InclusionRewriter.cpp revision 3432b7800d447267f4353e70a319f6f5a0bb4aab
1//===--- InclusionRewriter.cpp - Rewrite includes into their expansions ---===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This code rewrites include invocations into their expansions. This gives you 11// a file with all included files merged into it. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Rewrite/Frontend/Rewriters.h" 16#include "clang/Basic/SourceManager.h" 17#include "clang/Frontend/PreprocessorOutputOptions.h" 18#include "clang/Lex/HeaderSearch.h" 19#include "clang/Lex/Pragma.h" 20#include "clang/Lex/Preprocessor.h" 21#include "llvm/ADT/SmallString.h" 22#include "llvm/Support/raw_ostream.h" 23 24using namespace clang; 25using namespace llvm; 26 27namespace { 28 29class InclusionRewriter : public PPCallbacks { 30 /// Information about which #includes were actually performed, 31 /// created by preprocessor callbacks. 32 struct FileChange { 33 const Module *Mod; 34 SourceLocation From; 35 FileID Id; 36 SrcMgr::CharacteristicKind FileType; 37 FileChange(SourceLocation From, const Module *Mod) : Mod(Mod), From(From) { 38 } 39 }; 40 Preprocessor &PP; ///< Used to find inclusion directives. 41 SourceManager &SM; ///< Used to read and manage source files. 42 raw_ostream &OS; ///< The destination stream for rewritten contents. 43 const llvm::MemoryBuffer *PredefinesBuffer; ///< The preprocessor predefines. 44 bool ShowLineMarkers; ///< Show #line markers. 45 bool UseLineDirective; ///< Use of line directives or line markers. 46 typedef std::map<unsigned, FileChange> FileChangeMap; 47 FileChangeMap FileChanges; ///< Tracks which files were included where. 48 /// Used transitively for building up the FileChanges mapping over the 49 /// various \c PPCallbacks callbacks. 50 FileChangeMap::iterator LastInsertedFileChange; 51public: 52 InclusionRewriter(Preprocessor &PP, raw_ostream &OS, bool ShowLineMarkers); 53 bool Process(FileID FileId, SrcMgr::CharacteristicKind FileType); 54 void setPredefinesBuffer(const llvm::MemoryBuffer *Buf) { 55 PredefinesBuffer = Buf; 56 } 57private: 58 virtual void FileChanged(SourceLocation Loc, FileChangeReason Reason, 59 SrcMgr::CharacteristicKind FileType, 60 FileID PrevFID); 61 virtual void FileSkipped(const FileEntry &ParentFile, 62 const Token &FilenameTok, 63 SrcMgr::CharacteristicKind FileType); 64 virtual void InclusionDirective(SourceLocation HashLoc, 65 const Token &IncludeTok, 66 StringRef FileName, 67 bool IsAngled, 68 CharSourceRange FilenameRange, 69 const FileEntry *File, 70 StringRef SearchPath, 71 StringRef RelativePath, 72 const Module *Imported); 73 void WriteLineInfo(const char *Filename, int Line, 74 SrcMgr::CharacteristicKind FileType, 75 StringRef EOL, StringRef Extra = StringRef()); 76 void WriteImplicitModuleImport(const Module *Mod, StringRef EOL); 77 void OutputContentUpTo(const MemoryBuffer &FromFile, 78 unsigned &WriteFrom, unsigned WriteTo, 79 StringRef EOL, int &lines, 80 bool EnsureNewline = false); 81 void CommentOutDirective(Lexer &DirectivesLex, const Token &StartToken, 82 const MemoryBuffer &FromFile, StringRef EOL, 83 unsigned &NextToWrite, int &Lines); 84 bool HandleHasInclude(FileID FileId, Lexer &RawLex, 85 const DirectoryLookup *Lookup, Token &Tok, 86 bool &FileExists); 87 const FileChange *FindFileChangeLocation(SourceLocation Loc) const; 88 StringRef NextIdentifierName(Lexer &RawLex, Token &RawToken); 89}; 90 91} // end anonymous namespace 92 93/// Initializes an InclusionRewriter with a \p PP source and \p OS destination. 