1 //===--- SourceManager.cpp - Track and cache source files -----------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the SourceManager interface. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/Basic/SourceManager.h" 15 #include "clang/Basic/Diagnostic.h" 16 #include "clang/Basic/FileManager.h" 17 #include "clang/Basic/SourceManagerInternals.h" 18 #include "llvm/ADT/Optional.h" 19 #include "llvm/ADT/STLExtras.h" 20 #include "llvm/ADT/StringSwitch.h" 21 #include "llvm/Support/Capacity.h" 22 #include "llvm/Support/Compiler.h" 23 #include "llvm/Support/MemoryBuffer.h" 24 #include "llvm/Support/Path.h" 25 #include "llvm/Support/raw_ostream.h" 26 #include <algorithm> 27 #include <cstring> 28 #include <string> 29 30 using namespace clang; 31 using namespace SrcMgr; 32 using llvm::MemoryBuffer; 33 34 //===----------------------------------------------------------------------===// 35 // SourceManager Helper Classes 36 //===----------------------------------------------------------------------===// 37 38 ContentCache::~ContentCache() { 39 if (shouldFreeBuffer()) 40 delete Buffer.getPointer(); 41 } 42 43 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this 44 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded. 45 unsigned ContentCache::getSizeBytesMapped() const { 46 return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0; 47 } 48 49 /// Returns the kind of memory used to back the memory buffer for 50 /// this content cache. This is used for performance analysis. 51 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const { 52 assert(Buffer.getPointer()); 53 54 // Should be unreachable, but keep for sanity. 55 if (!Buffer.getPointer()) 56 return llvm::MemoryBuffer::MemoryBuffer_Malloc; 57 58 llvm::MemoryBuffer *buf = Buffer.getPointer(); 59 return buf->getBufferKind(); 60 } 61 62 /// getSize - Returns the size of the content encapsulated by this ContentCache. 63 /// This can be the size of the source file or the size of an arbitrary 64 /// scratch buffer. If the ContentCache encapsulates a source file, that 65 /// file is not lazily brought in from disk to satisfy this query. 66 unsigned ContentCache::getSize() const { 67 return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize() 68 : (unsigned) ContentsEntry->getSize(); 69 } 70 71 void ContentCache::replaceBuffer(llvm::MemoryBuffer *B, bool DoNotFree) { 72 if (B && B == Buffer.getPointer()) { 73 assert(0 && "Replacing with the same buffer"); 74 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0); 75 return; 76 } 77 78 if (shouldFreeBuffer()) 79 delete Buffer.getPointer(); 80 Buffer.setPointer(B); 81 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0); 82 } 83 84 llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag, 85 const SourceManager &SM, 86 SourceLocation Loc, 87 bool *Invalid) const { 88 // Lazily create the Buffer for ContentCaches that wrap files. If we already 89 // computed it, just return what we have. 90 if (Buffer.getPointer() || !ContentsEntry) { 91 if (Invalid) 92 *Invalid = isBufferInvalid(); 93 94 return Buffer.getPointer(); 95 } 96 97 bool isVolatile = SM.userFilesAreVolatile() && !IsSystemFile; 98 auto BufferOrError = 99 SM.getFileManager().getBufferForFile(ContentsEntry, isVolatile); 100 101 // If we were unable to open the file, then we are in an inconsistent 102 // situation where the content cache referenced a file which no longer 103 // exists. Most likely, we were using a stat cache with an invalid entry but 104 // the file could also have been removed during processing. Since we can't 105 // really deal with this situation, just create an empty buffer. 106 // 107 // FIXME: This is definitely not ideal, but our immediate clients can't 108 // currently handle returning a null entry here. Ideally we should detect 109 // that we are in an inconsistent situation and error out as quickly as 110 // possible. 