#include #include #include #include #include #include #include #include "libclangmm/clangmm.h" #include "libclangmm/Utility.h" #include "utils.h" #include #include #include #include //#include // While indexing, we should refer to symbols by USR. When joining into the db, we can have optimized access. struct TypeDef; struct FuncDef; struct VarDef; /* template struct Id { uint64_t file_id; uint64_t local_id; Id() : file_id(0), local_id(0) {} // Needed for containers. Do not use directly. Id(uint64_t file_id, uint64_t local_id) : file_id(file_id), local_id(local_id) {} }; */ template struct LocalId { uint64_t local_id; LocalId() : local_id(0) {} // Needed for containers. Do not use directly. explicit LocalId(uint64_t local_id) : local_id(local_id) {} }; using TypeId = LocalId; using FuncId = LocalId; using VarId = LocalId; template struct Ref { LocalId id; clang::SourceLocation loc; Ref(LocalId id, clang::SourceLocation loc) : id(id), loc(loc) {} }; using TypeRef = Ref; using FuncRef = Ref; using VarRef = Ref; // NOTE: declaration is empty if there is no forward declaration! struct TypeDef { // General metadata. TypeId id; std::string usr; std::string short_name; std::string qualified_name; // While a class/type can technically have a separate declaration/definition, // it doesn't really happen in practice. The declaration never contains // comments or insightful information. The user always wants to jump from // the declaration to the definition - never the other way around like in // functions and (less often) variables. // // It's also difficult to identify a `class Foo;` statement with the clang // indexer API (it's doable using cursor AST traversal), so we don't bother // supporting the feature. std::optional definition; // If set, then this is the same underlying type as the given value (ie, this // type comes from a using or typedef statement). std::optional alias_of; // Immediate parent and immediate derived types. std::vector parents; std::vector derived; // Types, functions, and variables defined in this type. std::vector types; std::vector funcs; std::vector vars; // Every usage, useful for things like renames. std::vector all_uses; // Usages that a user is probably interested in. std::vector interesting_uses; TypeDef(TypeId id, const std::string& usr) : id(id), usr(usr) { assert(usr.size() > 0); //std::cout << "Creating type with usr " << usr << std::endl; } }; struct FuncDef { // General metadata. FuncId id; std::string usr; std::string short_name; std::string qualified_name; std::optional declaration; std::optional definition; // Type which declares this one (ie, it is a method) std::optional declaring_type; // Method this method overrides. std::optional base; // Methods which directly override this one. std::vector derived; // Local variables defined in this function. std::vector locals; // Functions which call this one. // TODO: Functions can get called outside of just functions - for example, // they can get called in static context (maybe redirect to main?) // or in class initializer list (redirect to class ctor?) // - Right now those usages will not get listed here (but they should be // inside of all_uses). std::vector callers; // Functions that this function calls. std::vector callees; // All usages. For interesting usages, see callees. std::vector all_uses; FuncDef(FuncId id, const std::string& usr) : id(id), usr(usr) { assert(usr.size() > 0); } }; struct VarDef { // General metadata. VarId id; std::string usr; std::string short_name; std::string qualified_name; std::optional declaration; // TODO: definitions should be a list of locations, since there can be more // than one. std::optional definition; // Type of the variable. std::optional variable_type; // Type which declares this one (ie, it is a method) std::optional declaring_type; // Usages. std::vector all_uses; VarDef(VarId id, const std::string& usr) : id(id), usr(usr) { assert(usr.size() > 0); } }; struct ParsingDatabase { // NOTE: Every Id is resolved to a file_id of 0. The correct file_id needs // to get fixed up when inserting into the real db. std::unordered_map usr_to_type_id; std::unordered_map usr_to_func_id; std::unordered_map usr_to_var_id; std::vector types; std::vector funcs; std::vector vars; ParsingDatabase(); TypeId ToTypeId(const std::string& usr); FuncId ToFuncId(const std::string& usr); VarId ToVarId(const std::string& usr); TypeId ToTypeId(const CXCursor& usr); FuncId ToFuncId(const CXCursor& usr); VarId ToVarId(const