94InclusionRewriter::InclusionRewriter(Preprocessor &PP, raw_ostream &OS, 95 bool ShowLineMarkers) 96 : PP(PP), SM(PP.getSourceManager()), OS(OS), PredefinesBuffer(0), 97 ShowLineMarkers(ShowLineMarkers), 98 LastInsertedFileChange(FileChanges.end()) { 99 // If we're in microsoft mode, use normal #line instead of line markers. 100 UseLineDirective = PP.getLangOpts().MicrosoftExt; 101} 102 103/// Write appropriate line information as either #line directives or GNU line 104/// markers depending on what mode we're in, including the \p Filename and 105/// \p Line we are located at, using the specified \p EOL line separator, and 106/// any \p Extra context specifiers in GNU line directives. 107void InclusionRewriter::WriteLineInfo(const char *Filename, int Line, 108 SrcMgr::CharacteristicKind FileType, 109 StringRef EOL, StringRef Extra) { 110 if (!ShowLineMarkers) 111 return; 112 if (UseLineDirective) { 113 OS << "#line" << ' ' << Line << ' ' << '"' << Filename << '"'; 114 } else { 115 // Use GNU linemarkers as described here: 116 // http://gcc.gnu.org/onlinedocs/cpp/Preprocessor-Output.html 117 OS << '#' << ' ' << Line << ' ' << '"'; 118 OS.write_escaped(Filename); 119 OS << '"'; 120 if (!Extra.empty()) 121 OS << Extra; 122 if (FileType == SrcMgr::C_System) 123 // "`3' This indicates that the following text comes from a system header 124 // file, so certain warnings should be suppressed." 125 OS << " 3"; 126 else if (FileType == SrcMgr::C_ExternCSystem) 127 // as above for `3', plus "`4' This indicates that the following text 128 // should be treated as being wrapped in an implicit extern "C" block." 129 OS << " 3 4"; 130 } 131 OS << EOL; 132} 133 134void InclusionRewriter::WriteImplicitModuleImport(const Module *Mod, 135 StringRef EOL) { 136 OS << "@import " << Mod->getFullModuleName() << ";" 137 << " /* clang -frewrite-includes: implicit import */" << EOL; 138} 139 140/// FileChanged - Whenever the preprocessor enters or exits a #include file 141/// it invokes this handler. 142void InclusionRewriter::FileChanged(SourceLocation Loc, 143 FileChangeReason Reason, 144 SrcMgr::CharacteristicKind NewFileType, 145 FileID) { 146 if (Reason != EnterFile) 147 return; 148 if (LastInsertedFileChange == FileChanges.end()) 149 // we didn't reach this file (eg: the main file) via an inclusion directive 150 return; 151 LastInsertedFileChange->second.Id = FullSourceLoc(Loc, SM).getFileID(); 152 LastInsertedFileChange->second.FileType = NewFileType; 153 LastInsertedFileChange = FileChanges.end(); 154} 155 156/// Called whenever an inclusion is skipped due to canonical header protection 157/// macros. 158void InclusionRewriter::FileSkipped(const FileEntry &/*ParentFile*/, 159 const Token &/*FilenameTok*/, 160 SrcMgr::CharacteristicKind /*FileType*/) { 161 assert(LastInsertedFileChange != FileChanges.end() && "A file, that wasn't " 162 "found via an inclusion directive, was skipped"); 163 FileChanges.erase(LastInsertedFileChange); 164 LastInsertedFileChange = FileChanges.end(); 165} 166 167/// This should be called whenever the preprocessor encounters include 168/// directives. It does not say whether the file has been included, but it 169/// provides more information about the directive (hash location instead 170/// of location inside the included file). It is assumed that the matching 171/// FileChanged() or