Mercurial > hg > CbC > CbC_llvm
comparison clang/lib/Lex/PPLexerChange.cpp @ 150:1d019706d866
LLVM10
author | anatofuz |
---|---|
date | Thu, 13 Feb 2020 15:10:13 +0900 |
parents | |
children | e8a9b4f4d755 0572611fdcc8 |
comparison
equal
deleted
inserted
replaced
147:c2174574ed3a | 150:1d019706d866 |
---|---|
1 //===--- PPLexerChange.cpp - Handle changing lexers in the preprocessor ---===// | |
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 implements pieces of the Preprocessor interface that manage the | |
10 // current lexer stack. | |
11 // | |
12 //===----------------------------------------------------------------------===// | |
13 | |
14 #include "clang/Lex/Preprocessor.h" | |
15 #include "clang/Lex/PreprocessorOptions.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; | |
26 | |
27 PPCallbacks::~PPCallbacks() {} | |
28 | |
29 //===----------------------------------------------------------------------===// | |
30 // Miscellaneous Methods. | |
31 //===----------------------------------------------------------------------===// | |
32 | |
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 { | |
36 if (IsFileLexer()) | |
37 return IncludeMacroStack.empty(); | |
38 | |
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 return std::none_of( | |
43 IncludeMacroStack.begin() + 1, IncludeMacroStack.end(), | |
44 [&](const IncludeStackInfo &ISI) -> bool { return IsFileLexer(ISI); }); | |
45 } | |
46 | |
47 /// getCurrentLexer - Return the current file lexer being lexed from. Note | |
48 /// that this ignores any potentially active macro expansions and _Pragma | |
49 /// expansions going on at the time. | |
50 PreprocessorLexer *Preprocessor::getCurrentFileLexer() const { | |
51 if (IsFileLexer()) | |
52 return CurPPLexer; | |
53 | |
54 // Look for a stacked lexer. | |
55 for (const IncludeStackInfo &ISI : llvm::reverse(IncludeMacroStack)) { | |
56 if (IsFileLexer(ISI)) | |
57 return ISI.ThePPLexer; | |
58 } | |
59 return nullptr; | |
60 } | |
61 | |
62 | |
63 //===----------------------------------------------------------------------===// | |
64 // Methods for Entering and Callbacks for leaving various contexts | |
65 //===----------------------------------------------------------------------===// | |
66 | |
67 /// EnterSourceFile - Add a source file to the top of the include stack and | |
68 /// start lexing tokens from it instead of the current buffer. | |
69 bool Preprocessor::EnterSourceFile(FileID FID, const DirectoryLookup *CurDir, | |
70 SourceLocation Loc) { | |
71 assert(!CurTokenLexer && "Cannot #include a file inside a macro!"); | |
72 ++NumEnteredSourceFiles; | |
73 | |
74 if (MaxIncludeStackDepth < IncludeMacroStack.size()) | |
75 MaxIncludeStackDepth = IncludeMacroStack.size(); | |
76 | |
77 // Get the MemoryBuffer for this FID, if it fails, we fail. | |
78 bool Invalid = false; | |
79 const llvm::MemoryBuffer *InputFile = | |
80 getSourceManager().getBuffer(FID, Loc, &Invalid); | |
81 if (Invalid) { | |
82 SourceLocation FileStart = SourceMgr.getLocForStartOfFile(FID); | |
83 Diag(Loc, diag::err_pp_error_opening_file) | |
84 << std::string(SourceMgr.getBufferName(FileStart)) << ""; | |
85 return true; | |
86 } | |
87 | |
88 if (isCodeCompletionEnabled() && | |
89 SourceMgr.getFileEntryForID(FID) == CodeCompletionFile) { | |
90 CodeCompletionFileLoc = SourceMgr.getLocForStartOfFile(FID); | |
91 CodeCompletionLoc = | |
92 CodeCompletionFileLoc.getLocWithOffset(CodeCompletionOffset); | |
93 } | |
94 | |
95 EnterSourceFileWithLexer(new Lexer(FID, InputFile, *this), CurDir); | |
96 return false; | |
97 } | |
98 | |
99 /// EnterSourceFileWithLexer - Add a source file to the top of the include stack | |
100 /// and start lexing tokens from it instead of the current buffer. | |
101 void Preprocessor::EnterSourceFileWithLexer(Lexer *TheLexer, | |
102 const DirectoryLookup *CurDir) { | |
103 | |
104 // Add the current lexer to the include stack. | |
105 if (CurPPLexer || CurTokenLexer) | |
106 PushIncludeMacroStack(); | |
107 | |
108 CurLexer.reset(TheLexer); | |
109 CurPPLexer = TheLexer; | |
110 CurDirLookup = CurDir; | |
111 CurLexerSubmodule = nullptr; | |
112 if (CurLexerKind != CLK_LexAfterModuleImport) | |
113 CurLexerKind = CLK_Lexer; | |
114 | |
115 // Notify the client, if desired, that we are in a new source file. | |
116 if (Callbacks && !CurLexer->Is_PragmaLexer) { | |