CXCursor& usr); TypeDef* Resolve(TypeId id); FuncDef* Resolve(FuncId id); VarDef* Resolve(VarId id); std::string ToString(); }; ParsingDatabase::ParsingDatabase() {} // TODO: Optimize for const char*? TypeId ParsingDatabase::ToTypeId(const std::string& usr) { auto it = usr_to_type_id.find(usr); if (it != usr_to_type_id.end()) return it->second; TypeId id(types.size()); types.push_back(TypeDef(id, usr)); usr_to_type_id[usr] = id; return id; } FuncId ParsingDatabase::ToFuncId(const std::string& usr) { auto it = usr_to_func_id.find(usr); if (it != usr_to_func_id.end()) return it->second; FuncId id(funcs.size()); funcs.push_back(FuncDef(id, usr)); usr_to_func_id[usr] = id; return id; } VarId ParsingDatabase::ToVarId(const std::string& usr) { auto it = usr_to_var_id.find(usr); if (it != usr_to_var_id.end()) return it->second; VarId id(vars.size()); vars.push_back(VarDef(id, usr)); usr_to_var_id[usr] = id; return id; } TypeId ParsingDatabase::ToTypeId(const CXCursor& cursor) { return ToTypeId(clang::Cursor(cursor).get_usr()); } FuncId ParsingDatabase::ToFuncId(const CXCursor& cursor) { return ToFuncId(clang::Cursor(cursor).get_usr()); } VarId ParsingDatabase::ToVarId(const CXCursor& cursor) { return ToVarId(clang::Cursor(cursor).get_usr()); } TypeDef* ParsingDatabase::Resolve(TypeId id) { return &types[id.local_id]; } FuncDef* ParsingDatabase::Resolve(FuncId id) { return &funcs[id.local_id]; } VarDef* ParsingDatabase::Resolve(VarId id) { return &vars[id.local_id]; } using Writer = rapidjson::PrettyWriter; void Write(Writer& writer, const char* key, clang::SourceLocation location) { if (key) writer.Key(key); std::string s = location.ToString(); writer.String(s.c_str()); } void Write(Writer& writer, const char* key, std::optional location) { if (location) { Write(writer, key, location.value()); } //else { // if (key) writer.Key(key); // writer.Null(); //} } void Write(Writer& writer, const char* key, const std::vector& locs) { if (locs.size() == 0) return; if (key) writer.Key(key); writer.StartArray(); for (const clang::SourceLocation& loc : locs) Write(writer, nullptr, loc); writer.EndArray(); } template void Write(Writer& writer, const char* key, LocalId id) { if (key) writer.Key(key); writer.Uint64(id.local_id); } template void Write(Writer& writer, const char* key, std::optional> id) { if (id) { Write(writer, key, id.value()); } //else { // if (key) writer.Key(key); // writer.Null(); //} } template void Write(Writer& writer, const char* key, const std::vector>& ids) { if (ids.size() == 0) return; if (key) writer.Key(key); writer.StartArray(); for (LocalId id : ids) Write(writer, nullptr, id); writer.EndArray(); } template void Write(Writer& writer, const char* key, Ref ref) { if (key) writer.Key(key); std::string s = std::to_string(ref.id.local_id) + "@" + ref.loc.ToString(); writer.String(s.c_str()); } template void Write(Writer& writer, const char* key, const std::vector>& refs) { if (refs.size() == 0) return; if (key) writer.Key(key); writer.StartArray(); for (Ref ref : refs) Write(writer, nullptr, ref); writer.EndArray(); } void Write(Writer& writer, const char* key, const std::string& value) { if (value.size() == 0) return; if (key) writer.Key(key); writer.String(value.c_str()); } void Write(Writer& writer, const char* key, uint64_t value) { if (key) writer.Key(key); writer.Uint64(value); } std::string ParsingDatabase::ToString() { auto it = usr_to_type_id.find(""); if (it != usr_to_type_id.end()) { Resolve(it->second)->short_name = ""; assert(Resolve(it->second)->all_uses.size() == 0); assert(Resolve(it->second)->interesting_uses.size() == 0); } #define WRITE(name) Write(writer, #name, def.name) rapidjson::StringBuffer output; rapidjson::PrettyWriter writer(output); writer.SetFormatOptions( rapidjson::PrettyFormatOptions::kFormatSingleLineArray); writer.SetIndent(' ', 2); writer.StartObject(); // Types writer.Key("types"); writer.StartArray(); for (TypeDef& def : types) { writer.StartObject(); WRITE(id); WRITE(usr); WRITE(short_name); WRITE(qualified_name); WRITE(definition); WRITE(alias_of); WRITE(parents); WRITE(derived); WRITE(types); WRITE(funcs); WRITE(vars); WRITE(all_uses); WRITE(interesting_uses); writer.EndObject(); } writer.EndArray(); // Functions writer.Key("functions"); writer.StartArray(); for (FuncDef& def : funcs) { writer.StartObject(); WRITE(id); WRITE(usr); WRITE(short_name); WRITE(qualified_name); WRITE(declaration); WRITE(definition); WRITE(declaring_type); WRITE(base); WRITE(derived); WRITE(locals); WRITE(callers); WRITE(callees); WRITE(all_uses); writer.EndObject(); } writer.EndArray(); // Variables