111 if (!BufferOrError) { 112 StringRef FillStr("<<<MISSING SOURCE FILE>>>\n"); 113 Buffer.setPointer(MemoryBuffer::getNewUninitMemBuffer( 114 ContentsEntry->getSize(), "<invalid>").release()); 115 char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart()); 116 for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i) 117 Ptr[i] = FillStr[i % FillStr.size()]; 118 119 if (Diag.isDiagnosticInFlight()) 120 Diag.SetDelayedDiagnostic(diag::err_cannot_open_file, 121 ContentsEntry->getName(), 122 BufferOrError.getError().message()); 123 else 124 Diag.Report(Loc, diag::err_cannot_open_file) 125 << ContentsEntry->getName() << BufferOrError.getError().message(); 126 127 Buffer.setInt(Buffer.getInt() | InvalidFlag); 128 129 if (Invalid) *Invalid = true; 130 return Buffer.getPointer(); 131 } 132 133 Buffer.setPointer(BufferOrError->release()); 134 135 // Check that the file's size is the same as in the file entry (which may 136 // have come from a stat cache). 137 if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) { 138 if (Diag.isDiagnosticInFlight()) 139 Diag.SetDelayedDiagnostic(diag::err_file_modified, 140 ContentsEntry->getName()); 141 else 142 Diag.Report(Loc, diag::err_file_modified) 143 << ContentsEntry->getName(); 144 145 Buffer.setInt(Buffer.getInt() | InvalidFlag); 146 if (Invalid) *Invalid = true; 147 return Buffer.getPointer(); 148 } 149 150 // If the buffer is valid, check to see if it has a UTF Byte Order Mark 151 // (BOM). We only support UTF-8 with and without a BOM right now. See 152 // http://en.wikipedia.org/wiki/Byte_order_mark for more information. 153 StringRef BufStr = Buffer.getPointer()->getBuffer(); 154 const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr) 155 .StartsWith("\xFE\xFF", "UTF-16 (BE)") 156 .StartsWith("\xFF\xFE", "UTF-16 (LE)") 157 .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)") 158 .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)") 159 .StartsWith("\x2B\x2F\x76", "UTF-7") 160 .StartsWith("\xF7\x64\x4C", "UTF-1") 161 .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC") 162 .StartsWith("\x0E\xFE\xFF", "SDSU") 163 .StartsWith("\xFB\xEE\x28", "BOCU-1") 164 .StartsWith("\x84\x31\x95\x33", "GB-18030") 165 .Default(nullptr); 166 167 if (InvalidBOM) { 168 Diag.Report(Loc, diag::err_unsupported_bom) 169 << InvalidBOM << ContentsEntry->getName(); 170 Buffer.setInt(Buffer.getInt() | InvalidFlag); 171 } 172 173 if (Invalid) 174 *Invalid = isBufferInvalid(); 175 176 return Buffer.getPointer(); 177 } 178 179 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) { 180 auto IterBool = 181 FilenameIDs.insert(std::make_pair(Name, FilenamesByID.size())); 182 if (IterBool.second) 183 FilenamesByID.push_back(&*IterBool.first); 184 return IterBool.first->second; 185 } 186 187 /// AddLineNote - Add a line note to the line table that indicates that there 188 /// is a \#line at the specified FID/Offset location which changes the presumed 189 /// location to LineNo/FilenameID. 190 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset, 191 unsigned LineNo, int FilenameID) { 192 std::vector<LineEntry> &Entries = LineEntries[FID]; 193 194 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 195 "Adding line entries out of order!"); 196 197 SrcMgr::CharacteristicKind Kind = SrcMgr::C_User; 198 unsigned IncludeOffset = 0; 199 200 if (!Entries.empty()) { 201 // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember 202 // that we are still in "foo.h". 203 if (FilenameID == -1) 204 FilenameID = Entries.back().FilenameID; 205 206 // If we are after a line marker that switched us to system header mode, or 207 // that set #include information, preserve it. 208 Kind = Entries.back().FileKind; 209 IncludeOffset = Entries.back().IncludeOffset; 210 } 211 212 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind, 213 IncludeOffset)); 214 } 215 216 /// AddLineNote This is the same as the previous version of AddLineNote, but is 217 /// used for GNU line markers. If EntryExit is 0, then this doesn't change the 218 /// presumed \#include stack. If it is 1, this is a file entry, if it is 2 then 219 /// this is a file exit. FileKind specifies whether this is a system header or 220 /// extern C system header. 