117 SrcMgr::CharacteristicKind FileType = | |
118 SourceMgr.getFileCharacteristic(CurLexer->getFileLoc()); | |
119 | |
120 Callbacks->FileChanged(CurLexer->getFileLoc(), | |
121 PPCallbacks::EnterFile, FileType); | |
122 } | |
123 } | |
124 | |
125 /// EnterMacro - Add a Macro to the top of the include stack and start lexing | |
126 /// tokens from it instead of the current buffer. | |
127 void Preprocessor::EnterMacro(Token &Tok, SourceLocation ILEnd, | |
128 MacroInfo *Macro, MacroArgs *Args) { | |
129 std::unique_ptr<TokenLexer> TokLexer; | |
130 if (NumCachedTokenLexers == 0) { | |
131 TokLexer = std::make_unique<TokenLexer>(Tok, ILEnd, Macro, Args, *this); | |
132 } else { | |
133 TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]); | |
134 TokLexer->Init(Tok, ILEnd, Macro, Args); | |
135 } | |
136 | |
137 PushIncludeMacroStack(); | |
138 CurDirLookup = nullptr; | |
139 CurTokenLexer = std::move(TokLexer); | |
140 if (CurLexerKind != CLK_LexAfterModuleImport) | |
141 CurLexerKind = CLK_TokenLexer; | |
142 } | |
143 | |
144 /// EnterTokenStream - Add a "macro" context to the top of the include stack, | |
145 /// which will cause the lexer to start returning the specified tokens. | |
146 /// | |
147 /// If DisableMacroExpansion is true, tokens lexed from the token stream will | |
148 /// not be subject to further macro expansion. Otherwise, these tokens will | |
149 /// be re-macro-expanded when/if expansion is enabled. | |
150 /// | |
151 /// If OwnsTokens is false, this method assumes that the specified stream of | |
152 /// tokens has a permanent owner somewhere, so they do not need to be copied. | |
153 /// If it is true, it assumes the array of tokens is allocated with new[] and | |
154 /// must be freed. | |
155 /// | |
156 void Preprocessor::EnterTokenStream(const Token *Toks, unsigned NumToks, | |
157 bool DisableMacroExpansion, bool OwnsTokens, | |
158 bool IsReinject) { | |
159 if (CurLexerKind == CLK_CachingLexer) { | |
160 if (CachedLexPos < CachedTokens.size()) { | |
161 assert(IsReinject && "new tokens in the middle of cached stream"); | |
162 // We're entering tokens into the middle of our cached token stream. We | |
163 // can't represent that, so just insert the tokens into the buffer. | |
164 CachedTokens.insert(CachedTokens.begin() + CachedLexPos, | |
165 Toks, Toks + NumToks); | |
166 if (OwnsTokens) | |
167 delete [] Toks; | |
168 return; | |
169 } | |
170 | |
171 // New tokens are at the end of the cached token sequnece; insert the | |
172 // token stream underneath the caching lexer. | |
173 ExitCachingLexMode(); | |
174 EnterTokenStream(Toks, NumToks, DisableMacroExpansion, OwnsTokens, | |
175 IsReinject); | |
176 EnterCachingLexMode(); | |
177 return; | |
178 } | |
179 | |
180 // Create a macro expander to expand from the specified token stream. | |
181 std::unique_ptr<TokenLexer> TokLexer; | |
182 if (NumCachedTokenLexers == 0) { | |
183 TokLexer = std::make_unique<TokenLexer>( | |
184 Toks, NumToks, DisableMacroExpansion, OwnsTokens, IsReinject, *this); | |
185 } else { | |
186 TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]); | |
187 TokLexer->Init(Toks, NumToks, DisableMacroExpansion, OwnsTokens, | |
188 IsReinject); | |
189 } | |
190 | |
191 // Save our current state. | |
192 PushIncludeMacroStack(); | |
193 CurDirLookup = nullptr; | |
194 CurTokenLexer = std::move(TokLexer); | |
195 if (CurLexerKind != CLK_LexAfterModuleImport) | |
196 CurLexerKind = CLK_TokenLexer; | |
197 } | |
198 | |
199 /// Compute the relative path that names the given file relative to | |
200 /// the given directory. | |
201 static void computeRelativePath(FileManager &FM, const DirectoryEntry *Dir, | |
202 const FileEntry *File, | |
203 SmallString<128> &Result) { | |
204 Result.clear(); | |
205 | |
206 StringRef FilePath = File->getDir()->getName(); | |
207 StringRef Path = FilePath; | |
208 while (!Path.empty()) { | |
209 if (auto CurDir = FM.getDirectory(Path)) { | |
210 if (*CurDir == Dir) { | |
211 Result = FilePath.substr(Path.size()); | |
212 llvm::sys::path::append(Result, | |
213 llvm::sys::path::filename(File->getName())); | |
214 return; | |
215 } | |
216 } | |
217 | |
218 Path = llvm::sys::path::parent_path(Path); | |
219 } | |
220 | |
221 Result = File->getName(); | |
222 } | |
223 | |
224 void Preprocessor::PropagateLineStartLeadingSpaceInfo(Token &Result) { | |
225 if (CurTokenLexer) { | |
