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/YAMLParser.h" 33 #include "llvm/Support/raw_ostream.h" 34 #include <cassert> 35 #include <memory> 36 #include <string> 37 #include <system_error> 38 #include <tuple> 39 #include <utility> 40 #include <vector> 41 42 using namespace clang; 43 using namespace tooling; 44 45 namespace { 46 47 /// A parser for escaped strings of command line arguments. 48 /// 49 /// Assumes \-escaping for quoted arguments (see the documentation of 50 /// unescapeCommandLine(...)). 51 class CommandLineArgumentParser { 52 public: 53 CommandLineArgumentParser(StringRef CommandLine) 54 : Input(CommandLine), Position(Input.begin()-1) {} 55 56 std::vector<std::string> parse() { 57 bool HasMoreInput = true; 58 while (HasMoreInput && nextNonWhitespace()) { 59 std::string Argument; 60 HasMoreInput = parseStringInto(Argument); 61 CommandLine.push_back(Argument); 62 } 63 return CommandLine; 64 } 65 66 private: 67 // All private methods return true if there is more input available. 68 69 bool parseStringInto(std::string &String) { 70 do { 71 if (*Position == '"') { 72 if (!parseDoubleQuotedStringInto(String)) return false; 73 } else if (*Position == '\'') { 74 if (!parseSingleQuotedStringInto(String)) return false; 75 } else { 76 if (!parseFreeStringInto(String)) return false; 77 } 78 } while (*Position != ' '); 79 return true; 80 } 81 82 bool parseDoubleQuotedStringInto(std::string &String) { 83 if (!next()) return false; 84 while (*Position != '"') { 85 if (!skipEscapeCharacter()) return false; 86 String.push_back(*Position); 87 if (!next()) return false; 88 } 89 return next(); 90 } 91 92 bool parseSingleQuotedStringInto(std::string &String) { 93 if (!next()) return false; 94 while (*Position != '\'') { 95 String.push_back(*Position); 96 if (!next()) return false; 97 } 98 return next(); 99 } 100 101 bool parseFreeStringInto(std::string &String) { 102 do { 103 if (!skipEscapeCharacter()) return false; 104 String.push_back(*Position); 105 if (!next()) return false; 106 } while (*Position != ' ' && *Position != '"' && *Position != '\''); 107 return true; 108 } 109 110 bool skipEscapeCharacter() { 111 if (*Position == '\\') { 112 return next(); 113 } 114 return true; 115 } 116 117 bool nextNonWhitespace() { 118 do { 119 if (!next()) return false; 120 } while (*Position == ' '); 121 return true; 122 } 123 124 bool next() { 125 ++Position; 126 return Position != Input.end(); 127 } 128 129 const StringRef Input; 130 StringRef::iterator Position; 131 std::vector<std::string> CommandLine; 132 }; 133 134 std::vector<std::string> unescapeCommandLine(JSONCommandLineSyntax Syntax, 135 StringRef EscapedCommandLine) { 136 if (Syntax == JSONCommandLineSyntax::AutoDetect) { 137 Syntax = JSONCommandLineSyntax::Gnu; 138 llvm::Triple Triple(llvm::sys::getProcessTriple()); 139 if (Triple.getOS() == llvm::Triple::OSType::Win32) { 140 // Assume Windows command line parsing on Win32 unless the triple 141 // explicitly tells us otherwise. 142 if (!Triple.hasEnvironment() || 143 Triple.getEnvironment() == llvm::Triple::EnvironmentType::MSVC) 144 Syntax = JSONCommandLineSyntax::Windows; 145 } 146 } 147 148 if (Syntax == JSONCommandLineSyntax::Windows) { 149 llvm::BumpPtrAllocator Alloc; 150 llvm::StringSaver Saver(Alloc); 151 llvm::SmallVector<const char *, 64> T; 152 llvm::cl::TokenizeWindowsCommandLine(EscapedCommandLine, Saver, T); 153 std::vector<std::string> Result(T.begin(), T.end()); 154 return Result; 155 } 156 assert(Syntax == JSONCommandLineSyntax::Gnu); 157 CommandLineArgumentParser parser(EscapedCommandLine); 158 return parser.parse(); 159 } 160 161 // This plugin locates a nearby compile_command.json file, and also infers 162 // compile commands for files not present in the database. 