FileSkipped() is called after this. 172void InclusionRewriter::InclusionDirective(SourceLocation HashLoc, 173 const Token &/*IncludeTok*/, 174 StringRef /*FileName*/, 175 bool /*IsAngled*/, 176 CharSourceRange /*FilenameRange*/, 177 const FileEntry * /*File*/, 178 StringRef /*SearchPath*/, 179 StringRef /*RelativePath*/, 180 const Module *Imported) { 181 assert(LastInsertedFileChange == FileChanges.end() && "Another inclusion " 182 "directive was found before the previous one was processed"); 183 std::pair<FileChangeMap::iterator, bool> p = FileChanges.insert( 184 std::make_pair(HashLoc.getRawEncoding(), FileChange(HashLoc, Imported))); 185 assert(p.second && "Unexpected revisitation of the same include directive"); 186 if (!Imported) 187 LastInsertedFileChange = p.first; 188} 189 190/// Simple lookup for a SourceLocation (specifically one denoting the hash in 191/// an inclusion directive) in the map of inclusion information, FileChanges. 192const InclusionRewriter::FileChange * 193InclusionRewriter::FindFileChangeLocation(SourceLocation Loc) const { 194 FileChangeMap::const_iterator I = FileChanges.find(Loc.getRawEncoding()); 195 if (I != FileChanges.end()) 196 return &I->second; 197 return NULL; 198} 199 200/// Detect the likely line ending style of \p FromFile by examining the first 201/// newline found within it. 202static StringRef DetectEOL(const MemoryBuffer &FromFile) { 203 // detect what line endings the file uses, so that added content does not mix 204 // the style 205 const char *Pos = strchr(FromFile.getBufferStart(), '\n'); 206 if (Pos == NULL) 207 return "\n"; 208 if (Pos + 1 < FromFile.getBufferEnd() && Pos[1] == '\r') 209 return "\n\r"; 210 if (Pos - 1 >= FromFile.getBufferStart() && Pos[-1] == '\r') 211 return "\r\n"; 212 return "\n"; 213} 214 215/// Writes out bytes from \p FromFile, starting at \p NextToWrite and ending at 216/// \p WriteTo - 1. 217void InclusionRewriter::OutputContentUpTo(const MemoryBuffer &FromFile, 218 unsigned &WriteFrom, unsigned WriteTo, 219 StringRef EOL, int &Line, 220 bool EnsureNewline) { 221 if (WriteTo <= WriteFrom) 222 return; 223 if (&FromFile == PredefinesBuffer) { 224 // Ignore the #defines of the predefines buffer. 225 WriteFrom = WriteTo; 226 return; 227 } 228 OS.write(FromFile.getBufferStart() + WriteFrom, WriteTo - WriteFrom); 229 // count lines manually, it's faster than getPresumedLoc() 230 Line += std::count(FromFile.getBufferStart() + WriteFrom, 231 FromFile.getBufferStart() + WriteTo, '\n'); 232 if (EnsureNewline) { 233 char LastChar = FromFile.getBufferStart()[WriteTo - 1]; 234 if (LastChar != '\n' && LastChar != '\r') 235 OS << EOL; 236 } 237 WriteFrom = WriteTo; 238} 239 240/// Print characters from \p FromFile starting at \p NextToWrite up until the 241/// inclusion directive at \p StartToken, then print out the inclusion 242/// inclusion directive disabled by a #if directive, updating \p NextToWrite 243/// and \p Line to track the number of source lines visited and the progress 244/// through the \p FromFile buffer. 