226 CurTokenLexer->PropagateLineStartLeadingSpaceInfo(Result); | |
227 return; | |
228 } | |
229 if (CurLexer) { | |
230 CurLexer->PropagateLineStartLeadingSpaceInfo(Result); | |
231 return; | |
232 } | |
233 // FIXME: Handle other kinds of lexers? It generally shouldn't matter, | |
234 // but it might if they're empty? | |
235 } | |
236 | |
237 /// Determine the location to use as the end of the buffer for a lexer. | |
238 /// | |
239 /// If the file ends with a newline, form the EOF token on the newline itself, | |
240 /// rather than "on the line following it", which doesn't exist. This makes | |
241 /// diagnostics relating to the end of file include the last file that the user | |
242 /// actually typed, which is goodness. | |
243 const char *Preprocessor::getCurLexerEndPos() { | |
244 const char *EndPos = CurLexer->BufferEnd; | |
245 if (EndPos != CurLexer->BufferStart && | |
246 (EndPos[-1] == '\n' || EndPos[-1] == '\r')) { | |
247 --EndPos; | |
248 | |
249 // Handle \n\r and \r\n: | |
250 if (EndPos != CurLexer->BufferStart && | |
251 (EndPos[-1] == '\n' || EndPos[-1] == '\r') && | |
252 EndPos[-1] != EndPos[0]) | |
253 --EndPos; | |
254 } | |
255 | |
256 return EndPos; | |
257 } | |
258 | |
259 static void collectAllSubModulesWithUmbrellaHeader( | |
260 const Module &Mod, SmallVectorImpl<const Module *> &SubMods) { | |
261 if (Mod.getUmbrellaHeader()) | |
262 SubMods.push_back(&Mod); | |
263 for (auto *M : Mod.submodules()) | |
264 collectAllSubModulesWithUmbrellaHeader(*M, SubMods); | |
265 } | |
266 | |
267 void Preprocessor::diagnoseMissingHeaderInUmbrellaDir(const Module &Mod) { | |
268 assert(Mod.getUmbrellaHeader() && "Module must use umbrella header"); | |
269 SourceLocation StartLoc = | |
270 SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID()); | |
271 if (getDiagnostics().isIgnored(diag::warn_uncovered_module_header, StartLoc)) | |
272 return; | |
273 | |
274 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap(); | |
275 const DirectoryEntry *Dir = Mod.getUmbrellaDir().Entry; | |
276 llvm::vfs::FileSystem &FS = FileMgr.getVirtualFileSystem(); | |
277 std::error_code EC; | |
278 for (llvm::vfs::recursive_directory_iterator Entry(FS, Dir->getName(), EC), | |
279 End; | |
280 Entry != End && !EC; Entry.increment(EC)) { | |
281 using llvm::StringSwitch; | |
282 | |
283 // Check whether this entry has an extension typically associated with | |
284 // headers. | |
285 if (!StringSwitch<bool>(llvm::sys::path::extension(Entry->path())) | |
286 .Cases(".h", ".H", ".hh", ".hpp", true) | |
287 .Default(false)) | |
288 continue; | |
289 | |
290 if (auto Header = getFileManager().getFile(Entry->path())) | |
291 if (!getSourceManager().hasFileInfo(*Header)) { | |
292 if (!ModMap.isHeaderInUnavailableModule(*Header)) { | |
293 // Find the relative path that would access this header. | |
294 SmallString<128> RelativePath; | |
295 computeRelativePath(FileMgr, Dir, *Header, RelativePath); | |
296 Diag(StartLoc, diag::warn_uncovered_module_header) | |
297 << Mod.getFullModuleName() << RelativePath; | |
298 } | |
299 } | |
300 } | |
301 } | |
302 | |
303 /// HandleEndOfFile - This callback is invoked when the lexer hits the end of | |
304 /// the current file. This either returns the EOF token or pops a level off | |
305 /// the include stack and keeps going. | |
306 bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) { | |
307 assert(!CurTokenLexer && | |
308 "Ending a file when currently in a macro!"); | |
309 | |
310 // If we have an unclosed module region from a pragma at the end of a | |
311 // module, complain and close it now. | |
312 const bool LeavingSubmodule = CurLexer && CurLexerSubmodule; | |
313 if ((LeavingSubmodule || IncludeMacroStack.empty()) && | |
314 !BuildingSubmoduleStack.empty() && | |
315 BuildingSubmoduleStack.back().IsPragma) { | |
316 Diag(BuildingSubmoduleStack.back().ImportLoc, | |
317 diag::err_pp_module_begin_without_module_end); | |
318 Module *M = LeaveSubmodule(/*ForPragma*/true); | |
319 | |
320 Result.startToken(); | |
321 const char *EndPos = getCurLexerEndPos(); | |
322 CurLexer->BufferPtr = EndPos; | |
323 CurLexer->FormTokenWithChars(Result, EndPos, tok::annot_module_end); | |
324 Result.setAnnotationEndLoc(Result.getLocation()); | |
325 Result.setAnnotationValue(M); | |
326 return true; | |
327 } | |
328 | |
329 // See if this file had a controlling macro. | |
