[clang][DebugInfo] Emit unified (Itanium) mangled name to structor declarations (#154142)

Depends on #154137

This patch is motivated by
#149827, where we plan on using
mangled names on structor declarations to find the exact structor
definition that LLDB's expression evaluator should call.

Given a `DW_TAG_subprogram` for a function declaration, the most
convenient way for a debugger to find the corresponding definition is to
use the `DW_AT_linkage_name` (i.e., the mangled name). However, we
currently can't do that for constructors/destructors because Clang
doesn't attach linkage names to them. This is because, depending on ABI,
there can be multiple definitions for a single constructor/destructor
declaration. The way GCC works around this is by producing a `C4`/`D4`
"unified" mangling for structor declarations (see
[godbolt](https://godbolt.org/z/Wds6cja9K)). GDB uses this to locate the
relevant definitions.

This patch aligns Clang with GCC's DWARF output and allows us to
implement the same lookup scheme in LLDB.

Author: Michael137

clang/include/clang/Basic/ABI.h

@@ -27,14 +27,16 @@ enum CXXCtorType {
   Ctor_Comdat,         ///< The COMDAT used for ctors
   Ctor_CopyingClosure, ///< Copying closure variant of a ctor
   Ctor_DefaultClosure, ///< Default closure variant of a ctor
+  Ctor_Unified,        ///< GCC-style unified dtor
 };
 
 /// C++ destructor types.
 enum CXXDtorType {
-    Dtor_Deleting, ///< Deleting dtor
-    Dtor_Complete, ///< Complete object dtor
-    Dtor_Base,     ///< Base object dtor
-    Dtor_Comdat    ///< The COMDAT used for dtors
+  Dtor_Deleting, ///< Deleting dtor
+  Dtor_Complete, ///< Complete object dtor
+  Dtor_Base,     ///< Base object dtor
+  Dtor_Comdat,   ///< The COMDAT used for dtors
+  Dtor_Unified,  ///< GCC-style unified dtor
 };
 
 } // end namespace clang

clang/include/clang/Basic/DebugOptions.def

@@ -125,6 +125,12 @@ DEBUGOPT(DebugNameTable, 2, 0, Compatible)
 /// Whether to use DWARF base address specifiers in .debug_ranges.
 DEBUGOPT(DebugRangesBaseAddress, 1, 0, Compatible)
 
+/// Whether to add linkage names to constructor/destructor declarations.
+/// This is an escape hatch for cases where attaching the additional linkage
+/// names would increase debug-info size (particularly the .debug_str section)
+/// too much.
+DEBUGOPT(DebugStructorDeclLinkageNames, 1, 0, Benign)
+
 /// Whether to embed source in DWARF debug line section.
 DEBUGOPT(EmbedSource, 1, 0, Compatible)
 

clang/include/clang/Driver/Options.td

@@ -4789,6 +4789,18 @@ def gembed_source : Flag<["-"], "gembed-source">, Group<g_flags_Group>,
 def gno_embed_source : Flag<["-"], "gno-embed-source">, Group<g_flags_Group>,
     Flags<[NoXarchOption]>,
     HelpText<"Restore the default behavior of not embedding source text in DWARF debug sections">;
+defm structor_decl_linkage_names
+    : BoolGOption<"structor-decl-linkage-names",
+                  CodeGenOpts<"DebugStructorDeclLinkageNames">, DefaultTrue,
+                  NegFlag<SetFalse>,
+                  PosFlag<SetTrue, [], [],
+                          "Attach linkage names to C++ constructor/destructor "
+                          "declarations in DWARF."
+                          "Implies -g.">,
+                  BothFlags<[], [ClangOption, CLOption, CC1Option]>>,
+                  DocBrief<[{On some ABIs (e.g., Itanium), constructors and destructors may have multiple variants. Historically, when generating DWARF, Clang did not attach ``DW_AT_linkage_name``s to structor DIEs because there were multiple possible manglings (depending on the structor variant) that could be used. With ``-gstructor-decl-linkage-names``, for ABIs with structor variants, we attach a "unified" mangled name to structor declarations DIEs which debuggers can use to look up all the definitions for a structor declaration. E.g., a "unified" mangled name ``_ZN3FooC4Ev`` may have multiple definitions associated with it such as ``_ZN3FooC1Ev`` and ``_ZN3FooC2Ev``.
+
+Enabling this flag results in a better interactive debugging experience (both GDB and LLDB have support for understanding these "unified" linkage names). However, it comes with a significant increase in debug-info size (particularly the `.debug_str` section). As an escape hatch, users can disable this feature using ``-gno-structor-decl-linkage-names``.}]>;
 defm key_instructions : BoolGOption<"key-instructions",
     CodeGenOpts<"DebugKeyInstructions">, DefaultFalse,
     NegFlag<SetFalse>, PosFlag<SetTrue, [], [],