221 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset, 222 unsigned LineNo, int FilenameID, 223 unsigned EntryExit, 224 SrcMgr::CharacteristicKind FileKind) { 225 assert(FilenameID != -1 && "Unspecified filename should use other accessor"); 226 227 std::vector<LineEntry> &Entries = LineEntries[FID]; 228 229 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 230 "Adding line entries out of order!"); 231 232 unsigned IncludeOffset = 0; 233 if (EntryExit == 0) { // No #include stack change. 234 IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset; 235 } else if (EntryExit == 1) { 236 IncludeOffset = Offset-1; 237 } else if (EntryExit == 2) { 238 assert(!Entries.empty() && Entries.back().IncludeOffset && 239 "PPDirectives should have caught case when popping empty include stack"); 240 241 // Get the include loc of the last entries' include loc as our include loc. 242 IncludeOffset = 0; 243 if (const LineEntry *PrevEntry = 244 FindNearestLineEntry(FID, Entries.back().IncludeOffset)) 245 IncludeOffset = PrevEntry->IncludeOffset; 246 } 247 248 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind, 249 IncludeOffset)); 250 } 251 252 253 /// FindNearestLineEntry - Find the line entry nearest to FID that is before 254 /// it. If there is no line entry before Offset in FID, return null. 255 const LineEntry *LineTableInfo::FindNearestLineEntry(FileID FID, 256 unsigned Offset) { 257 const std::vector<LineEntry> &Entries = LineEntries[FID]; 258 assert(!Entries.empty() && "No #line entries for this FID after all!"); 259 260 // It is very common for the query to be after the last #line, check this 261 // first. 262 if (Entries.back().FileOffset <= Offset) 263 return &Entries.back(); 264 265 // Do a binary search to find the maximal element that is still before Offset. 266 std::vector<LineEntry>::const_iterator I = 267 std::upper_bound(Entries.begin(), Entries.end(), Offset); 268 if (I == Entries.begin()) return nullptr; 269 return &*--I; 270 } 271 272 /// \brief Add a new line entry that has already been encoded into 273 /// the internal representation of the line table. 274 void LineTableInfo::AddEntry(FileID FID, 275 const std::vector<LineEntry> &Entries) { 276 LineEntries[FID] = Entries; 277 } 278 279 /// getLineTableFilenameID - Return the uniqued ID for the specified filename. 280 /// 281 unsigned SourceManager::getLineTableFilenameID(StringRef Name) { 282 return getLineTable().getLineTableFilenameID(Name); 283 } 284 285 286 /// AddLineNote - Add a line note to the line table for the FileID and offset 287 /// specified by Loc. If FilenameID is -1, it is considered to be 288 /// unspecified. 289 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 290 int FilenameID) { 291 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 292 293 bool Invalid = false; 294 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 295 if (!Entry.isFile() || Invalid) 296 return; 297 298 const SrcMgr::FileInfo &FileInfo = Entry.getFile(); 299 300 // Remember that this file has #line directives now if it doesn't already. 301 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 302 303 getLineTable().AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID); 304 } 305 306 /// AddLineNote - Add a GNU line marker to the line table. 307 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 308 int FilenameID, bool IsFileEntry, 309 bool IsFileExit, bool IsSystemHeader, 310 bool IsExternCHeader) { 311 // If there is no filename and no flags, this is treated just like a #line, 312 // which does not change the flags of the previous line marker. 313 if (FilenameID == -1) { 314 assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader && 315 "Can't set flags without setting the filename!"); 316 return AddLineNote(Loc, LineNo, FilenameID); 317 } 318 319 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 320 321 bool Invalid = false; 322 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 323 if (!Entry.isFile() || Invalid) 324 return; 325 326 const SrcMgr::FileInfo &FileInfo = Entry.getFile(); 327 328 // Remember that this file has #line directives now if it doesn't already. 