330 if (CurPPLexer) { // Not ending a macro, ignore it. | |
331 if (const IdentifierInfo *ControllingMacro = | |
332 CurPPLexer->MIOpt.GetControllingMacroAtEndOfFile()) { | |
333 // Okay, this has a controlling macro, remember in HeaderFileInfo. | |
334 if (const FileEntry *FE = CurPPLexer->getFileEntry()) { | |
335 HeaderInfo.SetFileControllingMacro(FE, ControllingMacro); | |
336 if (MacroInfo *MI = | |
337 getMacroInfo(const_cast<IdentifierInfo*>(ControllingMacro))) | |
338 MI->setUsedForHeaderGuard(true); | |
339 if (const IdentifierInfo *DefinedMacro = | |
340 CurPPLexer->MIOpt.GetDefinedMacro()) { | |
341 if (!isMacroDefined(ControllingMacro) && | |
342 DefinedMacro != ControllingMacro && | |
343 HeaderInfo.FirstTimeLexingFile(FE)) { | |
344 | |
345 // If the edit distance between the two macros is more than 50%, | |
346 // DefinedMacro may not be header guard, or can be header guard of | |
347 // another header file. Therefore, it maybe defining something | |
348 // completely different. This can be observed in the wild when | |
349 // handling feature macros or header guards in different files. | |
350 | |
351 const StringRef ControllingMacroName = ControllingMacro->getName(); | |
352 const StringRef DefinedMacroName = DefinedMacro->getName(); | |
353 const size_t MaxHalfLength = std::max(ControllingMacroName.size(), | |
354 DefinedMacroName.size()) / 2; | |
355 const unsigned ED = ControllingMacroName.edit_distance( | |
356 DefinedMacroName, true, MaxHalfLength); | |
357 if (ED <= MaxHalfLength) { | |
358 // Emit a warning for a bad header guard. | |
359 Diag(CurPPLexer->MIOpt.GetMacroLocation(), | |
360 diag::warn_header_guard) | |
361 << CurPPLexer->MIOpt.GetMacroLocation() << ControllingMacro; | |
362 Diag(CurPPLexer->MIOpt.GetDefinedLocation(), | |
363 diag::note_header_guard) | |
364 << CurPPLexer->MIOpt.GetDefinedLocation() << DefinedMacro | |
365 << ControllingMacro | |
366 << FixItHint::CreateReplacement( | |
367 CurPPLexer->MIOpt.GetDefinedLocation(), | |
368 ControllingMacro->getName()); | |
369 } | |
370 } | |
371 } | |
372 } | |
373 } | |
374 } | |
375 | |
376 // Complain about reaching a true EOF within arc_cf_code_audited. | |
377 // We don't want to complain about reaching the end of a macro | |
378 // instantiation or a _Pragma. | |
379 if (PragmaARCCFCodeAuditedInfo.second.isValid() && !isEndOfMacro && | |
380 !(CurLexer && CurLexer->Is_PragmaLexer)) { | |
381 Diag(PragmaARCCFCodeAuditedInfo.second, | |
382 diag::err_pp_eof_in_arc_cf_code_audited); | |
383 | |
384 // Recover by leaving immediately. | |
385 PragmaARCCFCodeAuditedInfo = {nullptr, SourceLocation()}; | |
386 } | |
387 | |
388 // Complain about reaching a true EOF within assume_nonnull. | |
389 // We don't want to complain about reaching the end of a macro | |
390 // instantiation or a _Pragma. | |
391 if (PragmaAssumeNonNullLoc.isValid() && | |
392 !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) { | |
393 Diag(PragmaAssumeNonNullLoc, diag::err_pp_eof_in_assume_nonnull); | |
394 | |
395 // Recover by leaving immediately. | |
396 PragmaAssumeNonNullLoc = SourceLocation(); | |
397 } | |
398 | |
399 bool LeavingPCHThroughHeader = false; | |
400 | |
401 // If this is a #include'd file, pop it off the include stack and continue | |
402 // lexing the #includer file. | |
403 if (!IncludeMacroStack.empty()) { | |
404 | |
405 // If we lexed the code-completion file, act as if we reached EOF. | |
406 if (isCodeCompletionEnabled() && CurPPLexer && | |
407 SourceMgr.getLocForStartOfFile(CurPPLexer->getFileID()) == | |
408 CodeCompletionFileLoc) { | |
409 assert(CurLexer && "Got EOF but no current lexer set!"); | |
410 Result.startToken(); | |
411 CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof); | |
412 CurLexer.reset(); | |
413 | |
414 CurPPLexer = nullptr; | |
415 recomputeCurLexerKind(); | |
416 return true; | |
417 } | |
418 | |
419 if (!isEndOfMacro && CurPPLexer && | |
420 SourceMgr.getIncludeLoc(CurPPLexer->getFileID()).isValid()) { | |
421 // Notify SourceManager to record the number of FileIDs that were created | |
422 // during lexing of the #include'd file. | |
423 unsigned NumFIDs = | |
424 SourceMgr.local_sloc_entry_size() - | |
425 CurPPLexer->getInitialNumSLocEntries() + 1/*#include'd file*/; | |
426 SourceMgr.setNumCreatedFIDsForFileID(CurPPLexer->getFileID(), NumFIDs); | |
427 } | |
428 | |
429 bool ExitedFromPredefinesFile = false; | |
