1 //===--- PPLexerChange.cpp - Handle changing lexers in the preprocessor ---===// 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 file implements pieces of the Preprocessor interface that manage the 11 // current lexer stack. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "clang/Lex/Preprocessor.h" 16 #include "clang/Lex/HeaderSearch.h" 17 #include "clang/Lex/MacroInfo.h" 18 #include "clang/Lex/LexDiagnostic.h" 19 #include "clang/Basic/FileManager.h" 20 #include "clang/Basic/SourceManager.h" 21 #include "llvm/Support/MemoryBuffer.h" 22 using namespace clang; 23 24 PPCallbacks::~PPCallbacks() {} 25 26 //===----------------------------------------------------------------------===// 27 // Miscellaneous Methods. 28 //===----------------------------------------------------------------------===// 29 30 /// isInPrimaryFile - Return true if we're in the top-level file, not in a 31 /// #include. This looks through macro expansions and active _Pragma lexers. 32 bool Preprocessor::isInPrimaryFile() const { 33 if (IsFileLexer()) 34 return IncludeMacroStack.empty(); 35 36 // If there are any stacked lexers, we're in a #include. 37 assert(IsFileLexer(IncludeMacroStack[0]) && 38 "Top level include stack isn't our primary lexer?"); 39 for (unsigned i = 1, e = IncludeMacroStack.size(); i != e; ++i) 40 if (IsFileLexer(IncludeMacroStack[i])) 41 return false; 42 return true; 43 } 44 45 /// getCurrentLexer - Return the current file lexer being lexed from. Note 46 /// that this ignores any potentially active macro expansions and _Pragma 47 /// expansions going on at the time. 48 PreprocessorLexer *Preprocessor::getCurrentFileLexer() const { 49 if (IsFileLexer()) 50 return CurPPLexer; 51 52 // Look for a stacked lexer. 53 for (unsigned i = IncludeMacroStack.size(); i != 0; --i) { 54 const IncludeStackInfo& ISI = IncludeMacroStack[i-1]; 55 if (IsFileLexer(ISI)) 56 return ISI.ThePPLexer; 57 } 58 return 0; 59 } 60 61 62 //===----------------------------------------------------------------------===// 63 // Methods for Entering and Callbacks for leaving various contexts 64 //===----------------------------------------------------------------------===// 65 66 /// EnterSourceFile - Add a source file to the top of the include stack and 67 /// start lexing tokens from it instead of the current buffer. 68 void Preprocessor::EnterSourceFile(FileID FID, const DirectoryLookup *CurDir, 69 SourceLocation Loc) { 70 assert(CurTokenLexer == 0 && "Cannot #include a file inside a macro!"); 71 ++NumEnteredSourceFiles; 72 73 if (MaxIncludeStackDepth < IncludeMacroStack.size()) 74 MaxIncludeStackDepth = IncludeMacroStack.size(); 75 76 if (PTH) { 77 if (PTHLexer *PL = PTH->CreateLexer(FID)) { 78 EnterSourceFileWithPTH(PL, CurDir); 79 return; 80 } 81 } 82 83 // Get the MemoryBuffer for this FID, if it fails, we fail. 84 bool Invalid = false; 85 const llvm::MemoryBuffer *InputFile = 86 getSourceManager().getBuffer(FID, Loc, &Invalid); 87 if (Invalid) { 88 SourceLocation FileStart = SourceMgr.getLocForStartOfFile(FID); 89 Diag(Loc, diag::err_pp_error_opening_file) 90 << std::string(SourceMgr.getBufferName(FileStart)) << ""; 91 return; 92 } 93 94 if (isCodeCompletionEnabled() && 95 SourceMgr.getFileEntryForID(FID) == CodeCompletionFile) { 96 CodeCompletionFileLoc = SourceMgr.getLocForStartOfFile(FID); 97 CodeCompletionLoc = 98 CodeCompletionFileLoc.getLocWithOffset(CodeCompletionOffset); 99 } 100 101 EnterSourceFileWithLexer(new Lexer(FID, InputFile, *this), CurDir); 102 return; 103 } 104 105 /// EnterSourceFileWithLexer - Add a source file to the top of the include stack 106 /// and start lexing tokens from it instead of the current buffer. 