writer.Key("variables"); writer.StartArray(); for (VarDef& def : vars) { writer.StartObject(); WRITE(id); WRITE(usr); WRITE(short_name); WRITE(qualified_name); WRITE(declaration); WRITE(definition); WRITE(variable_type); WRITE(declaring_type); //WRITE(initializations); WRITE(all_uses); writer.EndObject(); } writer.EndArray(); writer.EndObject(); return output.GetString(); #undef WRITE } struct FileDef { uint64_t id; std::string path; std::vector types; std::vector funcs; std::vector vars; }; /* struct Database { std::unordered_map usr_to_type_id; std::unordered_map usr_to_func_id; std::unordered_map usr_to_var_id; std::vector files; TypeId ToTypeId(const std::string& usr); FuncId ToFuncId(const std::string& usr); VarId ToVarId(const std::string& usr); }; TypeId Database::ToTypeId(const std::string& usr) { auto it = usr_to_type_id.find(usr); assert(it != usr_to_type_id.end() && "Usr is not registered"); return it->second; } FuncId Database::ToFuncId(const std::string& usr) { auto it = usr_to_func_id.find(usr); assert(it != usr_to_func_id.end() && "Usr is not registered"); return it->second; } VarId Database::ToVarId(const std::string& usr) { auto it = usr_to_var_id.find(usr); assert(it != usr_to_var_id.end() && "Usr is not registered"); return it->second; } TypeDef* Resolve(FileDef* file, TypeId id) { assert(file->id == id.file_id); return &file->types[id.local_id]; } FuncDef* Resolve(FileDef* file, FuncId id) { assert(file->id == id.file_id); return &file->funcs[id.local_id]; } VarDef* Resolve(FileDef* file, VarId id) { assert(file->id == id.file_id); return &file->vars[id.local_id]; } TypeDef* Resolve(Database* db, TypeId id) { return Resolve(&db->files[id.file_id], id); } FuncDef* Resolve(Database* db, FuncId id) { return Resolve(&db->files[id.file_id], id); } VarDef* Resolve(Database* db, VarId id) { return Resolve(&db->files[id.file_id], id); } */ int abortQuery(CXClientData client_data, void *reserved) { // 0 -> continue return 0; } void diagnostic(CXClientData client_data, CXDiagnosticSet, void *reserved) {} CXIdxClientFile enteredMainFile(CXClientData client_data, CXFile mainFile, void *reserved) { return nullptr; } CXIdxClientFile ppIncludedFile(CXClientData client_data, const CXIdxIncludedFileInfo *) { return nullptr; } CXIdxClientASTFile importedASTFile(CXClientData client_data, const CXIdxImportedASTFileInfo *) { return nullptr; } CXIdxClientContainer startedTranslationUnit(CXClientData client_data, void *reserved) { return nullptr; } clang::VisiterResult DumpVisitor(clang::Cursor cursor, clang::Cursor parent, int* level) { for (int i = 0; i < *level; ++i) std::cout << " "; std::cout << clang::ToString(cursor.get_kind()) << " " << cursor.get_spelling() << std::endl; *level += 1; cursor.VisitChildren(&DumpVisitor, level); *level -= 1; return clang::VisiterResult::Continue; } void Dump(clang::Cursor cursor) { int level = 0; cursor.VisitChildren(&DumpVisitor, &level); } struct FindChildOfKindParam { CXCursorKind target_kind; std::optional result; FindChildOfKindParam(CXCursorKind target_kind) : target_kind(target_kind) {} }; clang::VisiterResult FindChildOfKindVisitor(clang::Cursor cursor, clang::Cursor parent, FindChildOfKindParam* param) { if (cursor.get_kind() == param->target_kind) { param->result = cursor; return clang::VisiterResult::Break; } return clang::VisiterResult::Recurse; } std::optional FindChildOfKind(clang::Cursor cursor, CXCursorKind kind) { FindChildOfKindParam param(kind); cursor.VisitChildren(&FindChildOfKindVisitor, ¶m); return param.result; } clang::VisiterResult FindTypeVisitor(clang::Cursor cursor, clang::Cursor parent, std::optional* result) { switch (cursor.get_kind()) { case CXCursor_TypeRef: case CXCursor_TemplateRef: *result = cursor; return clang::VisiterResult::Break; } return clang::VisiterResult::Recurse; } std::optional FindType(clang::Cursor cursor) { std::optional result; cursor.VisitChildren(&FindTypeVisitor, &result); return result; } struct NamespaceHelper { std::unordered_map container_usr_to_qualified_name; void RegisterQualifiedName(std::string usr, const CXIdxContainerInfo* container, std::string qualified_name) { if (container) { std::string container_usr = clang::Cursor(container->cursor).get_usr(); auto it = container_usr_to_qualified_name.find(container_usr); if (it != container_usr_to_qualified_name.end()) { container_usr_to_qualified_name[usr] = it->second + qualified_name + "::"; return; } } container_usr_to_qualified_name[usr] = qualified_name + "::"; } std::string QualifiedName(const CXIdxContainerInfo* container, std::string unqualified_name) { if (container) { std::string container_usr = clang::Cursor(container->cursor).get_usr(); auto it = container_usr_to_qualified_name.find(container_usr); if (it != container_usr_to_qualified_name.end()) return it->second + unqualified_name; // Anonymous namespaces are not processed by indexDeclaration. If we // encounter one insert it into map. if (container->cursor.kind == CXCursor_Namespace) { assert(clang::Cursor(container->cursor).get_spelling() == ""); container_usr_to_qualified_name[container_usr] = "::"; return "::" + unqualified_name; } } return unqualified_name; } }; struct IndexParam { ParsingDatabase* db; NamespaceHelper* ns; // Record the last type usage location we recorded. Clang will sometimes // visit the same expression twice so we wan't to avoid double-reporting // usage information for those locations. clang::SourceLocation last_type_usage_location; clang::SourceLocation last_func_usage_location; IndexParam(ParsingDatabase* db, NamespaceHelper* ns) : db(db), ns(ns) {} }; /* std::string GetNamespacePrefx(const CXIdxDeclInfo* decl) { const CXIdxContainerInfo* container = decl->lexicalContainer; while (container) { } } */ bool HasUsage(const std::vector& usages, const clang::SourceLocation& usage) { for (int i = usages.size() - 1; i >= 0; --i) { if (usages[i] == usage) return true; } return false; } bool IsTypeDefinition(const CXIdxContainerInfo* container) { if (!container) return false; switch (container->cursor.kind) { case CXCursor_StructDecl: case CXCursor_ClassDecl: return true; default: return false; } } struct VisitDeclForTypeUsageParam { ParsingDatabase* db; bool is_interesting; int has_processed_any = false; std::optional previous_cursor; std::optional initial_type; VisitDeclForTypeUsageParam(ParsingDatabase* db, bool is_interesting) : db(db), is_interesting(is_interesting) {} }; void VisitDeclForTypeUsageVisitorHandler(clang::Cursor cursor, VisitDeclForTypeUsageParam* param) { param->has_processed_any = true; ParsingDatabase* db = param->db; TypeId ref_type_id = db->ToTypeId(cursor.get_referenced().get_usr()); if (!param->initial_type) param->initial_type = ref_type_id; if (param->is_interesting) { TypeDef* ref_type_def = db->Resolve(ref_type_id); ref_type_def->interesting_uses.push_back(cursor.get_source_location()); } } clang::VisiterResult VisitDeclForTypeUsageVisitor(clang::Cursor cursor, clang::Cursor parent, VisitDeclForTypeUsageParam* param) { switch (cursor.get_kind()) { case CXCursor_TemplateRef: case CXCursor_TypeRef: if (param->previous_cursor) { VisitDeclForTypeUsageVisitorHandler(param->previous_cursor.value(), param); // This if is inside the above if because if there are multiple TypeRefs, // we always want to process the first one. If we did not always process // the first one, we cannot tell if there are more TypeRefs after it and // logic for fetching the return type breaks. This happens in ParmDecl // instances which only have one TypeRef child but are not interesting // usages. if (!param->is_interesting) return clang::VisiterResult::Break; } param->previous_cursor = cursor; } return clang::VisiterResult::Continue; } std::optional ResolveDeclToType(ParsingDatabase* db, clang::Cursor decl_cursor, const CXIdxContainerInfo* semantic_container, const CXIdxContainerInfo* lexical_container, bool is_interesting) { // // The general AST format for definitions follows this pattern: // // template // struct Container; // // struct S1; // struct S2; // // Container, S2> foo; // // => // // VarDecl // TemplateRef Container // TemplateRef Container // TypeRef struct S1 // TypeRef struct S2 // TypeRef struct S2 // // We skip the last type reference for methods/variables which are defined // out-of-line w.r.t. the parent type. // // S1* Foo::foo() {} // // The above example looks like this in the AST: // // CXXMethod foo // TypeRef struct S1 // TypeRef class Foo // CompoundStmt // ... // // The second TypeRef is an uninteresting usage. bool process_last_type_ref = true; if (IsTypeDefinition(semantic_container) && !IsTypeDefinition(lexical_container)) { assert(decl_cursor.is_definition()); process_last_type_ref = false; } VisitDeclForTypeUsageParam param(db, is_interesting); decl_cursor.VisitChildren(&VisitDeclForTypeUsageVisitor, ¶m); // VisitDeclForTypeUsageVisitor guarantees that