163 class JSONCompilationDatabasePlugin : public CompilationDatabasePlugin { 164 std::unique_ptr<CompilationDatabase> 165 loadFromDirectory(StringRef Directory, std::string &ErrorMessage) override { 166 SmallString<1024> JSONDatabasePath(Directory); 167 llvm::sys::path::append(JSONDatabasePath, "compile_commands.json"); 168 auto Base = JSONCompilationDatabase::loadFromFile( 169 JSONDatabasePath, ErrorMessage, JSONCommandLineSyntax::AutoDetect); 170 return Base ? inferTargetAndDriverMode( 171 inferMissingCompileCommands(std::move(Base))) 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, /*FileSize=*/-1, 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::getMemBuffer(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 // Most of these wrappers support being invoked 3 ways: 275 // `distcc g++ file.c` This is the mode we're trying to match. 276 // We need to drop `distcc`. 277 // `distcc file.c` This acts like compiler is cc or similar. 278 // Clang's driver can handle this, no change needed. 279 // `g++ file.c` g++ is a symlink to distcc. 280 // We don't even notice this case, and all is well. 281 // 282 // We need to distinguish between the first and second case. 283 // The wrappers themselves don't take flags, so Args[1] is a compiler flag, 284 // an input file, or a compiler. Inputs have extensions, compilers don't. 285 bool HasCompiler = 286 (Args[1][0] != '-') && 287 !llvm::sys::path::has_extension(stripExecutableExtension(Args[1])); 288 if (HasCompiler) { 289 Args.erase(Args.begin()); 290 return true; 291 } 292 // If !HasCompiler, wrappers act like GCC. Fine: so do we. 293 } 294 return false; 295 } 296 297 static std::vector<std::string> 298 nodeToCommandLine(JSONCommandLineSyntax Syntax, 299 const std::vector<llvm::yaml::ScalarNode *> &Nodes) { 300 SmallString<1024> Storage; 301 std::vector<std::string> Arguments; 302 if (Nodes.size() == 1) 303 Arguments = unescapeCommandLine(Syntax, Nodes[0]->getValue(Storage)); 304 else 305 for (const auto *Node : Nodes) 306 Arguments.push_back(Node->getValue(Storage)); 307 // There may be multiple wrappers: using distcc and ccache together is common. 308 while (unwrapCommand(Arguments)) 309 ; 310 return Arguments; 311 } 312 313 void JSONCompilationDatabase::getCommands( 314 ArrayRef<CompileCommandRef> CommandsRef, 315 std::vector<CompileCommand> &Commands) const { 316 for (const auto &CommandRef : CommandsRef) { 317 SmallString<8> DirectoryStorage; 318 SmallString<32> FilenameStorage; 319 SmallString<32> OutputStorage; 320 auto Output = std::get<3>(CommandRef); 321 Commands.emplace_back( 322 std::get<0>(CommandRef)->getValue(DirectoryStorage), 323 std::get<1>(CommandRef)->getValue(FilenameStorage), 324 nodeToCommandLine(Syntax, std::get<2>(CommandRef)), 325 Output ? Output->getValue(OutputStorage) : ""); 326 } 327 } 328 329 bool JSONCompilationDatabase::parse(std::string &ErrorMessage) { 330 llvm::yaml::document_iterator I = YAMLStream.begin(); 331 if (I == YAMLStream.end()) { 332 ErrorMessage = "Error