1 //===--- PPLexerChange.cpp - Handle changing lexers in the preprocessor ---===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements pieces of the Preprocessor interface that manage the
11 // current lexer stack.
13 //===----------------------------------------------------------------------===//
15 #include "clang/Lex/Preprocessor.h"
16 #include "clang/Basic/FileManager.h"
17 #include "clang/Basic/SourceManager.h"
18 #include "clang/Lex/HeaderSearch.h"
19 #include "clang/Lex/LexDiagnostic.h"
20 #include "clang/Lex/MacroInfo.h"
21 #include "llvm/ADT/StringSwitch.h"
22 #include "llvm/Support/FileSystem.h"
23 #include "llvm/Support/MemoryBuffer.h"
24 #include "llvm/Support/Path.h"
25 using namespace clang;
27 PPCallbacks::~PPCallbacks() {}
29 //===----------------------------------------------------------------------===//
30 // Miscellaneous Methods.
31 //===----------------------------------------------------------------------===//
33 /// isInPrimaryFile - Return true if we're in the top-level file, not in a
34 /// \#include. This looks through macro expansions and active _Pragma lexers.
35 bool Preprocessor::isInPrimaryFile() const {
37 return IncludeMacroStack.empty();
39 // If there are any stacked lexers, we're in a #include.
40 assert(IsFileLexer(IncludeMacroStack[0]) &&
41 "Top level include stack isn't our primary lexer?");
42 for (unsigned i = 1, e = IncludeMacroStack.size(); i != e; ++i)
43 if (IsFileLexer(IncludeMacroStack[i]))
48 /// getCurrentLexer - Return the current file lexer being lexed from. Note
49 /// that this ignores any potentially active macro expansions and _Pragma
50 /// expansions going on at the time.
51 PreprocessorLexer *Preprocessor::getCurrentFileLexer() const {
55 // Look for a stacked lexer.
56 for (unsigned i = IncludeMacroStack.size(); i != 0; --i) {
57 const IncludeStackInfo& ISI = IncludeMacroStack[i-1];
59 return ISI.ThePPLexer;
65 //===----------------------------------------------------------------------===//
66 // Methods for Entering and Callbacks for leaving various contexts
67 //===----------------------------------------------------------------------===//
69 /// EnterSourceFile - Add a source file to the top of the include stack and
70 /// start lexing tokens from it instead of the current buffer.
71 void Preprocessor::EnterSourceFile(FileID FID, const DirectoryLookup *CurDir,
73 assert(!CurTokenLexer && "Cannot #include a file inside a macro!");
74 ++NumEnteredSourceFiles;
76 if (MaxIncludeStackDepth < IncludeMacroStack.size())
77 MaxIncludeStackDepth = IncludeMacroStack.size();
80 if (PTHLexer *PL = PTH->CreateLexer(FID)) {
81 EnterSourceFileWithPTH(PL, CurDir);
86 // Get the MemoryBuffer for this FID, if it fails, we fail.
88 const llvm::MemoryBuffer *InputFile =
89 getSourceManager().getBuffer(FID, Loc, &Invalid);
91 SourceLocation FileStart = SourceMgr.getLocForStartOfFile(FID);
92 Diag(Loc, diag::err_pp_error_opening_file)
93 << std::string(SourceMgr.getBufferName(FileStart)) << "";
97 if (isCodeCompletionEnabled() &&
98 SourceMgr.getFileEntryForID(FID) == CodeCompletionFile) {
99 CodeCompletionFileLoc = SourceMgr.getLocForStartOfFile(FID);
101 CodeCompletionFileLoc.getLocWithOffset(CodeCompletionOffset);
104 EnterSourceFileWithLexer(new Lexer(FID, InputFile, *this), CurDir);
108 /// EnterSourceFileWithLexer - Add a source file to the top of the include stack
109 /// and start lexing tokens from it instead of the current buffer.
110 void Preprocessor::EnterSourceFileWithLexer(Lexer *TheLexer,
111 const DirectoryLookup *CurDir) {
113 // Add the current lexer to the include stack.
114 if (CurPPLexer || CurTokenLexer)
115 PushIncludeMacroStack();
117 CurLexer.reset(TheLexer);
118 CurPPLexer = TheLexer;
119 CurDirLookup = CurDir;
120 if (CurLexerKind != CLK_LexAfterModuleImport)
121 CurLexerKind = CLK_Lexer;
123 // Notify the client, if desired, that we are in a new source file.