430 FileID ExitedFID; | |
431 if (!isEndOfMacro && CurPPLexer) { | |
432 ExitedFID = CurPPLexer->getFileID(); | |
433 | |
434 assert(PredefinesFileID.isValid() && | |
435 "HandleEndOfFile is called before PredefinesFileId is set"); | |
436 ExitedFromPredefinesFile = (PredefinesFileID == ExitedFID); | |
437 } | |
438 | |
439 if (LeavingSubmodule) { | |
440 // We're done with this submodule. | |
441 Module *M = LeaveSubmodule(/*ForPragma*/false); | |
442 | |
443 // Notify the parser that we've left the module. | |
444 const char *EndPos = getCurLexerEndPos(); | |
445 Result.startToken(); | |
446 CurLexer->BufferPtr = EndPos; | |
447 CurLexer->FormTokenWithChars(Result, EndPos, tok::annot_module_end); | |
448 Result.setAnnotationEndLoc(Result.getLocation()); | |
449 Result.setAnnotationValue(M); | |
450 } | |
451 | |
452 bool FoundPCHThroughHeader = false; | |
453 if (CurPPLexer && creatingPCHWithThroughHeader() && | |
454 isPCHThroughHeader( | |
455 SourceMgr.getFileEntryForID(CurPPLexer->getFileID()))) | |
456 FoundPCHThroughHeader = true; | |
457 | |
458 // We're done with the #included file. | |
459 RemoveTopOfLexerStack(); | |
460 | |
461 // Propagate info about start-of-line/leading white-space/etc. | |
462 PropagateLineStartLeadingSpaceInfo(Result); | |
463 | |
464 // Notify the client, if desired, that we are in a new source file. | |
465 if (Callbacks && !isEndOfMacro && CurPPLexer) { | |
466 SrcMgr::CharacteristicKind FileType = | |
467 SourceMgr.getFileCharacteristic(CurPPLexer->getSourceLocation()); | |
468 Callbacks->FileChanged(CurPPLexer->getSourceLocation(), | |
469 PPCallbacks::ExitFile, FileType, ExitedFID); | |
470 } | |
471 | |
472 // Restore conditional stack from the preamble right after exiting from the | |
473 // predefines file. | |
474 if (ExitedFromPredefinesFile) | |
475 replayPreambleConditionalStack(); | |
476 | |
477 if (!isEndOfMacro && CurPPLexer && FoundPCHThroughHeader && | |
478 (isInPrimaryFile() || | |
479 CurPPLexer->getFileID() == getPredefinesFileID())) { | |
480 // Leaving the through header. Continue directly to end of main file | |
481 // processing. | |
482 LeavingPCHThroughHeader = true; | |
483 } else { | |
484 // Client should lex another token unless we generated an EOM. | |
485 return LeavingSubmodule; | |
486 } | |
487 } | |
488 | |
489 // If this is the end of the main file, form an EOF token. | |
490 assert(CurLexer && "Got EOF but no current lexer set!"); | |
491 const char *EndPos = getCurLexerEndPos(); | |
492 Result.startToken(); | |
493 CurLexer->BufferPtr = EndPos; | |
494 CurLexer->FormTokenWithChars(Result, EndPos, tok::eof); | |
495 | |
496 if (isCodeCompletionEnabled()) { | |
497 // Inserting the code-completion point increases the source buffer by 1, | |
498 // but the main FileID was created before inserting the point. | |
499 // Compensate by reducing the EOF location by 1, otherwise the location | |
500 // will point to the next FileID. | |
501 // FIXME: This is hacky, the code-completion point should probably be | |
502 // inserted before the main FileID is created. | |
503 if (CurLexer->getFileLoc() == CodeCompletionFileLoc) | |
504 Result.setLocation(Result.getLocation().getLocWithOffset(-1)); | |
505 } | |
506 | |
507 if (creatingPCHWithThroughHeader() && !LeavingPCHThroughHeader) { | |
508 // Reached the end of the compilation without finding the through header. | |
509 Diag(CurLexer->getFileLoc(), diag::err_pp_through_header_not_seen) | |
510 << PPOpts->PCHThroughHeader << 0; | |
511 } | |
512 | |
513 if (!isIncrementalProcessingEnabled()) | |
514 // We're done with lexing. | |
515 CurLexer.reset(); | |
516 | |
517 if (!isIncrementalProcessingEnabled()) | |
518 CurPPLexer = nullptr; | |
519 | |
520 if (TUKind == TU_Complete) { | |
521 // This is the end of the top-level file. 'WarnUnusedMacroLocs' has | |
522 // collected all macro locations that we need to warn because they are not | |
523 // used. | |
524 for (WarnUnusedMacroLocsTy::iterator | |
525 I=WarnUnusedMacroLocs.begin(), E=WarnUnusedMacroLocs.end(); | |
526 I!=E; ++I) | |
527 Diag(*I, diag::pp_macro_not_used); | |
528 } | |
529 | |
530 // If we are building a module that has an umbrella header, make sure that | |
531 // each of the headers within the directory, including all submodules, is | |
532 // covered by the umbrella header was actually included by the umbrella | |