if there are multiple TypeRef // children, the first one will always be visited. if (param.previous_cursor && process_last_type_ref) { VisitDeclForTypeUsageVisitorHandler(param.previous_cursor.value(), ¶m); } else { // If we are not processing the last type ref, it *must* be a TypeRef (ie, // and not a TemplateRef). assert(!param.previous_cursor.has_value() || param.previous_cursor.value().get_kind() == CXCursor_TypeRef); } return param.initial_type; } void indexDeclaration(CXClientData client_data, const CXIdxDeclInfo* decl) { IndexParam* param = static_cast(client_data); ParsingDatabase* db = param->db; NamespaceHelper* ns = param->ns; switch (decl->entityInfo->kind) { case CXIdxEntity_CXXNamespace: { ns->RegisterQualifiedName(decl->entityInfo->USR, decl->semanticContainer, decl->entityInfo->name); break; } case CXIdxEntity_Field: case CXIdxEntity_Variable: case CXIdxEntity_CXXStaticVariable: { clang::Cursor decl_cursor = decl->cursor; VarId var_id = db->ToVarId(decl->entityInfo->USR); VarDef* var_def = db->Resolve(var_id); // TODO: Eventually run with this if. Right now I want to iron out bugs this may shadow. // TODO: Verify this gets called multiple times //if (!decl->isRedeclaration) { var_def->short_name = decl->entityInfo->name; var_def->qualified_name = ns->QualifiedName(decl->semanticContainer, var_def->short_name); //} if (decl->isDefinition) var_def->definition = decl->loc; else var_def->declaration = decl->loc; var_def->all_uses.push_back(decl->loc); std::optional var_type = ResolveDeclToType(db, decl_cursor, decl->semanticContainer, decl->lexicalContainer, decl_cursor.get_kind() != CXCursor_ParmDecl); if (var_type.has_value()) var_def->variable_type = var_type.value(); // Declaring variable type information. /* TypeResolution var_type = ResolveToType(db, decl_cursor.get_type()); if (var_type.resolved_type) { var_def->variable_type = var_type.resolved_type.value(); // Insert an interesting type usage for variable declarations. Parameters // are handled when a function is declared because clang doesn't provide // parameter declarations for unnamed parameters. if (decl_cursor.get_kind() != CXCursor_ParmDecl) AddInterestingUsageToType(db, var_type, FindLocationOfTypeSpecifier(decl_cursor)); } */ if (decl->isDefinition && IsTypeDefinition(decl->semanticContainer)) { TypeId declaring_type_id = db->ToTypeId(decl->semanticContainer->cursor); TypeDef* declaring_type_def = db->Resolve(declaring_type_id); var_def->declaring_type = declaring_type_id; declaring_type_def->vars.push_back(var_id); } // std::optional declaration; // std::vector initializations; // std::optional variable_type; // std::optional declaring_type; // std::vector uses; break; } case CXIdxEntity_Function: case CXIdxEntity_CXXConstructor: case CXIdxEntity_CXXDestructor: case CXIdxEntity_CXXInstanceMethod: case CXIdxEntity_CXXStaticMethod: { clang::Cursor decl_cursor = decl->cursor; FuncId func_id = db->ToFuncId(decl->entityInfo->USR); FuncDef* func_def = db->Resolve(func_id); // TODO: Eventually run with this if. Right now I want to iron out bugs this may shadow. //if (!decl->isRedeclaration) { func_def->short_name = decl->entityInfo->name; func_def->qualified_name = ns->QualifiedName(decl->semanticContainer, func_def->short_name); //} if (decl->isDefinition) func_def->definition = decl->loc; else func_def->declaration = decl->loc; func_def->all_uses.push_back(decl->loc); bool is_pure_virtual = clang_CXXMethod_isPureVirtual(decl->cursor); // Add function usage information. We only want to do it once per // definition/declaration. Do it on definition since there should only ever // be one of those in the entire program. if ((decl->isDefinition || is_pure_virtual) && IsTypeDefinition(decl->semanticContainer)) { TypeId declaring_type_id = db->ToTypeId(decl->semanticContainer->cursor); TypeDef* declaring_type_def = db->Resolve(declaring_type_id); func_def->declaring_type = declaring_type_id; declaring_type_def->funcs.push_back(func_id); } // We don't actually need to know the return type, but we need to mark it // as an interesting usage. ResolveDeclToType(db, decl_cursor, decl->semanticContainer, decl->lexicalContainer, true /*is_interesting*/); //TypeResolution ret_type = ResolveToType(db, decl_cursor.get_type().get_return_type()); //if (ret_type.resolved_type) // AddInterestingUsageToType(db, ret_type, FindLocationOfTypeSpecifier(decl_cursor)); if (decl->isDefinition || is_pure_virtual) { // Mark type usage for parameters as interesting. We handle