#include "indexer.h" #include "clang_cursor.h" #include "clang_utils.h" #include "platform.h" #include "serializer.h" #include "timer.h" #include #include #include #include // TODO: See if we can use clang_indexLoc_getFileLocation to get a type ref on // |Foobar| in DISALLOW_COPY(Foobar) namespace { constexpr bool kIndexStdDeclarations = true; // For typedef/using spanning less than or equal to (this number) of lines, // display their declarations on hover. constexpr int kMaxLinesDisplayTypeAliasDeclarations = 3; void AddFuncRef(std::vector* result, IndexFuncRef ref) { if (!result->empty() && (*result)[result->size() - 1] == ref) return; result->push_back(ref); } Range Resolve(const CXSourceRange& range, CXFile* cx_file = nullptr) { CXSourceLocation start = clang_getRangeStart(range); CXSourceLocation end = clang_getRangeEnd(range); unsigned int start_line, start_column; clang_getSpellingLocation(start, cx_file, &start_line, &start_column, nullptr); unsigned int end_line, end_column; clang_getSpellingLocation(end, nullptr, &end_line, &end_column, nullptr); return Range(Position((int16_t)start_line, (int16_t)start_column) /*start*/, Position((int16_t)end_line, (int16_t)end_column) /*end*/); } Range ResolveSpelling(const CXCursor& cx_cursor, CXFile* cx_file = nullptr) { CXSourceRange cx_range = clang_Cursor_getSpellingNameRange(cx_cursor, 0, 0); return Resolve(cx_range, cx_file); } Range ResolveExtent(const CXCursor& cx_cursor, CXFile* cx_file = nullptr) { CXSourceRange cx_range = clang_getCursorExtent(cx_cursor); return Resolve(cx_range, cx_file); } bool IsScopeSemanticContainer(CXCursorKind kind) { switch (kind) { case CXCursor_Namespace: case CXCursor_TranslationUnit: case CXCursor_StructDecl: case CXCursor_UnionDecl: case CXCursor_ClassDecl: case CXCursor_EnumDecl: // TODO Add more Objective-C containers case CXCursor_ObjCInterfaceDecl: case CXCursor_ObjCImplementationDecl: return false; default: return true; } } struct NamespaceHelper { std::unordered_map container_cursor_to_qualified_name; void RegisterQualifiedName(std::string usr, const CXIdxContainerInfo* container, std::string qualified_name) {} std::string QualifiedName(const CXIdxContainerInfo* container, std::string unqualified_name) { if (!container) return unqualified_name; // Anonymous namespaces are not processed by indexDeclaration. We trace // nested namespaces bottom-up through clang_getCursorSemanticParent until // one that we know its qualified name. Then do another trace top-down and // put their names into a map of USR -> qualified_name. ClangCursor cursor = container->cursor; std::vector namespaces; std::string qualifier; while (cursor.get_kind() != CXCursor_TranslationUnit && !IsScopeSemanticContainer(cursor.get_kind())) { auto it = container_cursor_to_qualified_name.find(cursor); if (it != container_cursor_to_qualified_name.end()) { qualifier = it->second; break; } namespaces.push_back(cursor); cursor = clang_getCursorSemanticParent(cursor.cx_cursor); } for (size_t i = namespaces.size(); i > 0;) { i--; std::string name = namespaces[i].get_spelling(); // Empty name indicates unnamed namespace, anonymous struct, anonymous // union, ... if (name.size()) qualifier += name; else switch (namespaces[i].get_kind()) { case CXCursor_ClassDecl: qualifier += "(anon class)"; break; case CXCursor_EnumDecl: qualifier += "(anon enum)"; break; case CXCursor_StructDecl: qualifier += "(anon struct)"; break; case CXCursor_UnionDecl: qualifier += "(anon union)"; break; default: qualifier += "(anon)"; break; } qualifier += "::"; container_cursor_to_qualified_name[namespaces[i]] = qualifier; } return qualifier + unqualified_name; } }; // Caches all instances of constructors, regardless if they are indexed or not. // The constructor may have a make_unique call associated with it that we need // to export. If we do not capture the parameter type description for the // constructor we will not be able to attribute the constructor call correctly. struct ConstructorCache { using Usr = std::string; struct Constructor { Usr usr; std::vector param_type_desc; }; std::unordered_map> constructors_; // This should be called whenever there is a constructor declaration. void NotifyConstructor(ClangCursor ctor_cursor) { auto build_type_desc = [](ClangCursor cursor) { std::vector type_desc; for (ClangCursor arg : cursor.get_arguments()) { if (arg.get_kind() == CXCursor_ParmDecl) type_desc.push_back(arg.get_type_description()); } return type_desc; }; Constructor ctor{ctor_cursor.get_usr(), build_type_desc(ctor_cursor)}; // Insert into |constructors_|. std::string type_usr = ctor_cursor.get_semantic_parent().get_usr(); auto existing_ctors = constructors_.find(type_usr); if (existing_ctors != constructors_.end()) { existing_ctors->second.push_back(ctor); } else { constructors_[type_usr] = {ctor}; } } // Tries to lookup a constructor in |type_usr| that takes arguments most // closely aligned to |param_type_desc|. optional TryFindConstructorUsr( const std::string& type_usr, const std::vector& param_type_desc) { auto count_matching_prefix_length = [](const char* a, const char* b) { int matched = 0; while (*a && *b) { if (*a != *b) break; ++a; ++b; ++matched; } // Additional score if the strings were the same length, which makes // "a"/"a" match higher than "a"/"a&" if (*a == *b) matched += 1; return matched; }; // Try to find constructors for the type. If there are no constructors // available, return an empty result. auto ctors_it = constructors_.find(type_usr); if (ctors_it == constructors_.end()) return nullopt; const std::vector& ctors = ctors_it->second; if (ctors.empty()) return nullopt; std::string best_usr; int best_score = INT_MIN; // Scan constructors for the best possible match. for (const Constructor& ctor : ctors) { // If |param_type_desc| is empty and the constructor is as well, we don't // need to bother searching, as this is the match. if (param_type_desc.empty() && ctor.param_type_desc.empty()) { best_usr = ctor.usr; break; } // Weight matching parameter length heavily, as it is more accurate than // the fuzzy type matching approach. int score = 0; if (param_type_desc.size() == ctor.param_type_desc.size()) score += param_type_desc.size() * 1000; // Do prefix-based match on parameter type description. This works well in // practice because clang appends qualifiers to the end of the type, ie, // |foo *&&| for (int i = 0; i < std::min(param_type_desc.size(), ctor.param_type_desc.size()); ++i) { score += count_matching_prefix_length(param_type_desc[i].c_str(), ctor.param_type_desc[i].c_str()); } if (score > best_score) { best_usr = ctor.usr; best_score = score; } } assert(!best_usr.empty()); return