124 if (Callbacks && !CurLexer->Is_PragmaLexer) {
125 SrcMgr::CharacteristicKind FileType =
126 SourceMgr.getFileCharacteristic(CurLexer->getFileLoc());
128 Callbacks->FileChanged(CurLexer->getFileLoc(),
129 PPCallbacks::EnterFile, FileType);
133 /// EnterSourceFileWithPTH - Add a source file to the top of the include stack
134 /// and start getting tokens from it using the PTH cache.
135 void Preprocessor::EnterSourceFileWithPTH(PTHLexer *PL,
136 const DirectoryLookup *CurDir) {
138 if (CurPPLexer || CurTokenLexer)
139 PushIncludeMacroStack();
141 CurDirLookup = CurDir;
142 CurPTHLexer.reset(PL);
143 CurPPLexer = CurPTHLexer.get();
144 if (CurLexerKind != CLK_LexAfterModuleImport)
145 CurLexerKind = CLK_PTHLexer;
147 // Notify the client, if desired, that we are in a new source file.
149 FileID FID = CurPPLexer->getFileID();
150 SourceLocation EnterLoc = SourceMgr.getLocForStartOfFile(FID);
151 SrcMgr::CharacteristicKind FileType =
152 SourceMgr.getFileCharacteristic(EnterLoc);
153 Callbacks->FileChanged(EnterLoc, PPCallbacks::EnterFile, FileType);
157 /// EnterMacro - Add a Macro to the top of the include stack and start lexing
158 /// tokens from it instead of the current buffer.
159 void Preprocessor::EnterMacro(Token &Tok, SourceLocation ILEnd,
160 MacroInfo *Macro, MacroArgs *Args) {
161 TokenLexer *TokLexer;
162 if (NumCachedTokenLexers == 0) {
163 TokLexer = new TokenLexer(Tok, ILEnd, Macro, Args, *this);
165 TokLexer = TokenLexerCache[--NumCachedTokenLexers];
166 TokLexer->Init(Tok, ILEnd, Macro, Args);
169 PushIncludeMacroStack();
171 CurTokenLexer.reset(TokLexer);
172 if (CurLexerKind != CLK_LexAfterModuleImport)
173 CurLexerKind = CLK_TokenLexer;
176 /// EnterTokenStream - Add a "macro" context to the top of the include stack,
177 /// which will cause the lexer to start returning the specified tokens.
179 /// If DisableMacroExpansion is true, tokens lexed from the token stream will
180 /// not be subject to further macro expansion. Otherwise, these tokens will
181 /// be re-macro-expanded when/if expansion is enabled.
183 /// If OwnsTokens is false, this method assumes that the specified stream of
184 /// tokens has a permanent owner somewhere, so they do not need to be copied.
185 /// If it is true, it assumes the array of tokens is allocated with new[] and
188 void Preprocessor::EnterTokenStream(const Token *Toks, unsigned NumToks,
189 bool DisableMacroExpansion,
191 // Create a macro expander to expand from the specified token stream.
192 TokenLexer *TokLexer;
193 if (NumCachedTokenLexers == 0) {
194 TokLexer = new TokenLexer(Toks, NumToks, DisableMacroExpansion,
197 TokLexer = TokenLexerCache[--NumCachedTokenLexers];
198 TokLexer->Init(Toks, NumToks, DisableMacroExpansion, OwnsTokens);
201 // Save our current state.
202 PushIncludeMacroStack();
204 CurTokenLexer.reset(TokLexer);
205 if (CurLexerKind != CLK_LexAfterModuleImport)
206 CurLexerKind = CLK_TokenLexer;
209 /// \brief Compute the relative path that names the given file relative to
210 /// the given directory.
211 static void computeRelativePath(FileManager &FM, const DirectoryEntry *Dir,
212 const FileEntry *File,
213 SmallString<128> &Result) {
216 StringRef FilePath = File->getDir()->getName();
217 StringRef Path = FilePath;
218 while (!Path.empty()) {
219 if (const DirectoryEntry *CurDir = FM.getDirectory(Path)) {
221 Result = FilePath.substr(Path.size());
222 llvm::sys::path::append(Result,
223 llvm::sys::path::filename(File->getName()));
228 Path = llvm::sys::path::parent_path(Path);
231 Result = File->getName();
234 void Preprocessor::PropagateLineStartLeadingSpaceInfo(Token &Result) {
236 CurTokenLexer->PropagateLineStartLeadingSpaceInfo(Result);
240 CurLexer->PropagateLineStartLeadingSpaceInfo(Result);
243 // FIXME: Handle other kinds of lexers? It generally shouldn't matter,
244 // but it might if they're empty?
