1 //===- JSONCompilationDatabase.cpp ----------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file contains the implementation of the JSONCompilationDatabase. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "clang/Tooling/JSONCompilationDatabase.h" 14 #include "clang/Basic/LLVM.h" 15 #include "clang/Tooling/CompilationDatabase.h" 16 #include "clang/Tooling/CompilationDatabasePluginRegistry.h" 17 #include "clang/Tooling/Tooling.h" 18 #include "llvm/ADT/SmallVector.h" 19 #include "llvm/ADT/StringRef.h" 20 #include "llvm/Support/Allocator.h" 21 #include "llvm/Support/CommandLine.h" 22 #include "llvm/Support/ErrorOr.h" 23 #include "llvm/Support/MemoryBuffer.h" 24 #include "llvm/Support/Path.h" 25 #include "llvm/Support/StringSaver.h" 26 #include "llvm/Support/VirtualFileSystem.h" 27 #include "llvm/Support/YAMLParser.h" 28 #include "llvm/Support/raw_ostream.h" 29 #include "llvm/TargetParser/Host.h" 30 #include <cassert> 31 #include <memory> 32 #include <optional> 33 #include <string> 34 #include <system_error> 35 #include <utility> 36 #include <vector> 37 38 using namespace clang; 39 using namespace tooling; 40 41 namespace { 42 43 /// A parser for escaped strings of command line arguments. 44 /// 45 /// Assumes \-escaping for quoted arguments (see the documentation of 46 /// unescapeCommandLine(...)). 47 class CommandLineArgumentParser { 48 public: 49 CommandLineArgumentParser(StringRef CommandLine) 50 : Input(CommandLine), Position(Input.begin()-1) {} 51 52 std::vector<std::string> parse() { 53 bool HasMoreInput = true; 54 while (HasMoreInput && nextNonWhitespace()) { 55 std::string Argument; 56 HasMoreInput = parseStringInto(Argument); 57 CommandLine.push_back(Argument); 58 } 59 return CommandLine; 60 } 61 62 private: 63 // All private methods return true if there is more input available. 64 65 bool parseStringInto(std::string &String) { 66 do { 67 if (*Position == '"') { 68 if (!parseDoubleQuotedStringInto(String)) return false; 69 } else if (*Position == '\'') { 70 if (!parseSingleQuotedStringInto(String)) return false; 71 } else { 72 if (!parseFreeStringInto(String)) return false; 73 } 74 } while (*Position != ' '); 75 return true; 76 } 77 78 bool parseDoubleQuotedStringInto(std::string &String) { 79 if (!next()) return false; 80 while (*Position != '"') { 81 if (!skipEscapeCharacter()) return false; 82 String.push_back(*Position); 83 if (!next()) return false; 84 } 85 return next(); 86 } 87 88 bool parseSingleQuotedStringInto(std::string &String) { 89 if (!next()) return false; 90 while (*Position != '\'') { 91 String.push_back(*Position); 92 if (!next()) return false; 93 } 94 return next(); 95 } 96 97 bool parseFreeStringInto(std::string &String) { 98 do { 99 if (!skipEscapeCharacter()) return false; 100 String.push_back(*Position); 101 if (!next()) return false; 102 } while (*Position != ' ' && *Position != '"' && *Position != '\''); 103 return true; 104 } 105 106 bool skipEscapeCharacter() { 107 if (*Position == '\\') { 108 return next(); 109 } 110 return true; 111 } 112 113 bool nextNonWhitespace() { 114 do { 115 if (!next()) return false; 116 } while (*Position == ' '); 117 return true; 118 } 119 120 bool next() { 121 ++Position; 122 return Position != Input.end(); 123 } 124 125 const StringRef Input; 126 StringRef::iterator Position; 127 std::vector<std::string> CommandLine; 128 }; 129 130 std::vector<std::string> unescapeCommandLine(JSONCommandLineSyntax Syntax, 131 StringRef EscapedCommandLine) { 132 if (Syntax == JSONCommandLineSyntax::AutoDetect) { 133 #ifdef _WIN32 134 // Assume Windows command