clang/lib/AST/ItaniumMangle.cpp

@@ -6026,6 +6026,8 @@ void CXXNameMangler::mangleCXXCtorType(CXXCtorType T,
   //                  ::= CI2 <type> # base inheriting constructor
   //
   // In addition, C5 is a comdat name with C1 and C2 in it.
+  // C4 represents a ctor declaration and is used by debuggers to look up
+  // the various ctor variants.
   Out << 'C';
   if (InheritedFrom)
     Out << 'I';
@@ -6036,6 +6038,9 @@ void CXXNameMangler::mangleCXXCtorType(CXXCtorType T,
   case Ctor_Base:
     Out << '2';
     break;
+  case Ctor_Unified:
+    Out << '4';
+    break;
   case Ctor_Comdat:
     Out << '5';
     break;
@@ -6053,6 +6058,8 @@ void CXXNameMangler::mangleCXXDtorType(CXXDtorType T) {
   //                  ::= D2  # base object destructor
   //
   // In addition, D5 is a comdat name with D1, D2 and, if virtual, D0 in it.
+  // D4 represents a dtor declaration and is used by debuggers to look up
+  // the various dtor variants.
   switch (T) {
   case Dtor_Deleting:
     Out << "D0";
@@ -6063,6 +6070,9 @@ void CXXNameMangler::mangleCXXDtorType(CXXDtorType T) {
   case Dtor_Base:
     Out << "D2";
     break;
+  case Dtor_Unified:
+    Out << "D4";
+    break;
   case Dtor_Comdat:
     Out << "D5";
     break;

clang/lib/AST/MicrosoftMangle.cpp

@@ -1496,6 +1496,8 @@ void MicrosoftCXXNameMangler::mangleCXXDtorType(CXXDtorType T) {
   // it.
   case Dtor_Comdat:
     llvm_unreachable("not expecting a COMDAT");
+  case Dtor_Unified:
+    llvm_unreachable("not expecting a unified dtor type");
   }
   llvm_unreachable("Unsupported dtor type?");
 }

clang/lib/CodeGen/CGClass.cpp

@@ -1481,6 +1481,8 @@ void CodeGenFunction::EmitDestructorBody(FunctionArgList &Args) {
   // we'd introduce *two* handler blocks.  In the Microsoft ABI, we
   // always delegate because we might not have a definition in this TU.
   switch (DtorType) {
+  case Dtor_Unified:
+    llvm_unreachable("not expecting a unified dtor");
   case Dtor_Comdat: llvm_unreachable("not expecting a COMDAT");
   case Dtor_Deleting: llvm_unreachable("already handled deleting case");
 

clang/lib/CodeGen/CGDebugInfo.cpp

@@ -2177,24 +2177,47 @@ static bool isFunctionLocalClass(const CXXRecordDecl *RD) {
   return false;
 }
 
