "      instance.SetRNTupleSoARecord(\"" << cl.RequestedRNTupleSoARecord() << "\");" << "\n";
+   }
+
    //---------------------------------------------------------------------------
    // Pass the schema evolution rules to TGenericClassInfo
    /////////////////////////////////////////////////////////////////////////////

core/dictgen/res/ClassSelectionRule.h

@@ -47,6 +47,8 @@ class ClassSelectionRule final : public BaseSelectionRule
    int  fRequestedVersionNumber; // Explicit request for a specific version number (default to no request with -1).
    // Explicit request for unsplit (-1) or split (=1), defaults to unset (=0)
    int  fRequestedRNTupleSerializationMode = 0;
+   // If not empty, marks the class as an RNTuple SoA layout for the underlying given record class
+   std::string fRequestedRNTupleSoARecord;
    // clang-format on
 
 public:
@@ -83,6 +85,7 @@ class ClassSelectionRule final : public BaseSelectionRule
    void SetRequestPrivate(bool val);
    void SetRequestedVersionNumber(int version);
    void SetRequestedRNTupleSerializationMode(int serializationMode);
+   void SetRequestedRNTupleSoARecord(const std::string &recordName);
 
    bool RequestOnlyTClass() const;      // True if the user want the TClass intiliazer but *not* the interpreter meta data
    bool RequestNoStreamer() const;      // Request no Streamer function in the dictionary
@@ -92,6 +95,7 @@ class ClassSelectionRule final : public BaseSelectionRule
    bool RequestPrivate() const;
    int  RequestedVersionNumber() const;
    int RequestedRNTupleSerializationMode() const;
+   const std::string &RequestedRNTupleSoARecord() const;
 };
 
 #endif

core/dictgen/src/ClassSelectionRule.cxx

@@ -162,6 +162,11 @@ void ClassSelectionRule::SetRequestedRNTupleSerializationMode(int serializationM
    fRequestedRNTupleSerializationMode = serializationMode;
 }
 
+void ClassSelectionRule::SetRequestedRNTupleSoARecord(const std::string &recordName)
+{
+   fRequestedRNTupleSoARecord = recordName;
+}
+
 bool ClassSelectionRule::RequestOnlyTClass() const
 {
    return fRequestOnlyTClass;
@@ -186,3 +191,8 @@ int ClassSelectionRule::RequestedRNTupleSerializationMode() const
 {
    return fRequestedRNTupleSerializationMode;
 }
+
+const std::string &ClassSelectionRule::RequestedRNTupleSoARecord() const
+{
+   return fRequestedRNTupleSoARecord;
+}

core/dictgen/src/LinkdefReader.cxx

@@ -27,6 +27,7 @@
 
 #include <iostream>
 #include <memory>
+#include <string>
 #include "LinkdefReader.h"
 #include "SelectionRules.h"
 #include "RConversionRuleParser.h"
@@ -57,6 +58,7 @@ struct LinkdefReader::Options {
    };
    int fVersionNumber;
    int fRNTupleSerializationMode; // 0: unset, -1: enforce streamed, 1: enforce native
+   std::string fRNTupleSoARecord;
 };
 
 /*
@@ -409,6 +411,7 @@ bool LinkdefReader::AddRule(const std::string& ruletype,
                   if (options->fVersionNumber >= 0) csr.SetRequestedVersionNumber(options->fVersionNumber);
                   if (options->fRNTupleSerializationMode != 0)
                      csr.SetRequestedRNTupleSerializationMode(options->fRNTupleSerializationMode);
+                  csr.SetRequestedRNTupleSoARecord(options->fRNTupleSoARecord);
                }
                if (csr.RequestStreamerInfo() && csr.RequestNoStreamer()) {
                   std::cerr << "Warning: " << localIdentifier << " option + mutual exclusive with -, + prevails\n";
@@ -663,7 +666,8 @@ class LinkdefReaderPragmaHandler : public clang::PragmaHandler {
        *   nomap: (ignored by roocling; prevents entry in ROOT's rootmap file)
        *   stub: (ignored by rootcling was a directly for CINT code generation)
        *   version(x): sets the version number of the class to x
-       *   rntuple[un]split: enforce split/unsplit encoding in RNTuple
+       *   rntupleStreamerMode(true|false): enforce/prevent use of the RNTuple streamer field
+       *   rntupleSoARecord(x): marks the class as an RNTuple SoA layout for the underlying record class x
        */
 