245void InclusionRewriter::CommentOutDirective(Lexer &DirectiveLex, 246 const Token &StartToken, 247 const MemoryBuffer &FromFile, 248 StringRef EOL, 249 unsigned &NextToWrite, int &Line) { 250 OutputContentUpTo(FromFile, NextToWrite, 251 SM.getFileOffset(StartToken.getLocation()), EOL, Line); 252 Token DirectiveToken; 253 do { 254 DirectiveLex.LexFromRawLexer(DirectiveToken); 255 } while (!DirectiveToken.is(tok::eod) && DirectiveToken.isNot(tok::eof)); 256 OS << "#if 0 /* expanded by -frewrite-includes */" << EOL; 257 OutputContentUpTo(FromFile, NextToWrite, 258 SM.getFileOffset(DirectiveToken.getLocation()) + DirectiveToken.getLength(), 259 EOL, Line); 260 OS << "#endif /* expanded by -frewrite-includes */" << EOL; 261} 262 263/// Find the next identifier in the pragma directive specified by \p RawToken. 264StringRef InclusionRewriter::NextIdentifierName(Lexer &RawLex, 265 Token &RawToken) { 266 RawLex.LexFromRawLexer(RawToken); 267 if (RawToken.is(tok::raw_identifier)) 268 PP.LookUpIdentifierInfo(RawToken); 269 if (RawToken.is(tok::identifier)) 270 return RawToken.getIdentifierInfo()->getName(); 271 return StringRef(); 272} 273 274// Expand __has_include and __has_include_next if possible. If there's no 275// definitive answer return false. 276bool InclusionRewriter::HandleHasInclude( 277 FileID FileId, Lexer &RawLex, const DirectoryLookup *Lookup, Token &Tok, 278 bool &FileExists) { 279 // Lex the opening paren. 280 RawLex.LexFromRawLexer(Tok); 281 if (Tok.isNot(tok::l_paren)) 282 return false; 283 284 RawLex.LexFromRawLexer(Tok); 285 286 SmallString<128> FilenameBuffer; 287 StringRef Filename; 288 // Since the raw lexer doesn't give us angle_literals we have to parse them 289 // ourselves. 290 // FIXME: What to do if the file name is a macro? 291 if (Tok.is(tok::less)) { 292 RawLex.LexFromRawLexer(Tok); 293 294 FilenameBuffer += '<'; 295 do { 296 if (Tok.is(tok::eod)) // Sanity check. 297 return false; 298 299 if (Tok.is(tok::raw_identifier)) 300 PP.LookUpIdentifierInfo(Tok); 301 302 // Get the string piece. 303 SmallVector<char, 128> TmpBuffer; 304 bool Invalid = false; 305 StringRef TmpName = PP.getSpelling(Tok, TmpBuffer, &Invalid); 306 if (Invalid) 307 return false; 308 309 FilenameBuffer += TmpName; 310 311 RawLex.LexFromRawLexer(Tok); 312 } while (Tok.isNot(tok::greater)); 313 314 FilenameBuffer += '>'; 315 Filename = FilenameBuffer; 316 } else { 317 if (Tok.isNot(tok::string_literal)) 318 return false; 319 320 bool Invalid = false; 321 Filename = PP.getSpelling(Tok, FilenameBuffer, &Invalid); 322 if (Invalid) 323 return false; 324 } 325 326 // Lex the closing paren. 327 RawLex.LexFromRawLexer(Tok); 328 if (Tok.isNot(tok::r_paren)) 329 return false; 330 331 // Now ask HeaderInfo if it knows about the header. 332 // FIXME: Subframeworks aren't handled here. Do we care? 333 bool isAngled = PP.GetIncludeFilenameSpelling(Tok.getLocation(), Filename); 334 const DirectoryLookup *CurDir; 335 const FileEntry *File = PP.getHeaderSearchInfo().LookupFile( 336 Filename, isAngled, 0, CurDir, 337 PP.getSourceManager().getFileEntryForID(FileId), 0, 0, 0, false); 338 339 FileExists = File != 0; 340 return true; 341} 342 343/// Use a raw lexer to analyze \p FileId, inccrementally copying parts of it 344/// and including content of included files recursively. 345bool InclusionRewriter::Process(FileID FileId, 346 SrcMgr::CharacteristicKind FileType) 347{ 348 bool Invalid; 349 const MemoryBuffer &FromFile = *SM.getBuffer(FileId, &Invalid); 350 if (Invalid) // invalid inclusion 351 return false; 352 const char *FileName = FromFile.getBufferIdentifier(); 353 Lexer RawLex(FileId, &FromFile, PP.getSourceManager(), PP.getLangOpts()); 354 RawLex.SetCommentRetentionState(false); 355 356 StringRef EOL = DetectEOL(FromFile); 357 358 // Per the GNU docs: "1" indicates the start of a new file. 359 WriteLineInfo(FileName, 1, FileType, EOL, " 1"); 360 361 if (SM.getFileIDSize(FileId) == 0) 362 return false; 363 364 // The next byte to be copied from the source file 365 unsigned NextToWrite = 0; 366 int Line = 1; // The current input file line number. 