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/Basic/FileManager.h" |
15 | #include "clang/Basic/SourceManager.h" |
16 | #include "clang/Lex/HeaderSearch.h" |
17 | #include "clang/Lex/LexDiagnostic.h" |
18 | #include "clang/Lex/MacroInfo.h" |
19 | #include "clang/Lex/Preprocessor.h" |
20 | #include "clang/Lex/PreprocessorOptions.h" |
21 | #include "llvm/ADT/StringSwitch.h" |
22 | #include "llvm/Support/FileSystem.h" |
23 | #include "llvm/Support/MemoryBufferRef.h" |
24 | #include "llvm/Support/Path.h" |
25 | #include <optional> |
26 | |
27 | using namespace clang; |
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 llvm::none_of( |
43 | Range: llvm::drop_begin(RangeOrContainer: IncludeMacroStack), |
44 | P: [&](const IncludeStackInfo &ISI) -> bool { return IsFileLexer(I: 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(C: IncludeMacroStack)) { |
56 | if (IsFileLexer(I: 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, ConstSearchDirIterator CurDir, |
70 | SourceLocation Loc, |
71 | bool IsFirstIncludeOfFile) { |
72 | assert(!CurTokenLexer && "Cannot #include a file inside a macro!" ); |
73 | ++NumEnteredSourceFiles; |
74 | |
75 | if (MaxIncludeStackDepth < IncludeMacroStack.size()) |
76 | MaxIncludeStackDepth = IncludeMacroStack.size(); |
77 | |
78 | // Get the MemoryBuffer for this FID, if it fails, we fail. |
79 | std::optional<llvm::MemoryBufferRef> InputFile = |
80 | getSourceManager().getBufferOrNone(FID, Loc); |
81 | if (!InputFile) { |
82 | SourceLocation FileStart = SourceMgr.getLocForStartOfFile(FID); |
83 | Diag(Loc, DiagID: diag::err_pp_error_opening_file) |
84 | << std::string(SourceMgr.getBufferName(Loc: FileStart)) << "" ; |
85 | return true; |
86 | } |
87 | |
88 | if (isCodeCompletionEnabled() && |
89 | SourceMgr.getFileEntryForID(FID) == CodeCompletionFile) { |
90 | CodeCompletionFileLoc = SourceMgr.getLocForStartOfFile(FID); |
91 | CodeCompletionLoc = |
92 | CodeCompletionFileLoc.getLocWithOffset(Offset: CodeCompletionOffset); |
93 | } |
94 | |
95 | Lexer *TheLexer = new Lexer(FID, *InputFile, *this, IsFirstIncludeOfFile); |
96 | if (getPreprocessorOpts().DependencyDirectivesForFile && |
97 | FID != PredefinesFileID) { |
98 | if (OptionalFileEntryRef File = SourceMgr.getFileEntryRefForID(FID)) { |
99 | if (std::optional<ArrayRef<dependency_directives_scan::Directive>> |
100 | DepDirectives = |
101 | getPreprocessorOpts().DependencyDirectivesForFile(*File)) { |
102 | TheLexer->DepDirectives = *DepDirectives; |
103 | } |
104 | } |
105 | } |
106 | |
107 | EnterSourceFileWithLexer(TheLexer, Dir: CurDir); |
108 | return false; |
109 | } |
110 | |
111 | /// EnterSourceFileWithLexer - Add a source file to the top of the include stack |
112 | /// and start lexing tokens from it instead of the current buffer. |
113 | void Preprocessor::EnterSourceFileWithLexer(Lexer *TheLexer, |
114 | ConstSearchDirIterator CurDir) { |
115 | PreprocessorLexer *PrevPPLexer = CurPPLexer; |
116 | |
117 | // Add the current lexer to the include stack. |
118 | if (CurPPLexer || CurTokenLexer) |
119 | PushIncludeMacroStack(); |
120 | |
121 | CurLexer.reset(p: TheLexer); |
122 | CurPPLexer = TheLexer; |
123 | CurDirLookup = CurDir; |
124 | CurLexerSubmodule = nullptr; |
125 | if (CurLexerCallback != CLK_LexAfterModuleImport) |
126 | CurLexerCallback = TheLexer->isDependencyDirectivesLexer() |
127 | ? CLK_DependencyDirectivesLexer |
128 | : CLK_Lexer; |
129 | |
130 | // Notify the client, if desired, that we are in a new source file. |
131 | if (Callbacks && !CurLexer->Is_PragmaLexer) { |
132 | SrcMgr::CharacteristicKind FileType = |
133 | SourceMgr.getFileCharacteristic(Loc: CurLexer->getFileLoc()); |
134 | |
135 | FileID PrevFID; |
136 | SourceLocation EnterLoc; |
137 | if (PrevPPLexer) { |
138 | PrevFID = PrevPPLexer->getFileID(); |
139 | EnterLoc = PrevPPLexer->getSourceLocation(); |
140 | } |
141 | Callbacks->FileChanged(Loc: CurLexer->getFileLoc(), Reason: PPCallbacks::EnterFile, |
142 | FileType, PrevFID); |
143 | Callbacks->LexedFileChanged(FID: CurLexer->getFileID(), |
144 | Reason: PPCallbacks::LexedFileChangeReason::EnterFile, |
145 | FileType, PrevFID, Loc: EnterLoc); |
146 | } |
147 | } |
148 | |
149 | /// EnterMacro - Add a Macro to the top of the include stack and start lexing |
150 | /// tokens from it instead of the current buffer. |
151 | void Preprocessor::EnterMacro(Token &Tok, SourceLocation ILEnd, |
152 | MacroInfo *Macro, MacroArgs *Args) { |
153 | std::unique_ptr<TokenLexer> TokLexer; |
154 | if (NumCachedTokenLexers == 0) { |
155 | TokLexer = std::make_unique<TokenLexer>(args&: Tok, args&: ILEnd, args&: Macro, args&: Args, args&: *this); |
