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/Lex/PreprocessorOptions.h"
17 #include "clang/Basic/FileManager.h"
18 #include "clang/Basic/SourceManager.h"
19 #include "clang/Lex/HeaderSearch.h"
20 #include "clang/Lex/LexDiagnostic.h"
21 #include "clang/Lex/MacroInfo.h"
22 #include "clang/Lex/PTHManager.h"
23 #include "llvm/ADT/StringSwitch.h"
24 #include "llvm/Support/FileSystem.h"
25 #include "llvm/Support/MemoryBuffer.h"
26 #include "llvm/Support/Path.h"
27 using namespace clang;
29 PPCallbacks::~PPCallbacks() {}
31 //===----------------------------------------------------------------------===//
32 // Miscellaneous Methods.
33 //===----------------------------------------------------------------------===//
35 /// isInPrimaryFile - Return true if we're in the top-level file, not in a
36 /// \#include. This looks through macro expansions and active _Pragma lexers.
37 bool Preprocessor::isInPrimaryFile() const {
39 return IncludeMacroStack.empty();
41 // If there are any stacked lexers, we're in a #include.
42 assert(IsFileLexer(IncludeMacroStack[0]) &&
43 "Top level include stack isn't our primary lexer?");
45 IncludeMacroStack.begin() + 1, IncludeMacroStack.end(),
46 [&](const IncludeStackInfo &ISI) -> bool { return IsFileLexer(ISI); });
49 /// getCurrentLexer - Return the current file lexer being lexed from. Note
50 /// that this ignores any potentially active macro expansions and _Pragma
51 /// expansions going on at the time.
52 PreprocessorLexer *Preprocessor::getCurrentFileLexer() const {
56 // Look for a stacked lexer.
57 for (const IncludeStackInfo &ISI : llvm::reverse(IncludeMacroStack)) {
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 bool 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 CurLexerSubmodule = nullptr;
121 if (CurLexerKind != CLK_LexAfterModuleImport)
122 CurLexerKind = CLK_Lexer;
124 // Notify the client, if desired, that we are in a new source file.
125 if (Callbacks && !CurLexer->Is_PragmaLexer) {
126 SrcMgr::CharacteristicKind FileType =
127 SourceMgr.getFileCharacteristic(CurLexer->getFileLoc());
129 Callbacks->FileChanged(CurLexer->getFileLoc(),
130 PPCallbacks::EnterFile, FileType);
134 /// EnterSourceFileWithPTH - Add a source file to the top of the include stack
135 /// and start getting tokens from it using the PTH cache.
136 void Preprocessor::EnterSourceFileWithPTH(PTHLexer *PL,
137 const DirectoryLookup *CurDir) {
139 if (CurPPLexer || CurTokenLexer)
140 PushIncludeMacroStack();
142 CurDirLookup = CurDir;
143 CurPTHLexer.reset(PL);
144 CurPPLexer = CurPTHLexer.get();
145 CurLexerSubmodule = nullptr;
146 if (CurLexerKind != CLK_LexAfterModuleImport)
147 CurLexerKind = CLK_PTHLexer;
149 // Notify the client, if desired, that we are in a new source file.
151 FileID FID = CurPPLexer->getFileID();
152 SourceLocation EnterLoc = SourceMgr.getLocForStartOfFile(FID);
153 SrcMgr::CharacteristicKind FileType =
154 SourceMgr.getFileCharacteristic(EnterLoc);
155 Callbacks->FileChanged(EnterLoc, PPCallbacks::EnterFile, FileType);
159 /// EnterMacro - Add a Macro to the top of the include stack and start lexing
160 /// tokens from it instead of the current buffer.