533 // header. | |
534 if (Module *Mod = getCurrentModule()) { | |
535 llvm::SmallVector<const Module *, 4> AllMods; | |
536 collectAllSubModulesWithUmbrellaHeader(*Mod, AllMods); | |
537 for (auto *M : AllMods) | |
538 diagnoseMissingHeaderInUmbrellaDir(*M); | |
539 } | |
540 | |
541 return true; | |
542 } | |
543 | |
544 /// HandleEndOfTokenLexer - This callback is invoked when the current TokenLexer | |
545 /// hits the end of its token stream. | |
546 bool Preprocessor::HandleEndOfTokenLexer(Token &Result) { | |
547 assert(CurTokenLexer && !CurPPLexer && | |
548 "Ending a macro when currently in a #include file!"); | |
549 | |
550 if (!MacroExpandingLexersStack.empty() && | |
551 MacroExpandingLexersStack.back().first == CurTokenLexer.get()) | |
552 removeCachedMacroExpandedTokensOfLastLexer(); | |
553 | |
554 // Delete or cache the now-dead macro expander. | |
555 if (NumCachedTokenLexers == TokenLexerCacheSize) | |
556 CurTokenLexer.reset(); | |
557 else | |
558 TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer); | |
559 | |
560 // Handle this like a #include file being popped off the stack. | |
561 return HandleEndOfFile(Result, true); | |
562 } | |
563 | |
564 /// RemoveTopOfLexerStack - Pop the current lexer/macro exp off the top of the | |
565 /// lexer stack. This should only be used in situations where the current | |
566 /// state of the top-of-stack lexer is unknown. | |
567 void Preprocessor::RemoveTopOfLexerStack() { | |
568 assert(!IncludeMacroStack.empty() && "Ran out of stack entries to load"); | |
569 | |
570 if (CurTokenLexer) { | |
571 // Delete or cache the now-dead macro expander. | |
572 if (NumCachedTokenLexers == TokenLexerCacheSize) | |
573 CurTokenLexer.reset(); | |
574 else | |
575 TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer); | |
576 } | |
577 | |
578 PopIncludeMacroStack(); | |
579 } | |
580 | |
581 /// HandleMicrosoftCommentPaste - When the macro expander pastes together a | |
582 /// comment (/##/) in microsoft mode, this method handles updating the current | |
583 /// state, returning the token on the next source line. | |
584 void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) { | |
585 assert(CurTokenLexer && !CurPPLexer && | |
586 "Pasted comment can only be formed from macro"); | |
587 // We handle this by scanning for the closest real lexer, switching it to | |
588 // raw mode and preprocessor mode. This will cause it to return \n as an | |
589 // explicit EOD token. | |
590 PreprocessorLexer *FoundLexer = nullptr; | |
591 bool LexerWasInPPMode = false; | |
592 for (const IncludeStackInfo &ISI : llvm::reverse(IncludeMacroStack)) { | |
593 if (ISI.ThePPLexer == nullptr) continue; // Scan for a real lexer. | |
594 | |
595 // Once we find a real lexer, mark it as raw mode (disabling macro | |
596 // expansions) and preprocessor mode (return EOD). We know that the lexer | |
597 // was *not* in raw mode before, because the macro that the comment came | |
598 // from was expanded. However, it could have already been in preprocessor | |
599 // mode (#if COMMENT) in which case we have to return it to that mode and | |
600 // return EOD. | |
601 FoundLexer = ISI.ThePPLexer; | |
602 FoundLexer->LexingRawMode = true; | |
603 LexerWasInPPMode = FoundLexer->ParsingPreprocessorDirective; | |
604 FoundLexer->ParsingPreprocessorDirective = true; | |
605 break; | |
606 } | |
607 | |
608 // Okay, we either found and switched over the lexer, or we didn't find a | |
609 // lexer. In either case, finish off the macro the comment came from, getting | |
610 // the next token. | |
611 if (!HandleEndOfTokenLexer(Tok)) Lex(Tok); | |
612 | |
613 // Discarding comments as long as we don't have EOF or EOD. This 'comments | |
614 // out' the rest of the line, including any tokens that came from other macros | |
615 // that were active, as in: | |
616 // #define submacro a COMMENT b | |
617 // submacro c | |
618 // which should lex to 'a' only: 'b' and 'c' should be removed. | |
619 while (Tok.isNot(tok::eod) && Tok.isNot(tok::eof)) | |
620 Lex(Tok); | |
621 | |
622 // If we got an eod token, then we successfully found the end of the line. | |
623 if (Tok.is(tok::eod)) { | |
624 assert(FoundLexer && "Can't get end of line without an active lexer"); | |
625 // Restore the lexer back to normal mode instead of raw mode. | |
626 FoundLexer->LexingRawMode = false; | |
627 | |