367 368 Token RawToken; 369 RawLex.LexFromRawLexer(RawToken); 370 371 // TODO: Consider adding a switch that strips possibly unimportant content, 372 // such as comments, to reduce the size of repro files. 373 while (RawToken.isNot(tok::eof)) { 374 if (RawToken.is(tok::hash) && RawToken.isAtStartOfLine()) { 375 RawLex.setParsingPreprocessorDirective(true); 376 Token HashToken = RawToken; 377 RawLex.LexFromRawLexer(RawToken); 378 if (RawToken.is(tok::raw_identifier)) 379 PP.LookUpIdentifierInfo(RawToken); 380 if (RawToken.getIdentifierInfo() != NULL) { 381 switch (RawToken.getIdentifierInfo()->getPPKeywordID()) { 382 case tok::pp_include: 383 case tok::pp_include_next: 384 case tok::pp_import: { 385 CommentOutDirective(RawLex, HashToken, FromFile, EOL, NextToWrite, 386 Line); 387 StringRef LineInfoExtra; 388 if (const FileChange *Change = FindFileChangeLocation( 389 HashToken.getLocation())) { 390 if (Change->Mod) { 391 WriteImplicitModuleImport(Change->Mod, EOL); 392 393 // else now include and recursively process the file 394 } else if (Process(Change->Id, Change->FileType)) { 395 // and set lineinfo back to this file, if the nested one was 396 // actually included 397 // `2' indicates returning to a file (after having included 398 // another file. 399 LineInfoExtra = " 2"; 400 } 401 } 402 // fix up lineinfo (since commented out directive changed line 403 // numbers) for inclusions that were skipped due to header guards 404 WriteLineInfo(FileName, Line, FileType, EOL, LineInfoExtra); 405 break; 406 } 407 case tok::pp_pragma: { 408 StringRef Identifier = NextIdentifierName(RawLex, RawToken); 409 if (Identifier == "clang" || Identifier == "GCC") { 410 if (NextIdentifierName(RawLex, RawToken) == "system_header") { 411 // keep the directive in, commented out 412 CommentOutDirective(RawLex, HashToken, FromFile, EOL, 413 NextToWrite, Line); 414 // update our own type 415 FileType = SM.getFileCharacteristic(RawToken.getLocation()); 416 WriteLineInfo(FileName, Line, FileType, EOL); 417 } 418 } else if (Identifier == "once") { 419 // keep the directive in, commented out 420 CommentOutDirective(RawLex, HashToken, FromFile, EOL, 421 NextToWrite, Line); 422 WriteLineInfo(FileName, Line, FileType, EOL); 423 } 424 break; 425 } 426 case tok::pp_if: 427 case tok::pp_elif: { 428 bool elif = (RawToken.getIdentifierInfo()->getPPKeywordID() == 429 tok::pp_elif); 430 // Rewrite special builtin macros to avoid pulling in host details. 431 do { 432 // Walk over the directive. 433 RawLex.LexFromRawLexer(RawToken); 434 if (RawToken.is(tok::raw_identifier)) 435 PP.LookUpIdentifierInfo(RawToken); 436 437 if (RawToken.is(tok::identifier)) { 438 bool HasFile; 439 SourceLocation Loc = RawToken.getLocation(); 440 441 // Rewrite __has_include(x) 442 if (RawToken.getIdentifierInfo()->isStr("__has_include")) { 443 if (!HandleHasInclude(FileId, RawLex, 0, RawToken, HasFile)) 444 continue; 445 // Rewrite __has_include_next(x) 446 } else if (RawToken.getIdentifierInfo()->isStr( 447 "__has_include_next")) { 448 const DirectoryLookup *Lookup = PP.GetCurDirLookup(); 449 if (Lookup) 450 ++Lookup; 451 452 if (!HandleHasInclude(FileId, RawLex, Lookup, RawToken, 453 HasFile)) 454 continue; 455 } else { 456 continue; 457 } 458 // Replace the macro with (0) or (1), followed by the commented 459 // out macro for reference. 