Cherry-pick "[Modules] Record whether VarDecl initializers contain side effects (#143739)"

clang/include/clang/AST/Decl.h

@@ -1339,6 +1339,11 @@ class VarDecl : public DeclaratorDecl, public Redeclarable<VarDecl> {
     return const_cast<VarDecl *>(this)->getInitializingDeclaration();
   }
 
+  /// Checks whether this declaration has an initializer with side effects,
+  /// without triggering deserialization if the initializer is not yet
+  /// deserialized.
+  bool hasInitWithSideEffects() const;
+
   /// Determine whether this variable's value might be usable in a
   /// constant expression, according to the relevant language standard.
   /// This only checks properties of the declaration, and does not check

clang/include/clang/AST/ExternalASTSource.h

@@ -51,6 +51,7 @@ class RecordDecl;
 class Selector;
 class Stmt;
 class TagDecl;
+class VarDecl;
 
 /// Abstract interface for external sources of AST nodes.
 ///
@@ -173,6 +174,10 @@ class ExternalASTSource : public RefCountedBase<ExternalASTSource> {
 
   virtual ExtKind hasExternalDefinitions(const Decl *D);
 
+  virtual bool hasInitializerWithSideEffects(const VarDecl *VD) const {
+    return false;
+  }
+
   /// Finds all declarations lexically contained within the given
   /// DeclContext, after applying an optional filter predicate.
   ///
@@ -407,6 +412,17 @@ struct LazyOffsetPtr {
     return GetPtr();
   }
 
+  /// Retrieve the pointer to the AST node that this lazy pointer points to,
+  /// if it can be done without triggering deserialization.
+  ///
+  /// \returns a pointer to the AST node, or null if not yet deserialized.
+  T *getWithoutDeserializing() const {
+    if (isOffset()) {
+      return nullptr;
+    }
+    return GetPtr();
+  }
+
   /// Retrieve the address of the AST node pointer. Deserializes the pointee if
   /// necessary.
   T **getAddressOfPointer(ExternalASTSource *Source) const {

clang/include/clang/Sema/MultiplexExternalSemaSource.h

@@ -92,6 +92,8 @@ class MultiplexExternalSemaSource : public ExternalSemaSource {
 
   ExtKind hasExternalDefinitions(const Decl *D) override;
 
+  bool hasInitializerWithSideEffects(const VarDecl *VD) const override;
+
   /// Find all declarations with the given name in the
   /// given context.
   bool FindExternalVisibleDeclsByName(const DeclContext *DC,

clang/include/clang/Serialization/ASTReader.h

@@ -1348,6 +1348,12 @@ class ASTReader
     const StringRef &operator*() && = delete;
   };
 
+  /// VarDecls with initializers containing side effects must be emitted,
+  /// but DeclMustBeEmitted is not allowed to deserialize the intializer.
+  /// FIXME: Lower memory usage by removing VarDecls once the initializer
+  /// is deserialized.
+  llvm::SmallPtrSet<Decl *, 16> InitSideEffectVars;
+
 public:
   /// Get the buffer for resolving paths.
   SmallString<0> &getPathBuf() { return PathBuf; }
@@ -2263,6 +2269,8 @@ class ASTReader
 
   ExtKind hasExternalDefinitions(const Decl *D) override;
 
+  bool hasInitializerWithSideEffects(const VarDecl *VD) const override;
+
   /// Retrieve a selector from the given module with its local ID
   /// number.
   Selector getLocalSelector(ModuleFile &M, unsigned LocalID);

clang/lib/AST/ASTContext.cpp

@@ -13262,9 +13262,7 @@ bool ASTContext::DeclMustBeEmitted(const Decl *D) {
     return true;
 
   // Variables that have initialization with side-effects are required.
-  if (VD->getInit() && VD->getInit()->HasSideEffects(*this) &&
-      // We can get a value-dependent initializer during error recovery.
-      (VD->getInit()->isValueDependent() || !VD->evaluateValue()))
+  if (VD->hasInitWithSideEffects())
     return true;
 