161 void Preprocessor::EnterMacro(Token &Tok, SourceLocation ILEnd,
162 MacroInfo *Macro, MacroArgs *Args) {
163 std::unique_ptr<TokenLexer> TokLexer;
164 if (NumCachedTokenLexers == 0) {
165 TokLexer = llvm::make_unique<TokenLexer>(Tok, ILEnd, Macro, Args, *this);
167 TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]);
168 TokLexer->Init(Tok, ILEnd, Macro, Args);
171 PushIncludeMacroStack();
172 CurDirLookup = nullptr;
173 CurTokenLexer = std::move(TokLexer);
174 if (CurLexerKind != CLK_LexAfterModuleImport)
175 CurLexerKind = CLK_TokenLexer;
178 /// EnterTokenStream - Add a "macro" context to the top of the include stack,
179 /// which will cause the lexer to start returning the specified tokens.
181 /// If DisableMacroExpansion is true, tokens lexed from the token stream will
182 /// not be subject to further macro expansion. Otherwise, these tokens will
183 /// be re-macro-expanded when/if expansion is enabled.
185 /// If OwnsTokens is false, this method assumes that the specified stream of
186 /// tokens has a permanent owner somewhere, so they do not need to be copied.
187 /// If it is true, it assumes the array of tokens is allocated with new[] and
190 void Preprocessor::EnterTokenStream(const Token *Toks, unsigned NumToks,
191 bool DisableMacroExpansion,
193 if (CurLexerKind == CLK_CachingLexer) {
194 if (CachedLexPos < CachedTokens.size()) {
195 // We're entering tokens into the middle of our cached token stream. We
196 // can't represent that, so just insert the tokens into the buffer.
197 CachedTokens.insert(CachedTokens.begin() + CachedLexPos,
198 Toks, Toks + NumToks);
204 // New tokens are at the end of the cached token sequnece; insert the
205 // token stream underneath the caching lexer.
206 ExitCachingLexMode();
207 EnterTokenStream(Toks, NumToks, DisableMacroExpansion, OwnsTokens);
208 EnterCachingLexMode();
212 // Create a macro expander to expand from the specified token stream.
213 std::unique_ptr<TokenLexer> TokLexer;
214 if (NumCachedTokenLexers == 0) {
215 TokLexer = llvm::make_unique<TokenLexer>(
216 Toks, NumToks, DisableMacroExpansion, OwnsTokens, *this);
218 TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]);
219 TokLexer->Init(Toks, NumToks, DisableMacroExpansion, OwnsTokens);
222 // Save our current state.
223 PushIncludeMacroStack();
224 CurDirLookup = nullptr;
225 CurTokenLexer = std::move(TokLexer);
226 if (CurLexerKind != CLK_LexAfterModuleImport)
227 CurLexerKind = CLK_TokenLexer;
230 /// Compute the relative path that names the given file relative to
231 /// the given directory.
232 static void computeRelativePath(FileManager &FM, const DirectoryEntry *Dir,
233 const FileEntry *File,
234 SmallString<128> &Result) {
237 StringRef FilePath = File->getDir()->getName();
238 StringRef Path = FilePath;
239 while (!Path.empty()) {
240 if (const DirectoryEntry *CurDir = FM.getDirectory(Path)) {
242 Result = FilePath.substr(Path.size());
243 llvm::sys::path::append(Result,
244 llvm::sys::path::filename(File->getName()));
249 Path = llvm::sys::path::parent_path(Path);
252 Result = File->getName();
255 void Preprocessor::PropagateLineStartLeadingSpaceInfo(Token &Result) {
257 CurTokenLexer->PropagateLineStartLeadingSpaceInfo(Result);
261 CurLexer->PropagateLineStartLeadingSpaceInfo(Result);
264 // FIXME: Handle other kinds of lexers? It generally shouldn't matter,
265 // but it might if they're empty?
268 /// Determine the location to use as the end of the buffer for a lexer.
270 /// If the file ends with a newline, form the EOF token on the newline itself,
271 /// rather than "on the line following it", which doesn't exist. This makes
272 /// diagnostics relating to the end of file include the last file that the user
273 /// actually typed, which is goodness.