247 /// HandleEndOfFile - This callback is invoked when the lexer hits the end of
248 /// the current file. This either returns the EOF token or pops a level off
249 /// the include stack and keeps going.
250 bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) {
251 assert(!CurTokenLexer &&
252 "Ending a file when currently in a macro!");
254 // See if this file had a controlling macro.
255 if (CurPPLexer) { // Not ending a macro, ignore it.
256 if (const IdentifierInfo *ControllingMacro =
257 CurPPLexer->MIOpt.GetControllingMacroAtEndOfFile()) {
258 // Okay, this has a controlling macro, remember in HeaderFileInfo.
259 if (const FileEntry *FE =
260 SourceMgr.getFileEntryForID(CurPPLexer->getFileID())) {
261 HeaderInfo.SetFileControllingMacro(FE, ControllingMacro);
262 if (const IdentifierInfo *DefinedMacro =
263 CurPPLexer->MIOpt.GetDefinedMacro()) {
264 if (!ControllingMacro->hasMacroDefinition() &&
265 DefinedMacro != ControllingMacro &&
266 HeaderInfo.FirstTimeLexingFile(FE)) {
268 // If the edit distance between the two macros is more than 50%,
269 // DefinedMacro may not be header guard, or can be header guard of
270 // another header file. Therefore, it maybe defining something
271 // completely different. This can be observed in the wild when
272 // handling feature macros or header guards in different files.
274 const StringRef ControllingMacroName = ControllingMacro->getName();
275 const StringRef DefinedMacroName = DefinedMacro->getName();
276 const size_t MaxHalfLength = std::max(ControllingMacroName.size(),
277 DefinedMacroName.size()) / 2;
278 const unsigned ED = ControllingMacroName.edit_distance(
279 DefinedMacroName, true, MaxHalfLength);
280 if (ED <= MaxHalfLength) {
281 // Emit a warning for a bad header guard.
282 Diag(CurPPLexer->MIOpt.GetMacroLocation(),
283 diag::warn_header_guard)
284 << CurPPLexer->MIOpt.GetMacroLocation() << ControllingMacro;
285 Diag(CurPPLexer->MIOpt.GetDefinedLocation(),
286 diag::note_header_guard)
287 << CurPPLexer->MIOpt.GetDefinedLocation() << DefinedMacro
289 << FixItHint::CreateReplacement(
290 CurPPLexer->MIOpt.GetDefinedLocation(),
291 ControllingMacro->getName());
299 // Complain about reaching a true EOF within arc_cf_code_audited.
300 // We don't want to complain about reaching the end of a macro
301 // instantiation or a _Pragma.
302 if (PragmaARCCFCodeAuditedLoc.isValid() &&
303 !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) {
304 Diag(PragmaARCCFCodeAuditedLoc, diag::err_pp_eof_in_arc_cf_code_audited);
306 // Recover by leaving immediately.
307 PragmaARCCFCodeAuditedLoc = SourceLocation();
310 // If this is a #include'd file, pop it off the include stack and continue
311 // lexing the #includer file.
312 if (!IncludeMacroStack.empty()) {
314 // If we lexed the code-completion file, act as if we reached EOF.
315 if (isCodeCompletionEnabled() && CurPPLexer &&
316 SourceMgr.getLocForStartOfFile(CurPPLexer->getFileID()) ==
317 CodeCompletionFileLoc) {
320 CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof);
323 assert(CurPTHLexer && "Got EOF but no current lexer set!");
324 CurPTHLexer->getEOF(Result);
332 if (!isEndOfMacro && CurPPLexer &&
333 SourceMgr.getIncludeLoc(CurPPLexer->getFileID()).isValid()) {
334 // Notify SourceManager to record the number of FileIDs that were created
335 // during lexing of the #include'd file.
337 SourceMgr.local_sloc_entry_size() -
338 CurPPLexer->getInitialNumSLocEntries() + 1/*#include'd file*/;
339 SourceMgr.setNumCreatedFIDsForFileID(CurPPLexer->getFileID(), NumFIDs);
343 if (Callbacks && !isEndOfMacro && CurPPLexer)
344 ExitedFID = CurPPLexer->getFileID();
346 // We're done with the #included file.
347 RemoveTopOfLexerStack();
349 // Propagate info about start-of-line/leading white-space/etc.
350 PropagateLineStartLeadingSpaceInfo(Result);
352 // Notify the client, if desired, that we are in a new source file.