best_usr; } }; struct IndexParam { std::unordered_set seen_cx_files; std::vector seen_files; std::unordered_map file_contents; std::unordered_map file_modification_times; // Only use this when strictly needed (ie, primary translation unit is // needed). Most logic should get the IndexFile instance via // |file_consumer|. // // This can be null if we're not generating an index for the primary // translation unit. IndexFile* primary_file = nullptr; ClangTranslationUnit* tu = nullptr; FileConsumer* file_consumer = nullptr; NamespaceHelper ns; ConstructorCache ctors; IndexParam(ClangTranslationUnit* tu, FileConsumer* file_consumer) : tu(tu), file_consumer(file_consumer) {} }; IndexFile* ConsumeFile(IndexParam* param, CXFile file) { bool is_first_ownership = false; IndexFile* db = param->file_consumer->TryConsumeFile(file, &is_first_ownership); // If this is the first time we have seen the file (ignoring if we are // generating an index for it): if (param->seen_cx_files.insert(file).second) { std::string file_name = FileName(file); // Sometimes the fill name will be empty. Not sure why. Not much we can do // with it. if (!file_name.empty()) { // Add to all files we have seen so we can generate proper dependency // graph. param->seen_files.push_back(file_name); // Set modification time. optional modification_time = GetLastModificationTime(file_name); LOG_IF_S(ERROR, !modification_time) << "Failed fetching modification time for " << file_name; if (modification_time) param->file_modification_times[file_name] = *modification_time; // Capture file contents in |param->file_contents| if it was not specified // at the start of indexing. if (db && !param->file_contents.count(file_name)) { optional content = ReadContent(file_name); if (content) param->file_contents.emplace(file_name, *content); else LOG_S(ERROR) << "[indexer] Failed to read file content for " << file_name; } } } if (is_first_ownership) { // Report skipped source range list. CXSourceRangeList* skipped = clang_getSkippedRanges(param->tu->cx_tu, file); for (unsigned i = 0; i < skipped->count; ++i) { Range range = Resolve(skipped->ranges[i]); // clang_getSkippedRanges reports start one token after the '#', move it // back so it starts at the '#' range.start.column -= 1; db->skipped_by_preprocessor.push_back(range); } clang_disposeSourceRangeList(skipped); } return db; } // Returns true if the given entity kind can be called implicitly, ie, without // actually being written in the source code. bool CanBeCalledImplicitly(CXIdxEntityKind kind) { switch (kind) { case CXIdxEntity_CXXConstructor: case CXIdxEntity_CXXConversionFunction: case CXIdxEntity_CXXDestructor: return true; default: return false; } } // Returns true if the cursor spelling contains the given string. This is // useful to check for implicit function calls. bool CursorSpellingContainsString(CXCursor cursor, CXTranslationUnit cx_tu, std::string scanning_for) { CXSourceRange range = clang_Cursor_getSpellingNameRange(cursor, 0, 0); CXToken* tokens; unsigned num_tokens; clang_tokenize(cx_tu, range, &tokens, &num_tokens); bool result = false; for (unsigned i = 0; i < num_tokens; ++i) { CXString name = clang_getTokenSpelling(cx_tu, tokens[i]); if (strcmp(clang_getCString(name), scanning_for.c_str()) == 0) { result = true; break; } clang_disposeString(name); } clang_disposeTokens(cx_tu, tokens, num_tokens); return result; } // Returns the document content for the given range. May not work perfectly // when there are tabs instead of spaces. std::string GetDocumentContentInRange(CXTranslationUnit cx_tu, CXSourceRange range) { std::string result; CXToken* tokens; unsigned num_tokens; clang_tokenize(cx_tu, range, &tokens, &num_tokens); optional previous_token_range; for (unsigned i = 0; i < num_tokens; ++i) { // Add whitespace between the previous token and this one. Range token_range = Resolve(clang_getTokenExtent(cx_tu, tokens[i])); if (previous_token_range) { // Insert newlines. int16_t line_delta = token_range.start.line - previous_token_range->end.line; assert(line_delta >= 0); if (line_delta > 0) { result.append((size_t)line_delta, '\n'); // Reset column so we insert starting padding. previous_token_range->end.column = 0; } // Insert spaces. int16_t column_delta = token_range.start.column - previous_token_range->end.column; assert(column_delta >= 0); result.append((size_t)column_delta, ' '); } previous_token_range = token_range; // Add token content. CXString spelling = clang_getTokenSpelling(cx_tu, tokens[i]); result += clang_getCString(spelling); clang_disposeString(spelling); } clang_disposeTokens(cx_tu, tokens, num_tokens); return result; } } // namespace // static int IndexFile::kCurrentVersion = 6; IndexFile::IndexFile(const std::string& path) : id_cache(path), path(path) { // TODO: Reconsider if we should still be reusing the same id_cache. // Preallocate any existing resolved ids. for (const auto& entry : id_cache.usr_to_type_id) types.push_back(IndexType(entry.second, entry.first)); for (const auto& entry : id_cache.usr_to_func_id) funcs.push_back(IndexFunc(entry.second, entry.first)); for (const auto& entry : id_cache.usr_to_var_id) vars.push_back(IndexVar(entry.second, entry.first)); } // TODO: Optimize for const char*? IndexTypeId IndexFile::ToTypeId(const std::string& usr) { auto it = id_cache.usr_to_type_id.find(usr); if (it != id_cache.usr_to_type_id.end()) return it->second; IndexTypeId id(types.size()); types.push_back(IndexType(id, usr)); id_cache.usr_to_type_id[usr] = id; id_cache.type_id_to_usr[id] = usr; return id; } IndexFuncId IndexFile::ToFuncId(const std::string& usr) { auto it = id_cache.usr_to_func_id.find(usr); if (it != id_cache.usr_to_func_id.end()) return it->second; IndexFuncId id(funcs.size()); funcs.push_back(IndexFunc(id, usr)); id_cache.usr_to_func_id[usr] = id; id_cache.func_id_to_usr[id] = usr; return id; } IndexVarId IndexFile::ToVarId(const std::string& usr) { auto it = id_cache.usr_to_var_id.find(usr); if (it != id_cache.usr_to_var_id.end()) return it->second; IndexVarId id(vars.size()); vars.push_back(IndexVar(id, usr)); id_cache.usr_to_var_id[usr] = id; id_cache.var_id_to_usr[id] = usr; return id; } IndexTypeId IndexFile::ToTypeId(const CXCursor& cursor) { return ToTypeId(ClangCursor(cursor).get_usr()); } IndexFuncId IndexFile::ToFuncId(const CXCursor& cursor) { return ToFuncId(ClangCursor(cursor).get_usr()); } IndexVarId IndexFile::ToVarId(const CXCursor& cursor) { return ToVarId(ClangCursor(cursor).get_usr()); } IndexType* IndexFile::Resolve(IndexTypeId id) { return &types[id.id]; } IndexFunc* IndexFile::Resolve(IndexFuncId id) { return &funcs[id.id]; } IndexVar* IndexFile::Resolve(IndexVarId id) { return &vars[id.id]; } std::string IndexFile::ToString() { return Serialize(*this); } IndexType::IndexType(IndexTypeId id, const std::string& usr) : usr(usr), id(id) { assert(usr.size() > 0); } void RemoveItem(std::vector& ranges, Range to_remove) { auto it = std::find(ranges.begin(), ranges.end(), to_remove); if (it != ranges.end()) ranges.erase(it); } template void UniqueAdd(std::vector& values, T value) { if (std::find(values.begin(), values.end(), value) == values.end()) values.push_back(value); } IdCache::IdCache(const std::string& primary_file) : primary_file(primary_file) {} template bool Contains(const std::vector& vec, const T& element) { for (const T& entry : vec) { if (entry == element) return true; } return false; } void OnIndexDiagnostic(CXClientData client_data, CXDiagnosticSet diagnostics, void* reserved) { IndexParam* param = static_cast(client_data); for (unsigned i = 0; i < clang_getNumDiagnosticsInSet(diagnostics); ++i) { CXDiagnostic diagnostic = clang_getDiagnosticInSet(diagnostics, i); // Skip diagnostics in system headers. CXSourceLocation diag_loc = clang_getDiagnosticLocation(diagnostic); if (clang_Location_isInSystemHeader(diag_loc)) continue; // Get db so we can attribute diagnostic to the right indexed file. CXFile file; unsigned int line, column; clang_getSpellingLocation(diag_loc, &file, &line, &column, nullptr); IndexFile* db = ConsumeFile(param, file); if (!db) continue; // Build diagnostic. optional ls_diagnostic = BuildAndDisposeDiagnostic(diagnostic, db->path); if (ls_diagnostic) db->diagnostics_.push_back(*ls_diagnostic); } } CXIdxClientFile OnIndexIncludedFile(CXClientData client_data, const CXIdxIncludedFileInfo* file) { IndexParam* param = static_cast(client_data); // file->hashLoc only has the position of the hash. We don't have the full // range for the include. CXSourceLocation hash_loc = clang_indexLoc_getCXSourceLocation(file->hashLoc); CXFile cx_file; unsigned int line; clang_getSpellingLocation(hash_loc, &cx_file, &line, nullptr, nullptr); IndexFile* db = ConsumeFile(param, cx_file); if (!db) return nullptr; IndexInclude include; include.line = line; include.resolved_path = FileName(file->file); if (!include.resolved_path.empty()) db->includes.push_back(include); return nullptr; } ClangCursor::VisitResult DumpVisitor(ClangCursor cursor, ClangCursor parent, int* level) { for (int i = 0; i < *level; ++i) std::cerr << " "; std::cerr << ToString(cursor.get_kind()) << " " << cursor.get_spelling() << std::endl; *level += 1; cursor.VisitChildren(&DumpVisitor, level); *level -= 1; return ClangCursor::VisitResult::Continue; } void Dump(ClangCursor cursor) { int level = 0; cursor.VisitChildren(&DumpVisitor, &level); } struct FindChildOfKindParam { CXCursorKind target_kind; optional result; FindChildOfKindParam(CXCursorKind target_kind) : target_kind(target_kind) {} }; ClangCursor::VisitResult FindChildOfKindVisitor(ClangCursor cursor, ClangCursor parent, FindChildOfKindParam* param) { if (cursor.get_kind() == param->target_kind) { param->result = cursor; return ClangCursor::VisitResult::Break; } return ClangCursor::VisitResult::Recurse; } optional FindChildOfKind(ClangCursor cursor, CXCursorKind kind) { FindChildOfKindParam param(kind); cursor.VisitChildren(&FindChildOfKindVisitor, ¶m); return param.result; } ClangCursor::VisitResult FindTypeVisitor(ClangCursor cursor, ClangCursor parent, optional* result) { switch (cursor.get_kind()) { case CXCursor_TypeRef: case CXCursor_TemplateRef: *result = cursor; return ClangCursor::VisitResult::Break; default: break; } return ClangCursor::VisitResult::Recurse; } optional FindType(ClangCursor cursor) { optional result; cursor.VisitChildren(&FindTypeVisitor, &result); return result; } bool IsGlobalContainer(const CXIdxContainerInfo* container) { if (!container) return false; switch (container->cursor.kind) { case CXCursor_Namespace: case CXCursor_TranslationUnit: return true; default: return false; } } bool IsTypeDefinition(const CXIdxContainerInfo* container) { if (!container) return false; switch (container->cursor.kind) { case CXCursor_EnumDecl: case CXCursor_UnionDecl: case CXCursor_StructDecl: case CXCursor_ClassDecl: return true; default: return false; } } struct VisitDeclForTypeUsageParam { IndexFile* db; int has_processed_any = false; optional previous_cursor; optional initial_type; VisitDeclForTypeUsageParam(IndexFile* db) : db(db) {} }; void VisitDeclForTypeUsageVisitorHandler(ClangCursor cursor, VisitDeclForTypeUsageParam* param) { param->has_processed_any = true; IndexFile* db = param->db; std::string referenced_usr = cursor.get_referenced() .template_specialization_to_template_definition() .get_usr(); // TODO: things in STL cause this to be empty. Figure out why and document it. if (referenced_usr == "") return; IndexTypeId ref_type_id = db->ToTypeId(referenced_usr); if (!param->initial_type) param->initial_type = ref_type_id; IndexType* ref_type_def = db->Resolve(ref_type_id); // TODO: Should we even be visiting this if the file is not from the main // def? Try adding assert on |loc| later. Range loc = ResolveSpelling(cursor.cx_cursor); UniqueAdd(ref_type_def->uses, loc); } ClangCursor::VisitResult VisitDeclForTypeUsageVisitor( ClangCursor cursor, ClangCursor parent, VisitDeclForTypeUsageParam* param) { switch (cursor.get_kind()) { case CXCursor_TemplateRef: case CXCursor_TypeRef: if (param->previous_cursor) { VisitDeclForTypeUsageVisitorHandler(param->previous_cursor.value(), param); } param->previous_cursor = cursor; return ClangCursor::VisitResult::Continue; // We do not want to recurse for everything, since if we do that we will end // up visiting method definition bodies/etc. Instead, we only recurse for // things that can logically appear as part of an inline variable // initializer, // ie, // // class Foo { // int x = (Foo)3; // } case CXCursor_CallExpr: case CXCursor_CStyleCastExpr: case CXCursor_CXXStaticCastExpr: case CXCursor_CXXReinterpretCastExpr: return ClangCursor::VisitResult::Recurse; default: return ClangCursor::VisitResult::Continue; } return ClangCursor::VisitResult::Continue; } // Finds the cursor associated with the declaration type of |cursor|. This // strips // qualifies from |cursor| (ie, Foo* => Foo) and removes template arguments // (ie, Foo => Foo<*,*>). optional ResolveToDeclarationType(IndexFile* db, ClangCursor cursor) { ClangCursor declaration = cursor.get_declaration(); declaration = declaration.template_specialization_to_template_definition(); std::string usr = declaration.get_usr(); if (usr != "") return db->ToTypeId(usr); return nullopt; } // Add usages to any seen TypeRef or TemplateRef under the given |decl_cursor|. // This returns the first seen TypeRef or TemplateRef value, which can be // useful if trying to figure out ie, what a using statement refers to. If // trying to generally resolve a cursor to a type, use // ResolveToDeclarationType, which works in more scenarios. optional AddDeclTypeUsages( IndexFile* db, ClangCursor decl_cursor, const CXIdxContainerInfo* semantic_container, const CXIdxContainerInfo* lexical_container) { // // 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 // // // Here is another example: // // enum A {}; // enum B {}; // // template // struct Foo { // struct Inner {}; // }; // // Foo::Inner a; // Foo b; // // => // // EnumDecl A // EnumDecl B // ClassTemplate Foo // TemplateTypeParameter T // StructDecl Inner // VarDecl a // TemplateRef Foo // TypeRef enum A // TypeRef struct Foo::Inner // CallExpr Inner // VarDecl b // TemplateRef Foo // TypeRef enum B // CallExpr Foo // // // Determining the actual type of the variable/declaration from just the // children is tricky. Doing so would require looking up the template // definition associated with a TemplateRef, figuring out how many children // it has, and then skipping that many TypeRef values. This also has to work // with the example below (skipping the last TypeRef). As a result, we // determine variable types using |ResolveToDeclarationType|. // // // 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)) { // // In some code, such as the following example, we receive a cursor which is // not // a definition and is not associated with a definition due to an error // condition. // In this case, it is the Foo::Foo constructor. // // struct Foo {}; // // template // Foo::Foo() {} // if (!decl_cursor.is_definition()) { ClangCursor def = decl_cursor.get_definition(); if (def.get_kind() != CXCursor_FirstInvalid) decl_cursor = def; } process_last_type_ref = false; } VisitDeclForTypeUsageParam param(db); 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 or // TemplateRef. // // We will not visit every child if the is_interseting is false, so // previous_cursor // may not point to the last TemplateRef. assert(param.previous_cursor.has_value() == false || (param.previous_cursor.value().get_kind() == CXCursor_TypeRef || param.previous_cursor.value().get_kind() == CXCursor_TemplateRef)); } return param.initial_type; } // Various versions of LLVM (ie, 4.0) will not visit inline variable references // for template arguments. ClangCursor::VisitResult AddDeclInitializerUsagesVisitor(ClangCursor cursor, ClangCursor parent, IndexFile* db) { /* We need to index the |DeclRefExpr| below (ie, |var| inside of Foo::var). template struct Foo { static constexpr int var = 3; }; int a = Foo::var; => VarDecl a UnexposedExpr var DeclRefExpr var TemplateRef Foo */ switch (cursor.get_kind()) { case CXCursor_DeclRefExpr: { if (cursor.get_referenced().get_kind() != CXCursor_VarDecl) break; // TODO: when we resolve the template type to the definition, we get a // different USR. // ClangCursor ref = // cursor.get_referenced().template_specialization_to_template_definition().get_type().strip_qualifiers().get_usr(); // std::string ref_usr = // cursor.get_referenced().template_specialization_to_template_definition().get_type().strip_qualifiers().get_usr(); std::string ref_usr = cursor.get_referenced() .template_specialization_to_template_definition() .get_usr(); // std::string ref_usr = ref.get_usr(); if (ref_usr == "") break; Range loc = ResolveSpelling(cursor.cx_cursor); IndexVarId ref_id = db->ToVarId(ref_usr); IndexVar* ref_def = db->Resolve(ref_id); UniqueAdd(ref_def->uses, loc); break; } default: break; } return ClangCursor::VisitResult::Recurse; } void AddDeclInitializerUsages(IndexFile* db, ClangCursor decl_cursor) { decl_cursor.VisitChildren(&AddDeclInitializerUsagesVisitor, db); } bool AreEqualLocations(CXIdxLoc loc, CXCursor cursor) { // clang_getCursorExtent // clang_Cursor_getSpellingNameRange return clang_equalLocations( clang_indexLoc_getCXSourceLocation(loc), // clang_getRangeStart(clang_getCursorExtent(cursor))); clang_getRangeStart(clang_Cursor_getSpellingNameRange(cursor, 0, 0))); } ClangCursor::VisitResult VisitMacroDefinitionAndExpansions(ClangCursor cursor, ClangCursor parent, IndexParam* param) { switch (cursor.get_kind()) { case CXCursor_MacroDefinition: case CXCursor_MacroExpansion: { // Resolve location, find IndexFile instance. CXSourceRange cx_source_range = clang_Cursor_getSpellingNameRange(cursor.cx_cursor, 0, 0); CXSourceLocation start = clang_getRangeStart(cx_source_range); if (clang_Location_isInSystemHeader(start)) break; CXFile file; Range decl_loc_spelling = Resolve(cx_source_range, &file); IndexFile* db = ConsumeFile(param, file); if (!db) break; // TODO: Considering checking clang_Cursor_isMacroFunctionLike, but the // only real difference will be that we show 'callers' instead of 'refs' // (especially since macros cannot have overrides) std::string decl_usr; if (cursor.get_kind() == CXCursor_MacroDefinition) decl_usr = cursor.get_usr(); else decl_usr = cursor.get_referenced().get_usr(); IndexVarId var_id = db->ToVarId(decl_usr); IndexVar* var_def = db->Resolve(var_id); UniqueAdd(var_def->uses, decl_loc_spelling); if (cursor.get_kind() == CXCursor_MacroDefinition) { CXSourceRange cx_extent = clang_getCursorExtent(cursor.cx_cursor); var_def->def.short_name = cursor.get_display_name(); var_def->def.detailed_name = cursor.get_display_name(); var_def->def.hover = "#define " + GetDocumentContentInRange(param->tu->cx_tu, cx_extent); var_def->def.cls = VarClass::Macro; var_def->def.definition_spelling = decl_loc_spelling; var_def->def.definition_extent = Resolve(cx_extent, nullptr); } break; } default: break; } return ClangCursor::VisitResult::Continue; } void OnIndexDeclaration(CXClientData client_data, const CXIdxDeclInfo* decl) { if (!kIndexStdDeclarations && clang_Location_isInSystemHeader( clang_indexLoc_getCXSourceLocation(decl->loc))) return; IndexParam* param = static_cast(client_data); // Track all constructor declarations, as we may need to use it to manually // associate std::make_unique and the like as constructor invocations. if (decl->entityInfo->kind == CXIdxEntity_CXXConstructor) { param->ctors.NotifyConstructor(decl->cursor); } CXFile file; clang_getSpellingLocation(clang_indexLoc_getCXSourceLocation(decl->loc), &file, nullptr, nullptr, nullptr); IndexFile* db = ConsumeFile(param, file); if (!db) return; // The language of this declaration LanguageId decl_lang = [decl]() { switch (clang_getCursorLanguage(decl->cursor)) { case CXLanguage_C: return LanguageId::C; case CXLanguage_CPlusPlus: return LanguageId::Cpp; case CXLanguage_ObjC: return LanguageId::ObjC; default: return