274 const char *Preprocessor::getCurLexerEndPos() {
275 const char *EndPos = CurLexer->BufferEnd;
276 if (EndPos != CurLexer->BufferStart &&
277 (EndPos[-1] == '\n' || EndPos[-1] == '\r')) {
280 // Handle \n\r and \r\n:
281 if (EndPos != CurLexer->BufferStart &&
282 (EndPos[-1] == '\n' || EndPos[-1] == '\r') &&
283 EndPos[-1] != EndPos[0])
290 static void collectAllSubModulesWithUmbrellaHeader(
291 const Module &Mod, SmallVectorImpl<const Module *> &SubMods) {
292 if (Mod.getUmbrellaHeader())
293 SubMods.push_back(&Mod);
294 for (auto *M : Mod.submodules())
295 collectAllSubModulesWithUmbrellaHeader(*M, SubMods);
298 void Preprocessor::diagnoseMissingHeaderInUmbrellaDir(const Module &Mod) {
299 assert(Mod.getUmbrellaHeader() && "Module must use umbrella header");
300 SourceLocation StartLoc =
301 SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID());
302 if (getDiagnostics().isIgnored(diag::warn_uncovered_module_header, StartLoc))
305 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
306 const DirectoryEntry *Dir = Mod.getUmbrellaDir().Entry;
307 vfs::FileSystem &FS = *FileMgr.getVirtualFileSystem();
309 for (vfs::recursive_directory_iterator Entry(FS, Dir->getName(), EC), End;
310 Entry != End && !EC; Entry.increment(EC)) {
311 using llvm::StringSwitch;
313 // Check whether this entry has an extension typically associated with
315 if (!StringSwitch<bool>(llvm::sys::path::extension(Entry->getName()))
316 .Cases(".h", ".H", ".hh", ".hpp", true)
320 if (const FileEntry *Header = getFileManager().getFile(Entry->getName()))
321 if (!getSourceManager().hasFileInfo(Header)) {
322 if (!ModMap.isHeaderInUnavailableModule(Header)) {
323 // Find the relative path that would access this header.
324 SmallString<128> RelativePath;
325 computeRelativePath(FileMgr, Dir, Header, RelativePath);
326 Diag(StartLoc, diag::warn_uncovered_module_header)
327 << Mod.getFullModuleName() << RelativePath;
333 /// HandleEndOfFile - This callback is invoked when the lexer hits the end of
334 /// the current file. This either returns the EOF token or pops a level off
335 /// the include stack and keeps going.
336 bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) {
337 assert(!CurTokenLexer &&
338 "Ending a file when currently in a macro!");
340 // If we have an unclosed module region from a pragma at the end of a
341 // module, complain and close it now.
342 // FIXME: This is not correct if we are building a module from PTH.
343 const bool LeavingSubmodule = CurLexer && CurLexerSubmodule;
344 if ((LeavingSubmodule || IncludeMacroStack.empty()) &&
345 !BuildingSubmoduleStack.empty() &&
346 BuildingSubmoduleStack.back().IsPragma) {
347 Diag(BuildingSubmoduleStack.back().ImportLoc,
348 diag::err_pp_module_begin_without_module_end);
349 Module *M = LeaveSubmodule(/*ForPragma*/true);
352 const char *EndPos = getCurLexerEndPos();
353 CurLexer->BufferPtr = EndPos;
354 CurLexer->FormTokenWithChars(Result, EndPos, tok::annot_module_end);
355 Result.setAnnotationEndLoc(Result.getLocation());
356 Result.setAnnotationValue(M);
360 // See if this file had a controlling macro.
361 if (CurPPLexer) { // Not ending a macro, ignore it.
362 if (const IdentifierInfo *ControllingMacro =
363 CurPPLexer->MIOpt.GetControllingMacroAtEndOfFile()) {
364 // Okay, this has a controlling macro, remember in HeaderFileInfo.