156 | } else { |
157 | TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]); |
158 | TokLexer->Init(Tok, ELEnd: ILEnd, MI: Macro, Actuals: Args); |
159 | } |
160 | |
161 | PushIncludeMacroStack(); |
162 | CurDirLookup = nullptr; |
163 | CurTokenLexer = std::move(TokLexer); |
164 | if (CurLexerCallback != CLK_LexAfterModuleImport) |
165 | CurLexerCallback = CLK_TokenLexer; |
166 | } |
167 | |
168 | /// EnterTokenStream - Add a "macro" context to the top of the include stack, |
169 | /// which will cause the lexer to start returning the specified tokens. |
170 | /// |
171 | /// If DisableMacroExpansion is true, tokens lexed from the token stream will |
172 | /// not be subject to further macro expansion. Otherwise, these tokens will |
173 | /// be re-macro-expanded when/if expansion is enabled. |
174 | /// |
175 | /// If OwnsTokens is false, this method assumes that the specified stream of |
176 | /// tokens has a permanent owner somewhere, so they do not need to be copied. |
177 | /// If it is true, it assumes the array of tokens is allocated with new[] and |
178 | /// must be freed. |
179 | /// |
180 | void Preprocessor::EnterTokenStream(const Token *Toks, unsigned NumToks, |
181 | bool DisableMacroExpansion, bool OwnsTokens, |
182 | bool IsReinject) { |
183 | if (CurLexerCallback == CLK_CachingLexer) { |
184 | if (CachedLexPos < CachedTokens.size()) { |
185 | assert(IsReinject && "new tokens in the middle of cached stream" ); |
186 | // We're entering tokens into the middle of our cached token stream. We |
187 | // can't represent that, so just insert the tokens into the buffer. |
188 | CachedTokens.insert(I: CachedTokens.begin() + CachedLexPos, |
189 | From: Toks, To: Toks + NumToks); |
190 | if (OwnsTokens) |
191 | delete [] Toks; |
192 | return; |
193 | } |
194 | |
195 | // New tokens are at the end of the cached token sequnece; insert the |
196 | // token stream underneath the caching lexer. |
197 | ExitCachingLexMode(); |
198 | EnterTokenStream(Toks, NumToks, DisableMacroExpansion, OwnsTokens, |
199 | IsReinject); |
200 | EnterCachingLexMode(); |
201 | return; |
202 | } |
203 | |
204 | // Create a macro expander to expand from the specified token stream. |
205 | std::unique_ptr<TokenLexer> TokLexer; |
206 | if (NumCachedTokenLexers == 0) { |
207 | TokLexer = std::make_unique<TokenLexer>( |
208 | args&: Toks, args&: NumToks, args&: DisableMacroExpansion, args&: OwnsTokens, args&: IsReinject, args&: *this); |
209 | } else { |
210 | TokLexer = std::move(TokenLexerCache[--NumCachedTokenLexers]); |
211 | TokLexer->Init(TokArray: Toks, NumToks, DisableMacroExpansion, OwnsTokens, |
212 | IsReinject); |
213 | } |
214 | |
215 | // Save our current state. |
216 | PushIncludeMacroStack(); |
217 | CurDirLookup = nullptr; |
218 | CurTokenLexer = std::move(TokLexer); |
219 | if (CurLexerCallback != CLK_LexAfterModuleImport) |
220 | CurLexerCallback = CLK_TokenLexer; |
221 | } |
222 | |
223 | /// Compute the relative path that names the given file relative to |
224 | /// the given directory. |
225 | static void computeRelativePath(FileManager &FM, const DirectoryEntry *Dir, |
226 | FileEntryRef File, SmallString<128> &Result) { |
227 | Result.clear(); |
228 | |
229 | StringRef FilePath = File.getDir().getName(); |
230 | StringRef Path = FilePath; |
231 | while (!Path.empty()) { |
232 | if (auto CurDir = FM.getDirectory(DirName: Path)) { |
233 | if (*CurDir == Dir) { |
234 | Result = FilePath.substr(Start: Path.size()); |
235 | llvm::sys::path::append(path&: Result, |
236 | a: llvm::sys::path::filename(path: File.getName())); |
237 | return; |
238 | } |
239 | } |
240 | |
241 | Path = llvm::sys::path::parent_path(path: Path); |
242 | } |
243 | |
244 | Result = File.getName(); |
245 | } |
246 | |
247 | void Preprocessor::PropagateLineStartLeadingSpaceInfo(Token &Result) { |
248 | if (CurTokenLexer) { |
249 | CurTokenLexer->PropagateLineStartLeadingSpaceInfo(Result); |
250 | return; |
251 | } |
252 | if (CurLexer) { |
253 | CurLexer->PropagateLineStartLeadingSpaceInfo(Result); |
254 | return; |
255 | } |
256 | // FIXME: Handle other kinds of lexers? It generally shouldn't matter, |
257 | // but it might if they're empty? |
258 | } |
259 | |
260 | /// Determine the location to use as the end of the buffer for a lexer. |
261 | /// |
262 | /// If the file ends with a newline, form the EOF token on the newline itself, |
263 | /// rather than "on the line following it", which doesn't exist. This makes |
264 | /// diagnostics relating to the end of file include the last file that the user |
265 | /// actually typed, which is goodness. |
266 | const char *Preprocessor::getCurLexerEndPos() { |