628 // If the lexer was already in preprocessor mode, just return the EOD token | |
629 // to finish the preprocessor line. | |
630 if (LexerWasInPPMode) return; | |
631 | |
632 // Otherwise, switch out of PP mode and return the next lexed token. | |
633 FoundLexer->ParsingPreprocessorDirective = false; | |
634 return Lex(Tok); | |
635 } | |
636 | |
637 // If we got an EOF token, then we reached the end of the token stream but | |
638 // didn't find an explicit \n. This can only happen if there was no lexer | |
639 // active (an active lexer would return EOD at EOF if there was no \n in | |
640 // preprocessor directive mode), so just return EOF as our token. | |
641 assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode"); | |
642 } | |
643 | |
644 void Preprocessor::EnterSubmodule(Module *M, SourceLocation ImportLoc, | |
645 bool ForPragma) { | |
646 if (!getLangOpts().ModulesLocalVisibility) { | |
647 // Just track that we entered this submodule. | |
648 BuildingSubmoduleStack.push_back( | |
649 BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState, | |
650 PendingModuleMacroNames.size())); | |
651 if (Callbacks) | |
652 Callbacks->EnteredSubmodule(M, ImportLoc, ForPragma); | |
653 return; | |
654 } | |
655 | |
656 // Resolve as much of the module definition as we can now, before we enter | |
657 // one of its headers. | |
658 // FIXME: Can we enable Complain here? | |
659 // FIXME: Can we do this when local visibility is disabled? | |
660 ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap(); | |
661 ModMap.resolveExports(M, /*Complain=*/false); | |
662 ModMap.resolveUses(M, /*Complain=*/false); | |
663 ModMap.resolveConflicts(M, /*Complain=*/false); | |
664 | |
665 // If this is the first time we've entered this module, set up its state. | |
666 auto R = Submodules.insert(std::make_pair(M, SubmoduleState())); | |
667 auto &State = R.first->second; | |
668 bool FirstTime = R.second; | |
669 if (FirstTime) { | |
670 // Determine the set of starting macros for this submodule; take these | |
671 // from the "null" module (the predefines buffer). | |
672 // | |
673 // FIXME: If we have local visibility but not modules enabled, the | |
674 // NullSubmoduleState is polluted by #defines in the top-level source | |
675 // file. | |
676 auto &StartingMacros = NullSubmoduleState.Macros; | |
677 | |
678 // Restore to the starting state. | |
679 // FIXME: Do this lazily, when each macro name is first referenced. | |
680 for (auto &Macro : StartingMacros) { | |
681 // Skip uninteresting macros. | |
682 if (!Macro.second.getLatest() && | |
683 Macro.second.getOverriddenMacros().empty()) | |
684 continue; | |
685 | |
686 MacroState MS(Macro.second.getLatest()); | |
687 MS.setOverriddenMacros(*this, Macro.second.getOverriddenMacros()); | |
688 State.Macros.insert(std::make_pair(Macro.first, std::move(MS))); | |
689 } | |
690 } | |
691 | |
692 // Track that we entered this module. | |
693 BuildingSubmoduleStack.push_back( | |
694 BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState, | |
695 PendingModuleMacroNames.size())); | |
696 | |
697 if (Callbacks) | |
698 Callbacks->EnteredSubmodule(M, ImportLoc, ForPragma); | |
699 | |
700 // Switch to this submodule as the current submodule. | |
701 CurSubmoduleState = &State; | |
702 | |
703 // This module is visible to itself. | |
704 if (FirstTime) | |
705 makeModuleVisible(M, ImportLoc); | |
706 } | |
707 | |
708 bool Preprocessor::needModuleMacros() const { | |
709 // If we're not within a submodule, we never need to create ModuleMacros. | |
710 if (BuildingSubmoduleStack.empty()) | |
711 return false; | |
712 // If we are tracking module macro visibility even for textually-included | |
713 // headers, we need ModuleMacros. | |
714 if (getLangOpts().ModulesLocalVisibility) | |
715 return true; | |
716 // Otherwise, we only need module macros if we're actually compiling a module | |
717 // interface. | |
718 return getLangOpts().isCompilingModule(); | |
719 } | |
720 | |
721 Module *Preprocessor::LeaveSubmodule(bool ForPragma) { | |
722 if (BuildingSubmoduleStack.empty() || | |
723 BuildingSubmoduleStack.back().IsPragma != ForPragma) { | |
724 assert(ForPragma && "non-pragma module enter/leave mismatch"); | |
725 return nullptr; | |
726 } | |
727 | |
728 auto &Info = BuildingSubmoduleStack.back(); | |
729 | |
730 Module *LeavingMod = Info.M; | |
731 SourceLocation ImportLoc = Info.ImportLoc; | |