   // Likewise, variables with tuple-like bindings are required if their

clang/lib/AST/Decl.cpp

@@ -2440,6 +2440,30 @@ VarDecl *VarDecl::getInitializingDeclaration() {
   return Def;
 }
 
+bool VarDecl::hasInitWithSideEffects() const {
+  if (!hasInit())
+    return false;
+
+  // Check if we can get the initializer without deserializing
+  const Expr *E = nullptr;
+  if (auto *S = dyn_cast<Stmt *>(Init)) {
+    E = cast<Expr>(S);
+  } else {
+    E = cast_or_null<Expr>(getEvaluatedStmt()->Value.getWithoutDeserializing());
+  }
+
+  if (E)
+    return E->HasSideEffects(getASTContext()) &&
+           // We can get a value-dependent initializer during error recovery.
+           (E->isValueDependent() || !evaluateValue());
+
+  assert(getEvaluatedStmt()->Value.isOffset());
+  // ASTReader tracks this without having to deserialize the initializer
+  if (auto Source = getASTContext().getExternalSource())
+    return Source->hasInitializerWithSideEffects(this);
+  return false;
+}
+
 bool VarDecl::isOutOfLine() const {
   if (Decl::isOutOfLine())
     return true;

clang/lib/AST/ExprConstant.cpp

@@ -16268,6 +16268,12 @@ static bool EvaluateInPlace(APValue &Result, EvalInfo &Info, const LValue &This,
                             const Expr *E, bool AllowNonLiteralTypes) {
   assert(!E->isValueDependent());
 
+  // Normally expressions passed to EvaluateInPlace have a type, but not when
+  // a VarDecl initializer is evaluated before the untyped ParenListExpr is
+  // replaced with a CXXConstructExpr. This can happen in LLDB.
+  if (E->getType().isNull())
+    return false;
+
   if (!AllowNonLiteralTypes && !CheckLiteralType(Info, E, &This))
     return false;
 

clang/lib/Sema/MultiplexExternalSemaSource.cpp

@@ -107,6 +107,14 @@ MultiplexExternalSemaSource::hasExternalDefinitions(const Decl *D) {
   return EK_ReplyHazy;
 }
 
+bool MultiplexExternalSemaSource::hasInitializerWithSideEffects(
+    const VarDecl *VD) const {
+  for (const auto &S : Sources)
+    if (S->hasInitializerWithSideEffects(VD))
+      return true;
+  return false;
+}
+
 bool MultiplexExternalSemaSource::
 FindExternalVisibleDeclsByName(const DeclContext *DC, DeclarationName Name) {
   bool AnyDeclsFound = false;

clang/lib/Serialization/ASTReader.cpp

@@ -9268,6 +9268,10 @@ ExternalASTSource::ExtKind ASTReader::hasExternalDefinitions(const Decl *FD) {
   return I->second ? EK_Never : EK_Always;
 }
 
+bool ASTReader::hasInitializerWithSideEffects(const VarDecl *VD) const {
+  return InitSideEffectVars.count(VD);
+}
+
 Selector ASTReader::getLocalSelector(ModuleFile &M, unsigned LocalID) {
   return DecodeSelector(getGlobalSelectorID(M, LocalID));
 }

clang/lib/Serialization/ASTReaderDecl.cpp

@@ -1622,6 +1622,9 @@ ASTDeclReader::RedeclarableResult ASTDeclReader::VisitVarDeclImpl(VarDecl *VD) {
     VD->NonParmVarDeclBits.PreviousDeclInSameBlockScope =
         VarDeclBits.getNextBit();
 