267 | const char *EndPos = CurLexer->BufferEnd; |
268 | if (EndPos != CurLexer->BufferStart && |
269 | (EndPos[-1] == '\n' || EndPos[-1] == '\r')) { |
270 | --EndPos; |
271 | |
272 | // Handle \n\r and \r\n: |
273 | if (EndPos != CurLexer->BufferStart && |
274 | (EndPos[-1] == '\n' || EndPos[-1] == '\r') && |
275 | EndPos[-1] != EndPos[0]) |
276 | --EndPos; |
277 | } |
278 | |
279 | return EndPos; |
280 | } |
281 | |
282 | static void ( |
283 | const Module &Mod, SmallVectorImpl<const Module *> &SubMods) { |
284 | if (Mod.getUmbrellaHeaderAsWritten()) |
285 | SubMods.push_back(Elt: &Mod); |
286 | for (auto *M : Mod.submodules()) |
287 | collectAllSubModulesWithUmbrellaHeader(Mod: *M, SubMods); |
288 | } |
289 | |
290 | void Preprocessor::(const Module &Mod) { |
291 | std::optional<Module::Header> = |
292 | Mod.getUmbrellaHeaderAsWritten(); |
293 | assert(UmbrellaHeader && "Module must use umbrella header" ); |
294 | const FileID &File = SourceMgr.translateFile(SourceFile: UmbrellaHeader->Entry); |
295 | SourceLocation = SourceMgr.getLocForEndOfFile(FID: File); |
296 | if (getDiagnostics().isIgnored(DiagID: diag::warn_uncovered_module_header, |
297 | Loc: ExpectedHeadersLoc)) |
298 | return; |
299 | |
300 | ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap(); |
301 | OptionalDirectoryEntryRef Dir = Mod.getEffectiveUmbrellaDir(); |
302 | llvm::vfs::FileSystem &FS = FileMgr.getVirtualFileSystem(); |
303 | std::error_code EC; |
304 | for (llvm::vfs::recursive_directory_iterator Entry(FS, Dir->getName(), EC), |
305 | End; |
306 | Entry != End && !EC; Entry.increment(EC)) { |
307 | using llvm::StringSwitch; |
308 | |
309 | // Check whether this entry has an extension typically associated with |
310 | // headers. |
311 | if (!StringSwitch<bool>(llvm::sys::path::extension(path: Entry->path())) |
312 | .Cases(S0: ".h" , S1: ".H" , S2: ".hh" , S3: ".hpp" , Value: true) |
313 | .Default(Value: false)) |
314 | continue; |
315 | |
316 | if (auto = getFileManager().getOptionalFileRef(Filename: Entry->path())) |
317 | if (!getSourceManager().hasFileInfo(File: *Header)) { |
318 | if (!ModMap.isHeaderInUnavailableModule(Header: *Header)) { |
319 | // Find the relative path that would access this header. |
320 | SmallString<128> RelativePath; |
321 | computeRelativePath(FM&: FileMgr, Dir: *Dir, File: *Header, Result&: RelativePath); |
322 | Diag(Loc: ExpectedHeadersLoc, DiagID: diag::warn_uncovered_module_header) |
323 | << Mod.getFullModuleName() << RelativePath; |
324 | } |
325 | } |
326 | } |
327 | } |
328 | |
329 | /// HandleEndOfFile - This callback is invoked when the lexer hits the end of |
330 | /// the current file. This either returns the EOF token or pops a level off |
331 | /// the include stack and keeps going. |
332 | bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) { |
333 | assert(!CurTokenLexer && |
334 | "Ending a file when currently in a macro!" ); |
335 | |
336 | SourceLocation UnclosedSafeBufferOptOutLoc; |
337 | |
338 | if (IncludeMacroStack.empty() && |
339 | isPPInSafeBufferOptOutRegion(StartLoc&: UnclosedSafeBufferOptOutLoc)) { |
340 | // To warn if a "-Wunsafe-buffer-usage" opt-out region is still open by the |
341 | // end of a file. |
342 | Diag(Loc: UnclosedSafeBufferOptOutLoc, |
343 | DiagID: diag::err_pp_unclosed_pragma_unsafe_buffer_usage); |
344 | } |
345 | // If we have an unclosed module region from a pragma at the end of a |
346 | // module, complain and close it now. |
347 | const bool LeavingSubmodule = CurLexer && CurLexerSubmodule; |
348 | if ((LeavingSubmodule || IncludeMacroStack.empty()) && |
349 | !BuildingSubmoduleStack.empty() && |
350 | BuildingSubmoduleStack.back().IsPragma) { |
351 | Diag(Loc: BuildingSubmoduleStack.back().ImportLoc, |
352 | DiagID: diag::err_pp_module_begin_without_module_end); |
353 | Module *M = LeaveSubmodule(/*ForPragma*/true); |
354 | |
355 | Result.startToken(); |
356 | const char *EndPos = getCurLexerEndPos(); |
357 | CurLexer->BufferPtr = EndPos; |
358 | CurLexer->FormTokenWithChars(Result, TokEnd: EndPos, Kind: tok::annot_module_end); |
359 | Result.setAnnotationEndLoc(Result.getLocation()); |
360 | Result.setAnnotationValue(M); |
361 | return true; |
362 | } |
363 | |
364 | // See if this file had a controlling macro. |
365 | if (CurPPLexer) { // Not ending a macro, ignore it. |
366 | if (const IdentifierInfo *ControllingMacro = |
367 | CurPPLexer->MIOpt.GetControllingMacroAtEndOfFile()) { |
368 | // Okay, this has a controlling macro, remember in HeaderFileInfo. |
369 | if (OptionalFileEntryRef FE = CurPPLexer->getFileEntry()) { |
370 | HeaderInfo.SetFileControllingMacro(File: *FE, ControllingMacro); |