LanguageId::Unknown; }; }(); // Only update the file language if the new language is "greater" than the old if (decl_lang > db->language) { db->language = decl_lang; } NamespaceHelper* ns = ¶m->ns; switch (decl->entityInfo->kind) { case CXIdxEntity_CXXNamespace: { ns->RegisterQualifiedName(decl->entityInfo->USR, decl->semanticContainer, decl->entityInfo->name); break; } case CXIdxEntity_ObjCProperty: case CXIdxEntity_ObjCIvar: case CXIdxEntity_EnumConstant: case CXIdxEntity_Field: case CXIdxEntity_Variable: case CXIdxEntity_CXXStaticVariable: { Range decl_loc_spelling = ResolveSpelling(decl->cursor); ClangCursor decl_cursor = decl->cursor; // Do not index implicit template instantiations. if (decl_cursor != decl_cursor.template_specialization_to_template_definition()) break; std::string decl_usr = decl_cursor.get_usr(); IndexVarId var_id = db->ToVarId(decl->entityInfo->USR); IndexVar* var = 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; std::string type_name = ToString(clang_getTypeSpelling(clang_getCursorType(decl->cursor))); // clang may report "(lambda at foo.cc)" which end up being a very long // string. Shorten it to just "lambda". if (type_name.find("(lambda at") != std::string::npos) type_name = "lambda"; { std::string qualified_name = ns->QualifiedName(decl->semanticContainer, var->def.short_name); if (decl->entityInfo->kind == CXIdxEntity_EnumConstant) var->def.detailed_name = std::move(qualified_name); else var->def.detailed_name = type_name + " " + std::move(qualified_name); } bool is_system = clang_Location_isInSystemHeader( clang_indexLoc_getCXSourceLocation(decl->loc)); if (is_system) var->def.cls = VarClass::Unknown; else { if (IsGlobalContainer(decl->semanticContainer)) var->def.cls = VarClass::Global; else if (IsTypeDefinition(decl->semanticContainer)) var->def.cls = VarClass::Member; else var->def.cls = VarClass::Local; } //} if (decl->isDefinition) { var->def.definition_spelling = ResolveSpelling(decl->cursor); var->def.definition_extent = ResolveExtent(decl->cursor); ; } else { var->def.declaration = ResolveSpelling(decl->cursor); } UniqueAdd(var->uses, decl_loc_spelling); AddDeclInitializerUsages(db, decl_cursor); var = db->Resolve(var_id); // Declaring variable type information. Note that we do not insert an // interesting reference for parameter declarations - that is handled when // the function declaration is encountered since we won't receive ParmDecl // declarations for unnamed parameters. // TODO: See if we can remove this function call. AddDeclTypeUsages(db, decl_cursor, decl->semanticContainer, decl->lexicalContainer); // We don't need to assign declaring type multiple times if this variable // has already been seen. if (!decl->isRedeclaration) { optional var_type = ResolveToDeclarationType(db, decl_cursor); if (var_type.has_value()) { // Don't treat enum definition variables as instantiations. bool is_enum_member = decl->semanticContainer && decl->semanticContainer->cursor.kind == CXCursor_EnumDecl; if (!is_enum_member) db->Resolve(var_type.value())->instances.push_back(var_id); var->def.variable_type = var_type.value(); } } // TODO: Refactor handlers so more things are under 'if // (!decl->isRedeclaration)' if (decl->isDefinition && IsTypeDefinition(decl->semanticContainer)) { IndexTypeId declaring_type_id = db->ToTypeId(decl->semanticContainer->cursor); IndexType* declaring_type_def = db->Resolve(declaring_type_id); var->def.declaring_type = declaring_type_id; declaring_type_def->def.vars.push_back(var_id); } break; } case CXIdxEntity_ObjCInstanceMethod: case CXIdxEntity_ObjCClassMethod: case CXIdxEntity_Function: case CXIdxEntity_CXXConstructor: case CXIdxEntity_CXXDestructor: case CXIdxEntity_CXXInstanceMethod: case CXIdxEntity_CXXStaticMethod: case CXIdxEntity_CXXConversionFunction: { Range decl_spelling = ResolveSpelling(decl->cursor); Range decl_extent = ResolveExtent(decl->cursor); ClangCursor decl_cursor = decl->cursor; ClangCursor decl_cursor_resolved = decl_cursor.template_specialization_to_template_definition(); bool is_template_specialization = decl_cursor != decl_cursor_resolved; IndexFuncId func_id = db->ToFuncId(decl_cursor_resolved.cx_cursor); IndexFunc* func = db->Resolve(func_id); // We don't actually need to know the return type, but we need to mark it // as an interesting usage. AddDeclTypeUsages(db, decl_cursor, decl->semanticContainer, decl->lexicalContainer); // Add definition or declaration. This is a bit tricky because we treat // template specializations as declarations, even though they are // technically definitions. // TODO: Support multiple function definitions, which is common for // template specializations. if (decl->isDefinition && !is_template_specialization) { // assert(!func->def.definition_spelling); // assert(!func->def.definition_extent); func->def.definition_spelling = decl_spelling; func->def.definition_extent = decl_extent; } else { IndexFunc::Declaration declaration; declaration.spelling = decl_spelling; declaration.extent = decl_extent; declaration.content = GetDocumentContentInRange( param->tu->cx_tu, clang_getCursorExtent(decl->cursor)); // Add parameters. for (ClangCursor arg : decl_cursor.get_arguments()) { switch (arg.get_kind()) { case CXCursor_ParmDecl: { Range param_spelling = ResolveSpelling(arg.cx_cursor); // If the name is empty (which is common for parameters), clang // will report a range with length 1, which is not correct. if (param_spelling.start.column == (param_spelling.end.column - 1) && arg.get_display_name().empty()) { param_spelling.end.column -= 1; } declaration.param_spellings.push_back(param_spelling); break; } default: break; } } func->declarations.push_back(declaration); } // Emit definition data for the function. We do this even if it isn't a // definition because there can be, for example, interfaces, or a class // declaration that doesn't have a definition yet. If we never end up // indexing the definition, then there will not be any (ie) outline // information. if (!is_template_specialization) { func->def.short_name = decl->entityInfo->name; // Set the |is_operator| flag to true if the function name starts with // "operator" func->def.is_operator = func->def.short_name.compare(0, 8, "operator") == 0; // Build detailed name. The type desc looks like void (void *). We // insert the qualified name before the first '('. std::string qualified_name = ns->QualifiedName(decl->semanticContainer, func->def.short_name); std::string type_desc = decl_cursor.get_type_description(); { size_t balance = 0, offset; for (offset = type_desc.size(); offset;) { offset--; if (type_desc[offset] == ')') balance++; // Balanced paren pair that may appear before the paren enclosing // function parameters, see clang/lib/AST/TypePrinter.cpp else if (type_desc[offset] == '(' && --balance == 0 && !