353 if (Callbacks && !isEndOfMacro && CurPPLexer) {
354 SrcMgr::CharacteristicKind FileType =
355 SourceMgr.getFileCharacteristic(CurPPLexer->getSourceLocation());
356 Callbacks->FileChanged(CurPPLexer->getSourceLocation(),
357 PPCallbacks::ExitFile, FileType, ExitedFID);
360 // Client should lex another token.
364 // If the file ends with a newline, form the EOF token on the newline itself,
365 // rather than "on the line following it", which doesn't exist. This makes
366 // diagnostics relating to the end of file include the last file that the user
367 // actually typed, which is goodness.
369 const char *EndPos = CurLexer->BufferEnd;
370 if (EndPos != CurLexer->BufferStart &&
371 (EndPos[-1] == '\n' || EndPos[-1] == '\r')) {
374 // Handle \n\r and \r\n:
375 if (EndPos != CurLexer->BufferStart &&
376 (EndPos[-1] == '\n' || EndPos[-1] == '\r') &&
377 EndPos[-1] != EndPos[0])
382 CurLexer->BufferPtr = EndPos;
383 CurLexer->FormTokenWithChars(Result, EndPos, tok::eof);
385 if (isCodeCompletionEnabled()) {
386 // Inserting the code-completion point increases the source buffer by 1,
387 // but the main FileID was created before inserting the point.
388 // Compensate by reducing the EOF location by 1, otherwise the location
389 // will point to the next FileID.
390 // FIXME: This is hacky, the code-completion point should probably be
391 // inserted before the main FileID is created.
392 if (CurLexer->getFileLoc() == CodeCompletionFileLoc)
393 Result.setLocation(Result.getLocation().getLocWithOffset(-1));
396 if (!isIncrementalProcessingEnabled())
397 // We're done with lexing.
400 assert(CurPTHLexer && "Got EOF but no current lexer set!");
401 CurPTHLexer->getEOF(Result);
405 if (!isIncrementalProcessingEnabled())
408 // This is the end of the top-level file. 'WarnUnusedMacroLocs' has collected
409 // all macro locations that we need to warn because they are not used.
410 for (WarnUnusedMacroLocsTy::iterator
411 I=WarnUnusedMacroLocs.begin(), E=WarnUnusedMacroLocs.end(); I!=E; ++I)
412 Diag(*I, diag::pp_macro_not_used);
414 // If we are building a module that has an umbrella header, make sure that
415 // each of the headers within the directory covered by the umbrella header
416 // was actually included by the umbrella header.
417 if (Module *Mod = getCurrentModule()) {
418 if (Mod->getUmbrellaHeader()) {
419 SourceLocation StartLoc
420 = SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID());
422 if (getDiagnostics().getDiagnosticLevel(
423 diag::warn_uncovered_module_header,
424 StartLoc) != DiagnosticsEngine::Ignored) {
425 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
426 typedef llvm::sys::fs::recursive_directory_iterator
427 recursive_directory_iterator;
428 const DirectoryEntry *Dir = Mod->getUmbrellaDir();
430 for (recursive_directory_iterator Entry(Dir->getName(), EC), End;
431 Entry != End && !EC; Entry.increment(EC)) {
432 using llvm::StringSwitch;
434 // Check whether this entry has an extension typically associated with
436 if (!StringSwitch<bool>(llvm::sys::path::extension(Entry->path()))
437 .Cases(".h", ".H", ".hh", ".hpp", true)
441 if (const FileEntry *Header = getFileManager().getFile(Entry->path()))
442 if (!getSourceManager().hasFileInfo(Header)) {
443 if (!ModMap.isHeaderInUnavailableModule(Header)) {
444 // Find the relative path that would access this header.
445 SmallString<128> RelativePath;
446 computeRelativePath(FileMgr, Dir, Header, RelativePath);
447 Diag(StartLoc, diag::warn_uncovered_module_header)
448 << Mod->getFullModuleName() << RelativePath;
455 // Check whether there are any headers that were included, but not
456 // mentioned at all in the module map. Such headers
457 SourceLocation StartLoc
458 = SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID());
459 if (getDiagnostics().getDiagnosticLevel(diag::warn_forgotten_module_header,
461 != DiagnosticsEngine::Ignored) {
462 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
463 for (unsigned I = 0, N = SourceMgr.local_sloc_entry_size(); I != N; ++I) {
464 // We only care about file entries.
465 const SrcMgr::SLocEntry &Entry = SourceMgr.getLocalSLocEntry(I);
469 // Dig out the actual file.
470 const FileEntry *File = Entry.getFile().getContentCache()->OrigEntry;
474 // If it's not part of a module and not unknown, complain.