line parsing on Win32 135 Syntax = JSONCommandLineSyntax::Windows; 136 #else 137 Syntax = JSONCommandLineSyntax::Gnu; 138 #endif 139 } 140 141 if (Syntax == JSONCommandLineSyntax::Windows) { 142 llvm::BumpPtrAllocator Alloc; 143 llvm::StringSaver Saver(Alloc); 144 llvm::SmallVector<const char *, 64> T; 145 llvm::cl::TokenizeWindowsCommandLine(EscapedCommandLine, Saver, T); 146 std::vector<std::string> Result(T.begin(), T.end()); 147 return Result; 148 } 149 assert(Syntax == JSONCommandLineSyntax::Gnu); 150 CommandLineArgumentParser parser(EscapedCommandLine); 151 return parser.parse(); 152 } 153 154 // This plugin locates a nearby compile_command.json file, and also infers 155 // compile commands for files not present in the database. 156 class JSONCompilationDatabasePlugin : public CompilationDatabasePlugin { 157 std::unique_ptr<CompilationDatabase> 158 loadFromDirectory(StringRef Directory, std::string &ErrorMessage) override { 159 SmallString<1024> JSONDatabasePath(Directory); 160 llvm::sys::path::append(JSONDatabasePath, "compile_commands.json"); 161 auto Base = JSONCompilationDatabase::loadFromFile( 162 JSONDatabasePath, ErrorMessage, JSONCommandLineSyntax::AutoDetect); 163 return Base ? inferTargetAndDriverMode( 164 inferMissingCompileCommands(expandResponseFiles( 165 std::move(Base), llvm::vfs::getRealFileSystem()))) 166 : nullptr; 167 } 168 }; 169 170 } // namespace 171 172 // Register the JSONCompilationDatabasePlugin with the 173 // CompilationDatabasePluginRegistry using this statically initialized variable. 174 static CompilationDatabasePluginRegistry::Add<JSONCompilationDatabasePlugin> 175 X("json-compilation-database", "Reads JSON formatted compilation databases"); 176 177 namespace clang { 178 namespace tooling { 179 180 // This anchor is used to force the linker to link in the generated object file 181 // and thus register the JSONCompilationDatabasePlugin. 182 volatile int JSONAnchorSource = 0; 183 184 } // namespace tooling 185 } // namespace clang 186 187 std::unique_ptr<JSONCompilationDatabase> 188 JSONCompilationDatabase::loadFromFile(StringRef FilePath, 189 std::string &ErrorMessage, 190 JSONCommandLineSyntax Syntax) { 191 // Don't mmap: if we're a long-lived process, the build system may overwrite. 192 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> DatabaseBuffer = 193 llvm::MemoryBuffer::getFile(FilePath, /*IsText=*/false, 194 /*RequiresNullTerminator=*/true, 195 /*IsVolatile=*/true); 196 if (std::error_code Result = DatabaseBuffer.getError()) { 197 ErrorMessage = "Error while opening JSON database: " + Result.message(); 198 return nullptr; 199 } 200 std::unique_ptr<JSONCompilationDatabase> Database( 201 new JSONCompilationDatabase(std::move(*DatabaseBuffer), Syntax)); 202 if (!Database->parse(ErrorMessage)) 203 return nullptr; 204 return Database; 205 } 206 207 std::unique_ptr<JSONCompilationDatabase> 208 JSONCompilationDatabase::loadFromBuffer(StringRef DatabaseString, 209 std::string &ErrorMessage, 210 JSONCommandLineSyntax Syntax) { 211 std::unique_ptr<llvm::MemoryBuffer> DatabaseBuffer( 212 llvm::MemoryBuffer::getMemBufferCopy(DatabaseString)); 213 std::unique_ptr<JSONCompilationDatabase> Database( 214 new JSONCompilationDatabase(std::move(DatabaseBuffer), Syntax)); 215 if (!Database->parse(ErrorMessage)) 216 return nullptr; 217 return Database; 218 } 219 220 std::vector<CompileCommand> 221 JSONCompilationDatabase::getCompileCommands(StringRef FilePath) const { 222 SmallString<128> NativeFilePath; 223 llvm::sys::path::native(FilePath, NativeFilePath); 224 225 std::string