365 if (const FileEntry *FE = CurPPLexer->getFileEntry()) {
366 HeaderInfo.SetFileControllingMacro(FE, ControllingMacro);
368 getMacroInfo(const_cast<IdentifierInfo*>(ControllingMacro)))
369 MI->setUsedForHeaderGuard(true);
370 if (const IdentifierInfo *DefinedMacro =
371 CurPPLexer->MIOpt.GetDefinedMacro()) {
372 if (!isMacroDefined(ControllingMacro) &&
373 DefinedMacro != ControllingMacro &&
374 HeaderInfo.FirstTimeLexingFile(FE)) {
376 // If the edit distance between the two macros is more than 50%,
377 // DefinedMacro may not be header guard, or can be header guard of
378 // another header file. Therefore, it maybe defining something
379 // completely different. This can be observed in the wild when
380 // handling feature macros or header guards in different files.
382 const StringRef ControllingMacroName = ControllingMacro->getName();
383 const StringRef DefinedMacroName = DefinedMacro->getName();
384 const size_t MaxHalfLength = std::max(ControllingMacroName.size(),
385 DefinedMacroName.size()) / 2;
386 const unsigned ED = ControllingMacroName.edit_distance(
387 DefinedMacroName, true, MaxHalfLength);
388 if (ED <= MaxHalfLength) {
389 // Emit a warning for a bad header guard.
390 Diag(CurPPLexer->MIOpt.GetMacroLocation(),
391 diag::warn_header_guard)
392 << CurPPLexer->MIOpt.GetMacroLocation() << ControllingMacro;
393 Diag(CurPPLexer->MIOpt.GetDefinedLocation(),
394 diag::note_header_guard)
395 << CurPPLexer->MIOpt.GetDefinedLocation() << DefinedMacro
397 << FixItHint::CreateReplacement(
398 CurPPLexer->MIOpt.GetDefinedLocation(),
399 ControllingMacro->getName());
407 // Complain about reaching a true EOF within arc_cf_code_audited.
408 // We don't want to complain about reaching the end of a macro
409 // instantiation or a _Pragma.
410 if (PragmaARCCFCodeAuditedLoc.isValid() &&
411 !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) {
412 Diag(PragmaARCCFCodeAuditedLoc, diag::err_pp_eof_in_arc_cf_code_audited);
414 // Recover by leaving immediately.
415 PragmaARCCFCodeAuditedLoc = SourceLocation();
418 // Complain about reaching a true EOF within assume_nonnull.
419 // We don't want to complain about reaching the end of a macro
420 // instantiation or a _Pragma.
421 if (PragmaAssumeNonNullLoc.isValid() &&
422 !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) {
423 Diag(PragmaAssumeNonNullLoc, diag::err_pp_eof_in_assume_nonnull);
425 // Recover by leaving immediately.
426 PragmaAssumeNonNullLoc = SourceLocation();
429 bool LeavingPCHThroughHeader = false;
431 // If this is a #include'd file, pop it off the include stack and continue
432 // lexing the #includer file.
433 if (!IncludeMacroStack.empty()) {
435 // If we lexed the code-completion file, act as if we reached EOF.
436 if (isCodeCompletionEnabled() && CurPPLexer &&
437 SourceMgr.getLocForStartOfFile(CurPPLexer->getFileID()) ==
438 CodeCompletionFileLoc) {
441 CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof);
444 assert(CurPTHLexer && "Got EOF but no current lexer set!");
445 CurPTHLexer->getEOF(Result);
449 CurPPLexer = nullptr;
450 recomputeCurLexerKind();
454 if (!isEndOfMacro && CurPPLexer &&
455 SourceMgr.getIncludeLoc(CurPPLexer->getFileID()).isValid()) {
456 // Notify SourceManager to record the number of FileIDs that were created
457 // during lexing of the #include'd file.
459 SourceMgr.local_sloc_entry_size() -
460 CurPPLexer->getInitialNumSLocEntries() + 1/*#include'd file*/;
461 SourceMgr.setNumCreatedFIDsForFileID(CurPPLexer->getFileID(), NumFIDs);
464 bool ExitedFromPredefinesFile = false;
466 if (!isEndOfMacro && CurPPLexer) {
467 ExitedFID = CurPPLexer->getFileID();
469 assert(PredefinesFileID.isValid() &&
470 "HandleEndOfFile is called before PredefinesFileId is set");
471 ExitedFromPredefinesFile = (PredefinesFileID == ExitedFID);
474 if (LeavingSubmodule) {
475 // We're done with this submodule.
476 Module *M = LeaveSubmodule(/*ForPragma*/false);
478 // Notify the parser that we've left the module.
479 const char *EndPos = getCurLexerEndPos();
481 CurLexer->BufferPtr = EndPos;
482 CurLexer->FormTokenWithChars(Result, EndPos, tok::annot_module_end);
483 Result.setAnnotationEndLoc(Result.getLocation());
484 Result.setAnnotationValue(M);
487 bool FoundPCHThroughHeader = false;
488 if (CurPPLexer && creatingPCHWithThroughHeader() &&
490 SourceMgr.getFileEntryForID(CurPPLexer->getFileID())))
491 FoundPCHThroughHeader = true;
493 // We're done with the #included file.
494 RemoveTopOfLexerStack();
496 // Propagate info about start-of-line/leading white-space/etc.
497 PropagateLineStartLeadingSpaceInfo(Result);
499 // Notify the client, if desired, that we are in a new source file.
500 if (Callbacks && !isEndOfMacro && CurPPLexer) {
501 SrcMgr::CharacteristicKind FileType =
502 SourceMgr.getFileCharacteristic(CurPPLexer->getSourceLocation());
503 Callbacks->FileChanged(CurPPLexer->getSourceLocation(),
504 PPCallbacks::ExitFile, FileType, ExitedFID);
507 // Restore conditional stack from the preamble right after exiting from the
509 if (ExitedFromPredefinesFile)
510 replayPreambleConditionalStack();
512 if (!isEndOfMacro && CurPPLexer && FoundPCHThroughHeader &&
513 (isInPrimaryFile() ||
514 CurPPLexer->getFileID() == getPredefinesFileID())) {
515 // Leaving the through header. Continue directly to end of main file
517 LeavingPCHThroughHeader = true;
519 // Client should lex another token unless we generated an EOM.
520 return LeavingSubmodule;
524 // If this is the end of the main file, form an EOF token.
526 const char *EndPos = getCurLexerEndPos();
528 CurLexer->BufferPtr = EndPos;
529 CurLexer->FormTokenWithChars(Result, EndPos, tok::eof);
531 if (isCodeCompletionEnabled()) {
532 // Inserting the code-completion point increases the source buffer by 1,
533 // but the main FileID was created before inserting the point.
534 // Compensate by reducing the EOF location by 1, otherwise the location
535 // will point to the next FileID.
536 // FIXME: This is hacky, the code-completion point should probably be
537 // inserted before the main FileID is created.
538 if (CurLexer->getFileLoc() == CodeCompletionFileLoc)
539 Result.setLocation(Result.getLocation().getLocWithOffset(-1));
542 if (creatingPCHWithThroughHeader() && !LeavingPCHThroughHeader) {
543 // Reached the end of the compilation without finding the through header.
544 Diag(CurLexer->getFileLoc(), diag::err_pp_through_header_not_seen)
545 << PPOpts->PCHThroughHeader << 0;
548 if (!isIncrementalProcessingEnabled())
549 // We're done with lexing.
552 assert(CurPTHLexer && "Got EOF but no current lexer set!");
553 CurPTHLexer->getEOF(Result);
557 if (!isIncrementalProcessingEnabled())
558 CurPPLexer = nullptr;
560 if (TUKind == TU_Complete) {
561 // This is the end of the top-level file. 'WarnUnusedMacroLocs' has
562 // collected all macro locations that we need to warn because they are not
564 for (WarnUnusedMacroLocsTy::iterator
565 I=WarnUnusedMacroLocs.begin(), E=WarnUnusedMacroLocs.end();
567 Diag(*I, diag::pp_macro_not_used);
570 // If we are building a module that has an umbrella header, make sure that
571 // each of the headers within the directory, including all submodules, is
572 // covered by the umbrella header was actually included by the umbrella
574 if (Module *Mod = getCurrentModule()) {
575 llvm::SmallVector<const Module *, 4> AllMods;
576 collectAllSubModulesWithUmbrellaHeader(*Mod, AllMods);
577 for (auto *M : AllMods)
578 diagnoseMissingHeaderInUmbrellaDir(*M);
584 /// HandleEndOfTokenLexer - This callback is invoked when the current TokenLexer
585 /// hits the end of its token stream.
586 bool Preprocessor::HandleEndOfTokenLexer(Token &Result) {
587 assert(CurTokenLexer && !CurPPLexer &&
588 "Ending a macro when currently in a #include file!");
590 if (!MacroExpandingLexersStack.empty() &&
591 MacroExpandingLexersStack.back().first == CurTokenLexer.get())
592 removeCachedMacroExpandedTokensOfLastLexer();
594 // Delete or cache the now-dead macro expander.
595 if (NumCachedTokenLexers == TokenLexerCacheSize)
596 CurTokenLexer.reset();
598 TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer);
600 // Handle this like a #include file being popped off the stack.
601 return HandleEndOfFile(Result, true);
604 /// RemoveTopOfLexerStack - Pop the current lexer/macro exp off the top of the
605 /// lexer stack. This should only be used in situations where the current
606 /// state of the top-of-stack lexer is unknown.
607 void Preprocessor::RemoveTopOfLexerStack() {
608 assert(!IncludeMacroStack.empty() && "Ran out of stack entries to load");
611 // Delete or cache the now-dead macro expander.
612 if (NumCachedTokenLexers == TokenLexerCacheSize)
613 CurTokenLexer.reset();
615 TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer);
618 PopIncludeMacroStack();
621 /// HandleMicrosoftCommentPaste - When the macro expander pastes together a
622 /// comment (/##/) in microsoft mode, this method handles updating the current
623 /// state, returning the token on the next source line.
624 void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) {
625 assert(CurTokenLexer && !CurPPLexer &&
626 "Pasted comment can only be formed from macro");
627 // We handle this by scanning for the closest real lexer, switching it to
628 // raw mode and preprocessor mode. This will cause it to return \n as an
629 // explicit EOD token.
630 PreprocessorLexer *FoundLexer = nullptr;
631 bool LexerWasInPPMode = false;
632 for (const IncludeStackInfo &ISI : llvm::reverse(IncludeMacroStack)) {
633 if (ISI.ThePPLexer == nullptr) continue; // Scan for a real lexer.
635 // Once we find a real lexer, mark it as raw mode (disabling macro
636 // expansions) and preprocessor mode (return EOD). We know that the lexer
637 // was *not* in raw mode before, because the macro that the comment came
638 // from was expanded. However, it could have already been in preprocessor
639 // mode (#if COMMENT) in which case we have to return it to that mode and
641 FoundLexer = ISI.ThePPLexer;
642 FoundLexer->LexingRawMode = true;
643 LexerWasInPPMode = FoundLexer->ParsingPreprocessorDirective;
644 FoundLexer->ParsingPreprocessorDirective = true;
648 // Okay, we either found and switched over the lexer, or we didn't find a
649 // lexer. In either case, finish off the macro the comment came from, getting
651 if (!HandleEndOfTokenLexer(Tok)) Lex(Tok);
653 // Discarding comments as long as we don't have EOF or EOD. This 'comments
654 // out' the rest of the line, including any tokens that came from other macros
655 // that were active, as in:
656 // #define submacro a COMMENT b
658 // which should lex to 'a' only: 'b' and 'c' should be removed.