371 | if (MacroInfo *MI = getMacroInfo(II: ControllingMacro)) |
372 | MI->setUsedForHeaderGuard(true); |
373 | if (const IdentifierInfo *DefinedMacro = |
374 | CurPPLexer->MIOpt.GetDefinedMacro()) { |
375 | if (!isMacroDefined(II: ControllingMacro) && |
376 | DefinedMacro != ControllingMacro && |
377 | CurLexer->isFirstTimeLexingFile()) { |
378 | |
379 | // If the edit distance between the two macros is more than 50%, |
380 | // DefinedMacro may not be header guard, or can be header guard of |
381 | // another header file. Therefore, it maybe defining something |
382 | // completely different. This can be observed in the wild when |
383 | // handling feature macros or header guards in different files. |
384 | |
385 | const StringRef ControllingMacroName = ControllingMacro->getName(); |
386 | const StringRef DefinedMacroName = DefinedMacro->getName(); |
387 | const size_t MaxHalfLength = std::max(a: ControllingMacroName.size(), |
388 | b: DefinedMacroName.size()) / 2; |
389 | const unsigned ED = ControllingMacroName.edit_distance( |
390 | Other: DefinedMacroName, AllowReplacements: true, MaxEditDistance: MaxHalfLength); |
391 | if (ED <= MaxHalfLength) { |
392 | // Emit a warning for a bad header guard. |
393 | Diag(Loc: CurPPLexer->MIOpt.GetMacroLocation(), |
394 | DiagID: diag::warn_header_guard) |
395 | << CurPPLexer->MIOpt.GetMacroLocation() << ControllingMacro; |
396 | Diag(Loc: CurPPLexer->MIOpt.GetDefinedLocation(), |
397 | DiagID: diag::note_header_guard) |
398 | << CurPPLexer->MIOpt.GetDefinedLocation() << DefinedMacro |
399 | << ControllingMacro |
400 | << FixItHint::CreateReplacement( |
401 | RemoveRange: CurPPLexer->MIOpt.GetDefinedLocation(), |
402 | Code: ControllingMacro->getName()); |
403 | } |
404 | } |
405 | } |
406 | } |
407 | } |
408 | } |
409 | |
410 | // Complain about reaching a true EOF within arc_cf_code_audited. |
411 | // We don't want to complain about reaching the end of a macro |
412 | // instantiation or a _Pragma. |
413 | if (PragmaARCCFCodeAuditedInfo.second.isValid() && !isEndOfMacro && |
414 | !(CurLexer && CurLexer->Is_PragmaLexer)) { |
415 | Diag(Loc: PragmaARCCFCodeAuditedInfo.second, |
416 | DiagID: diag::err_pp_eof_in_arc_cf_code_audited); |
417 | |
418 | // Recover by leaving immediately. |
419 | PragmaARCCFCodeAuditedInfo = {nullptr, SourceLocation()}; |
420 | } |
421 | |
422 | // Complain about reaching a true EOF within assume_nonnull. |
423 | // We don't want to complain about reaching the end of a macro |
424 | // instantiation or a _Pragma. |
425 | if (PragmaAssumeNonNullLoc.isValid() && |
426 | !isEndOfMacro && !(CurLexer && CurLexer->Is_PragmaLexer)) { |
427 | // If we're at the end of generating a preamble, we should record the |
428 | // unterminated \#pragma clang assume_nonnull so we can restore it later |
429 | // when the preamble is loaded into the main file. |
430 | if (isRecordingPreamble() && isInPrimaryFile()) |
431 | PreambleRecordedPragmaAssumeNonNullLoc = PragmaAssumeNonNullLoc; |
432 | else |
433 | Diag(Loc: PragmaAssumeNonNullLoc, DiagID: diag::err_pp_eof_in_assume_nonnull); |
434 | // Recover by leaving immediately. |
435 | PragmaAssumeNonNullLoc = SourceLocation(); |
436 | } |
437 | |
438 | bool = false; |
439 | |
440 | // If this is a #include'd file, pop it off the include stack and continue |
441 | // lexing the #includer file. |
442 | if (!IncludeMacroStack.empty()) { |
443 | |
444 | // If we lexed the code-completion file, act as if we reached EOF. |
445 | if (isCodeCompletionEnabled() && CurPPLexer && |
446 | SourceMgr.getLocForStartOfFile(FID: CurPPLexer->getFileID()) == |
447 | CodeCompletionFileLoc) { |
448 | assert(CurLexer && "Got EOF but no current lexer set!" ); |
449 | Result.startToken(); |
450 | CurLexer->FormTokenWithChars(Result, TokEnd: CurLexer->BufferEnd, Kind: tok::eof); |
451 | CurLexer.reset(); |
452 | |
453 | CurPPLexer = nullptr; |
454 | recomputeCurLexerKind(); |
455 | return true; |
456 | } |
457 | |
458 | if (!isEndOfMacro && CurPPLexer && |
459 | (SourceMgr.getIncludeLoc(FID: CurPPLexer->getFileID()).isValid() || |
460 | // Predefines file doesn't have a valid include location. |
461 | (PredefinesFileID.isValid() && |
462 | CurPPLexer->getFileID() == PredefinesFileID))) { |
463 | // Notify SourceManager to record the number of FileIDs that were created |
464 | // during lexing of the #include'd file. |
465 | unsigned NumFIDs = |
466 | SourceMgr.local_sloc_entry_size() - |
467 | CurPPLexer->getInitialNumSLocEntries() + 1/*#include'd file*/; |
468 | SourceMgr.setNumCreatedFIDsForFileID(FID: CurPPLexer->getFileID(), NumFIDs); |
469 | } |
470 | |