((offset >= 5 && !type_desc.compare(offset - 5, 5, "throw")) || (offset >= 6 && !type_desc.compare(offset - 6, 6, "typeof")) || (offset >= 7 && !type_desc.compare(offset - 7, 7, "_Atomic")) || (offset >= 8 && !type_desc.compare(offset - 8, 8, "decltype")) || (offset >= 8 && !type_desc.compare(offset - 8, 8, "noexcept")) || (offset >= 13 && !type_desc.compare(offset - 13, 13, "__attribute__")))) break; } if (offset > 0) { type_desc.insert(offset, qualified_name); func->def.detailed_name = type_desc; } else { // type_desc is probably the name of a typedef. func->def.detailed_name = type_desc + " " + qualified_name; } } // 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 (IsTypeDefinition(decl->semanticContainer)) { IndexTypeId declaring_type_id = db->ToTypeId(decl->semanticContainer->cursor); IndexType* declaring_type_def = db->Resolve(declaring_type_id); func->def.declaring_type = declaring_type_id; // Mark a type reference at the ctor/dtor location. if (decl->entityInfo->kind == CXIdxEntity_CXXConstructor) UniqueAdd(declaring_type_def->uses, decl_spelling); if (decl->entityInfo->kind == CXIdxEntity_CXXDestructor) { Range dtor_type_range = decl_spelling; dtor_type_range.start.column += 1; // Don't count the leading ~ UniqueAdd(declaring_type_def->uses, dtor_type_range); } // Add function to declaring type. UniqueAdd(declaring_type_def->def.funcs, func_id); } // Process inheritance. if (clang_CXXMethod_isVirtual(decl->cursor)) { CXCursor* overridden; unsigned int num_overridden; clang_getOverriddenCursors(decl->cursor, &overridden, &num_overridden); for (unsigned i = 0; i < num_overridden; ++i) { ClangCursor parent = ClangCursor(overridden[i]) .template_specialization_to_template_definition(); IndexFuncId parent_id = db->ToFuncId(parent.get_usr()); IndexFunc* parent_def = db->Resolve(parent_id); func = db->Resolve(func_id); // ToFuncId invalidated func_def func->def.base.push_back(parent_id); parent_def->derived.push_back(func_id); } clang_disposeOverriddenCursors(overridden); } } break; } case CXIdxEntity_Typedef: case CXIdxEntity_CXXTypeAlias: { Range decl_loc_spelling = ResolveSpelling(decl->cursor); // Note we want to fetch the first TypeRef. Running // ResolveCursorType(decl->cursor) would return // the type of the typedef/using, not the type of the referenced type. optional alias_of = AddDeclTypeUsages( db, decl->cursor, decl->semanticContainer, decl->lexicalContainer); IndexTypeId type_id = db->ToTypeId(decl->entityInfo->USR); IndexType* type = db->Resolve(type_id); if (alias_of) type->def.alias_of = alias_of.value(); Range spell = ResolveSpelling(decl->cursor); Range extent = ResolveExtent(decl->cursor); type->def.definition_spelling = spell; type->def.definition_extent = extent; type->def.short_name = decl->entityInfo->name; type->def.detailed_name = ns->QualifiedName(decl->semanticContainer, type->def.short_name); // For Typedef/CXXTypeAlias spanning a few lines, display the declaration // line, with spelling name replaced with qualified name. // TODO Think how to display multi-line declaration like `typedef struct { // ... } foo;` https://github.com/jacobdufault/cquery/issues/29 if (extent.end.line - extent.start.line < kMaxLinesDisplayTypeAliasDeclarations) { FileContentsWithOffsets& fc = param->file_contents[db->path]; optional extent_start = fc.ToOffset(extent.start), spell_start = fc.ToOffset(spell.start), spell_end = fc.ToOffset(spell.end), extent_end = fc.ToOffset(extent.end); if (extent_start && spell_start && spell_end && extent_end) { type->def.hover = fc.contents.substr(*extent_start, *spell_start - *extent_start) + type->def.detailed_name + fc.contents.substr(*spell_end, *extent_end - *spell_end); } } UniqueAdd(type->uses, decl_loc_spelling); break; } case CXIdxEntity_ObjCProtocol: case CXIdxEntity_ObjCCategory: case CXIdxEntity_ObjCClass: case CXIdxEntity_Enum: case CXIdxEntity_Union: case CXIdxEntity_Struct: case CXIdxEntity_CXXClass: { Range decl_loc_spelling = ResolveSpelling(decl->cursor); IndexTypeId type_id = db->ToTypeId(decl->entityInfo->USR); IndexType* type = 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) { // name can be null in an anonymous struct (see // tests/types/anonymous_struct.cc). if (decl->entityInfo->name) { ns->RegisterQualifiedName(decl->entityInfo->USR, decl->semanticContainer, decl->entityInfo->name); type->def.short_name = decl->entityInfo->name; } else { type->def.short_name = ""; } type->def.detailed_name = ns->QualifiedName(decl->semanticContainer, type->def.short_name); // } if (decl->isDefinition) { type->def.definition_spelling = ResolveSpelling(decl->cursor); type->def.definition_extent = ResolveExtent(decl->cursor); } UniqueAdd(type->uses, decl_loc_spelling); // 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); if (class_info) { for (unsigned int i = 0; i < class_info->numBases; ++i) { const CXIdxBaseClassInfo* base_class = class_info->bases[i]; AddDeclTypeUsages(db, base_class->cursor, decl->semanticContainer, decl->lexicalContainer); optional parent_type_id = ResolveToDeclarationType(db, base_class->cursor); // type_def ptr could be invalidated by ResolveToDeclarationType. type = db->Resolve(type_id); if (parent_type_id) { IndexType* parent_type_def = db->Resolve(parent_type_id.value()); parent_type_def->derived.push_back(type_id); type->def.parents.push_back(parent_type_id.value()); } } } break; } default: std::cerr << "!! Unhandled indexDeclaration: " << ClangCursor(decl->cursor).ToString() << " at " << ResolveSpelling(decl->cursor).start.ToString() << std::endl; std::cerr << " entityInfo->kind = " << decl->entityInfo->kind << std::endl; std::cerr << " entityInfo->USR = " << decl->entityInfo->USR << std::endl; if (decl->declAsContainer) std::cerr << " declAsContainer = " << ClangCursor(decl->declAsContainer->cursor).ToString() << std::endl; if (decl->semanticContainer) std::cerr << " semanticContainer = " << ClangCursor(decl->semanticContainer->cursor).ToString() << std::endl; if (decl->lexicalContainer) std::cerr << " lexicalContainer = " << ClangCursor(decl->lexicalContainer->cursor).get_usr() << std::endl; break; } } bool IsFunctionCallContext(CXCursorKind kind) { switch (kind) { case CXCursor_FunctionDecl: case CXCursor_CXXMethod: case CXCursor_Constructor: case CXCursor_Destructor: case CXCursor_ConversionFunction: case CXCursor_FunctionTemplate: case CXCursor_OverloadedDeclRef: // TODO: we need to test lambdas case CXCursor_LambdaExpr: return true; default: break; } return false; } void OnIndexReference(CXClientData client_data, const