Error; 226 llvm::raw_string_ostream ES(Error); 227 StringRef Match = MatchTrie.findEquivalent(NativeFilePath, ES); 228 if (Match.empty()) 229 return {}; 230 const auto CommandsRefI = IndexByFile.find(Match); 231 if (CommandsRefI == IndexByFile.end()) 232 return {}; 233 std::vector<CompileCommand> Commands; 234 getCommands(CommandsRefI->getValue(), Commands); 235 return Commands; 236 } 237 238 std::vector<std::string> 239 JSONCompilationDatabase::getAllFiles() const { 240 std::vector<std::string> Result; 241 for (const auto &CommandRef : IndexByFile) 242 Result.push_back(CommandRef.first().str()); 243 return Result; 244 } 245 246 std::vector<CompileCommand> 247 JSONCompilationDatabase::getAllCompileCommands() const { 248 std::vector<CompileCommand> Commands; 249 getCommands(AllCommands, Commands); 250 return Commands; 251 } 252 253 static llvm::StringRef stripExecutableExtension(llvm::StringRef Name) { 254 Name.consume_back(".exe"); 255 return Name; 256 } 257 258 // There are compiler-wrappers (ccache, distcc) that take the "real" 259 // compiler as an argument, e.g. distcc gcc -O3 foo.c. 260 // These end up in compile_commands.json when people set CC="distcc gcc". 261 // Clang's driver doesn't understand this, so we need to unwrap. 262 static bool unwrapCommand(std::vector<std::string> &Args) { 263 if (Args.size() < 2) 264 return false; 265 StringRef Wrapper = 266 stripExecutableExtension(llvm::sys::path::filename(Args.front())); 267 if (Wrapper == "distcc" || Wrapper == "ccache" || Wrapper == "sccache") { 268 // Most of these wrappers support being invoked 3 ways: 269 // `distcc g++ file.c` This is the mode we're trying to match. 270 // We need to drop `distcc`. 271 // `distcc file.c` This acts like compiler is cc or similar. 272 // Clang's driver can handle this, no change needed. 273 // `g++ file.c` g++ is a symlink to distcc. 274 // We don't even notice this case, and all is well. 275 // 276 // We need to distinguish between the first and second case. 277 // The wrappers themselves don't take flags, so Args[1] is a compiler flag, 278 // an input file, or a compiler. Inputs have extensions, compilers don't. 279 bool HasCompiler = 280 (Args[1][0] != '-') && 281 !llvm::sys::path::has_extension(stripExecutableExtension(Args[1])); 282 if (HasCompiler) { 283 Args.erase(Args.begin()); 284 return true; 285 } 286 // If !HasCompiler, wrappers act like GCC. Fine: so do we. 287 } 288 return false; 289 } 290 291 static std::vector<std::string> 292 nodeToCommandLine(JSONCommandLineSyntax Syntax, 293 const std::vector<llvm::yaml::ScalarNode *> &Nodes) { 294 SmallString<1024> Storage; 295 std::vector<std::string> Arguments; 296 if (Nodes.size() == 1) 297 Arguments = unescapeCommandLine(Syntax, Nodes[0]->getValue(Storage)); 298 else 299 for (const auto *Node : Nodes) 300 Arguments.push_back(std::string(Node->getValue(Storage))); 301 // There may be multiple wrappers: using distcc and ccache together is common. 302 while (unwrapCommand(Arguments)) 303 ; 304 return Arguments; 305 } 306 307 void JSONCompilationDatabase::getCommands( 308 ArrayRef<CompileCommandRef> CommandsRef, 309 std::vector<CompileCommand> &Commands) const { 310 for (const auto &CommandRef : CommandsRef) { 311 SmallString<8> DirectoryStorage; 312 SmallString<32> FilenameStorage; 313 SmallString<32> OutputStorage; 314 auto Output = std::get<3>(CommandRef); 315 Commands.emplace_back( 316 std::get<0>(CommandRef)->getValue(DirectoryStorage), 317 std::get<1>(CommandRef)->getValue(FilenameStorage), 318 nodeToCommandLine(Syntax, std::get<2>(CommandRef)), 319 Output ? Output->getValue(OutputStorage) : ""); 320 } 321 } 322 323 bool JSONCompilationDatabase::parse(std::string &ErrorMessage) { 324 llvm::yaml::document_iterator I = YAMLStream.begin(); 325 if (I == YAMLStream.end()) { 326 ErrorMessage = "Error while parsing YAML."; 327 return false; 328 } 329 llvm::yaml::Node *Root = I->getRoot(); 330 if (!Root) { 331 ErrorMessage = "Error while parsing YAML."; 332 return false; 333 } 334 auto *Array = dyn_cast<llvm::yaml::SequenceNode>(Root); 335 if (!Array) { 336 ErrorMessage = "Expected array."; 337 return false; 338 } 339 for (auto &NextObject : *Array) { 340 auto *Object = dyn_cast<llvm::yaml::MappingNode>(&NextObject); 341 if (!Object) { 342 ErrorMessage = "Expected object."; 343 return false; 344 } 345 llvm::yaml::ScalarNode *Directory = nullptr; 346 std::optional<std::vector<llvm::yaml::ScalarNode *>> Command; 347 llvm::yaml::ScalarNode *File = nullptr; 348 llvm::yaml::ScalarNode *Output = nullptr; 349 for (auto& NextKeyValue : *Object) { 350 auto *KeyString = dyn_cast<llvm::yaml::ScalarNode>(NextKeyValue.getKey()); 351 if (!KeyString) { 352 ErrorMessage = "Expected strings as key."; 353 return false; 354 } 355 SmallString<10> KeyStorage; 356 StringRef KeyValue = KeyString->getValue(KeyStorage); 357 llvm::yaml::Node *Value = NextKeyValue.getValue(); 358 if (!Value) { 359 ErrorMessage = "Expected value."; 360 return false; 361 } 362 auto *ValueString = dyn_cast<llvm::yaml::ScalarNode>(Value); 363 auto *SequenceString = dyn_cast<llvm::yaml::SequenceNode>(Value); 364 if (KeyValue == "arguments") { 365 if (!SequenceString) { 366 ErrorMessage = "Expected sequence as value."; 367 return false; 368 } 369 Command = std::vector<llvm::yaml::ScalarNode *>(); 370 for (auto &Argument : *SequenceString) { 371 auto *Scalar = dyn_cast<llvm::yaml::ScalarNode>(&Argument); 372 if (!Scalar) { 373 ErrorMessage = "Only strings are allowed in 'arguments'."; 374 return false; 375 } 376 Command->push_back(Scalar); 377 } 378 } else { 379 if (!ValueString) { 380 ErrorMessage = "Expected string as value."; 381 return false; 382 } 383 if (KeyValue == "directory") { 384 Directory = ValueString; 385 } else if (KeyValue == "command") { 386 if (!Command) 387 Command = std::vector<llvm::yaml::ScalarNode *>(1, ValueString); 388 } else if (KeyValue == "file") { 389 File = ValueString; 390 } else if (KeyValue == "output") { 391 Output = ValueString; 392 } else { 393 ErrorMessage = 394 ("Unknown key: \"" + KeyString->getRawValue() + "\"").str(); 395 return false; 396 } 397 } 398 } 399 if (!File) { 400 ErrorMessage = "Missing key: \"file\"."; 401 return false; 402 } 403 if (!Command) { 404 ErrorMessage = "Missing key: \"command\" or \"arguments\"."; 405 return false; 406 } 407 if (!Directory) { 408 ErrorMessage = "Missing key: \"directory\"."; 409 return false; 410 } 411 SmallString<8> FileStorage; 412 StringRef FileName = File->getValue(FileStorage); 413 SmallString<128> NativeFilePath; 414 if (llvm::sys::path::is_relative(FileName)) { 415 SmallString<8> DirectoryStorage; 416 SmallString<128> AbsolutePath(Directory->getValue(DirectoryStorage)); 417 llvm::sys::path::append(AbsolutePath, FileName); 418 llvm::sys::path::native(AbsolutePath, NativeFilePath); 419 } else { 420 llvm::sys::path::native(FileName, NativeFilePath); 421 } 422 llvm::sys::path::remove_dots(NativeFilePath, /*remove_dot_dot=*/true); 423 auto Cmd = CompileCommandRef(Directory, File, *Command, Output); 424 IndexByFile[NativeFilePath].push_back(Cmd); 425 AllCommands.push_back(Cmd); 426 MatchTrie.insert(NativeFilePath); 427 } 428 return true; 429 } 430