329 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 330 331 (void) getLineTable(); 332 333 SrcMgr::CharacteristicKind FileKind; 334 if (IsExternCHeader) 335 FileKind = SrcMgr::C_ExternCSystem; 336 else if (IsSystemHeader) 337 FileKind = SrcMgr::C_System; 338 else 339 FileKind = SrcMgr::C_User; 340 341 unsigned EntryExit = 0; 342 if (IsFileEntry) 343 EntryExit = 1; 344 else if (IsFileExit) 345 EntryExit = 2; 346 347 LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID, 348 EntryExit, FileKind); 349 } 350 351 LineTableInfo &SourceManager::getLineTable() { 352 if (!LineTable) 353 LineTable = new LineTableInfo(); 354 return *LineTable; 355 } 356 357 //===----------------------------------------------------------------------===// 358 // Private 'Create' methods. 359 //===----------------------------------------------------------------------===// 360 361 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr, 362 bool UserFilesAreVolatile) 363 : Diag(Diag), FileMgr(FileMgr), OverridenFilesKeepOriginalName(true), 364 UserFilesAreVolatile(UserFilesAreVolatile), FilesAreTransient(false), 365 ExternalSLocEntries(nullptr), LineTable(nullptr), NumLinearScans(0), 366 NumBinaryProbes(0) { 367 clearIDTables(); 368 Diag.setSourceManager(this); 369 } 370 371 SourceManager::~SourceManager() { 372 delete LineTable; 373 374 // Delete FileEntry objects corresponding to content caches. Since the actual 375 // content cache objects are bump pointer allocated, we just have to run the 376 // dtors, but we call the deallocate method for completeness. 377 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) { 378 if (MemBufferInfos[i]) { 379 MemBufferInfos[i]->~ContentCache(); 380 ContentCacheAlloc.Deallocate(MemBufferInfos[i]); 381 } 382 } 383 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator 384 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) { 385 if (I->second) { 386 I->second->~ContentCache(); 387 ContentCacheAlloc.Deallocate(I->second); 388 } 389 } 390 391 llvm::DeleteContainerSeconds(MacroArgsCacheMap); 392 } 393 394 void SourceManager::clearIDTables() { 395 MainFileID = FileID(); 396 LocalSLocEntryTable.clear(); 397 LoadedSLocEntryTable.clear(); 398 SLocEntryLoaded.clear(); 399 LastLineNoFileIDQuery = FileID(); 400 LastLineNoContentCache = nullptr; 401 LastFileIDLookup = FileID(); 402 403 if (LineTable) 404 LineTable->clear(); 405 406 // Use up FileID #0 as an invalid expansion. 407 NextLocalOffset = 0; 408 CurrentLoadedOffset = MaxLoadedOffset; 409 createExpansionLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1); 410 } 411 412 /// getOrCreateContentCache - Create or return a cached ContentCache for the 413 /// specified file. 414 const ContentCache * 415 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt, 416 bool isSystemFile) { 417 assert(FileEnt && "Didn't specify a file entry to use?"); 418 419 // Do we already have information about this file? 420 ContentCache *&Entry = FileInfos[FileEnt]; 421 if (Entry) return Entry; 422 423 // Nope, create a new Cache entry. 424 Entry = ContentCacheAlloc.Allocate<ContentCache>(); 425 426 if (OverriddenFilesInfo) { 427 // If the file contents are overridden with contents from another file, 428 // pass that file to ContentCache. 429 llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator 430 overI = OverriddenFilesInfo->OverriddenFiles.find(FileEnt); 431 if (overI == OverriddenFilesInfo->OverriddenFiles.end()) 432 new (Entry) ContentCache(FileEnt); 433 else 434 new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt 435 : overI->second, 436 overI->second); 437 } else { 438 new (Entry) ContentCache(FileEnt); 439 } 440 441 Entry->IsSystemFile = isSystemFile; 442 Entry->IsTransient = FilesAreTransient; 443 444 return Entry; 445 } 446 447 448 /// createMemBufferContentCache - Create a new ContentCache for the specified 449 /// memory buffer. This does no caching. 