460 OutputContentUpTo(FromFile, NextToWrite, SM.getFileOffset(Loc), 461 EOL, Line); 462 OS << '(' << (int) HasFile << ")/*"; 463 OutputContentUpTo(FromFile, NextToWrite, 464 SM.getFileOffset(RawToken.getLocation()) + 465 RawToken.getLength(), 466 EOL, Line); 467 OS << "*/"; 468 } 469 } while (RawToken.isNot(tok::eod)); 470 if (elif) { 471 OutputContentUpTo(FromFile, NextToWrite, 472 SM.getFileOffset(RawToken.getLocation()) + 473 RawToken.getLength(), 474 EOL, Line, /*EnsureNewLine*/ true); 475 WriteLineInfo(FileName, Line, FileType, EOL); 476 } 477 break; 478 } 479 case tok::pp_endif: 480 case tok::pp_else: { 481 // We surround every #include by #if 0 to comment it out, but that 482 // changes line numbers. These are fixed up right after that, but 483 // the whole #include could be inside a preprocessor conditional 484 // that is not processed. So it is necessary to fix the line 485 // numbers one the next line after each #else/#endif as well. 486 RawLex.SetKeepWhitespaceMode(true); 487 do { 488 RawLex.LexFromRawLexer(RawToken); 489 } while (RawToken.isNot(tok::eod) && RawToken.isNot(tok::eof)); 490 OutputContentUpTo( 491 FromFile, NextToWrite, 492 SM.getFileOffset(RawToken.getLocation()) + RawToken.getLength(), 493 EOL, Line, /*EnsureNewLine*/ true); 494 WriteLineInfo(FileName, Line, FileType, EOL); 495 RawLex.SetKeepWhitespaceMode(false); 496 } 497 default: 498 break; 499 } 500 } 501 RawLex.setParsingPreprocessorDirective(false); 502 } 503 RawLex.LexFromRawLexer(RawToken); 504 } 505 OutputContentUpTo(FromFile, NextToWrite, 506 SM.getFileOffset(SM.getLocForEndOfFile(FileId)), EOL, Line, 507 /*EnsureNewline*/true); 508 return true; 509} 510 511/// InclusionRewriterInInput - Implement -frewrite-includes mode. 512void clang::RewriteIncludesInInput(Preprocessor &PP, raw_ostream *OS, 513 const PreprocessorOutputOptions &Opts) { 514 SourceManager &SM = PP.getSourceManager(); 515 InclusionRewriter *Rewrite = new InclusionRewriter(PP, *OS, 516 Opts.ShowLineMarkers); 517 PP.addPPCallbacks(Rewrite); 518 // Ignore all pragmas, otherwise there will be warnings about unknown pragmas 519 // (because there's nothing to handle them). 520 PP.AddPragmaHandler(new EmptyPragmaHandler()); 521 // Ignore also all pragma in all namespaces created 522 // in Preprocessor::RegisterBuiltinPragmas(). 523 PP.AddPragmaHandler("GCC", new EmptyPragmaHandler()); 524 PP.AddPragmaHandler("clang", new EmptyPragmaHandler()); 525 526 // First let the preprocessor process the entire file and call callbacks. 527 // Callbacks will record which #include's were actually performed. 528 PP.EnterMainSourceFile(); 529 Token Tok; 530 // Only preprocessor directives matter here, so disable macro expansion 531 // everywhere else as an optimization. 532 // TODO: It would be even faster if the preprocessor could be switched 533 // to a mode where it would parse only preprocessor directives and comments, 534 // nothing else matters for parsing or processing. 535 PP.SetMacroExpansionOnlyInDirectives(); 536 do { 537 PP.Lex(Tok); 538 } while (Tok.isNot(tok::eof)); 539 Rewrite->setPredefinesBuffer(SM.getBuffer(PP.getPredefinesFileID())); 540 Rewrite->Process(PP.getPredefinesFileID(), SrcMgr::C_User); 541 Rewrite->Process(SM.getMainFileID(), SrcMgr::C_User); 542 OS->flush(); 543} 544