Introduce special decl names

Special DeclNames represent names that do not have an identifier in the
surface language. This implies serializing the information about whether
a name is special together with its identifier (if it is not special)
in both the module file and the swift lookup table.

ValueDecls are not guaranteed to have a name anymore and thus getNameStr
is moved from ValueDecl to subclasses that guarantee to be backed by
identifiers. Other calls need to handle special names if the context
doesn't guarantee that the declaration's name is non-special.

Author: ahoppen

include/swift/AST/Decl.h

@@ -2104,13 +2104,6 @@ class ValueDecl : public Decl {
   bool hasName() const { return bool(Name); }
   bool isOperator() const { return Name.isOperator(); }
 
-  /// Returns the string for the base name, or "_" if this is unnamed.
-  StringRef getNameStr() const {
-    // TODO: Check if this function is called for special names
-    assert(!Name.isSpecial() && "Cannot get string for special names");
-    return hasName() ? Name.getBaseName().getIdentifier().str() : "_";
-  }
-
   /// Retrieve the full name of the declaration.
   /// TODO: Rename to getName?
   DeclName getFullName() const { return Name; }
@@ -2329,6 +2322,12 @@ class TypeDecl : public ValueDecl {
 public:
   Identifier getName() const { return getFullName().getBaseIdentifier(); }
 
+  /// Returns the string for the base name, or "_" if this is unnamed.
+  StringRef getNameStr() const {
+    assert(!getFullName().isSpecial() && "Cannot get string for special names");
+    return hasName() ? getBaseName().getIdentifier().str() : "_";
+  }
+
   /// The type of this declaration's values. For the type of the
   /// declaration itself, use getInterfaceType(), which returns a
   /// metatype.
@@ -4215,6 +4214,12 @@ class VarDecl : public AbstractStorageDecl {
 
   Identifier getName() const { return getFullName().getBaseIdentifier(); }
 
+  /// Returns the string for the base name, or "_" if this is unnamed.
+  StringRef getNameStr() const {
+    assert(!getFullName().isSpecial() && "Cannot get string for special names");
+    return hasName() ? getBaseName().getIdentifier().str() : "_";
+  }
+
   TypeLoc &getTypeLoc() { return typeLoc; }
   TypeLoc getTypeLoc() const { return typeLoc; }
 
@@ -4684,6 +4689,12 @@ class AbstractFunctionDecl : public ValueDecl, public GenericContext {
 public:
   Identifier getName() const { return getFullName().getBaseIdentifier(); }
 
+  /// Returns the string for the base name, or "_" if this is unnamed.
+  StringRef getNameStr() const {
+    assert(!getFullName().isSpecial() && "Cannot get string for special names");
+    return hasName() ? getBaseName().getIdentifier().str() : "_";
+  }
+
   /// \brief Should this declaration be treated as if annotated with transparent
   /// attribute.
   bool isTransparent() const;
@@ -5307,6 +5318,12 @@ class EnumElementDecl : public ValueDecl {
 
   Identifier getName() const { return getFullName().getBaseIdentifier(); }
 
+  /// Returns the string for the base name, or "_" if this is unnamed.
+  StringRef getNameStr() const {
+    assert(!getFullName().isSpecial() && "Cannot get string for special names");
+    return hasName() ? getBaseName().getIdentifier().str() : "_";
+  }
+
   /// \returns false if there was an error during the computation rendering the
   /// EnumElementDecl invalid, true otherwise.
   bool computeType();

include/swift/AST/Identifier.h

@@ -51,6 +51,8 @@ enum class DeclRefKind {
 /// ASTContext.  It just wraps a nul-terminated "const char*".
 class Identifier {
   friend class ASTContext;
+  friend class DeclBaseName;
+
   const char *Pointer;
 
   /// Constructor, only accessible by ASTContext, which handles the uniquing.
@@ -206,15 +208,45 @@ namespace swift {
 
 /// Wrapper that may either be an Identifier or a special name
 /// (e.g. for subscripts)
+///
+/// The raw values of this enum are used in serialization by both
+/// BaseNameToEntitiesTable(Reader|Writer)Info and DeclTableInfo. For every
+/// reordering, deletion or insertion of any case, the major (resp. minor)
+/// version of both SwiftLookupTable AND ModuleFormat must thus be incremented.
 class DeclBaseName {
+public:
+  enum class Kind : uint8_t {
+    Normal = 0,
+    Subscript
+  };
+  
+private:
+  /// In a special DeclName represenenting a subscript, this opaque pointer
+  /// is used as the data of the base name identifier.
+  /// This is an implementation detail that should never leak outside of
+  /// DeclName.
+  static void *SubscriptIdentifierData;
+
   Identifier Ident;
 
 public:
   DeclBaseName() : DeclBaseName(Identifier()) {}
 
   DeclBaseName(Identifier I) : Ident(I) {}
 
-  bool isSpecial() const { return false; }
+  static DeclBaseName createSubscript() {
+    return DeclBaseName(Identifier((const char *)SubscriptIdentifierData));
+  }
+
+  Kind getKind() const {
+    if (Ident.get() == SubscriptIdentifierData) {
+      return Kind::Subscript;
+    } else {
+      return Kind::Normal;
+    }
+  }
+
+  bool isSpecial() const { return getKind() != Kind::Normal; }
 
   /// Return the identifier backing the name. Assumes that the name is not
   /// special.
@@ -233,9 +265,19 @@ class DeclBaseName {
     return !isSpecial() && getIdentifier().isEditorPlaceholder();
   }
 
+  /// A representation of the name to be displayed to users. May be ambiguous
+  /// between identifiers and special names
+  StringRef userFacingStr() const {
+    switch (getKind()) {
+    case Kind::Normal:
+      return getIdentifier().str();
+    case Kind::Subscript:
+      return "subscript";
+    }
+  }
+
   int compare(DeclBaseName other) const {
-    // TODO: Sort special names cleverly
-    return getIdentifier().compare(other.getIdentifier());
+    return userFacingStr().compare(other.userFacingStr());
   }
 
   bool operator==(StringRef Str) const {
@@ -253,7 +295,6 @@ class DeclBaseName {
   const void *getAsOpaquePointer() const { return Ident.get(); }
 
   static DeclBaseName getFromOpaquePointer(void *P) {
-    // TODO: Check if P is a special name
     return Identifier::getFromOpaquePointer(P);
   }
 
@@ -387,7 +428,6 @@ class DeclName {
     auto baseName = getBaseName();
     assert(!baseName.isSpecial() &&
            "Can't retrieve the identifier of a special base name");
-    // FIXME: Do we need this method?
     return baseName.getIdentifier();
   }
 

include/swift/Serialization/ModuleFile.h

@@ -737,7 +737,11 @@ class ModuleFile : public LazyMemberLoader {
   /// Returns the type with the given ID, deserializing it if needed.
   Type getType(serialization::TypeID TID);
 
-  /// Returns the identifier with the given ID, deserializing it if needed.
+  /// Returns the base name with the given ID, deserializing it if needed.
+  DeclBaseName getDeclBaseName(serialization::IdentifierID IID);
+
+  /// Convenience method to retrieve the identifier backing the name with
+  /// given ID. Asserts that the name with this ID is not special.
   Identifier getIdentifier(serialization::IdentifierID IID);
 
   /// Returns the decl with the given ID, deserializing it if needed.

include/swift/Serialization/ModuleFormat.h

@@ -54,7 +54,7 @@ const uint16_t VERSION_MAJOR = 0;
 /// in source control, you should also update the comment to briefly
 /// describe what change you made. The content of this comment isn't important;
 /// it just ensures a conflict if two people change the module format.
-const uint16_t VERSION_MINOR = 325; // Last change: unchecked_ownership_conver
+const uint16_t VERSION_MINOR = 326; // Last change: special decl names
 
 using DeclID = PointerEmbeddedInt<unsigned, 31>;
 using DeclIDField = BCFixed<31>;
@@ -344,6 +344,18 @@ enum SpecialModuleID : uint8_t {
   NUM_SPECIAL_MODULES
 };
 
+// Continuation of the special module IDs above for special names that don't
+// have an identifier. As above, these IDs must \em not be renumbered or
+// reordered without incrementing VERSION_MAJOR.
+enum SpecialNameID : uint8_t {
+  SUBSCRIPT_ID = NUM_SPECIAL_MODULES,
+
+  /// The number of special modules and IDs. This value should never be encoded;
+  /// it should only be used to count the number of names above. As such, it
+  /// is correct and necessary to add new values above this one.
+  NUM_SPECIAL_IDS
+};
+
 // These IDs must \em not be renumbered or reordered without incrementing
 // VERSION_MAJOR.
 enum class EnumElementRawValueKind : uint8_t {

lib/AST/ASTMangler.cpp

@@ -391,8 +391,15 @@ void ASTMangler::appendDeclName(const ValueDecl *decl) {
         break;
     }
   } else {
-    // TODO: Handle special names
-    appendIdentifier(decl->getBaseName().getIdentifier().str());
+    // FIXME: Should a mangled subscript name contain the string "substring"?
+    switch (decl->getBaseName().getKind()) {
+    case DeclBaseName::Kind::Normal:
+      appendIdentifier(decl->getBaseName().getIdentifier().str());
+      break;
+    case DeclBaseName::Kind::Subscript:
+      appendIdentifier("subscript");
+      break;
+    }
   }
 
   if (decl->getDeclContext()->isLocalContext()) {

lib/AST/ASTVerifier.cpp

@@ -2933,7 +2933,8 @@ bool swift::shouldVerify(const Decl *D, const ASTContext &Context) {
     return true;
   }
 
-  size_t Hash = llvm::hash_value(VD->getNameStr());
+  size_t Hash =
+      llvm::DenseMapInfo<DeclBaseName>::getHashValue(VD->getBaseName());
   return Hash % ProcessCount == ProcessId;
 #else
   return false;

lib/AST/Decl.cpp

@@ -449,7 +449,8 @@ bool Decl::isPrivateStdlibDecl(bool whitelistProtocols) const {
     return false;
 
   // If the name has leading underscore then it's a private symbol.
-  if (VD->getNameStr().startswith("_"))
+  if (!VD->getBaseName().isSpecial() &&
+      VD->getBaseName().getIdentifier().str().startswith("_"))
     return true;
 
   return false;

lib/AST/DiagnosticEngine.cpp

@@ -700,7 +700,7 @@ void DiagnosticEngine::emitDiagnostic(const Diagnostic &diagnostic) {
           }
 
           if (auto value = dyn_cast<ValueDecl>(ppDecl)) {
-            bufferName += value->getNameStr();
+            bufferName += value->getBaseName().userFacingStr();
           } else if (auto ext = dyn_cast<ExtensionDecl>(ppDecl)) {
             bufferName += ext->getExtendedType().getString();
           }

lib/AST/Identifier.cpp

@@ -20,6 +20,8 @@
 #include "llvm/Support/ConvertUTF.h"
 using namespace swift;
 
+void *DeclBaseName::SubscriptIdentifierData =
+    &DeclBaseName::SubscriptIdentifierData;
 
 raw_ostream &llvm::operator<<(raw_ostream &OS, Identifier I) {
   if (I.get() == nullptr)
@@ -28,8 +30,9 @@ raw_ostream &llvm::operator<<(raw_ostream &OS, Identifier I) {
 }
 
 raw_ostream &llvm::operator<<(raw_ostream &OS, DeclBaseName I) {
-  // TODO: Handle special names
-  return OS << I.getIdentifier();
+  if (I.empty())
+    return OS << "_";
+  return OS << I.userFacingStr();
 }
 
 raw_ostream &llvm::operator<<(raw_ostream &OS, DeclName I) {

lib/AST/Mangle.cpp

@@ -105,9 +105,17 @@ void Mangler::mangleIdentifier(Identifier ident, OperatorFixity fixity) {
 
 /// Mangle an identifier into the buffer.
 void Mangler::mangleIdentifier(DeclBaseName ident, OperatorFixity fixity) {
-  // TODO: Handle special names
-  StringRef str = ident.getIdentifier().str();
-  assert(!str.empty() && "mangling an empty identifier!");
+  StringRef str;
+  // FIXME: Should a serialized subscript name contain the string "substring"?
+  switch (ident.getKind()) {
+  case DeclBaseName::Kind::Normal:
+    str = ident.getIdentifier().str();
+    break;
+  case DeclBaseName::Kind::Subscript:
+    str = "subscript";
+    break;
+  }
+  assert(!str.empty() && "mangling an empty name!");
   return mangleIdentifier(str, fixity, ident.isOperator());
 }
 

lib/AST/NameLookup.cpp

@@ -138,9 +138,9 @@ bool swift::removeShadowedDecls(SmallVectorImpl<ValueDecl*> &decls,
                                 const ModuleDecl *curModule,
                                 LazyResolver *typeResolver) {
   // Category declarations by their signatures.
-  llvm::SmallDenseMap<std::pair<CanType, Identifier>,
+  llvm::SmallDenseMap<std::pair<CanType, DeclBaseName>,
                       llvm::TinyPtrVector<ValueDecl *>>
-    CollidingDeclGroups;
+      CollidingDeclGroups;
 
   /// Objective-C initializers are tracked by their context type and
   /// full name.
@@ -172,8 +172,8 @@ bool swift::removeShadowedDecls(SmallVectorImpl<ValueDecl*> &decls,
       signature = asd->getOverloadSignature().InterfaceType;
 
     // If we've seen a declaration with this signature before, note it.
-    auto &knownDecls = CollidingDeclGroups[std::make_pair(
-        signature, decl->getBaseName().getIdentifier())];
+    auto &knownDecls =
+        CollidingDeclGroups[std::make_pair(signature, decl->getBaseName())];
     if (!knownDecls.empty())
       anyCollisions = true;
 

lib/AST/SourceEntityWalker.cpp

@@ -93,9 +93,9 @@ bool SemaAnnotator::walkToDeclPre(Decl *D) {
   unsigned NameLen = 0;
 
   if (ValueDecl *VD = dyn_cast<ValueDecl>(D)) {
-    // TODO: Handle special names
-    if (VD->hasName())
-      NameLen = VD->getBaseName().getIdentifier().getLength();
+    if (VD->hasName()) {
+      NameLen = VD->getBaseName().userFacingStr().size();
+    }
 
   } else if (ExtensionDecl *ED = dyn_cast<ExtensionDecl>(D)) {
     SourceRange SR = ED->getExtendedTypeLoc().getSourceRange();

lib/AST/USRGeneration.cpp

@@ -195,7 +195,8 @@ bool ide::printDeclUSR(const ValueDecl *D, raw_ostream &OS) {
     auto &Importer = *D->getASTContext().getClangModuleLoader();
 
     auto ClangMacroInfo = ClangN.getAsMacro();
-    bool Ignore = clang::index::generateUSRForMacro(D->getNameStr(),
+    bool Ignore = clang::index::generateUSRForMacro(
+        D->getBaseName().getIdentifier().str(),
         ClangMacroInfo->getDefinitionLoc(),
         Importer.getClangASTContext().getSourceManager(), Buf);
     if (!Ignore)

lib/ClangImporter/ClangImporter.cpp

@@ -1802,7 +1802,9 @@ class DarwinBlacklistDeclConsumer : public swift::VisibleDeclConsumer {
       return false;
 
     if (clangModule->Name == "MacTypes") {
-      return llvm::StringSwitch<bool>(VD->getNameStr())
+      if (!VD->hasName() || VD->getBaseName().isSpecial())
+        return true;
+      return llvm::StringSwitch<bool>(VD->getBaseName().getIdentifier().str())
           .Cases("OSErr", "OSStatus", "OptionBits", false)
           .Cases("FourCharCode", "OSType", false)
           .Case("Boolean", false)
@@ -2706,7 +2708,7 @@ void ClangImporter::Implementation::lookupValue(
   auto &clangCtx = getClangASTContext();
   auto clangTU = clangCtx.getTranslationUnitDecl();
 
-  for (auto entry : table.lookup(name.getBaseIdentifier().str(), clangTU)) {
+  for (auto entry : table.lookup(name.getBaseName(), clangTU)) {
     // If the entry is not visible, skip it.
     if (!isVisibleClangEntry(clangCtx, entry)) continue;
 
@@ -2810,7 +2812,7 @@ void ClangImporter::Implementation::lookupVisibleDecls(
 
   // Look for namespace-scope entities with each base name.
   for (auto baseName : baseNames) {
-    lookupValue(table, SwiftContext.getIdentifier(baseName), consumer);
+    lookupValue(table, baseName.toDeclBaseName(SwiftContext), consumer);
   }
 }
 
@@ -2819,9 +2821,8 @@ void ClangImporter::Implementation::lookupObjCMembers(
        DeclName name,
        VisibleDeclConsumer &consumer) {
   auto &clangCtx = getClangASTContext();
-  auto baseName = name.getBaseIdentifier().str();
 
-  for (auto clangDecl : table.lookupObjCMembers(baseName)) {
+  for (auto clangDecl : table.lookupObjCMembers(name.getBaseName())) {
     // If the entry is not visible, skip it.
     if (!isVisibleClangEntry(clangCtx, clangDecl)) continue;
 
@@ -2868,7 +2869,7 @@ void ClangImporter::Implementation::lookupAllObjCMembers(
 
   // Look for Objective-C members with each base name.
   for (auto baseName : baseNames) {
-    lookupObjCMembers(table, SwiftContext.getIdentifier(baseName), consumer);
+    lookupObjCMembers(table, baseName.toDeclBaseName(SwiftContext), consumer);
   }
 }