CXIdxEntityRefInfo* ref) { // TODO: Use clang_getFileUniqueID CXFile file; clang_getSpellingLocation(clang_indexLoc_getCXSourceLocation(ref->loc), &file, nullptr, nullptr, nullptr); IndexParam* param = static_cast(client_data); IndexFile* db = ConsumeFile(param, file); if (!db) return; ClangCursor cursor(ref->cursor); switch (ref->referencedEntity->kind) { case CXIdxEntity_CXXNamespaceAlias: case CXIdxEntity_CXXNamespace: { // We don't index namespace usages. break; } case CXIdxEntity_ObjCProperty: case CXIdxEntity_ObjCIvar: case CXIdxEntity_EnumConstant: case CXIdxEntity_CXXStaticVariable: case CXIdxEntity_Variable: case CXIdxEntity_Field: { Range loc_spelling = ResolveSpelling(ref->cursor); ClangCursor referenced = ref->referencedEntity->cursor; referenced = referenced.template_specialization_to_template_definition(); IndexVarId var_id = db->ToVarId(referenced.get_usr()); IndexVar* var = db->Resolve(var_id); // Lambda paramaters are not processed by OnIndexDeclaration and // may not have a short_name yet. Note that we only process the lambda // parameter as a definition if it is in the same file as the reference, // as lambdas cannot be split across files. if (var->def.short_name.empty()) { CXFile referenced_file; Range spelling = ResolveSpelling(referenced.cx_cursor, &referenced_file); if (file == referenced_file) { var->def.definition_spelling = spelling; var->def.definition_extent = ResolveExtent(referenced.cx_cursor); // TODO Some of the logic here duplicates CXIdxEntity_Variable branch // of OnIndexDeclaration. But there `decl` is of type CXIdxDeclInfo // and has more information, thus not easy to reuse the code. var->def.short_name = referenced.get_spelling(); std::string type_name = ToString( clang_getTypeSpelling(clang_getCursorType(referenced.cx_cursor))); var->def.detailed_name = type_name + " " + var->def.short_name; var->def.cls = VarClass::Member; UniqueAdd(var->uses, ResolveSpelling(referenced.cx_cursor)); AddDeclInitializerUsages(db, referenced.cx_cursor); // TODO Use proper semantic_container and lexical_container. AddDeclTypeUsages(db, referenced.cx_cursor, nullptr, nullptr); // TODO Other logic in OnIndexDeclaration may need to be adapted. } } UniqueAdd(var->uses, loc_spelling); break; } case CXIdxEntity_CXXConversionFunction: case CXIdxEntity_CXXStaticMethod: case CXIdxEntity_CXXInstanceMethod: case CXIdxEntity_ObjCInstanceMethod: case CXIdxEntity_ObjCClassMethod: case CXIdxEntity_Function: case CXIdxEntity_CXXConstructor: case CXIdxEntity_CXXDestructor: { // TODO: Redirect container to constructor for the following example, ie, // we should be inserting an outgoing function call from the Foo // ctor. // // int Gen() { return 5; } // class Foo { // int x = Gen(); // } // TODO: search full history? Range loc_spelling = ResolveSpelling(ref->cursor); IndexFuncId called_id = db->ToFuncId(ref->referencedEntity->USR); IndexFunc* called = db->Resolve(called_id); // libclang doesn't provide a nice api to check if the given function // call is implicit. ref->kind should probably work (it's either direct // or implicit), but libclang only supports implicit for objective-c. bool is_implicit = CanBeCalledImplicitly(ref->referencedEntity->kind) && // Treats empty short_name as an implicit call like implicit move // constructor in `vector a = f();` (called->def.short_name.empty() || // For explicit destructor call, ref->cursor may be "~" while // called->def.short_name is "~A" // "~A" is not a substring of ref->cursor, but we should take this // case as not `is_implicit`. (called->def.short_name[0] != '~' && !CursorSpellingContainsString(ref->cursor, param->tu->cx_tu, called->def.short_name))); if (IsFunctionCallContext(ref->container->cursor.kind)) { IndexFuncId caller_id = db->ToFuncId(ref->container->cursor); IndexFunc* caller = db->Resolve(caller_id); // Calling db->ToFuncId invalidates the FuncDef* ptrs. called = db->Resolve(called_id); AddFuncRef(&caller->def.callees, IndexFuncRef(called_id, loc_spelling, is_implicit)); AddFuncRef(&called->callers, IndexFuncRef(caller_id, loc_spelling, is_implicit)); } else { AddFuncRef(&called->callers, IndexFuncRef(loc_spelling, is_implicit)); } // Checks if |str| starts with |start|. Ignores case. auto str_begin = [](const char* start, const char* str) { while (*start && *str) { char a = tolower(*start); char b = tolower(*str); if (a != b) return false; ++start; ++str; } return !*start; }; bool is_template = ref->referencedEntity->templateKind != CXIdxEntityCXXTemplateKind::CXIdxEntity_NonTemplate; if (is_template && str_begin("make", ref->referencedEntity->name)) { // Try to find the return type of called function. That type will have // the constructor function we add a usage to. optional opt_found_type = FindType(ref->cursor); if (opt_found_type) { std::string ctor_type_usr = opt_found_type->get_referenced().get_usr(); ClangCursor call_cursor = ref->cursor; // Build a type description from the parameters of the call, so we // can try to find a constructor with the same type description. std::vector call_type_desc; for (ClangType type : call_cursor.get_type().get_arguments()) { std::string type_desc = type.get_spelling(); if (!type_desc.empty()) call_type_desc.push_back(type_desc); } // Try to find the constructor and add a reference. optional ctor_usr = param->ctors.TryFindConstructorUsr(ctor_type_usr, call_type_desc); if (ctor_usr) { IndexFunc* ctor = db->Resolve(db->ToFuncId(*ctor_usr)); AddFuncRef(&ctor->callers, IndexFuncRef(loc_spelling, true /*is_implicit*/)); } } } break; } case CXIdxEntity_ObjCCategory: case CXIdxEntity_ObjCProtocol: case CXIdxEntity_ObjCClass: case CXIdxEntity_Typedef: case CXIdxEntity_CXXTypeAlias: case CXIdxEntity_Enum: case CXIdxEntity_Union: case CXIdxEntity_Struct: case CXIdxEntity_CXXClass: { ClangCursor referenced_cursor = ref->referencedEntity->cursor; referenced_cursor = referenced_cursor.template_specialization_to_template_definition(); IndexTypeId referenced_id = db->ToTypeId(referenced_cursor.get_usr()); IndexType* referenced = 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; // } // UniqueAdd(referenced->uses, ResolveSpelling(ref->cursor)); break; } default: std::cerr << "!! Unhandled indexEntityReference: " << cursor.ToString() << " at " << ResolveSpelling(ref->cursor).start.ToString() << std::endl; std::cerr << " ref->referencedEntity->kind = " << ref->referencedEntity->kind << std::endl; if (ref->parentEntity) std::cerr << " ref->parentEntity->kind = " << ref->parentEntity->kind << std::endl; std::cerr << " ref->loc = " << ResolveSpelling(ref->cursor).start.ToString() << std::endl; std::cerr << " ref->kind = " << ref->kind << std::endl; if (ref->parentEntity) std::cerr << " parentEntity = " << ClangCursor(ref->parentEntity->cursor).ToString() << std::endl; if (ref->referencedEntity) std::cerr << " referencedEntity = " << ClangCursor(ref->referencedEntity->cursor).ToString() << std::endl; if (ref->container) std::cerr << " container = " << ClangCursor(ref->container->cursor).ToString() << std::endl; break; } } FileContents::FileContents(const std::string& path, const std::string& content) : path(path), content(content) {} FileContentsWithOffsets::FileContentsWithOffsets() : line_offsets_{0} {} FileContentsWithOffsets::FileContentsWithOffsets(std::string s) { contents = s; line_offsets_.push_back(0); for (size_t i = 0; i < s.size(); i++) if (s[i] == '\n') line_offsets_.push_back(i + 1); } optional FileContentsWithOffsets::ToOffset(Position p) const { if (0 < p.line && size_t(p.line) <= line_offsets_.size()) { int ret = line_offsets_[p.line - 1] + p.column - 1; if (size_t(ret) <= contents.size()) return {ret}; } return nullopt; } optional FileContentsWithOffsets::ContentsInRange( Range range) const { optional start_offset = ToOffset(range.start), end_offset = ToOffset(range.end); if (start_offset && end_offset && *start_offset < *end_offset) return {contents.substr(*start_offset, *end_offset - *start_offset)}; return nullopt; } std::vector> Parse( Config* config, FileConsumer::SharedState* file_consumer_shared, std::string file, const std::vector& args, const std::vector& file_contents, PerformanceImportFile* perf, ClangIndex* index, bool dump_ast) { if (!config->enableIndexing) return {}; file = NormalizePath(file); Timer timer; std::vector unsaved_files; for (const FileContents& contents : file_contents) { CXUnsavedFile unsaved; unsaved.Filename = contents.path.c_str(); unsaved.Contents = contents.content.c_str(); unsaved.Length = (unsigned long)contents.content.size(); unsaved_files.push_back(unsaved); } std::unique_ptr tu = ClangTranslationUnit::Create( index, file, args, unsaved_files, CXTranslationUnit_KeepGoing | CXTranslationUnit_DetailedPreprocessingRecord); if (!tu) return {}; perf->index_parse = timer.ElapsedMicrosecondsAndReset(); if (dump_ast) Dump(clang_getTranslationUnitCursor(tu->cx_tu)); return ParseWithTu(file_consumer_shared, perf, tu.get(), index, file, args, unsaved_files); } std::vector> ParseWithTu( FileConsumer::SharedState* file_consumer_shared, PerformanceImportFile* perf, ClangTranslationUnit* tu, ClangIndex* index, const std::string& file, const std::vector& args, const std::vector& file_contents) { Timer timer; IndexerCallbacks callback = {0}; // Available callbacks: // - abortQuery // - enteredMainFile // - ppIncludedFile // - importedASTFile // - startedTranslationUnit callback.diagnostic = &OnIndexDiagnostic; callback.ppIncludedFile = &OnIndexIncludedFile; callback.indexDeclaration = &OnIndexDeclaration; callback.indexEntityReference = &OnIndexReference; FileConsumer file_consumer(file_consumer_shared, file); IndexParam param(tu, &file_consumer); for (const CXUnsavedFile& contents : file_contents) { param.file_contents.emplace( contents.Filename, std::string(contents.Contents, contents.Length)); } CXFile cx_file = clang_getFile(tu->cx_tu, file.c_str()); param.primary_file = ConsumeFile(¶m, cx_file); CXIndexAction index_action = clang_IndexAction_create(index->cx_index); // NOTE: libclang re-enables crash recovery whenever a new index is created. // To have clang crash toggle crash recovery right before calling // clang_indexTranslationUnit. // clang_toggleCrashRecovery(0); // |index_result| is a CXErrorCode instance. int index_result = clang_indexTranslationUnit( index_action, ¶m, &callback, sizeof(IndexerCallbacks), CXIndexOpt_IndexFunctionLocalSymbols | CXIndexOpt_SkipParsedBodiesInSession | CXIndexOpt_IndexImplicitTemplateInstantiations, tu->cx_tu); if (index_result != CXError_Success) { LOG_S(WARNING) << "Indexing " << file << " failed with errno=" << index_result; return {}; } clang_IndexAction_dispose(index_action); ClangCursor(clang_getTranslationUnitCursor(tu->cx_tu)) .VisitChildren(&VisitMacroDefinitionAndExpansions, ¶m); perf->index_build = timer.ElapsedMicrosecondsAndReset(); auto result = param.file_consumer->TakeLocalState(); for (std::unique_ptr& entry : result) { entry->import_file = file; entry->args = args; // Update file contents and modification time. entry->file_contents_ = param.file_contents[entry->path].contents; entry->last_modification_time = param.file_modification_times[entry->path]; // Update dependencies for the file. Do not include the file in its own // dependency set. entry->dependencies = param.seen_files; entry->dependencies.erase( std::remove(entry->dependencies.begin(), entry->dependencies.end(), entry->path), entry->dependencies.end()); // Make sure we are using correct file contents. for (const CXUnsavedFile& contents : file_contents) { if (entry->path == contents.Filename) entry->file_contents_ = std::string(contents.Contents, contents.Length); } } return result; } void IndexInit() { clang_enableStackTraces(); clang_toggleCrashRecovery(1); } void ClangSanityCheck() { std::vector args = {"clang", "tests/vars/class_member.cc"}; unsigned opts = 0; CXIndex index = clang_createIndex(0, 1); CXTranslationUnit tu; clang_parseTranslationUnit2FullArgv(index, nullptr, args.data(), args.size(), nullptr, 0, opts, &tu); assert(tu); IndexerCallbacks callback = {0}; callback.abortQuery = [](CXClientData client_data, void* reserved) { return 0; }; callback.diagnostic = [](CXClientData client_data, CXDiagnosticSet diagnostics, void* reserved) {}; callback.enteredMainFile = [](CXClientData client_data, CXFile mainFile, void* reserved) -> CXIdxClientFile { return nullptr; }; callback.ppIncludedFile = []( CXClientData client_data, const CXIdxIncludedFileInfo* file) -> CXIdxClientFile { return nullptr; }; callback.importedASTFile = []( CXClientData client_data, const CXIdxImportedASTFileInfo*) -> CXIdxClientASTFile { return nullptr; }; callback.startedTranslationUnit = [](CXClientData client_data, void* reserved) -> CXIdxClientContainer { return nullptr; }; callback.indexDeclaration = [](CXClientData client_data, const CXIdxDeclInfo* decl) {}; callback.indexEntityReference = [](CXClientData client_data, const CXIdxEntityRefInfo* ref) {}; const unsigned kIndexOpts = 0; CXIndexAction index_action = clang_IndexAction_create(index); int index_param = 0; clang_toggleCrashRecovery(0); clang_indexTranslationUnit(index_action, &index_param, &callback, sizeof(IndexerCallbacks), kIndexOpts, tu); clang_IndexAction_dispose(index_action); clang_disposeTranslationUnit(tu); clang_disposeIndex(index); } std::string GetClangVersion() { return ToString(clang_getClangVersion()); }