450 const ContentCache *SourceManager::createMemBufferContentCache( 451 std::unique_ptr<llvm::MemoryBuffer> Buffer) { 452 // Add a new ContentCache to the MemBufferInfos list and return it. 453 ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(); 454 new (Entry) ContentCache(); 455 MemBufferInfos.push_back(Entry); 456 Entry->setBuffer(std::move(Buffer)); 457 return Entry; 458 } 459 460 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index, 461 bool *Invalid) const { 462 assert(!SLocEntryLoaded[Index]); 463 if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) { 464 if (Invalid) 465 *Invalid = true; 466 // If the file of the SLocEntry changed we could still have loaded it. 467 if (!SLocEntryLoaded[Index]) { 468 // Try to recover; create a SLocEntry so the rest of clang can handle it. 469 LoadedSLocEntryTable[Index] = SLocEntry::get(0, 470 FileInfo::get(SourceLocation(), 471 getFakeContentCacheForRecovery(), 472 SrcMgr::C_User)); 473 } 474 } 475 476 return LoadedSLocEntryTable[Index]; 477 } 478 479 std::pair<int, unsigned> 480 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries, 481 unsigned TotalSize) { 482 assert(ExternalSLocEntries && "Don't have an external sloc source"); 483 // Make sure we're not about to run out of source locations. 484 if (CurrentLoadedOffset - TotalSize < NextLocalOffset) 485 return std::make_pair(0, 0); 486 LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries); 487 SLocEntryLoaded.resize(LoadedSLocEntryTable.size()); 488 CurrentLoadedOffset -= TotalSize; 489 int ID = LoadedSLocEntryTable.size(); 490 return std::make_pair(-ID - 1, CurrentLoadedOffset); 491 } 492 493 /// \brief As part of recovering from missing or changed content, produce a 494 /// fake, non-empty buffer. 495 llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const { 496 if (!FakeBufferForRecovery) 497 FakeBufferForRecovery = 498 llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>"); 499 500 return FakeBufferForRecovery.get(); 501 } 502 503 /// \brief As part of recovering from missing or changed content, produce a 504 /// fake content cache. 505 const SrcMgr::ContentCache * 506 SourceManager::getFakeContentCacheForRecovery() const { 507 if (!FakeContentCacheForRecovery) { 508 FakeContentCacheForRecovery = llvm::make_unique<SrcMgr::ContentCache>(); 509 FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(), 510 /*DoNotFree=*/true); 511 } 512 return FakeContentCacheForRecovery.get(); 513 } 514 515 /// \brief Returns the previous in-order FileID or an invalid FileID if there 516 /// is no previous one. 517 FileID SourceManager::getPreviousFileID(FileID FID) const { 518 if (FID.isInvalid()) 519 return FileID(); 520 521 int ID = FID.ID; 522 if (ID == -1) 523 return FileID(); 524 525 if (ID > 0) { 526 if (ID-1 == 0) 527 return FileID(); 528 } else if (unsigned(-(ID-1) - 2) >= LoadedSLocEntryTable.size()) { 529 return FileID(); 530 } 531 532 return FileID::get(ID-1); 533 } 534 535 /// \brief Returns the next in-order FileID or an invalid FileID if there is 536 /// no next one. 537 FileID SourceManager::getNextFileID(FileID FID) const { 538 if (FID.isInvalid()) 539 return FileID(); 540 541 int ID = FID.ID; 542 if (ID > 0) { 543 if (unsigned(ID+1) >= local_sloc_entry_size()) 544 return FileID(); 545 } else if (ID+1 >= -1) { 546 return FileID(); 547 } 548 549 return FileID::get(ID+1); 550 } 551 552 //===----------------------------------------------------------------------===// 553 // Methods to create new FileID's and macro expansions. 554 //===----------------------------------------------------------------------===// 555 556 /// createFileID - Create a new FileID for the specified ContentCache and 557 /// include position. This works regardless of whether the ContentCache 558 /// corresponds to a file or some other input source. 