       // We assume that the first toke in option or options
@@ -718,11 +722,36 @@ class LinkdefReaderPragmaHandler : public clang::PragmaHandler {
                         "(either rntupleStreamerMode(true) or rntupleStreamerMode(false))",
                         boolval, PP);
                } else {
+                  if (!options.fRNTupleSoARecord.empty()) {
+                     Error("Error: rntupleStreamerMode(true) and rntupleSoARecord are mutually exclusive", boolval, PP);
+                  }
                   options.fRNTupleSerializationMode = -1;
                }
             } else {
                Error("Error: Malformed rntupleStreamerMode option (either 'true' or 'false').", boolval, PP);
             }
+         } else if (tok.getIdentifierInfo()->getName() == "rntupleSoARecord") {
+            clang::Token start = tok;
+            PP.Lex(tok);
+            if (tok.isNot(clang::tok::l_paren)) {
+               Error("Error: missing left parenthesis after rntupleSoARecord.", start, PP);
+               return false;
+            }
+            PP.Lex(tok);
+            clang::Token strval = tok;
+            if (tok.isNot(clang::tok::eod))
+               PP.Lex(tok);
+            if (tok.isNot(clang::tok::r_paren)) {
+               Error("Error: missing right parenthesis after rntupleSoARecord.", start, PP);
+               return false;
+            }
+            if (!strval.getIdentifierInfo() || strval.getIdentifierInfo()->getName().empty()) {
+               Error("Error: Malformed rntupleSoARecord option.", strval, PP);
+            }
+            if (options.fRNTupleSerializationMode == -1) {
+               Error("Error: rntupleStreamerMode(true) and rntupleSoARecord are mutually exclusive", strval, PP);
+            }
+            options.fRNTupleSoARecord = strval.getIdentifierInfo()->getName();
          } else if (tok.getIdentifierInfo()->getName() == "stub") {
             // This was solely for CINT dictionary, ignore for now.
             // options.fUseStubs = 1;

core/dictgen/src/Scanner.cxx

@@ -563,6 +563,7 @@ int RScanner::AddAnnotatedRecordDecl(const ClassSelectionRule* selected,
                                     selected->RequestOnlyTClass(),
                                     selected->RequestedVersionNumber(),
                                     selected->RequestedRNTupleSerializationMode(),
+                                    selected->RequestedRNTupleSoARecord(),
                                     fInterpreter,
                                     fNormCtxt);
    } else {
@@ -574,6 +575,7 @@ int RScanner::AddAnnotatedRecordDecl(const ClassSelectionRule* selected,
                                     selected->RequestOnlyTClass(),
                                     selected->RequestedVersionNumber(),
                                     selected->RequestedRNTupleSerializationMode(),
+                                    selected->RequestedRNTupleSoARecord(),
                                     fInterpreter,
                                     fNormCtxt);
    }

core/dictgen/src/XMLReader.cxx

@@ -961,6 +961,13 @@ bool XMLReader::Parse(const std::string &fileName, SelectionRules& out)
                               "not both\n",
                               lineNumCharp, iAttrValue.c_str());
                         } else {
+                           if (!csr->RequestedRNTupleSoARecord().empty()) {
+                              ROOT::TMetaUtils::Error(
+                                 nullptr,
+                                 "XML at line %s: class attributes rntupleStreamerMode and rntupleSoARecord "
+                                 "are mutually exclusive\n",
+                                 lineNumCharp, iAttrValue.c_str());
+                           }
                            csr->SetRequestedRNTupleSerializationMode(-1);
                         }
                      } else {
@@ -972,6 +979,18 @@ bool XMLReader::Parse(const std::string &fileName, SelectionRules& out)
                      }
                   }
 
+                  // request RNTuple SoA record type name
+                  if (tagKind == kClass && csr && "rntupleSoARecord" == iAttrName) {
+                     if (csr->RequestedRNTupleSerializationMode() == -1) {
+                        ROOT::TMetaUtils::Error(
+                           nullptr,
+                           "XML at line %s: class attributes rntupleStreamerMode and rntupleSoARecord "
+                           "are mutually exclusive\n",
+                           lineNumCharp, iAttrValue.c_str());
+                     }
+                     csr->SetRequestedRNTupleSoARecord(iAttrValue);
+                  }
+
                   // request no input operator
                   if (tagKind == kClass && csr && "noInputOperator" == iAttrName){
                     if (iAttrValue == "true") {

core/dictgen/src/rootcling_impl.cxx

@@ -3930,7 +3930,7 @@ int RootClingMain(int argc,
 
    const char *etcDir = gDriverConfig->fTROOT__GetEtcDir();
    std::string llvmResourceDir = etcDir ? std::string(etcDir) + "/cling" : "";
-   
+
    if (gBareClingSubcommand) {
       std::vector<const char *> clingArgsC;
       clingArgsC.push_back(executableFileName);
@@ -4106,7 +4106,7 @@ int RootClingMain(int argc,
    // cling-only arguments
    if (etcDir)
       clingArgs.push_back(std::string("-I") + llvm::sys::path::convert_to_slash(etcDir));
-   
+
    // We do not want __ROOTCLING__ in the pch!
    if (!gOptGeneratePCH) {
       clingArgs.push_back("-D__ROOTCLING__");
@@ -4616,7 +4616,7 @@ int RootClingMain(int argc,
       // interpPragmaSource and we still need to process it.
 
       LinkdefReader ldefr(interp, constructorTypes);
-      
+
       if (!ldefr.Parse(selectionRules, interpPragmaSource, clingArgs,
                        llvmResourceDir.c_str())) {
          ROOT::TMetaUtils::Error(nullptr, "Parsing #pragma failed %s\n", linkdefFilename.c_str());
@@ -5532,6 +5532,7 @@ int GenReflexMain(int argc, char **argv)
       "        Default value is 'false'\n"
       "      - rntupleStreamerMode [true/false]: enforce streamed or native writing for RNTuple.\n"
       "        If unset, RNTuple stores classes in split mode or fails if the class cannot be split.\n"
+      "      - rntupleSoARecord [class name]: marks the class as an RNTuple SoA layout for the underlying record\n"
       "      - noInputOperator [true/false]: turns off input operator generation if set\n"
       "        to 'true'. Default value is 'false'\n"
       "      Example XML:\n"
@@ -5542,6 +5543,7 @@ int GenReflexMain(int argc, char **argv)
       "                 [id=\"xxxx\"] [noStreamer=\"true/false\"]\n"
       "                 [noInputOperator=\"true/false\"]\n"
       "                 [rntupleStreamerMode=\"true/false\"] />\n"
+      "                 [rntupleSoARecord=\"class_name\"] />\n"
       "          <class name=\"classname\" >\n"
       "            <field name=\"m_transient\" transient=\"true\"/>\n"
       "            <field name=\"m_anothertransient\" persistent=\"false\"/>\n"

core/meta/inc/TGenericClassInfo.h

@@ -68,6 +68,7 @@ namespace ROOT {
       TVirtualCollectionProxy    *fCollectionProxy;
       Int_t                       fSizeof;
       Int_t                       fPragmaBits;
+      std::string                 fRNTupleSoARecord;
       Detail::TCollectionProxyInfo *fCollectionProxyInfo;
       Detail::TCollectionProxyInfo *fCollectionStreamerInfo;
       std::vector<ROOT::Internal::TSchemaHelper>  fReadRules;
@@ -142,6 +143,7 @@ namespace ROOT {
       void                              SetStreamerFunc(ClassStreamerFunc_t);
       void                              SetConvStreamerFunc(ClassConvStreamerFunc_t);
       Short_t                           SetVersion(Short_t version);
+      void                              SetRNTupleSoARecord(const std::string &recordName);
 
       //   protected:
    private: