[Clang] [Sema] Diagnose unknown std::initializer_list layout in SemaInit (#95580)

This checks if the layout of `std::initializer_list` is something Clang
can handle much earlier and deduplicates the checks in
CodeGen/CGExprAgg.cpp and AST/ExprConstant.cpp

Also now diagnose `union initializer_list` (Fixes #95495), bit-field for
the size (Fixes a crash that would happen during codegen if it were
unnamed), base classes (that wouldn't be initialized) and polymorphic
classes (whose vtable pointer wouldn't be initialized).

Author: MitalAshok

clang-tools-extra/clangd/unittests/ASTTests.cpp

@@ -80,7 +80,7 @@ TEST(GetDeducedType, KwAutoKwDecltypeExpansion) {
               namespace std
               {
                 template<class _E>
-                class [[initializer_list]] {};
+                class [[initializer_list]] { const _E *a, *b; };
               }
 
               ^auto i = {1,2};

clang-tools-extra/clangd/unittests/HoverTests.cpp

@@ -2284,7 +2284,7 @@ TEST(Hover, All) {
             namespace std
             {
               template<class _E>
-              class initializer_list {};
+              class initializer_list { const _E *a, *b; };
             }
             void foo() {
               ^[[auto]] i = {1,2};

clang-tools-extra/clangd/unittests/InlayHintTests.cpp

@@ -945,7 +945,7 @@ TEST(ParameterHints, ConstructorStdInitList) {
   // Do not show hints for std::initializer_list constructors.
   assertParameterHints(R"cpp(
     namespace std {
-      template <typename> class initializer_list {};
+      template <typename E> class initializer_list { const E *a, *b; };
     }
     struct S {
       S(std::initializer_list<int> param);

clang-tools-extra/clangd/unittests/XRefsTests.cpp

@@ -771,7 +771,7 @@ TEST(LocateSymbol, All) {
         namespace std
         {
           template<class _E>
-          class [[initializer_list]] {};
+          class [[initializer_list]] { const _E *a, *b; };
         }
 
         ^auto i = {1,2};

clang-tools-extra/include-cleaner/unittests/WalkASTTest.cpp

@@ -534,7 +534,7 @@ TEST(WalkAST, Enums) {
 TEST(WalkAST, InitializerList) {
   testWalk(R"cpp(
        namespace std {
-        template <typename T> struct $implicit^initializer_list {};
+        template <typename T> struct $implicit^initializer_list { const T *a, *b; };
        })cpp",
            R"cpp(
        const char* s = "";

clang-tools-extra/test/clang-tidy/checkers/modernize/min-max-use-initializer-list.cpp

@@ -11,6 +11,7 @@ T max(T a, T b) {
 namespace std {
 template< class T >
 struct initializer_list {
+  const T *a, *b;
   initializer_list()=default;
   initializer_list(T*,int){}
   const T* begin() const {return nullptr;}

clang-tools-extra/test/clang-tidy/checkers/modernize/use-emplace-ignore-implicit-constructors.cpp

@@ -4,10 +4,11 @@
 // RUN:                true}}"
 
 namespace std {
-template <typename>
+template <typename E>
 class initializer_list
 {
 public:
+  const E *a, *b;
   initializer_list() noexcept {}
 };
 

clang-tools-extra/test/clang-tidy/checkers/modernize/use-emplace.cpp

@@ -8,9 +8,10 @@
 // RUN:                '::std::make_pair; ::std::make_tuple; ::test::MakeSingle'}}"
 
 namespace std {
-template <typename>
+template <typename E>
 class initializer_list {
 public:
+  const E *a, *b;
   initializer_list() noexcept {}
 };
 

clang-tools-extra/test/clang-tidy/checkers/performance/inefficient-vector-operation.cpp

@@ -5,7 +5,7 @@
 
 namespace std {
 
-typedef int size_t;
+typedef decltype(sizeof 0) size_t;
 
 template<class E> class initializer_list {
 public:
@@ -15,6 +15,8 @@ template<class E> class initializer_list {
   using size_type = size_t;
   using iterator = const E*;
   using const_iterator = const E*;
+  iterator p;
+  size_t sz;
   initializer_list();
   size_t size() const; // number of elements
   const E* begin() const; // first element

clang-tools-extra/test/clang-tidy/checkers/readability/isolate-declaration-cxx17.cpp

@@ -31,7 +31,7 @@ struct SomeClass {
 
 namespace std {
 template <typename T>
-class initializer_list {};
+class initializer_list { const T *a, *b; };
 
 template <typename T>
 class vector {

clang/docs/ReleaseNotes.rst

@@ -598,6 +598,10 @@ Improvements to Clang's diagnostics
 - Clang no longer emits a "declared here" note for a builtin function that has no declaration in source.
   Fixes #GH93369.
 
+- Clang now diagnoses unsupported class declarations for ``std::initializer_list<E>`` when they are
+  used rather than when they are needed for constant evaluation or when code is generated for them.
+  The check is now stricter to prevent crashes for some unsupported declarations (Fixes #GH95495).
+
 Improvements to Clang's time-trace
 ----------------------------------
 

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -12207,6 +12207,9 @@ def err_std_source_location_impl_not_found : Error<
 def err_std_source_location_impl_malformed : Error<
   "'std::source_location::__impl' must be standard-layout and have only two 'const char *' fields '_M_file_name' and '_M_function_name', and two integral fields '_M_line' and '_M_column'">;
 
+def err_std_initializer_list_malformed : Error<
+  "%0 layout not recognized. Must be a non-polymorphic class type with no bases and two fields: a 'const E *' and either another 'const E *' or a 'std::size_t'">;
+
 // HLSL Diagnostics
 def err_hlsl_attr_unsupported_in_stage : Error<"attribute %0 is unsupported in '%1' shaders, requires %select{|one of the following: }2%3">;
 def err_hlsl_attr_invalid_type : Error<

clang/lib/AST/ExprConstant.cpp

@@ -10545,48 +10545,37 @@ bool RecordExprEvaluator::VisitCXXStdInitializerListExpr(
   // Get a pointer to the first element of the array.
   Array.addArray(Info, E, ArrayType);
 
-  auto InvalidType = [&] {
-    Info.FFDiag(E, diag::note_constexpr_unsupported_layout)
-      << E->getType();
-    return false;
-  };
-
-  // FIXME: Perform the checks on the field types in SemaInit.
-  RecordDecl *Record = E->getType()->castAs<RecordType>()->getDecl();
-  RecordDecl::field_iterator Field = Record->field_begin();
-  if (Field == Record->field_end())
-    return InvalidType();
-
-  // Start pointer.
-  if (!Field->getType()->isPointerType() ||
-      !Info.Ctx.hasSameType(Field->getType()->getPointeeType(),
-                            ArrayType->getElementType()))
-    return InvalidType();
-
   // FIXME: What if the initializer_list type has base classes, etc?
   Result = APValue(APValue::UninitStruct(), 0, 2);
   Array.moveInto(Result.getStructField(0));
 
-  if (++Field == Record->field_end())
-    return InvalidType();
-
-  if (Field->getType()->isPointerType() &&
-      Info.Ctx.hasSameType(Field->getType()->getPointeeType(),
-                           ArrayType->getElementType())) {
+  RecordDecl *Record = E->getType()->castAs<RecordType>()->getDecl();
+  RecordDecl::field_iterator Field = Record->field_begin();
+  assert(Field != Record->field_end() &&
+         Info.Ctx.hasSameType(Field->getType()->getPointeeType(),
+                              ArrayType->getElementType()) &&
+         "Expected std::initializer_list first field to be const E *");
+  ++Field;
+  assert(Field != Record->field_end() &&
+         "Expected std::initializer_list to have two fields");
+
+  if (Info.Ctx.hasSameType(Field->getType(), Info.Ctx.getSizeType())) {
+    // Length.
+    Result.getStructField(1) = APValue(APSInt(ArrayType->getSize()));
+  } else {
     // End pointer.
+    assert(Info.Ctx.hasSameType(Field->getType()->getPointeeType(),
+                                ArrayType->getElementType()) &&
+           "Expected std::initializer_list second field to be const E *");
     if (!HandleLValueArrayAdjustment(Info, E, Array,
                                      ArrayType->getElementType(),
                                      ArrayType->getZExtSize()))
       return false;
     Array.moveInto(Result.getStructField(1));
-  } else if (Info.Ctx.hasSameType(Field->getType(), Info.Ctx.getSizeType()))
-    // Length.
-    Result.getStructField(1) = APValue(APSInt(ArrayType->getSize()));
-  else
-    return InvalidType();
+  }
 
-  if (++Field != Record->field_end())
-    return InvalidType();
+  assert(++Field == Record->field_end() &&
+         "Expected std::initializer_list to only have two fields");
 
   return true;
 }

clang/lib/CodeGen/CGExprAgg.cpp

@@ -426,53 +426,45 @@ AggExprEmitter::VisitCXXStdInitializerListExpr(CXXStdInitializerListExpr *E) {
       Ctx.getAsConstantArrayType(E->getSubExpr()->getType());
   assert(ArrayType && "std::initializer_list constructed from non-array");
 
-  // FIXME: Perform the checks on the field types in SemaInit.
   RecordDecl *Record = E->getType()->castAs<RecordType>()->getDecl();
   RecordDecl::field_iterator Field = Record->field_begin();
-  if (Field == Record->field_end()) {
-    CGF.ErrorUnsupported(E, "weird std::initializer_list");
-    return;
-  }
+  assert(Field != Record->field_end() &&
+         Ctx.hasSameType(Field->getType()->getPointeeType(),
+                         ArrayType->getElementType()) &&
+         "Expected std::initializer_list first field to be const E *");
 
   // Start pointer.
-  if (!Field->getType()->isPointerType() ||
-      !Ctx.hasSameType(Field->getType()->getPointeeType(),
-                       ArrayType->getElementType())) {
-    CGF.ErrorUnsupported(E, "weird std::initializer_list");
-    return;
-  }
-
   AggValueSlot Dest = EnsureSlot(E->getType());
   LValue DestLV = CGF.MakeAddrLValue(Dest.getAddress(), E->getType());
   LValue Start = CGF.EmitLValueForFieldInitialization(DestLV, *Field);
   llvm::Value *ArrayStart = ArrayPtr.emitRawPointer(CGF);
   CGF.EmitStoreThroughLValue(RValue::get(ArrayStart), Start);
   ++Field;
-
-  if (Field == Record->field_end()) {
-    CGF.ErrorUnsupported(E, "weird std::initializer_list");
-    return;
-  }
+  assert(Field != Record->field_end() &&
+         "Expected std::initializer_list to have two fields");
 
   llvm::Value *Size = Builder.getInt(ArrayType->getSize());
   LValue EndOrLength = CGF.EmitLValueForFieldInitialization(DestLV, *Field);
-  if (Field->getType()->isPointerType() &&
-      Ctx.hasSameType(Field->getType()->getPointeeType(),
-                      ArrayType->getElementType())) {
+  if (Ctx.hasSameType(Field->getType(), Ctx.getSizeType())) {
+    // Length.
+    CGF.EmitStoreThroughLValue(RValue::get(Size), EndOrLength);
+
+  } else {
     // End pointer.
+    assert(Field->getType()->isPointerType() &&
+           Ctx.hasSameType(Field->getType()->getPointeeType(),
+                           ArrayType->getElementType()) &&
+           "Expected std::initializer_list second field to be const E *");
     llvm::Value *Zero = llvm::ConstantInt::get(CGF.PtrDiffTy, 0);
     llvm::Value *IdxEnd[] = { Zero, Size };
     llvm::Value *ArrayEnd = Builder.CreateInBoundsGEP(
         ArrayPtr.getElementType(), ArrayPtr.emitRawPointer(CGF), IdxEnd,
         "arrayend");
     CGF.EmitStoreThroughLValue(RValue::get(ArrayEnd), EndOrLength);
-  } else if (Ctx.hasSameType(Field->getType(), Ctx.getSizeType())) {
-    // Length.
-    CGF.EmitStoreThroughLValue(RValue::get(Size), EndOrLength);
-  } else {
-    CGF.ErrorUnsupported(E, "weird std::initializer_list");
-    return;
   }
+
+  assert(++Field == Record->field_end() &&
+         "Expected std::initializer_list to only have two fields");
 }
 
 /// Determine if E is a trivial array filler, that is, one that is

clang/lib/Sema/SemaInit.cpp

@@ -9479,6 +9479,57 @@ ExprResult InitializationSequence::Perform(Sema &S,
       // Wrap it in a construction of a std::initializer_list<T>.
       CurInit = new (S.Context) CXXStdInitializerListExpr(Step->Type, MTE);
 