+    if (VarDeclBits.getNextBit())
+      Reader.InitSideEffectVars.insert(VD);
+
     VD->NonParmVarDeclBits.EscapingByref = VarDeclBits.getNextBit();
     HasDeducedType = VarDeclBits.getNextBit();
     VD->NonParmVarDeclBits.ImplicitParamKind =

clang/lib/Serialization/ASTWriterDecl.cpp

@@ -1158,6 +1158,7 @@ void ASTDeclWriter::VisitVarDecl(VarDecl *D) {
     VarDeclBits.addBit(D->isConstexpr());
     VarDeclBits.addBit(D->isInitCapture());
     VarDeclBits.addBit(D->isPreviousDeclInSameBlockScope());
+    VarDeclBits.addBit(D->hasInitWithSideEffects());
 
     VarDeclBits.addBit(D->isEscapingByref());
     HasDeducedType = D->getType()->getContainedDeducedType();
@@ -1207,10 +1208,11 @@ void ASTDeclWriter::VisitVarDecl(VarDecl *D) {
       !D->hasExtInfo() && D->getFirstDecl() == D->getMostRecentDecl() &&
       D->getKind() == Decl::Var && !D->isInline() && !D->isConstexpr() &&
       !D->isInitCapture() && !D->isPreviousDeclInSameBlockScope() &&
-      !D->isEscapingByref() && !HasDeducedType &&
-      D->getStorageDuration() != SD_Static && !D->getDescribedVarTemplate() &&
-      !D->getMemberSpecializationInfo() && !D->isObjCForDecl() &&
-      !isa<ImplicitParamDecl>(D) && !D->isEscapingByref())
+      !D->hasInitWithSideEffects() && !D->isEscapingByref() &&
+      !HasDeducedType && D->getStorageDuration() != SD_Static &&
+      !D->getDescribedVarTemplate() && !D->getMemberSpecializationInfo() &&
+      !D->isObjCForDecl() && !isa<ImplicitParamDecl>(D) &&
+      !D->isEscapingByref())
     AbbrevToUse = Writer.getDeclVarAbbrev();
 
   Code = serialization::DECL_VAR;
@@ -2529,12 +2531,12 @@ void ASTWriter::WriteDeclAbbrevs() {
   // VarDecl
   Abv->Add(BitCodeAbbrevOp(
       BitCodeAbbrevOp::Fixed,
-      21)); // Packed Var Decl bits:  Linkage, ModulesCodegen,
+      22)); // Packed Var Decl bits:  Linkage, ModulesCodegen,
             // SClass, TSCSpec, InitStyle,
             // isARCPseudoStrong, IsThisDeclarationADemotedDefinition,
             // isExceptionVariable, isNRVOVariable, isCXXForRangeDecl,
             // isInline, isInlineSpecified, isConstexpr,
-            // isInitCapture, isPrevDeclInSameScope,
+            // isInitCapture, isPrevDeclInSameScope, hasInitWithSideEffects,
             // EscapingByref, HasDeducedType, ImplicitParamKind, isObjCForDecl
   Abv->Add(BitCodeAbbrevOp(0));                         // VarKind (local enum)
   // Type Source Info

clang/test/Modules/var-init-side-effects.cpp

@@ -0,0 +1,37 @@
+// Tests referencing variable with initializer containing side effect across module boundary
+//
+// RUN: rm -rf %t
+// RUN: mkdir -p %t
+// RUN: split-file %s %t
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface %t/Foo.cppm \
+// RUN: -o %t/Foo.pcm
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface \
+// RUN: -fmodule-file=Foo=%t/Foo.pcm \
+// RUN: %t/Bar.cppm \
+// RUN: -o %t/Bar.pcm
+
+// RUN: %clang_cc1 -std=c++20 -emit-obj \
+// RUN: -main-file-name Bar.cppm \
+// RUN: -fmodule-file=Foo=%t/Foo.pcm \
+// RUN: -x pcm %t/Bar.pcm \
+// RUN: -o %t/Bar.o
+
+//--- Foo.cppm
+export module Foo;
+
+export {
+class S {};
+
+inline S s = S{};
+}
+
+//--- Bar.cppm
+export module Bar;
+import Foo;
+
+S bar() {
+  return s;
+}
+