while parsing YAML."; 333 return false; 334 } 335 llvm::yaml::Node *Root = I->getRoot(); 336 if (!Root) { 337 ErrorMessage = "Error while parsing YAML."; 338 return false; 339 } 340 auto *Array = dyn_cast<llvm::yaml::SequenceNode>(Root); 341 if (!Array) { 342 ErrorMessage = "Expected array."; 343 return false; 344 } 345 for (auto &NextObject : *Array) { 346 auto *Object = dyn_cast<llvm::yaml::MappingNode>(&NextObject); 347 if (!Object) { 348 ErrorMessage = "Expected object."; 349 return false; 350 } 351 llvm::yaml::ScalarNode *Directory = nullptr; 352 llvm::Optional<std::vector<llvm::yaml::ScalarNode *>> Command; 353 llvm::yaml::ScalarNode *File = nullptr; 354 llvm::yaml::ScalarNode *Output = nullptr; 355 for (auto& NextKeyValue : *Object) { 356 auto *KeyString = dyn_cast<llvm::yaml::ScalarNode>(NextKeyValue.getKey()); 357 if (!KeyString) { 358 ErrorMessage = "Expected strings as key."; 359 return false; 360 } 361 SmallString<10> KeyStorage; 362 StringRef KeyValue = KeyString->getValue(KeyStorage); 363 llvm::yaml::Node *Value = NextKeyValue.getValue(); 364 if (!Value) { 365 ErrorMessage = "Expected value."; 366 return false; 367 } 368 auto *ValueString = dyn_cast<llvm::yaml::ScalarNode>(Value); 369 auto *SequenceString = dyn_cast<llvm::yaml::SequenceNode>(Value); 370 if (KeyValue == "arguments" && !SequenceString) { 371 ErrorMessage = "Expected sequence as value."; 372 return false; 373 } else if (KeyValue != "arguments" && !ValueString) { 374 ErrorMessage = "Expected string as value."; 375 return false; 376 } 377 if (KeyValue == "directory") { 378 Directory = ValueString; 379 } else if (KeyValue == "arguments") { 380 Command = std::vector<llvm::yaml::ScalarNode *>(); 381 for (auto &Argument : *SequenceString) { 382 auto *Scalar = dyn_cast<llvm::yaml::ScalarNode>(&Argument); 383 if (!Scalar) { 384 ErrorMessage = "Only strings are allowed in 'arguments'."; 385 return false; 386 } 387 Command->push_back(Scalar); 388 } 389 } else if (KeyValue == "command") { 390 if (!Command) 391 Command = std::vector<llvm::yaml::ScalarNode *>(1, ValueString); 392 } else if (KeyValue == "file") { 393 File = ValueString; 394 } else if (KeyValue == "output") { 395 Output = ValueString; 396 } else { 397 ErrorMessage = ("Unknown key: \"" + 398 KeyString->getRawValue() + "\"").str(); 399 return false; 400 } 401 } 402 if (!File) { 403 ErrorMessage = "Missing key: \"file\"."; 404 return false; 405 } 406 if (!Command) { 407 ErrorMessage = "Missing key: \"command\" or \"arguments\"."; 408 return false; 409 } 410 if (!Directory) { 411 ErrorMessage = "Missing key: \"directory\"."; 412 return false; 413 } 414 SmallString<8> FileStorage; 415 StringRef FileName = File->getValue(FileStorage); 416 SmallString<128> NativeFilePath; 417 if (llvm::sys::path::is_relative(FileName)) { 418 SmallString<8> DirectoryStorage; 419 SmallString<128> AbsolutePath( 420 Directory->getValue(DirectoryStorage)); 421 llvm::sys::path::append(AbsolutePath, FileName); 422 llvm::sys::path::remove_dots(AbsolutePath, /*remove_dot_dot=*/ true); 423 llvm::sys::path::native(AbsolutePath, NativeFilePath); 424 } else { 425 llvm::sys::path::native(FileName, NativeFilePath); 426 } 427 auto Cmd = CompileCommandRef(Directory, File, *Command, Output); 428 IndexByFile[NativeFilePath].push_back(Cmd); 429 AllCommands.push_back(Cmd); 430 MatchTrie.insert(NativeFilePath); 431 } 432 return true; 433 } 434