+      if (!Step->Type->isDependentType()) {
+        QualType ElementType;
+        [[maybe_unused]] bool IsStdInitializerList =
+            S.isStdInitializerList(Step->Type, &ElementType);
+        assert(IsStdInitializerList &&
+               "StdInitializerList step to non-std::initializer_list");
+        const CXXRecordDecl *Record =
+            Step->Type->getAsCXXRecordDecl()->getDefinition();
+        assert(Record && Record->isCompleteDefinition() &&
+               "std::initializer_list should have already be "
+               "complete/instantiated by this point");
+
+        auto InvalidType = [&] {
+          S.Diag(Record->getLocation(),
+                 diag::err_std_initializer_list_malformed)
+              << Step->Type.getUnqualifiedType();
+          return ExprError();
+        };
+
+        if (Record->isUnion() || Record->getNumBases() != 0 ||
+            Record->isPolymorphic())
+          return InvalidType();
+
+        RecordDecl::field_iterator Field = Record->field_begin();
+        if (Field == Record->field_end())
+          return InvalidType();
+
+        // Start pointer
+        if (!Field->getType()->isPointerType() ||
+            !S.Context.hasSameType(Field->getType()->getPointeeType(),
+                                   ElementType.withConst()))
+          return InvalidType();
+
+        if (++Field == Record->field_end())
+          return InvalidType();
+
+        // Size or end pointer
+        if (const auto *PT = Field->getType()->getAs<PointerType>()) {
+          if (!S.Context.hasSameType(PT->getPointeeType(),
+                                     ElementType.withConst()))
+            return InvalidType();
+        } else {
+          if (Field->isBitField() ||
+              !S.Context.hasSameType(Field->getType(), S.Context.getSizeType()))
+            return InvalidType();
+        }
+
+        if (++Field != Record->field_end())
+          return InvalidType();
+      }
+
       // Bind the result, in case the library has given initializer_list a
       // non-trivial destructor.
       if (shouldBindAsTemporary(Entity))

clang/test/CXX/dcl.dcl/dcl.spec/dcl.type/dcl.spec.auto/p4.cpp

@@ -43,7 +43,7 @@ struct S {
 const int S::b;
 const auto S::c = 0;
 
-namespace std { template<typename T> struct initializer_list { initializer_list(); }; }
+namespace std { template<typename T> struct initializer_list { const T *a, *b; initializer_list(); }; }
 
 // In an initializer of the form ( expression-list ), the expression-list
 // shall be a single assigment-expression.

clang/test/CXX/dcl.dcl/dcl.spec/dcl.type/dcl.spec.auto/p7-cxx14.cpp

@@ -4,9 +4,7 @@
 
 namespace std {
   template<typename T> struct initializer_list {
-    const T *p;
-    unsigned long n;
-    initializer_list(const T *p, unsigned long n);
+    const T *a, *b;
   };
 }
 

clang/test/CodeCompletion/ctor-signature.cpp

@@ -17,7 +17,7 @@ void foo() {
 }
 
 namespace std {
-template <typename> struct initializer_list {};
+template <typename E> struct initializer_list { const E *a, *b; };
 } // namespace std
 
 struct Bar {

clang/test/Coverage/unresolved-ctor-expr.cpp

@@ -4,7 +4,7 @@
 // GH62105 demonstrated a crash with this example code when calculating
 // coverage mapping because some source location information was being dropped.
 // Demonstrate that we do not crash on this code.
-namespace std { template <typename> class initializer_list {}; }
+namespace std { template <typename E> class initializer_list { const E *a, *b; }; }
 
 template <typename> struct T {
   T(std::initializer_list<int>, int = int());

clang/test/Modules/Inputs/initializer_list/direct.h

@@ -2,7 +2,7 @@ namespace std {
   using size_t = decltype(sizeof(0));
 
   template<typename T> struct initializer_list {
-    initializer_list(T*, size_t);
+    const T* ptr; size_t sz;
   };
 
   template<typename T> int min(initializer_list<T>);

clang/test/Modules/pr60775.cppm

@@ -29,9 +29,10 @@
 namespace std {
   typedef decltype(sizeof(int)) size_t;
   template<typename T> struct initializer_list {
+    const T* ptr; size_t sz;
     initializer_list(const T *, size_t);
-    T* begin();
-    T* end();
+    const T* begin();
+    const T* end();
   };
 }
 

clang/test/OpenMP/declare_reduction_codegen_in_templates.cpp

@@ -16,7 +16,7 @@
 #ifndef HEADER
 #define HEADER
 
-typedef long unsigned a;
+typedef decltype(sizeof 0) a;
 namespace std {
 template <class> class initializer_list {
   const int *b;