471 | bool ExitedFromPredefinesFile = false; |
472 | FileID ExitedFID; |
473 | if (!isEndOfMacro && CurPPLexer) { |
474 | ExitedFID = CurPPLexer->getFileID(); |
475 | |
476 | assert(PredefinesFileID.isValid() && |
477 | "HandleEndOfFile is called before PredefinesFileId is set" ); |
478 | ExitedFromPredefinesFile = (PredefinesFileID == ExitedFID); |
479 | } |
480 | |
481 | if (LeavingSubmodule) { |
482 | // We're done with this submodule. |
483 | Module *M = LeaveSubmodule(/*ForPragma*/false); |
484 | |
485 | // Notify the parser that we've left the module. |
486 | const char *EndPos = getCurLexerEndPos(); |
487 | Result.startToken(); |
488 | CurLexer->BufferPtr = EndPos; |
489 | CurLexer->FormTokenWithChars(Result, TokEnd: EndPos, Kind: tok::annot_module_end); |
490 | Result.setAnnotationEndLoc(Result.getLocation()); |
491 | Result.setAnnotationValue(M); |
492 | } |
493 | |
494 | bool = false; |
495 | if (CurPPLexer && creatingPCHWithThroughHeader() && |
496 | isPCHThroughHeader( |
497 | FE: SourceMgr.getFileEntryForID(FID: CurPPLexer->getFileID()))) |
498 | FoundPCHThroughHeader = true; |
499 | |
500 | // We're done with the #included file. |
501 | RemoveTopOfLexerStack(); |
502 | |
503 | // Propagate info about start-of-line/leading white-space/etc. |
504 | PropagateLineStartLeadingSpaceInfo(Result); |
505 | |
506 | // Notify the client, if desired, that we are in a new source file. |
507 | if (Callbacks && !isEndOfMacro && CurPPLexer) { |
508 | SourceLocation Loc = CurPPLexer->getSourceLocation(); |
509 | SrcMgr::CharacteristicKind FileType = |
510 | SourceMgr.getFileCharacteristic(Loc); |
511 | Callbacks->FileChanged(Loc, Reason: PPCallbacks::ExitFile, FileType, PrevFID: ExitedFID); |
512 | Callbacks->LexedFileChanged(FID: CurPPLexer->getFileID(), |
513 | Reason: PPCallbacks::LexedFileChangeReason::ExitFile, |
514 | FileType, PrevFID: ExitedFID, Loc); |
515 | } |
516 | |
517 | // Restore conditional stack as well as the recorded |
518 | // \#pragma clang assume_nonnull from the preamble right after exiting |
519 | // from the predefines file. |
520 | if (ExitedFromPredefinesFile) { |
521 | replayPreambleConditionalStack(); |
522 | if (PreambleRecordedPragmaAssumeNonNullLoc.isValid()) |
523 | PragmaAssumeNonNullLoc = PreambleRecordedPragmaAssumeNonNullLoc; |
524 | } |
525 | |
526 | if (!isEndOfMacro && CurPPLexer && FoundPCHThroughHeader && |
527 | (isInPrimaryFile() || |
528 | CurPPLexer->getFileID() == getPredefinesFileID())) { |
529 | // Leaving the through header. Continue directly to end of main file |
530 | // processing. |
531 | LeavingPCHThroughHeader = true; |
532 | } else { |
533 | // Client should lex another token unless we generated an EOM. |
534 | return LeavingSubmodule; |
535 | } |
536 | } |
537 | // If this is the end of the main file, form an EOF token. |
538 | assert(CurLexer && "Got EOF but no current lexer set!" ); |
539 | const char *EndPos = getCurLexerEndPos(); |
540 | Result.startToken(); |
541 | CurLexer->BufferPtr = EndPos; |
542 | |
543 | if (getLangOpts().IncrementalExtensions) { |
544 | CurLexer->FormTokenWithChars(Result, TokEnd: EndPos, Kind: tok::annot_repl_input_end); |
545 | Result.setAnnotationEndLoc(Result.getLocation()); |
546 | Result.setAnnotationValue(nullptr); |
547 | } else { |
548 | CurLexer->FormTokenWithChars(Result, TokEnd: EndPos, Kind: tok::eof); |
549 | } |
550 | |
551 | if (isCodeCompletionEnabled()) { |
552 | // Inserting the code-completion point increases the source buffer by 1, |
553 | // but the main FileID was created before inserting the point. |
554 | // Compensate by reducing the EOF location by 1, otherwise the location |
555 | // will point to the next FileID. |
556 | // FIXME: This is hacky, the code-completion point should probably be |
557 | // inserted before the main FileID is created. |
558 | if (CurLexer->getFileLoc() == CodeCompletionFileLoc) |
559 | Result.setLocation(Result.getLocation().getLocWithOffset(Offset: -1)); |
560 | } |
561 | |
562 | if (creatingPCHWithThroughHeader() && !LeavingPCHThroughHeader) { |
563 | // Reached the end of the compilation without finding the through header. |
564 | Diag(Loc: CurLexer->getFileLoc(), DiagID: diag::err_pp_through_header_not_seen) |
565 | << PPOpts->PCHThroughHeader << 0; |
566 | } |
567 | |
568 | if (!isIncrementalProcessingEnabled()) |
569 | // We're done with lexing. |
570 | CurLexer.reset(); |
571 | |
572 | if (!isIncrementalProcessingEnabled()) |
573 | CurPPLexer = nullptr; |
574 | |
575 | if (TUKind == TU_Complete) { |
576 | // This is the end of the top-level file. 'WarnUnusedMacroLocs' has |
577 | // collected all macro locations that we need to warn because they are not |