559 FileID SourceManager::createFileID(const ContentCache *File, 560 SourceLocation IncludePos, 561 SrcMgr::CharacteristicKind FileCharacter, 562 int LoadedID, unsigned LoadedOffset) { 563 if (LoadedID < 0) { 564 assert(LoadedID != -1 && "Loading sentinel FileID"); 565 unsigned Index = unsigned(-LoadedID) - 2; 566 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range"); 567 assert(!SLocEntryLoaded[Index] && "FileID already loaded"); 568 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, 569 FileInfo::get(IncludePos, File, FileCharacter)); 570 SLocEntryLoaded[Index] = true; 571 return FileID::get(LoadedID); 572 } 573 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, 574 FileInfo::get(IncludePos, File, 575 FileCharacter))); 576 unsigned FileSize = File->getSize(); 577 assert(NextLocalOffset + FileSize + 1 > NextLocalOffset && 578 NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset && 579 "Ran out of source locations!"); 580 // We do a +1 here because we want a SourceLocation that means "the end of the 581 // file", e.g. for the "no newline at the end of the file" diagnostic. 582 NextLocalOffset += FileSize + 1; 583 584 // Set LastFileIDLookup to the newly created file. The next getFileID call is 585 // almost guaranteed to be from that file. 586 FileID FID = FileID::get(LocalSLocEntryTable.size()-1); 587 return LastFileIDLookup = FID; 588 } 589 590 SourceLocation 591 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc, 592 SourceLocation ExpansionLoc, 593 unsigned TokLength) { 594 ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc, 595 ExpansionLoc); 596 return createExpansionLocImpl(Info, TokLength); 597 } 598 599 SourceLocation 600 SourceManager::createExpansionLoc(SourceLocation SpellingLoc, 601 SourceLocation ExpansionLocStart, 602 SourceLocation ExpansionLocEnd, 603 unsigned TokLength, 604 int LoadedID, 605 unsigned LoadedOffset) { 606 ExpansionInfo Info = ExpansionInfo::create(SpellingLoc, ExpansionLocStart, 607 ExpansionLocEnd); 608 return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset); 609 } 610 611 SourceLocation 612 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info, 613 unsigned TokLength, 614 int LoadedID, 615 unsigned LoadedOffset) { 616 if (LoadedID < 0) { 617 assert(LoadedID != -1 && "Loading sentinel FileID"); 618 unsigned Index = unsigned(-LoadedID) - 2; 619 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range"); 620 assert(!SLocEntryLoaded[Index] && "FileID already loaded"); 621 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info); 622 SLocEntryLoaded[Index] = true; 623 return SourceLocation::getMacroLoc(LoadedOffset); 624 } 625 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info)); 626 assert(NextLocalOffset + TokLength + 1 > NextLocalOffset && 627 NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset && 628 "Ran out of source locations!"); 629 // See createFileID for that +1. 630 NextLocalOffset += TokLength + 1; 631 return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1)); 632 } 633 634 llvm::MemoryBuffer *SourceManager::getMemoryBufferForFile(const FileEntry *File, 635 bool *Invalid) { 636 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 637 assert(IR && "getOrCreateContentCache() cannot return NULL"); 638 return IR->getBuffer(Diag, *this, SourceLocation(), Invalid); 639 } 640 641 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 642 llvm::MemoryBuffer *Buffer, 643 bool DoNotFree) { 644 const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile); 645 assert(IR && "getOrCreateContentCache() cannot return NULL"); 646 647 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree); 648 const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true; 649 650 getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile); 651 } 652 653 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 654 const FileEntry *NewFile) { 655 assert(SourceFile->getSize() == NewFile->getSize() && 656 "Different sizes, use the FileManager to create a virtual file with " 657 "the correct size"); 658 assert(FileInfos.count(SourceFile) == 0 && 659 "This function should be called at the initialization stage, before " 660 "any parsing occurs."); 661 getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile; 662 } 663 664 void SourceManager::disableFileContentsOverride(const FileEntry *File) { 665 if (!isFileOverridden(File)) 666 return; 667 668 