732 | |
733 if (!needModuleMacros() || | |
734 (!getLangOpts().ModulesLocalVisibility && | |
735 LeavingMod->getTopLevelModuleName() != getLangOpts().CurrentModule)) { | |
736 // If we don't need module macros, or this is not a module for which we | |
737 // are tracking macro visibility, don't build any, and preserve the list | |
738 // of pending names for the surrounding submodule. | |
739 BuildingSubmoduleStack.pop_back(); | |
740 | |
741 if (Callbacks) | |
742 Callbacks->LeftSubmodule(LeavingMod, ImportLoc, ForPragma); | |
743 | |
744 makeModuleVisible(LeavingMod, ImportLoc); | |
745 return LeavingMod; | |
746 } | |
747 | |
748 // Create ModuleMacros for any macros defined in this submodule. | |
749 llvm::SmallPtrSet<const IdentifierInfo*, 8> VisitedMacros; | |
750 for (unsigned I = Info.OuterPendingModuleMacroNames; | |
751 I != PendingModuleMacroNames.size(); ++I) { | |
752 auto *II = const_cast<IdentifierInfo*>(PendingModuleMacroNames[I]); | |
753 if (!VisitedMacros.insert(II).second) | |
754 continue; | |
755 | |
756 auto MacroIt = CurSubmoduleState->Macros.find(II); | |
757 if (MacroIt == CurSubmoduleState->Macros.end()) | |
758 continue; | |
759 auto &Macro = MacroIt->second; | |
760 | |
761 // Find the starting point for the MacroDirective chain in this submodule. | |
762 MacroDirective *OldMD = nullptr; | |
763 auto *OldState = Info.OuterSubmoduleState; | |
764 if (getLangOpts().ModulesLocalVisibility) | |
765 OldState = &NullSubmoduleState; | |
766 if (OldState && OldState != CurSubmoduleState) { | |
767 // FIXME: It'd be better to start at the state from when we most recently | |
768 // entered this submodule, but it doesn't really matter. | |
769 auto &OldMacros = OldState->Macros; | |
770 auto OldMacroIt = OldMacros.find(II); | |
771 if (OldMacroIt == OldMacros.end()) | |
772 OldMD = nullptr; | |
773 else | |
774 OldMD = OldMacroIt->second.getLatest(); | |
775 } | |
776 | |
777 // This module may have exported a new macro. If so, create a ModuleMacro | |
778 // representing that fact. | |
779 bool ExplicitlyPublic = false; | |
780 for (auto *MD = Macro.getLatest(); MD != OldMD; MD = MD->getPrevious()) { | |
781 assert(MD && "broken macro directive chain"); | |
782 | |
783 if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(MD)) { | |
784 // The latest visibility directive for a name in a submodule affects | |
785 // all the directives that come before it. | |
786 if (VisMD->isPublic()) | |
787 ExplicitlyPublic = true; | |
788 else if (!ExplicitlyPublic) | |
789 // Private with no following public directive: not exported. | |
790 break; | |
791 } else { | |
792 MacroInfo *Def = nullptr; | |
793 if (DefMacroDirective *DefMD = dyn_cast<DefMacroDirective>(MD)) | |
794 Def = DefMD->getInfo(); | |
795 | |
796 // FIXME: Issue a warning if multiple headers for the same submodule | |
797 // define a macro, rather than silently ignoring all but the first. | |
798 bool IsNew; | |
799 // Don't bother creating a module macro if it would represent a #undef | |
800 // that doesn't override anything. | |
801 if (Def || !Macro.getOverriddenMacros().empty()) | |
802 addModuleMacro(LeavingMod, II, Def, | |
803 Macro.getOverriddenMacros(), IsNew); | |
804 | |
805 if (!getLangOpts().ModulesLocalVisibility) { | |
806 // This macro is exposed to the rest of this compilation as a | |
807 // ModuleMacro; we don't need to track its MacroDirective any more. | |
808 Macro.setLatest(nullptr); | |
809 Macro.setOverriddenMacros(*this, {}); | |
810 } | |
811 break; | |
812 } | |
813 } | |
814 } | |
815 PendingModuleMacroNames.resize(Info.OuterPendingModuleMacroNames); | |
816 | |
817 // FIXME: Before we leave this submodule, we should parse all the other | |
818 // headers within it. Otherwise, we're left with an inconsistent state | |
819 // where we've made the module visible but don't yet have its complete | |
820 // contents. | |
821 | |
822 // Put back the outer module's state, if we're tracking it. | |
823 if (getLangOpts().ModulesLocalVisibility) | |
824 CurSubmoduleState = Info.OuterSubmoduleState; | |
825 | |
826 BuildingSubmoduleStack.pop_back(); | |
827 | |
828 if (Callbacks) | |
829 Callbacks->LeftSubmodule(LeavingMod, ImportLoc, ForPragma); | |
830 | |
831 // A nested #include makes the included submodule visible. | |
832 makeModuleVisible(LeavingMod, ImportLoc); | |
833 return LeavingMod; | |
834 } |