this here // instead of inside var declaration because clang will not emit a var // declaration for an unnamed parameter, but we still want to mark the // usage as interesting. // TODO: Do a similar thing for function decl parameter usages. Mark // prototype params as interesting type usages but also relate mark // them as as usages on the primary variable - requires USR to be // the same. We can work around it by declaring which variables a // parameter has declared and update the USR in the definition. clang::Cursor cursor = decl->cursor; for (clang::Cursor arg : cursor.get_arguments()) { switch (arg.get_kind()) { case CXCursor_ParmDecl: // We don't need to know the arg type, but we do want to mark it as // an interesting usage. ResolveDeclToType(db, arg, decl->semanticContainer, decl->lexicalContainer, true /*is_interesting*/); //TypeResolution arg_type = ResolveToType(db, arg.get_type()); //if (arg_type.resolved_type) // AddInterestingUsageToType(db, arg_type, FindLocationOfTypeSpecifier(arg)); break; } } // Process inheritance. //void clang_getOverriddenCursors(CXCursor cursor, CXCursor **overridden, unsigned *num_overridden); //void clang_disposeOverriddenCursors(CXCursor *overridden); if (clang_CXXMethod_isVirtual(decl->cursor)) { CXCursor* overridden; unsigned int num_overridden; clang_getOverriddenCursors(decl->cursor, &overridden, &num_overridden); // TODO: How to handle multiple parent overrides?? for (unsigned int i = 0; i < num_overridden; ++i) { clang::Cursor parent = overridden[i]; FuncId parent_id = db->ToFuncId(parent.get_usr()); FuncDef* parent_def = db->Resolve(parent_id); func_def = db->Resolve(func_id); // ToFuncId invalidated func_def func_def->base = parent_id; parent_def->derived.push_back(func_id); } clang_disposeOverriddenCursors(overridden); } } /* std::optional base; std::vector derived; std::vector locals; std::vector callers; std::vector callees; std::vector uses; */ break; } case CXIdxEntity_Typedef: case CXIdxEntity_CXXTypeAlias: { TypeId type_id = db->ToTypeId(decl->entityInfo->USR); std::optional type_ref = FindChildOfKind(decl->cursor, CXCursor_TypeRef); assert(type_ref.has_value()); TypeId alias_of = db->ToTypeId(type_ref.value().get_referenced().get_usr()); TypeDef* type_def = db->Resolve(type_id); type_def->alias_of = alias_of; db->Resolve(alias_of)->interesting_uses.push_back(type_ref.value().get_source_location()); type_def->short_name = decl->entityInfo->name; type_def->qualified_name = ns->QualifiedName(decl->semanticContainer, type_def->short_name); type_def->definition = decl->loc; type_def->all_uses.push_back(decl->loc); break; } case CXIdxEntity_Struct: case CXIdxEntity_CXXClass: { ns->RegisterQualifiedName(decl->entityInfo->USR, decl->semanticContainer, decl->entityInfo->name); TypeId type_id = db->ToTypeId(decl->entityInfo->USR); TypeDef* type_def = db->Resolve(type_id); // TODO: Eventually run with this if. Right now I want to iron out bugs this may shadow. // TODO: For type section, verify if this ever runs for non definitions? //if (!decl->isRedeclaration) { type_def->short_name = decl->entityInfo->name; type_def->qualified_name = ns->QualifiedName(decl->semanticContainer, type_def->short_name); // } assert(decl->isDefinition); type_def->definition = decl->loc; type_def->all_uses.push_back(decl->loc); //type_def->alias_of //type_def->funcs //type_def->types //type_def->uses //type_def->vars // Add type-level inheritance information. CXIdxCXXClassDeclInfo const* class_info = clang_index_getCXXClassDeclInfo(decl); for (unsigned int i = 0; i < class_info->numBases; ++i) { const CXIdxBaseClassInfo* base_class = class_info->bases[i]; TypeId parent_type_id = db->ToTypeId(clang::Cursor(base_class->cursor).get_referenced().get_usr()); TypeDef* parent_type_def = db->Resolve(parent_type_id); TypeDef* type_def = db->Resolve(type_id); // type_def ptr could be invalidated by ToTypeId. parent_type_def->derived.push_back(type_id); type_def->parents.push_back(parent_type_id); } break; } default: std::cout << "!! Unhandled indexDeclaration: " << clang::Cursor(decl->cursor).ToString() << " at " << clang::SourceLocation(decl->loc).ToString() << std::endl; std::cout << " entityInfo->kind = " << decl->entityInfo->kind << std::endl; std::cout << " entityInfo->USR = " << decl->entityInfo->USR << std::endl; if (decl->declAsContainer) std::cout << " declAsContainer = " << clang::Cursor(decl->declAsContainer->cursor).ToString() << std::endl; if (decl->semanticContainer) std::cout << " semanticContainer = " << clang::Cursor(decl->semanticContainer->cursor).ToString() << std::endl; if (decl->lexicalContainer) std::cout << " lexicalContainer = " << clang::Cursor(decl->lexicalContainer->cursor).get_usr() << std::endl; break; } } bool IsFunction(CXCursorKind kind) { switch (kind) { case CXCursor_CXXMethod: case CXCursor_FunctionDecl: return true; } return false; } void indexEntityReference(CXClientData client_data, const CXIdxEntityRefInfo* ref) { IndexParam* param = static_cast(client_data); ParsingDatabase* db = param->db; clang::Cursor cursor(ref->cursor); // TODO: Index entity call/ctor creation, like Foo().x = 3 switch (ref->referencedEntity->kind) { case CXIdxEntity_CXXStaticVariable: case CXIdxEntity_Variable: case CXIdxEntity_Field: { VarId var_id = db->ToVarId(ref->referencedEntity->cursor); VarDef* var_def = db->Resolve(var_id); var_def->all_uses.push_back(ref->loc); break; } case CXIdxEntity_CXXStaticMethod: case CXIdxEntity_CXXInstanceMethod: case CXIdxEntity_Function: case CXIdxEntity_CXXConstructor: { // TODO: Redirect container to constructor for // int Gen() { return 5; } // class Foo { // int x = Gen(); // } // Don't report duplicate usages. // TODO: search full history? clang::SourceLocation loc = ref->loc; if (param->last_func_usage_location == loc) break; param->last_func_usage_location = loc; // Note: be careful, calling db->ToFuncId invalidates the FuncDef* ptrs. FuncId called_id = db->ToFuncId(ref->referencedEntity->USR); if (IsFunction(ref->container->cursor.kind)) { FuncId caller_id = db->ToFuncId(ref->container->cursor); FuncDef* caller_def = db->Resolve(caller_id); FuncDef* called_def = db->Resolve(called_id); caller_def->callees.push_back(FuncRef(called_id, loc)); called_def->callers.push_back(FuncRef(caller_id, loc)); called_def->all_uses.push_back(loc); } else { FuncDef* called_def = db->Resolve(called_id); called_def->all_uses.push_back(loc); } break; } case CXIdxEntity_Typedef: case CXIdxEntity_CXXTypeAlias: case CXIdxEntity_Struct: case CXIdxEntity_CXXClass: { TypeId referenced_id = db->ToTypeId(ref->referencedEntity->USR); TypeDef* referenced_def = db->Resolve(referenced_id); // // The following will generate two TypeRefs to Foo, both located at the // same spot (line 3, column 3). One of the parents will be set to // CXIdxEntity_Variable, the other will be CXIdxEntity_Function. There does // not appear to be a good way to disambiguate these references, as using // parent type alone breaks other indexing tasks. // // To work around this, we check to see if the usage location has been // inserted into all_uses previously. // // struct Foo {}; // void Make() { // Foo f; // } // clang::SourceLocation loc = ref->loc; if (!HasUsage(referenced_def->all_uses, loc)) referenced_def->all_uses.push_back(loc); break; } default: std::cout << "!! Unhandled indexEntityReference: " << cursor.ToString() << " at " << clang::SourceLocation(ref->loc).ToString() << std::endl; std::cout << " ref->referencedEntity->kind = " << ref->referencedEntity->kind << std::endl; if (ref->parentEntity) std::cout << " ref->parentEntity->kind = " << ref->parentEntity->kind << std::endl; std::cout << " ref->loc = " << clang::SourceLocation(ref->loc).ToString() << std::endl; std::cout << " ref->kind = " << ref->kind << std::endl; if (ref->parentEntity) std::cout << " parentEntity = " << clang::Cursor(ref->parentEntity->cursor).ToString() << std::endl; if (ref->referencedEntity) std::cout << " referencedEntity = " << clang::Cursor(ref->referencedEntity->cursor).ToString() << std::endl; if (ref->container) std::cout << " container = " << clang::Cursor(ref->container->cursor).ToString() << std::endl; break; } } ParsingDatabase Parse(std::string filename) { std::vector args; clang::Index index(0 /*excludeDeclarationsFromPCH*/, 0 /*displayDiagnostics*/); clang::TranslationUnit tu(index, filename, args); //Dump(tu.document_cursor()); CXIndexAction index_action = clang_IndexAction_create(index.cx_index); IndexerCallbacks callbacks[] = { { &abortQuery, &diagnostic, &enteredMainFile, &ppIncludedFile, &importedASTFile, &startedTranslationUnit, &indexDeclaration, &indexEntityReference } /* callbacks.abortQuery = &abortQuery; callbacks.diagnostic = &diagnostic; callbacks.enteredMainFile = &enteredMainFile; callbacks.ppIncludedFile = &ppIncludedFile; callbacks.importedASTFile = &importedASTFile; callbacks.startedTranslationUnit = &startedTranslationUnit; callbacks.indexDeclaration = &indexDeclaration; callbacks.indexEntityReference = &indexEntityReference; */ }; ParsingDatabase db; NamespaceHelper ns; IndexParam param(&db, &ns); clang_indexTranslationUnit(index_action, ¶m, callbacks, sizeof(callbacks), CXIndexOpt_IndexFunctionLocalSymbols, tu.cx_tu); clang_IndexAction_dispose(index_action); return db; } template bool AreEqual(const std::vector& a, const std::vector& b) { if (a.size() != b.size()) return false; for (int i = 0; i < a.size(); ++i) { if (a[i] != b[i]) return false; } return true; } void Write(const std::vector& strs) { for (const std::string& str : strs) { std::cout << str << std::endl; } } std::string ToString(const rapidjson::Document& document) { rapidjson::StringBuffer buffer; rapidjson::PrettyWriter writer(buffer); writer.SetFormatOptions( rapidjson::PrettyFormatOptions::kFormatSingleLineArray); writer.SetIndent(' ', 2); buffer.Clear(); document.Accept(writer); return buffer.GetString(); } std::vector split_string(const std::string& str, const std::string& delimiter) { // http://stackoverflow.com/a/13172514 std::vector strings; std::string::size_type pos = 0; std::string::size_type prev = 0; while ((pos = str.find(delimiter, prev)) != std::string::npos) { strings.push_back(str.substr(prev, pos - prev)); prev = pos + 1; } // To get the last substring (or only, if delimiter is not found) strings.push_back(str.substr(prev)); return strings; } void DiffDocuments(rapidjson::Document& expected, rapidjson::Document& actual) { std::vector actual_output; { std::string buffer = ToString(actual); actual_output = split_string(buffer, "\n"); } std::vector expected_output; { std::string buffer = ToString(expected); expected_output = split_string(buffer, "\n"); } int len = std::min(actual_output.size(), expected_output.size()); for (int i = 0; i < len; ++i) { if (actual_output[i] != expected_output[i]) { std::cout << "Line " << i << " differs:" << std::endl; std::cout << " expected: " << expected_output[i] << std::endl; std::cout << " actual: " << actual_output[i] << std::endl; } } if (actual_output.size() > len) { std::cout << "Additional output in actual:" << std::endl; for (int i = len; i < actual_output.size(); ++i) std::cout << " " << actual_output[i] << std::endl; } if (expected_output.size() > len) { std::cout << "Additional output in expected:" << std::endl; for (int i = len; i < expected_output.size(); ++i) std::cout << " " << expected_output[i] << std::endl; } } int main(int argc, char** argv) { /* ParsingDatabase db = Parse("tests/vars/function_local.cc"); std::cout << std::endl << "== Database ==" << std::endl; std::cout << db.ToString(); std::cin.get(); return 0; */ for (std::string path : GetFilesInFolder("tests")) { //if (path != "tests/declaration_vs_definition/class_member_static.cc") continue; //if (path != "tests/usage/type_usage_declare_param.cc") continue; //if (path == "tests/constructors/constructor.cc") continue; //if (path == "tests/constructors/destructor.cc") continue; //if (path == "tests/usage/func_usage_call_method.cc") continue; //if (path != "tests/usage/type_usage_as_template_parameter.cc") continue; //if (path != "tests/usage/type_usage_as_template_parameter_complex.cc") continue; //if (path != "tests/usage/type_usage_as_template_parameter_simple.cc") continue; //if (path != "tests/usage/type_usage_typedef_and_using.cc") continue; //if (path != "tests/usage/type_usage_declare_local.cc") continue; //if (path != "tests/usage/func_usage_addr_method.cc") continue; //if (path != "tests/usage/func_usage_template_func.cc") continue; //if (path != "tests/usage/usage_inside_of_call.cc") continue; //if (path != "tests/foobar.cc") continue; // Parse expected output from the test, parse it into JSON document. std::string expected_output; ParseTestExpectation(path, &expected_output); rapidjson::Document expected; expected.Parse(expected_output.c_str()); // Run test. std::cout << "[START] " << path << std::endl; ParsingDatabase db = Parse(path); std::string actual_output = db.ToString(); rapidjson::Document actual; actual.Parse(actual_output.c_str()); if (actual == expected) { std::cout << "[PASSED] " << path << std::endl; } else { std::cout << "[FAILED] " << path << std::endl; std::cout << "Expected output for " << path << ":" << std::endl; std::cout << expected_output; std::cout << "Actual output for " << path << ":" << std::endl; std::cout << actual_output; std::cout << std::endl; std::cout << std::endl; DiffDocuments(expected, actual); break; } } std::cin.get(); return 0; } // TODO: ctor/dtor, copy ctor