+llvm::StringRef
+CGDebugInfo::GetMethodLinkageName(const CXXMethodDecl *Method) const {
+  assert(Method);
+
+  const bool IsCtorOrDtor =
+      isa<CXXConstructorDecl>(Method) || isa<CXXDestructorDecl>(Method);
+
+  if (IsCtorOrDtor && !CGM.getCodeGenOpts().DebugStructorDeclLinkageNames)
+    return {};
+
+  // In some ABIs (particularly Itanium) a single ctor/dtor
+  // corresponds to multiple functions. Attach a "unified"
+  // linkage name for those (which is the convention GCC uses).
+  // Otherwise, attach no linkage name.
+  if (IsCtorOrDtor && !CGM.getTarget().getCXXABI().hasConstructorVariants())
+    return {};
+
+  if (const auto *Ctor = llvm::dyn_cast<CXXConstructorDecl>(Method))
+    return CGM.getMangledName(GlobalDecl(Ctor, CXXCtorType::Ctor_Unified));
+
+  if (const auto *Dtor = llvm::dyn_cast<CXXDestructorDecl>(Method))
+    return CGM.getMangledName(GlobalDecl(Dtor, CXXDtorType::Dtor_Unified));
+
+  return CGM.getMangledName(Method);
+}
+
 llvm::DISubprogram *CGDebugInfo::CreateCXXMemberFunction(
     const CXXMethodDecl *Method, llvm::DIFile *Unit, llvm::DIType *RecordTy) {
-  bool IsCtorOrDtor =
-      isa<CXXConstructorDecl>(Method) || isa<CXXDestructorDecl>(Method);
+  assert(Method);
 
   StringRef MethodName = getFunctionName(Method);
   llvm::DISubroutineType *MethodTy = getOrCreateMethodType(Method, Unit);
 
-  // Since a single ctor/dtor corresponds to multiple functions, it doesn't
-  // make sense to give a single ctor/dtor a linkage name.
   StringRef MethodLinkageName;
   // FIXME: 'isFunctionLocalClass' seems like an arbitrary/unintentional
   // property to use here. It may've been intended to model "is non-external
   // type" but misses cases of non-function-local but non-external classes such
   // as those in anonymous namespaces as well as the reverse - external types
   // that are function local, such as those in (non-local) inline functions.
-  if (!IsCtorOrDtor && !isFunctionLocalClass(Method->getParent()))
-    MethodLinkageName = CGM.getMangledName(Method);
+  if (!isFunctionLocalClass(Method->getParent()))
+    MethodLinkageName = GetMethodLinkageName(Method);
 
   // Get the location for the method.
   llvm::DIFile *MethodDefUnit = nullptr;

clang/lib/CodeGen/CGDebugInfo.h

@@ -899,6 +899,10 @@ class CGDebugInfo {
       std::memcpy(Data + A.size(), B.data(), B.size());
     return StringRef(Data, A.size() + B.size());
   }
+
+  /// If one exists, returns the linkage name of the specified \
+  /// (non-null) \c Method. Returns empty string otherwise.
+  llvm::StringRef GetMethodLinkageName(const CXXMethodDecl *Method) const;
 };
 
 /// A scoped helper to set the current debug location to the specified

clang/lib/CodeGen/ItaniumCXXABI.cpp

@@ -91,6 +91,8 @@ class ItaniumCXXABI : public CodeGen::CGCXXABI {
 
       case Dtor_Comdat:
         llvm_unreachable("emitting dtor comdat as function?");
+      case Dtor_Unified:
+        llvm_unreachable("emitting unified dtor as function?");
       }
       llvm_unreachable("bad dtor kind");
     }
@@ -108,6 +110,9 @@ class ItaniumCXXABI : public CodeGen::CGCXXABI {
 
       case Ctor_Comdat:
         llvm_unreachable("emitting ctor comdat as function?");
+
+      case Ctor_Unified:
+        llvm_unreachable("emitting unified ctor as function?");
       }
       llvm_unreachable("bad dtor kind");
     }

clang/lib/CodeGen/MicrosoftCXXABI.cpp

@@ -77,6 +77,8 @@ class MicrosoftCXXABI : public CGCXXABI {
         return false;
 
       case Dtor_Comdat: llvm_unreachable("emitting dtor comdat as function?");
+      case Dtor_Unified:
+        llvm_unreachable("unexpected unified dtor type");
       }
       llvm_unreachable("bad dtor kind");
     }
@@ -1417,6 +1419,8 @@ llvm::GlobalValue::LinkageTypes MicrosoftCXXABI::getCXXDestructorLinkage(
     // and are emitted everywhere they are used. They are internal if the class
     // is internal.
     return llvm::GlobalValue::LinkOnceODRLinkage;
+  case Dtor_Unified:
+    llvm_unreachable("MS C++ ABI does not support unified dtors");
   case Dtor_Comdat:
     llvm_unreachable("MS C++ ABI does not support comdat dtors");
   }