659 while (Tok.isNot(tok::eod) && Tok.isNot(tok::eof))
662 // If we got an eod token, then we successfully found the end of the line.
663 if (Tok.is(tok::eod)) {
664 assert(FoundLexer && "Can't get end of line without an active lexer");
665 // Restore the lexer back to normal mode instead of raw mode.
666 FoundLexer->LexingRawMode = false;
668 // If the lexer was already in preprocessor mode, just return the EOD token
669 // to finish the preprocessor line.
670 if (LexerWasInPPMode) return;
672 // Otherwise, switch out of PP mode and return the next lexed token.
673 FoundLexer->ParsingPreprocessorDirective = false;
677 // If we got an EOF token, then we reached the end of the token stream but
678 // didn't find an explicit \n. This can only happen if there was no lexer
679 // active (an active lexer would return EOD at EOF if there was no \n in
680 // preprocessor directive mode), so just return EOF as our token.
681 assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode");
684 void Preprocessor::EnterSubmodule(Module *M, SourceLocation ImportLoc,
686 if (!getLangOpts().ModulesLocalVisibility) {
687 // Just track that we entered this submodule.
688 BuildingSubmoduleStack.push_back(
689 BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState,
690 PendingModuleMacroNames.size()));
694 // Resolve as much of the module definition as we can now, before we enter
695 // one of its headers.
696 // FIXME: Can we enable Complain here?
697 // FIXME: Can we do this when local visibility is disabled?
698 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
699 ModMap.resolveExports(M, /*Complain=*/false);
700 ModMap.resolveUses(M, /*Complain=*/false);
701 ModMap.resolveConflicts(M, /*Complain=*/false);
703 // If this is the first time we've entered this module, set up its state.
704 auto R = Submodules.insert(std::make_pair(M, SubmoduleState()));
705 auto &State = R.first->second;
706 bool FirstTime = R.second;
708 // Determine the set of starting macros for this submodule; take these
709 // from the "null" module (the predefines buffer).
711 // FIXME: If we have local visibility but not modules enabled, the
712 // NullSubmoduleState is polluted by #defines in the top-level source
714 auto &StartingMacros = NullSubmoduleState.Macros;
716 // Restore to the starting state.
717 // FIXME: Do this lazily, when each macro name is first referenced.
718 for (auto &Macro : StartingMacros) {
719 // Skip uninteresting macros.
720 if (!Macro.second.getLatest() &&
721 Macro.second.getOverriddenMacros().empty())
724 MacroState MS(Macro.second.getLatest());
725 MS.setOverriddenMacros(*this, Macro.second.getOverriddenMacros());
726 State.Macros.insert(std::make_pair(Macro.first, std::move(MS)));
730 // Track that we entered this module.
731 BuildingSubmoduleStack.push_back(
732 BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState,
733 PendingModuleMacroNames.size()));
735 // Switch to this submodule as the current submodule.
736 CurSubmoduleState = &State;
738 // This module is visible to itself.
740 makeModuleVisible(M, ImportLoc);
743 bool Preprocessor::needModuleMacros() const {
744 // If we're not within a submodule, we never need to create ModuleMacros.
745 if (BuildingSubmoduleStack.empty())
747 // If we are tracking module macro visibility even for textually-included
748 // headers, we need ModuleMacros.
749 if (getLangOpts().ModulesLocalVisibility)
751 // Otherwise, we only need module macros if we're actually compiling a module
753 return getLangOpts().isCompilingModule();
756 Module *Preprocessor::LeaveSubmodule(bool ForPragma) {
757 if (BuildingSubmoduleStack.empty() ||
758 BuildingSubmoduleStack.back().IsPragma != ForPragma) {
759 assert(ForPragma && "non-pragma module enter/leave mismatch");
763 auto &Info = BuildingSubmoduleStack.back();
765 Module *LeavingMod = Info.M;
766 SourceLocation ImportLoc = Info.ImportLoc;
768 if (!needModuleMacros() ||
769 (!getLangOpts().ModulesLocalVisibility &&
770 LeavingMod->getTopLevelModuleName() != getLangOpts().CurrentModule)) {
771 // If we don't need module macros, or this is not a module for which we
772 // are tracking macro visibility, don't build any, and preserve the list
773 // of pending names for the surrounding submodule.