475 if (!ModMap.findModuleForHeader(File) &&
476 !ModMap.isHeaderInUnavailableModule(File)) {
477 Diag(StartLoc, diag::warn_forgotten_module_header)
478 << File->getName() << Mod->getFullModuleName();
487 /// HandleEndOfTokenLexer - This callback is invoked when the current TokenLexer
488 /// hits the end of its token stream.
489 bool Preprocessor::HandleEndOfTokenLexer(Token &Result) {
490 assert(CurTokenLexer && !CurPPLexer &&
491 "Ending a macro when currently in a #include file!");
493 if (!MacroExpandingLexersStack.empty() &&
494 MacroExpandingLexersStack.back().first == CurTokenLexer.get())
495 removeCachedMacroExpandedTokensOfLastLexer();
497 // Delete or cache the now-dead macro expander.
498 if (NumCachedTokenLexers == TokenLexerCacheSize)
499 CurTokenLexer.reset();
501 TokenLexerCache[NumCachedTokenLexers++] = CurTokenLexer.take();
503 // Handle this like a #include file being popped off the stack.
504 return HandleEndOfFile(Result, true);
507 /// RemoveTopOfLexerStack - Pop the current lexer/macro exp off the top of the
508 /// lexer stack. This should only be used in situations where the current
509 /// state of the top-of-stack lexer is unknown.
510 void Preprocessor::RemoveTopOfLexerStack() {
511 assert(!IncludeMacroStack.empty() && "Ran out of stack entries to load");
514 // Delete or cache the now-dead macro expander.
515 if (NumCachedTokenLexers == TokenLexerCacheSize)
516 CurTokenLexer.reset();
518 TokenLexerCache[NumCachedTokenLexers++] = CurTokenLexer.take();
521 PopIncludeMacroStack();
524 /// HandleMicrosoftCommentPaste - When the macro expander pastes together a
525 /// comment (/##/) in microsoft mode, this method handles updating the current
526 /// state, returning the token on the next source line.
527 void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) {
528 assert(CurTokenLexer && !CurPPLexer &&
529 "Pasted comment can only be formed from macro");
531 // We handle this by scanning for the closest real lexer, switching it to
532 // raw mode and preprocessor mode. This will cause it to return \n as an
533 // explicit EOD token.
534 PreprocessorLexer *FoundLexer = 0;
535 bool LexerWasInPPMode = false;
536 for (unsigned i = 0, e = IncludeMacroStack.size(); i != e; ++i) {
537 IncludeStackInfo &ISI = *(IncludeMacroStack.end()-i-1);
538 if (ISI.ThePPLexer == 0) continue; // Scan for a real lexer.
540 // Once we find a real lexer, mark it as raw mode (disabling macro
541 // expansions) and preprocessor mode (return EOD). We know that the lexer
542 // was *not* in raw mode before, because the macro that the comment came
543 // from was expanded. However, it could have already been in preprocessor
544 // mode (#if COMMENT) in which case we have to return it to that mode and
546 FoundLexer = ISI.ThePPLexer;
547 FoundLexer->LexingRawMode = true;
548 LexerWasInPPMode = FoundLexer->ParsingPreprocessorDirective;
549 FoundLexer->ParsingPreprocessorDirective = true;
553 // Okay, we either found and switched over the lexer, or we didn't find a
554 // lexer. In either case, finish off the macro the comment came from, getting
556 if (!HandleEndOfTokenLexer(Tok)) Lex(Tok);
558 // Discarding comments as long as we don't have EOF or EOD. This 'comments
559 // out' the rest of the line, including any tokens that came from other macros
560 // that were active, as in:
561 // #define submacro a COMMENT b
563 // which should lex to 'a' only: 'b' and 'c' should be removed.
564 while (Tok.isNot(tok::eod) && Tok.isNot(tok::eof))
567 // If we got an eod token, then we successfully found the end of the line.
568 if (Tok.is(tok::eod)) {
569 assert(FoundLexer && "Can't get end of line without an active lexer");
570 // Restore the lexer back to normal mode instead of raw mode.
571 FoundLexer->LexingRawMode = false;
573 // If the lexer was already in preprocessor mode, just return the EOD token
574 // to finish the preprocessor line.
575 if (LexerWasInPPMode) return;
577 // Otherwise, switch out of PP mode and return the next lexed token.
578 FoundLexer->ParsingPreprocessorDirective = false;
582 // If we got an EOF token, then we reached the end of the token stream but
583 // didn't find an explicit \n. This can only happen if there was no lexer
584 // active (an active lexer would return EOD at EOF if there was no \n in
585 // preprocessor directive mode), so just return EOF as our token.
586 assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode");