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/Optional.h" 19 #include "llvm/ADT/STLExtras.h" 20 #include "llvm/ADT/SmallString.h" 21 #include "llvm/ADT/SmallVector.h" 22 #include "llvm/ADT/StringRef.h" 23 #include "llvm/ADT/Triple.h" 24 #include "llvm/Support/Allocator.h" 25 #include "llvm/Support/Casting.h" 26 #include "llvm/Support/CommandLine.h" 27 #include "llvm/Support/ErrorOr.h" 28 #include "llvm/Support/Host.h" 29 #include "llvm/Support/MemoryBuffer.h" 30 #include "llvm/Support/Path.h" 31 #include "llvm/Support/StringSaver.h" 32 #include "llvm/Support/VirtualFileSystem.h" 33 #include "llvm/Support/YAMLParser.h" 34 #include "llvm/Support/raw_ostream.h" 35 #include <cassert> 36 #include <memory> 37 #include <optional> 38 #include <string> 39 #include <system_error> 40 #include <tuple> 41 #include <utility> 42 #include <vector> 43 44 using namespace clang; 45 using namespace tooling; 46 47 namespace { 48 49 /// A parser for escaped strings of command line arguments. 50 /// 51 /// Assumes \-escaping for quoted arguments (see the documentation of 52 /// unescapeCommandLine(...)). 53 class CommandLineArgumentParser { 54 public: 55 CommandLineArgumentParser(StringRef CommandLine) 56 : Input(CommandLine), Position(Input.begin()-1) {} 57 58 std::vector<std::string> parse() { 59 bool HasMoreInput = true; 60 while (HasMoreInput && nextNonWhitespace()) { 61 std::string Argument; 62 HasMoreInput = parseStringInto(Argument); 63 CommandLine.push_back(Argument); 64 } 65 return CommandLine; 66 } 67 68 private: 69 // All private methods return true if there is more input available. 70 71 bool parseStringInto(std::string &String) { 72 do { 73 if (*Position == '"') { 74 if (!parseDoubleQuotedStringInto(String)) return false; 75 } else if (*Position == '\'') { 76 if (!parseSingleQuotedStringInto(String)) return false; 77 } else { 78 if (!parseFreeStringInto(String)) return false; 79 } 80 } while (*Position != ' '); 81 return true; 82 } 83 84 bool parseDoubleQuotedStringInto(std::string &String) { 85 if (!next()) return false; 86 while (*Position != '"') { 87 if (!skipEscapeCharacter()) return false; 88 String.push_back(*Position); 89 if (!next()) return false; 90 } 91 return next(); 92 } 93 94 bool parseSingleQuotedStringInto(std::string &String) { 95 if (!next()) return false; 96 while (*Position != '\'') { 97 String.push_back(*Position); 98 if (!next()) return false; 99 } 100 return next(); 101 } 102 103 bool parseFreeStringInto(std::string &String) { 104 do { 105 if (!skipEscapeCharacter()) return false; 106 String.push_back(*Position); 107 if (!next()) return false; 108 } while (*Position != ' ' && *Position != '"' && *Position != '\''); 109 return true; 110 } 111 112 bool skipEscapeCharacter() { 113 if (*Position == '\\') { 114 return next(); 115 } 116 return true; 117 } 118 119 bool nextNonWhitespace() { 120 do { 121 if (!next()) return false; 122 } while (*Position == ' '); 123 return true; 124 } 125 126 bool next() { 127 ++Position; 128 return Position != Input.end(); 129 } 130 131 const StringRef Input; 132 StringRef::iterator Position; 133 std::vector<std::string> CommandLine; 134 }; 135 136 std::vector<std::string> unescapeCommandLine(JSONCommandLineSyntax Syntax, 137 StringRef EscapedCommandLine) { 138 if (Syntax == JSONCommandLineSyntax::AutoDetect) { 139 #ifdef _WIN32 140 // Assume Windows command line parsing on Win32 141 Syntax = JSONCommandLineSyntax::Windows; 142 #else 143 Syntax = JSONCommandLineSyntax::Gnu; 144 #endif 145 } 146 147 if (Syntax == JSONCommandLineSyntax::Windows) { 148 llvm::BumpPtrAllocator Alloc; 149 llvm::StringSaver Saver(Alloc); 150 llvm::SmallVector<const char *, 64> T; 151 llvm::cl::TokenizeWindowsCommandLine(EscapedCommandLine, Saver, T); 152 std::vector<std::string> Result(T.begin(), T.end()); 153 return Result; 154 } 155 assert(Syntax == JSONCommandLineSyntax::Gnu); 156 CommandLineArgumentParser parser(EscapedCommandLine); 157 return parser.parse(); 158 } 159 160 // This plugin locates a nearby compile_command.json file, and also infers 161 // compile commands for files not present in the database. 162 class JSONCompilationDatabasePlugin : public CompilationDatabasePlugin { 163 std::unique_ptr<CompilationDatabase> 164 loadFromDirectory(StringRef Directory, std::string &ErrorMessage) override { 165 SmallString<1024> JSONDatabasePath(Directory); 166 llvm::sys::path::append(JSONDatabasePath, "compile_commands.json"); 167 auto Base = JSONCompilationDatabase::loadFromFile( 168 JSONDatabasePath, ErrorMessage, JSONCommandLineSyntax::AutoDetect); 169 return Base ? inferTargetAndDriverMode( 170 inferMissingCompileCommands(expandResponseFiles( 171 std::move(Base), llvm::vfs::getRealFileSystem()))) 172 : nullptr; 173 } 174 }; 175 176 } // namespace 177 178 // Register the JSONCompilationDatabasePlugin with the 179 // CompilationDatabasePluginRegistry using this statically initialized variable. 180 static CompilationDatabasePluginRegistry::Add<JSONCompilationDatabasePlugin> 181 X("json-compilation-database", "Reads JSON formatted compilation databases"); 182 183 namespace clang { 184 namespace tooling { 185 186 // This anchor is used to force the linker to link in the generated object file 187 // and thus register the JSONCompilationDatabasePlugin. 188 volatile int JSONAnchorSource = 0; 189 190 } // namespace tooling 191 } // namespace clang 192 193 std::unique_ptr<JSONCompilationDatabase> 194 JSONCompilationDatabase::loadFromFile(StringRef FilePath, 195 std::string &ErrorMessage, 196 JSONCommandLineSyntax Syntax) { 197 // Don't mmap: if we're a long-lived process, the build system may overwrite. 198 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> DatabaseBuffer = 199 llvm::MemoryBuffer::getFile(FilePath, /*IsText=*/false, 200 /*RequiresNullTerminator=*/true, 201 /*IsVolatile=*/true); 202 if (std::error_code Result = DatabaseBuffer.getError()) { 203 ErrorMessage = "Error while opening JSON database: " + Result.message(); 204 return nullptr; 205 } 206 std::unique_ptr<JSONCompilationDatabase> Database( 207 new JSONCompilationDatabase(std::move(*DatabaseBuffer), Syntax)); 208 if (!Database->parse(ErrorMessage)) 209 return nullptr; 210 return Database; 211 } 212 213 std::unique_ptr<JSONCompilationDatabase> 214 JSONCompilationDatabase::loadFromBuffer(StringRef DatabaseString, 215 std::string &ErrorMessage, 216 JSONCommandLineSyntax Syntax) { 217 std::unique_ptr<llvm::MemoryBuffer> DatabaseBuffer( 218 llvm::MemoryBuffer::getMemBufferCopy(DatabaseString)); 219 std::unique_ptr<JSONCompilationDatabase> Database( 220 new JSONCompilationDatabase(std::move(DatabaseBuffer), Syntax)); 221 if (!Database->parse(ErrorMessage)) 222 return nullptr; 223 return Database; 224 } 225 226 std::vector<CompileCommand> 227 JSONCompilationDatabase::getCompileCommands(StringRef FilePath) const { 228 SmallString<128> NativeFilePath; 229 llvm::sys::path::native(FilePath, NativeFilePath); 230 231 std::string Error; 232 llvm::raw_string_ostream ES(Error); 233 StringRef Match = MatchTrie.findEquivalent(NativeFilePath, ES); 234 if (Match.empty()) 235 return {}; 236 const auto CommandsRefI = IndexByFile.find(Match); 237 if (CommandsRefI == IndexByFile.end()) 238 return {}; 239 std::vector<CompileCommand> Commands; 240 getCommands(CommandsRefI->getValue(), Commands); 241 return Commands; 242 } 243 244 std::vector<std::string> 245 JSONCompilationDatabase::getAllFiles() const { 246 std::vector<std::string> Result; 247 for (const auto &CommandRef : IndexByFile) 248 Result.push_back(CommandRef.first().str()); 249 return Result; 250 } 251 252 std::vector<CompileCommand> 253 JSONCompilationDatabase::getAllCompileCommands() const { 254 std::vector<CompileCommand> Commands; 255 getCommands(AllCommands, Commands); 256 return Commands; 257 } 258 259 static llvm::StringRef stripExecutableExtension(llvm::StringRef Name) { 260 Name.consume_back(".exe"); 261 return Name; 262 } 263 264 // There are compiler-wrappers (ccache, distcc, gomacc) that take the "real" 265 // compiler as an argument, e.g. distcc gcc -O3 foo.c. 266 // These end up in compile_commands.json when people set CC="distcc gcc". 267 // Clang's driver doesn't understand this, so we need to unwrap. 268 static bool unwrapCommand(std::vector<std::string> &Args) { 269 if (Args.size() < 2) 270 return false; 271 StringRef Wrapper = 272 stripExecutableExtension(llvm::sys::path::filename(Args.front())); 273 if (Wrapper == "distcc" || Wrapper == "gomacc" || Wrapper == "ccache" || 274 Wrapper == "sccache") { 275 // Most of these wrappers support being invoked 3 ways: 276 // `distcc g++ file.c` This is the mode we're trying to match. 277 // We need to drop `distcc`. 278 // `distcc file.c` This acts like compiler is cc or similar. 279 // Clang's driver can handle this, no change needed. 280 // `g++ file.c` g++ is a symlink to distcc. 281 // We don't even notice this case, and all is well. 282 // 283 // We need to distinguish between the first and second case. 284 // The wrappers themselves don't take flags, so Args[1] is a compiler flag, 285 // an input file, or a compiler. Inputs have extensions, compilers don't. 286 bool HasCompiler = 287 (Args[1][0] != '-') && 288 !llvm::sys::path::has_extension(stripExecutableExtension(Args[1])); 289 if (HasCompiler) { 290 Args.erase(Args.begin()); 291 return true; 292 } 293 // If !HasCompiler, wrappers act like GCC. Fine: so do we. 294 } 295 return false; 296 } 297 298 static std::vector<std::string> 299 nodeToCommandLine(JSONCommandLineSyntax Syntax, 300 const std::vector<llvm::yaml::ScalarNode *> &Nodes) { 301 SmallString<1024> Storage; 302 std::vector<std::string> Arguments; 303 if (Nodes.size() == 1) 304 Arguments = unescapeCommandLine(Syntax, Nodes[0]->getValue(Storage)); 305 else 306 for (const auto *Node : Nodes) 307 Arguments.push_back(std::string(Node->getValue(Storage))); 308 // There may be multiple wrappers: using distcc and ccache together is common. 309 while (unwrapCommand(Arguments)) 310 ; 311 return Arguments; 312 } 313 314 void JSONCompilationDatabase::getCommands( 315 ArrayRef<CompileCommandRef> CommandsRef, 316 std::vector<CompileCommand> &Commands) const { 317 for (const auto &CommandRef : CommandsRef) { 318 SmallString<8> DirectoryStorage; 319 SmallString<32> FilenameStorage; 320 SmallString<32> OutputStorage; 321 auto Output = std::get<3>(CommandRef); 322 Commands.emplace_back( 323 std::get<0>(CommandRef)->getValue(DirectoryStorage), 324 std::get<1>(CommandRef)->getValue(FilenameStorage), 325 nodeToCommandLine(Syntax, std::get<2>(CommandRef)), 326 Output ? Output->getValue(OutputStorage) : ""); 327 } 328 } 329 330 bool JSONCompilationDatabase::parse(std::string &ErrorMessage) { 331 llvm::yaml::document_iterator I = YAMLStream.begin(); 332 if (I == YAMLStream.end()) { 333 ErrorMessage = "Error while parsing YAML."; 334 return false; 335 } 336 llvm::yaml::Node *Root = I->getRoot(); 337 if (!Root) { 338 ErrorMessage = "Error while parsing YAML."; 339 return false; 340 } 341 auto *Array = dyn_cast<llvm::yaml::SequenceNode>(Root); 342 if (!Array) { 343 ErrorMessage = "Expected array."; 344 return false; 345 } 346 for (auto &NextObject : *Array) { 347 auto *Object = dyn_cast<llvm::yaml::MappingNode>(&NextObject); 348 if (!Object) { 349 ErrorMessage = "Expected object."; 350 return false; 351 } 352 llvm::yaml::ScalarNode *Directory = nullptr; 353 llvm::Optional<std::vector<llvm::yaml::ScalarNode *>> Command; 354 llvm::yaml::ScalarNode *File = nullptr; 355 llvm::yaml::ScalarNode *Output = nullptr; 356 for (auto& NextKeyValue : *Object) { 357 auto *KeyString = dyn_cast<llvm::yaml::ScalarNode>(NextKeyValue.getKey()); 358 if (!KeyString) { 359 ErrorMessage = "Expected strings as key."; 360 return false; 361 } 362 SmallString<10> KeyStorage; 363 StringRef KeyValue = KeyString->getValue(KeyStorage); 364 llvm::yaml::Node *Value = NextKeyValue.getValue(); 365 if (!Value) { 366 ErrorMessage = "Expected value."; 367 return false; 368 } 369 auto *ValueString = dyn_cast<llvm::yaml::ScalarNode>(Value); 370 auto *SequenceString = dyn_cast<llvm::yaml::SequenceNode>(Value); 371 if (KeyValue == "arguments") { 372 if (!SequenceString) { 373 ErrorMessage = "Expected sequence as value."; 374 return false; 375 } 376 Command = std::vector<llvm::yaml::ScalarNode *>(); 377 for (auto &Argument : *SequenceString) { 378 auto *Scalar = dyn_cast<llvm::yaml::ScalarNode>(&Argument); 379 if (!Scalar) { 380 ErrorMessage = "Only strings are allowed in 'arguments'."; 381 return false; 382 } 383 Command->push_back(Scalar); 384 } 385 } else { 386 if (!ValueString) { 387 ErrorMessage = "Expected string as value."; 388 return false; 389 } 390 if (KeyValue == "directory") { 391 Directory = ValueString; 392 } else if (KeyValue == "command") { 393 if (!Command) 394 Command = std::vector<llvm::yaml::ScalarNode *>(1, ValueString); 395 } else if (KeyValue == "file") { 396 File = ValueString; 397 } else if (KeyValue == "output") { 398 Output = ValueString; 399 } else { 400 ErrorMessage = 401 ("Unknown key: \"" + KeyString->getRawValue() + "\"").str(); 402 return false; 403 } 404 } 405 } 406 if (!File) { 407 ErrorMessage = "Missing key: \"file\"."; 408 return false; 409 } 410 if (!Command) { 411 ErrorMessage = "Missing key: \"command\" or \"arguments\"."; 412 return false; 413 } 414 if (!Directory) { 415 ErrorMessage = "Missing key: \"directory\"."; 416 return false; 417 } 418 SmallString<8> FileStorage; 419 StringRef FileName = File->getValue(FileStorage); 420 SmallString<128> NativeFilePath; 421 if (llvm::sys::path::is_relative(FileName)) { 422 SmallString<8> DirectoryStorage; 423 SmallString<128> AbsolutePath(Directory->getValue(DirectoryStorage)); 424 llvm::sys::path::append(AbsolutePath, FileName); 425 llvm::sys::path::native(AbsolutePath, NativeFilePath); 426 } else { 427 llvm::sys::path::native(FileName, NativeFilePath); 428 } 429 llvm::sys::path::remove_dots(NativeFilePath, /*remove_dot_dot=*/true); 430 auto Cmd = CompileCommandRef(Directory, File, *Command, Output); 431 IndexByFile[NativeFilePath].push_back(Cmd); 432 AllCommands.push_back(Cmd); 433 MatchTrie.insert(NativeFilePath); 434 } 435 return true; 436 } 437