774 BuildingSubmoduleStack.pop_back();
775 makeModuleVisible(LeavingMod, ImportLoc);
779 // Create ModuleMacros for any macros defined in this submodule.
780 llvm::SmallPtrSet<const IdentifierInfo*, 8> VisitedMacros;
781 for (unsigned I = Info.OuterPendingModuleMacroNames;
782 I != PendingModuleMacroNames.size(); ++I) {
783 auto *II = const_cast<IdentifierInfo*>(PendingModuleMacroNames[I]);
784 if (!VisitedMacros.insert(II).second)
787 auto MacroIt = CurSubmoduleState->Macros.find(II);
788 if (MacroIt == CurSubmoduleState->Macros.end())
790 auto &Macro = MacroIt->second;
792 // Find the starting point for the MacroDirective chain in this submodule.
793 MacroDirective *OldMD = nullptr;
794 auto *OldState = Info.OuterSubmoduleState;
795 if (getLangOpts().ModulesLocalVisibility)
796 OldState = &NullSubmoduleState;
797 if (OldState && OldState != CurSubmoduleState) {
798 // FIXME: It'd be better to start at the state from when we most recently
799 // entered this submodule, but it doesn't really matter.
800 auto &OldMacros = OldState->Macros;
801 auto OldMacroIt = OldMacros.find(II);
802 if (OldMacroIt == OldMacros.end())
805 OldMD = OldMacroIt->second.getLatest();
808 // This module may have exported a new macro. If so, create a ModuleMacro
809 // representing that fact.
810 bool ExplicitlyPublic = false;
811 for (auto *MD = Macro.getLatest(); MD != OldMD; MD = MD->getPrevious()) {
812 assert(MD && "broken macro directive chain");
814 if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(MD)) {
815 // The latest visibility directive for a name in a submodule affects
816 // all the directives that come before it.
817 if (VisMD->isPublic())
818 ExplicitlyPublic = true;
819 else if (!ExplicitlyPublic)
820 // Private with no following public directive: not exported.
823 MacroInfo *Def = nullptr;
824 if (DefMacroDirective *DefMD = dyn_cast<DefMacroDirective>(MD))
825 Def = DefMD->getInfo();
827 // FIXME: Issue a warning if multiple headers for the same submodule
828 // define a macro, rather than silently ignoring all but the first.
830 // Don't bother creating a module macro if it would represent a #undef
831 // that doesn't override anything.
832 if (Def || !Macro.getOverriddenMacros().empty())
833 addModuleMacro(LeavingMod, II, Def,
834 Macro.getOverriddenMacros(), IsNew);
836 if (!getLangOpts().ModulesLocalVisibility) {
837 // This macro is exposed to the rest of this compilation as a
838 // ModuleMacro; we don't need to track its MacroDirective any more.
839 Macro.setLatest(nullptr);
840 Macro.setOverriddenMacros(*this, {});
846 PendingModuleMacroNames.resize(Info.OuterPendingModuleMacroNames);
848 // FIXME: Before we leave this submodule, we should parse all the other
849 // headers within it. Otherwise, we're left with an inconsistent state
850 // where we've made the module visible but don't yet have its complete
853 // Put back the outer module's state, if we're tracking it.
854 if (getLangOpts().ModulesLocalVisibility)
855 CurSubmoduleState = Info.OuterSubmoduleState;
857 BuildingSubmoduleStack.pop_back();
859 // A nested #include makes the included submodule visible.
860 makeModuleVisible(LeavingMod, ImportLoc);