578 | // used. |
579 | for (WarnUnusedMacroLocsTy::iterator |
580 | I=WarnUnusedMacroLocs.begin(), E=WarnUnusedMacroLocs.end(); |
581 | I!=E; ++I) |
582 | Diag(Loc: *I, DiagID: diag::pp_macro_not_used); |
583 | } |
584 | |
585 | // If we are building a module that has an umbrella header, make sure that |
586 | // each of the headers within the directory, including all submodules, is |
587 | // covered by the umbrella header was actually included by the umbrella |
588 | // header. |
589 | if (Module *Mod = getCurrentModule()) { |
590 | llvm::SmallVector<const Module *, 4> AllMods; |
591 | collectAllSubModulesWithUmbrellaHeader(Mod: *Mod, SubMods&: AllMods); |
592 | for (auto *M : AllMods) |
593 | diagnoseMissingHeaderInUmbrellaDir(Mod: *M); |
594 | } |
595 | |
596 | return true; |
597 | } |
598 | |
599 | /// HandleEndOfTokenLexer - This callback is invoked when the current TokenLexer |
600 | /// hits the end of its token stream. |
601 | bool Preprocessor::HandleEndOfTokenLexer(Token &Result) { |
602 | assert(CurTokenLexer && !CurPPLexer && |
603 | "Ending a macro when currently in a #include file!" ); |
604 | |
605 | if (!MacroExpandingLexersStack.empty() && |
606 | MacroExpandingLexersStack.back().first == CurTokenLexer.get()) |
607 | removeCachedMacroExpandedTokensOfLastLexer(); |
608 | |
609 | // Delete or cache the now-dead macro expander. |
610 | if (NumCachedTokenLexers == TokenLexerCacheSize) |
611 | CurTokenLexer.reset(); |
612 | else |
613 | TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer); |
614 | |
615 | // Handle this like a #include file being popped off the stack. |
616 | return HandleEndOfFile(Result, isEndOfMacro: true); |
617 | } |
618 | |
619 | /// RemoveTopOfLexerStack - Pop the current lexer/macro exp off the top of the |
620 | /// lexer stack. This should only be used in situations where the current |
621 | /// state of the top-of-stack lexer is unknown. |
622 | void Preprocessor::RemoveTopOfLexerStack() { |
623 | assert(!IncludeMacroStack.empty() && "Ran out of stack entries to load" ); |
624 | |
625 | if (CurTokenLexer) { |
626 | // Delete or cache the now-dead macro expander. |
627 | if (NumCachedTokenLexers == TokenLexerCacheSize) |
628 | CurTokenLexer.reset(); |
629 | else |
630 | TokenLexerCache[NumCachedTokenLexers++] = std::move(CurTokenLexer); |
631 | } |
632 | |
633 | PopIncludeMacroStack(); |
634 | } |
635 | |
636 | /// HandleMicrosoftCommentPaste - When the macro expander pastes together a |
637 | /// comment (/##/) in microsoft mode, this method handles updating the current |
638 | /// state, returning the token on the next source line. |
639 | void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) { |
640 | assert(CurTokenLexer && !CurPPLexer && |
641 | "Pasted comment can only be formed from macro" ); |
642 | // We handle this by scanning for the closest real lexer, switching it to |
643 | // raw mode and preprocessor mode. This will cause it to return \n as an |
644 | // explicit EOD token. |
645 | PreprocessorLexer *FoundLexer = nullptr; |
646 | bool LexerWasInPPMode = false; |
647 | for (const IncludeStackInfo &ISI : llvm::reverse(C&: IncludeMacroStack)) { |
648 | if (ISI.ThePPLexer == nullptr) continue; // Scan for a real lexer. |
649 | |
650 | // Once we find a real lexer, mark it as raw mode (disabling macro |
651 | // expansions) and preprocessor mode (return EOD). We know that the lexer |
652 | // was *not* in raw mode before, because the macro that the comment came |
653 | // from was expanded. However, it could have already been in preprocessor |
654 | // mode (#if COMMENT) in which case we have to return it to that mode and |
655 | // return EOD. |
656 | FoundLexer = ISI.ThePPLexer; |
657 | FoundLexer->LexingRawMode = true; |
658 | LexerWasInPPMode = FoundLexer->ParsingPreprocessorDirective; |
659 | FoundLexer->ParsingPreprocessorDirective = true; |
660 | break; |
661 | } |
662 | |
663 | // Okay, we either found and switched over the lexer, or we didn't find a |
664 | // lexer. In either case, finish off the macro the comment came from, getting |
665 | // the next token. |
666 | if (!HandleEndOfTokenLexer(Result&: Tok)) Lex(Result&: Tok); |
667 | |
668 | // Discarding comments as long as we don't have EOF or EOD. This 'comments |
669 | // out' the rest of the line, including any tokens that came from other macros |
670 | // that were active, as in: |
671 | // #define submacro a COMMENT b |
672 | // submacro c |
673 | // which should lex to 'a' only: 'b' and 'c' should be removed. |
674 | while (Tok.isNot(K: tok::eod) && Tok.isNot(K: tok::eof)) |
675 | Lex(Result&: Tok); |
676 | |
677 | // If we got an eod token, then we successfully found the end of the line. |
678 | if (Tok.is(K: tok::eod)) { |
679 | assert(FoundLexer && "Can't get end of line without an active lexer" ); |
680 | // Restore the lexer back to normal mode instead of raw mode. |
681 | FoundLexer->LexingRawMode = false; |
682 | |
683 | // If the lexer was already in preprocessor mode, just return the EOD token |
684 | // to finish the preprocessor line. |
685 | if (LexerWasInPPMode) return; |
686 | |
687 | // Otherwise, switch out of PP mode and return the next lexed token. |
688 | FoundLexer->ParsingPreprocessorDirective = false; |
689 | return Lex(Result&: Tok); |
690 | } |
691 | |
692 | // If we got an EOF token, then we reached the end of the token stream but |
693 | // didn't find an explicit \n. This can only happen if there was no lexer |
694 | // active (an active lexer would return EOD at EOF if there was no \n in |
695 | // preprocessor directive mode), so just return EOF as our token. |
696 | assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode" ); |
697 | } |
698 | |
699 | void Preprocessor::EnterSubmodule(Module *M, SourceLocation ImportLoc, |
700 | bool ForPragma) { |
701 | if (!getLangOpts().ModulesLocalVisibility) { |
702 | // Just track that we entered this submodule. |
703 | BuildingSubmoduleStack.push_back( |
704 | Elt: BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState, |
705 | PendingModuleMacroNames.size())); |
706 | if (Callbacks) |
707 | Callbacks->EnteredSubmodule(M, ImportLoc, ForPragma); |
708 | return; |
709 | } |
710 | |
711 | // Resolve as much of the module definition as we can now, before we enter |
712 | // one of its headers. |
713 | // FIXME: Can we enable Complain here? |
714 | // FIXME: Can we do this when local visibility is disabled? |
715 | ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap(); |
716 | ModMap.resolveExports(Mod: M, /*Complain=*/false); |
717 | ModMap.resolveUses(Mod: M, /*Complain=*/false); |
718 | ModMap.resolveConflicts(Mod: M, /*Complain=*/false); |
719 | |
720 | // If this is the first time we've entered this module, set up its state. |
721 | auto R = Submodules.insert(x: std::make_pair(x&: M, y: SubmoduleState())); |
722 | auto &State = R.first->second; |
723 | bool FirstTime = R.second; |
724 | if (FirstTime) { |
725 | // Determine the set of starting macros for this submodule; take these |
726 | // from the "null" module (the predefines buffer). |
727 | // |
728 | // FIXME: If we have local visibility but not modules enabled, the |
729 | // NullSubmoduleState is polluted by #defines in the top-level source |
730 | // file. |
731 | auto &StartingMacros = NullSubmoduleState.Macros; |
732 | |
733 | // Restore to the starting state. |
734 | // FIXME: Do this lazily, when each macro name is first referenced. |
735 | for (auto &Macro : StartingMacros) { |
736 | // Skip uninteresting macros. |
737 | if (!Macro.second.getLatest() && |
738 | Macro.second.getOverriddenMacros().empty()) |
739 | continue; |
740 | |
741 | MacroState MS(Macro.second.getLatest()); |
742 | MS.setOverriddenMacros(PP&: *this, Overrides: Macro.second.getOverriddenMacros()); |
743 | State.Macros.insert(KV: std::make_pair(x&: Macro.first, y: std::move(MS))); |
744 | } |
745 | } |
746 | |
747 | // Track that we entered this module. |
748 | BuildingSubmoduleStack.push_back( |
749 | Elt: BuildingSubmoduleInfo(M, ImportLoc, ForPragma, CurSubmoduleState, |
750 | PendingModuleMacroNames.size())); |
751 | |
752 | if (Callbacks) |
753 | Callbacks->EnteredSubmodule(M, ImportLoc, ForPragma); |
754 | |
755 | // Switch to this submodule as the current submodule. |
756 | CurSubmoduleState = &State; |
757 | |
758 | // This module is visible to itself. |
759 | if (FirstTime) |
760 | makeModuleVisible(M, Loc: ImportLoc); |
761 | } |
762 | |
763 | bool Preprocessor::needModuleMacros() const { |
764 | // If we're not within a submodule, we never need to create ModuleMacros. |
765 | if (BuildingSubmoduleStack.empty()) |
766 | return false; |
767 | // If we are tracking module macro visibility even for textually-included |
768 | // headers, we need ModuleMacros. |
769 | if (getLangOpts().ModulesLocalVisibility) |
770 | return true; |
771 | // Otherwise, we only need module macros if we're actually compiling a module |
772 | // interface. |
773 | return getLangOpts().isCompilingModule(); |
774 | } |
775 | |
776 | Module *Preprocessor::LeaveSubmodule(bool ForPragma) { |
777 | if (BuildingSubmoduleStack.empty() || |
778 | BuildingSubmoduleStack.back().IsPragma != ForPragma) { |
779 | assert(ForPragma && "non-pragma module enter/leave mismatch" ); |
780 | return nullptr; |
781 | } |
782 | |
783 | auto &Info = BuildingSubmoduleStack.back(); |
784 | |
785 | Module *LeavingMod = Info.M; |
786 | SourceLocation ImportLoc = Info.ImportLoc; |
787 | |
788 | if (!needModuleMacros() || |
789 | (!getLangOpts().ModulesLocalVisibility && |
790 | LeavingMod->getTopLevelModuleName() != getLangOpts().CurrentModule)) { |
791 | // If we don't need module macros, or this is not a module for which we |
792 | // are tracking macro visibility, don't build any, and preserve the list |
793 | // of pending names for the surrounding submodule. |
794 | BuildingSubmoduleStack.pop_back(); |
795 | |
796 | if (Callbacks) |
797 | Callbacks->LeftSubmodule(M: LeavingMod, ImportLoc, ForPragma); |
798 | |
799 | makeModuleVisible(M: LeavingMod, Loc: ImportLoc); |
800 | return LeavingMod; |
801 | } |
802 | |
803 | // Create ModuleMacros for any macros defined in this submodule. |
804 | llvm::SmallPtrSet<const IdentifierInfo*, 8> VisitedMacros; |
805 | for (unsigned I = Info.OuterPendingModuleMacroNames; |
806 | I != PendingModuleMacroNames.size(); ++I) { |
807 | auto *II = PendingModuleMacroNames[I]; |
808 | if (!VisitedMacros.insert(Ptr: II).second) |
809 | continue; |
810 | |
811 | auto MacroIt = CurSubmoduleState->Macros.find(Val: II); |
812 | if (MacroIt == CurSubmoduleState->Macros.end()) |
813 | continue; |
814 | auto &Macro = MacroIt->second; |
815 | |
816 | // Find the starting point for the MacroDirective chain in this submodule. |
817 | MacroDirective *OldMD = nullptr; |
818 | auto *OldState = Info.OuterSubmoduleState; |
819 | if (getLangOpts().ModulesLocalVisibility) |
820 | OldState = &NullSubmoduleState; |
821 | if (OldState && OldState != CurSubmoduleState) { |
822 | // FIXME: It'd be better to start at the state from when we most recently |
823 | // entered this submodule, but it doesn't really matter. |
824 | auto &OldMacros = OldState->Macros; |
825 | auto OldMacroIt = OldMacros.find(Val: II); |
826 | if (OldMacroIt == OldMacros.end()) |
827 | OldMD = nullptr; |
828 | else |
829 | OldMD = OldMacroIt->second.getLatest(); |
830 | } |
831 | |
832 | // This module may have exported a new macro. If so, create a ModuleMacro |
833 | // representing that fact. |
834 | bool ExplicitlyPublic = false; |
835 | for (auto *MD = Macro.getLatest(); MD != OldMD; MD = MD->getPrevious()) { |
836 | assert(MD && "broken macro directive chain" ); |
837 | |
838 | if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(Val: MD)) { |
839 | // The latest visibility directive for a name in a submodule affects |
840 | // all the directives that come before it. |
841 | if (VisMD->isPublic()) |
842 | ExplicitlyPublic = true; |
843 | else if (!ExplicitlyPublic) |
844 | // Private with no following public directive: not exported. |
845 | break; |
846 | } else { |
847 | MacroInfo *Def = nullptr; |
848 | if (DefMacroDirective *DefMD = dyn_cast<DefMacroDirective>(Val: MD)) |
849 | Def = DefMD->getInfo(); |
850 | |
851 | // FIXME: Issue a warning if multiple headers for the same submodule |
852 | // define a macro, rather than silently ignoring all but the first. |
853 | bool IsNew; |
854 | // Don't bother creating a module macro if it would represent a #undef |
855 | // that doesn't override anything. |
856 | if (Def || !Macro.getOverriddenMacros().empty()) |
857 | addModuleMacro(Mod: LeavingMod, II, Macro: Def, Overrides: Macro.getOverriddenMacros(), |
858 | IsNew); |
859 | |
860 | if (!getLangOpts().ModulesLocalVisibility) { |
861 | // This macro is exposed to the rest of this compilation as a |
862 | // ModuleMacro; we don't need to track its MacroDirective any more. |
863 | Macro.setLatest(nullptr); |
864 | Macro.setOverriddenMacros(PP&: *this, Overrides: {}); |
865 | } |
866 | break; |
867 | } |
868 | } |
869 | } |
870 | PendingModuleMacroNames.resize(N: Info.OuterPendingModuleMacroNames); |
871 | |
872 | // FIXME: Before we leave this submodule, we should parse all the other |
873 | // headers within it. Otherwise, we're left with an inconsistent state |
874 | // where we've made the module visible but don't yet have its complete |
875 | // contents. |
876 | |
877 | // Put back the outer module's state, if we're tracking it. |
878 | if (getLangOpts().ModulesLocalVisibility) |
879 | CurSubmoduleState = Info.OuterSubmoduleState; |
880 | |
881 | BuildingSubmoduleStack.pop_back(); |
882 | |
883 | if (Callbacks) |
884 | Callbacks->LeftSubmodule(M: LeavingMod, ImportLoc, ForPragma); |
885 | |
886 | // A nested #include makes the included submodule visible. |
887 | makeModuleVisible(M: LeavingMod, Loc: ImportLoc); |
888 | return LeavingMod; |
889 | } |
890 | |