107 void Preprocessor::EnterSourceFileWithLexer(Lexer *TheLexer, 108 const DirectoryLookup *CurDir) { 109 110 // Add the current lexer to the include stack. 111 if (CurPPLexer || CurTokenLexer) 112 PushIncludeMacroStack(); 113 114 CurLexer.reset(TheLexer); 115 CurPPLexer = TheLexer; 116 CurDirLookup = CurDir; 117 if (CurLexerKind != CLK_LexAfterModuleImport) 118 CurLexerKind = CLK_Lexer; 119 120 // Notify the client, if desired, that we are in a new source file. 121 if (Callbacks && !CurLexer->Is_PragmaLexer) { 122 SrcMgr::CharacteristicKind FileType = 123 SourceMgr.getFileCharacteristic(CurLexer->getFileLoc()); 124 125 Callbacks->FileChanged(CurLexer->getFileLoc(), 126 PPCallbacks::EnterFile, FileType); 127 } 128 } 129 130 /// EnterSourceFileWithPTH - Add a source file to the top of the include stack 131 /// and start getting tokens from it using the PTH cache. 132 void Preprocessor::EnterSourceFileWithPTH(PTHLexer *PL, 133 const DirectoryLookup *CurDir) { 134 135 if (CurPPLexer || CurTokenLexer) 136 PushIncludeMacroStack(); 137 138 CurDirLookup = CurDir; 139 CurPTHLexer.reset(PL); 140 CurPPLexer = CurPTHLexer.get(); 141 if (CurLexerKind != CLK_LexAfterModuleImport) 142 CurLexerKind = CLK_PTHLexer; 143 144 // Notify the client, if desired, that we are in a new source file. 145 if (Callbacks) { 146 FileID FID = CurPPLexer->getFileID(); 147 SourceLocation EnterLoc = SourceMgr.getLocForStartOfFile(FID); 148 SrcMgr::CharacteristicKind FileType = 149 SourceMgr.getFileCharacteristic(EnterLoc); 150 Callbacks->FileChanged(EnterLoc, PPCallbacks::EnterFile, FileType); 151 } 152 } 153 154 /// EnterMacro - Add a Macro to the top of the include stack and start lexing 155 /// tokens from it instead of the current buffer. 156 void Preprocessor::EnterMacro(Token &Tok, SourceLocation ILEnd, 157 MacroArgs *Args) { 158 PushIncludeMacroStack(); 159 CurDirLookup = 0; 160 161 if (NumCachedTokenLexers == 0) { 162 CurTokenLexer.reset(new TokenLexer(Tok, ILEnd, Args, *this)); 163 } else { 164 CurTokenLexer.reset(TokenLexerCache[--NumCachedTokenLexers]); 165 CurTokenLexer->Init(Tok, ILEnd, Args); 166 } 167 if (CurLexerKind != CLK_LexAfterModuleImport) 168 CurLexerKind = CLK_TokenLexer; 169 } 170 171 /// EnterTokenStream - Add a "macro" context to the top of the include stack, 172 /// which will cause the lexer to start returning the specified tokens. 173 /// 174 /// If DisableMacroExpansion is true, tokens lexed from the token stream will 175 /// not be subject to further macro expansion. Otherwise, these tokens will 176 /// be re-macro-expanded when/if expansion is enabled. 177 /// 178 /// If OwnsTokens is false, this method assumes that the specified stream of 179 /// tokens has a permanent owner somewhere, so they do not need to be copied. 180 /// If it is true, it assumes the array of tokens is allocated with new[] and 181 /// must be freed. 182 /// 183 void Preprocessor::EnterTokenStream(const Token *Toks, unsigned NumToks, 184 bool DisableMacroExpansion, 185 bool OwnsTokens) { 186 // Save our current state. 187 PushIncludeMacroStack(); 188 CurDirLookup = 0; 189 190 // Create a macro expander to expand from the specified token stream. 191 if (NumCachedTokenLexers == 0) { 192 CurTokenLexer.reset(new TokenLexer(Toks, NumToks, DisableMacroExpansion, 193 OwnsTokens, *this)); 194 } else { 195 CurTokenLexer.reset(TokenLexerCache[--NumCachedTokenLexers]); 196 CurTokenLexer->Init(Toks, NumToks, DisableMacroExpansion, OwnsTokens); 197 } 198 if (CurLexerKind != CLK_LexAfterModuleImport) 199 CurLexerKind = CLK_TokenLexer; 200 } 201 202 /// HandleEndOfFile - This callback is invoked when the lexer hits the end of 203 /// the current file. This either returns the EOF token or pops a level off 204 /// the include stack and keeps going. 205 bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) { 206 assert(!CurTokenLexer && 207 "Ending a file when currently in a macro!"); 208 209 // See if this file had a controlling macro. 210 if (CurPPLexer) { // Not ending a macro, ignore it. 211 if (const IdentifierInfo *ControllingMacro = 212 CurPPLexer->MIOpt.GetControllingMacroAtEndOfFile()) { 213 // Okay, this has a controlling macro, remember in HeaderFileInfo. 214 if (const FileEntry *FE = 215 SourceMgr.getFileEntryForID(CurPPLexer->getFileID())) { 216 HeaderInfo.SetFileControllingMacro(FE, ControllingMacro); 217 218 // Controlling macros are implicitly private. 219 if (MacroInfo *MI = getMacroInfo( 220 const_cast<IdentifierInfo *>(ControllingMacro))) { 221 if (MI->getVisibilityLocation().isInvalid()) { 222 // FIXME: HACK! Mark controlling macros from system headers as 223 // exported, along with our own Clang headers. This is a gross 224 // hack to deal with the fact that system headers are included in 225 // many places within module headers, but are not themselves 226 // modularized. 227 if ((StringRef(FE->getName()).find("lib/clang") 228 == StringRef::npos) && 229 (StringRef(FE->getName()).find("usr/include") 230 == StringRef::npos) && 231 (StringRef(FE->getName()).find("usr/local/include") 232 == StringRef::npos)) 233 MI->setVisibility(false, SourceLocation()); 234 } 235 } 236 } 237 } 238 } 239 240 // Complain about reaching a true EOF within arc_cf_code_audited. 241 // We don't want to complain about reaching the end of a macro 242 // instantiation or a _Pragma. 243 if (PragmaARCCFCodeAuditedLoc.isValid() && 244 !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) { 245 Diag(PragmaARCCFCodeAuditedLoc, diag::err_pp_eof_in_arc_cf_code_audited); 246 247 // Recover by leaving immediately. 248 PragmaARCCFCodeAuditedLoc = SourceLocation(); 249 } 250 251 // If this is a #include'd file, pop it off the include stack and continue 252 // lexing the #includer file. 253 if (!IncludeMacroStack.empty()) { 254 255 // If we lexed the code-completion file, act as if we reached EOF. 256 if (isCodeCompletionEnabled() && CurPPLexer && 257 SourceMgr.getLocForStartOfFile(CurPPLexer->getFileID()) == 258 CodeCompletionFileLoc) { 259 if (CurLexer) { 260 Result.startToken(); 261 CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof); 262 CurLexer.reset(); 263 } else { 264 assert(CurPTHLexer && "Got EOF but no current lexer set!"); 265 CurPTHLexer->getEOF(Result); 266 CurPTHLexer.reset(); 267 } 268 269 CurPPLexer = 0; 270 return true; 271 } 272 273 if (!isEndOfMacro && CurPPLexer && 274 SourceMgr.getIncludeLoc(CurPPLexer->getFileID()).isValid()) { 275 // Notify SourceManager to record the number of FileIDs that were created 276 // during lexing of the #include'd file. 277 unsigned NumFIDs = 278 SourceMgr.local_sloc_entry_size() - 279 CurPPLexer->getInitialNumSLocEntries() + 1/*#include'd file*/; 280 SourceMgr.setNumCreatedFIDsForFileID(CurPPLexer->getFileID(), NumFIDs); 281 } 282 283 FileID ExitedFID; 284 if (Callbacks && !isEndOfMacro && CurPPLexer) 285 ExitedFID = CurPPLexer->getFileID(); 286 287 // We're done with the #included file. 288 RemoveTopOfLexerStack(); 289 290 // Notify the client, if desired, that we are in a new source file. 291 if (Callbacks && !isEndOfMacro && CurPPLexer) { 292 SrcMgr::CharacteristicKind FileType = 293 SourceMgr.getFileCharacteristic(CurPPLexer->getSourceLocation()); 294 Callbacks->FileChanged(CurPPLexer->getSourceLocation(), 295 PPCallbacks::ExitFile, FileType, ExitedFID); 296 } 297 298 // Client should lex another token. 299 return false; 300 } 301 302 // If the file ends with a newline, form the EOF token on the newline itself, 303 // rather than "on the line following it", which doesn't exist. This makes 304 // diagnostics relating to the end of file include the last file that the user 305 // actually typed, which is goodness. 306 if (CurLexer) { 307 const char *EndPos = CurLexer->BufferEnd; 308 if (EndPos != CurLexer->BufferStart && 309 (EndPos[-1] == '\n' || EndPos[-1] == '\r')) { 310 --EndPos; 311 312 // Handle \n\r and \r\n: 313 if (EndPos != CurLexer->BufferStart && 314 (EndPos[-1] == '\n' || EndPos[-1] == '\r') && 315 EndPos[-1] != EndPos[0]) 316 --EndPos; 317 } 318 319 Result.startToken(); 320 CurLexer->BufferPtr = EndPos; 321 CurLexer->FormTokenWithChars(Result, EndPos, tok::eof); 322 323 // We're done with the #included file. 324 CurLexer.reset(); 325 } else { 326 assert(CurPTHLexer && "Got EOF but no current lexer set!"); 327 CurPTHLexer->getEOF(Result); 328 CurPTHLexer.reset(); 329 } 330 331 CurPPLexer = 0; 332 333 // This is the end of the top-level file. 'WarnUnusedMacroLocs' has collected 334 // all macro locations that we need to warn because they are not used. 335 for (WarnUnusedMacroLocsTy::iterator 336 I=WarnUnusedMacroLocs.begin(), E=WarnUnusedMacroLocs.end(); I!=E; ++I) 337 Diag(*I, diag::pp_macro_not_used); 338 339 return true; 340 } 341 342 /// HandleEndOfTokenLexer - This callback is invoked when the current TokenLexer 343 /// hits the end of its token stream. 344 bool Preprocessor::HandleEndOfTokenLexer(Token &Result) { 345 assert(CurTokenLexer && !CurPPLexer && 346 "Ending a macro when currently in a #include file!"); 347 348 if (!MacroExpandingLexersStack.empty() && 349 MacroExpandingLexersStack.back().first == CurTokenLexer.get()) 350 removeCachedMacroExpandedTokensOfLastLexer(); 351 352 // Delete or cache the now-dead macro expander. 353 if (NumCachedTokenLexers == TokenLexerCacheSize) 354 CurTokenLexer.reset(); 355 else 356 TokenLexerCache[NumCachedTokenLexers++] = CurTokenLexer.take(); 357 358 // Handle this like a #include file being popped off the stack. 359 return HandleEndOfFile(Result, true); 360 } 361 362 /// RemoveTopOfLexerStack - Pop the current lexer/macro exp off the top of the 363 /// lexer stack. This should only be used in situations where the current 364 /// state of the top-of-stack lexer is unknown. 365 void Preprocessor::RemoveTopOfLexerStack() { 366 assert(!IncludeMacroStack.empty() && "Ran out of stack entries to load"); 367 368 if (CurTokenLexer) { 369 // Delete or cache the now-dead macro expander. 370 if (NumCachedTokenLexers == TokenLexerCacheSize) 371 CurTokenLexer.reset(); 372 else 373 TokenLexerCache[NumCachedTokenLexers++] = CurTokenLexer.take(); 374 } 375 376 PopIncludeMacroStack(); 377 } 378 379 /// HandleMicrosoftCommentPaste - When the macro expander pastes together a 380 /// comment (/##/) in microsoft mode, this method handles updating the current 381 /// state, returning the token on the next source line. 382 void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) { 383 assert(CurTokenLexer && !CurPPLexer && 384 "Pasted comment can only be formed from macro"); 385 386 // We handle this by scanning for the closest real lexer, switching it to 387 // raw mode and preprocessor mode. This will cause it to return \n as an 388 // explicit EOD token. 389 PreprocessorLexer *FoundLexer = 0; 390 bool LexerWasInPPMode = false; 391 for (unsigned i = 0, e = IncludeMacroStack.size(); i != e; ++i) { 392 IncludeStackInfo &ISI = *(IncludeMacroStack.end()-i-1); 393 if (ISI.ThePPLexer == 0) continue; // Scan for a real lexer. 394 395 // Once we find a real lexer, mark it as raw mode (disabling macro 396 // expansions) and preprocessor mode (return EOD). We know that the lexer 397 // was *not* in raw mode before, because the macro that the comment came 398 // from was expanded. However, it could have already been in preprocessor 399 // mode (#if COMMENT) in which case we have to return it to that mode and 400 // return EOD. 401 FoundLexer = ISI.ThePPLexer; 402 FoundLexer->LexingRawMode = true; 403 LexerWasInPPMode = FoundLexer->ParsingPreprocessorDirective; 404 FoundLexer->ParsingPreprocessorDirective = true; 405 break; 406 } 407 408 // Okay, we either found and switched over the lexer, or we didn't find a 409 // lexer. In either case, finish off the macro the comment came from, getting 410 // the next token. 411 if (!HandleEndOfTokenLexer(Tok)) Lex(Tok); 412 413 // Discarding comments as long as we don't have EOF or EOD. This 'comments 414 // out' the rest of the line, including any tokens that came from other macros 415 // that were active, as in: 416 // #define submacro a COMMENT b 417 // submacro c 418 // which should lex to 'a' only: 'b' and 'c' should be removed. 419 while (Tok.isNot(tok::eod) && Tok.isNot(tok::eof)) 420 Lex(Tok); 421 422 // If we got an eod token, then we successfully found the end of the line. 423 if (Tok.is(tok::eod)) { 424 assert(FoundLexer && "Can't get end of line without an active lexer"); 425 // Restore the lexer back to normal mode instead of raw mode. 426 FoundLexer->LexingRawMode = false; 427 428 // If the lexer was already in preprocessor mode, just return the EOD token 429 // to finish the preprocessor line. 430 if (LexerWasInPPMode) return; 431 432 // Otherwise, switch out of PP mode and return the next lexed token. 433 FoundLexer->ParsingPreprocessorDirective = false; 434 return Lex(Tok); 435 } 436 437 // If we got an EOF token, then we reached the end of the token stream but 438 // didn't find an explicit \n. This can only happen if there was no lexer 439 // active (an active lexer would return EOD at EOF if there was no \n in 440 // preprocessor directive mode), so just return EOF as our token. 441 assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode"); 442 } 443