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 669 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(nullptr); 670 const_cast<SrcMgr::ContentCache *>(IR)->ContentsEntry = IR->OrigEntry; 671 672 assert(OverriddenFilesInfo); 673 OverriddenFilesInfo->OverriddenFiles.erase(File); 674 OverriddenFilesInfo->OverriddenFilesWithBuffer.erase(File); 675 } 676 677 void SourceManager::setFileIsTransient(const FileEntry *File) { 678 const SrcMgr::ContentCache *CC = getOrCreateContentCache(File); 679 const_cast<SrcMgr::ContentCache *>(CC)->IsTransient = true; 680 } 681 682 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const { 683 bool MyInvalid = false; 684 const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid); 685 if (!SLoc.isFile() || MyInvalid) { 686 if (Invalid) 687 *Invalid = true; 688 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 689 } 690 691 llvm::MemoryBuffer *Buf = SLoc.getFile().getContentCache()->getBuffer( 692 Diag, *this, SourceLocation(), &MyInvalid); 693 if (Invalid) 694 *Invalid = MyInvalid; 695 696 if (MyInvalid) 697 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 698 699 return Buf->getBuffer(); 700 } 701 702 //===----------------------------------------------------------------------===// 703 // SourceLocation manipulation methods. 704 //===----------------------------------------------------------------------===// 705 706 /// \brief Return the FileID for a SourceLocation. 707 /// 708 /// This is the cache-miss path of getFileID. Not as hot as that function, but 709 /// still very important. It is responsible for finding the entry in the 710 /// SLocEntry tables that contains the specified location. 711 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const { 712 if (!SLocOffset) 713 return FileID::get(0); 714 715 // Now it is time to search for the correct file. See where the SLocOffset 716 // sits in the global view and consult local or loaded buffers for it. 717 if (SLocOffset < NextLocalOffset) 718 return getFileIDLocal(SLocOffset); 719 return getFileIDLoaded(SLocOffset); 720 } 721 722 /// \brief Return the FileID for a SourceLocation with a low offset. 723 /// 724 /// This function knows that the SourceLocation is in a local buffer, not a 725 /// loaded one. 726 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const { 727 assert(SLocOffset < NextLocalOffset && "Bad function choice"); 728 729 // After the first and second level caches, I see two common sorts of 730 // behavior: 1) a lot of searched FileID's are "near" the cached file 731 // location or are "near" the cached expansion location. 2) others are just 732 // completely random and may be a very long way away. 733 // 734 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly 735 // then we fall back to a less cache efficient, but more scalable, binary 736 // search to find the location. 737 738 // See if this is near the file point - worst case we start scanning from the 739 // most newly created FileID. 740 const SrcMgr::SLocEntry *I; 741 742 if (LastFileIDLookup.ID < 0 || 743 LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) { 744 // Neither loc prunes our search. 745 I = LocalSLocEntryTable.end(); 746 } else { 747 // Perhaps it is near the file point. 748 I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID; 749 } 750 751 // Find the FileID that contains this. "I" is an iterator that points to a 752 // FileID whose offset is known to be larger than SLocOffset. 753 unsigned NumProbes = 0; 754 while (1) { 755 --I; 756 if (I->getOffset() <= SLocOffset) { 757 FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin())); 758 759 // If this isn't an expansion, remember it. We have good locality across 760 // FileID lookups. 761 if (!I->isExpansion()) 762 LastFileIDLookup = Res; 763 NumLinearScans += NumProbes+1; 764 return Res; 765 } 766 if (++NumProbes == 8) 767 break; 768 } 769 770 // Convert "I" back into an index. We know that it is an entry whose index is 771 // larger than the offset we are looking for. 772 unsigned GreaterIndex = I - LocalSLocEntryTable.begin(); 773 // LessIndex - This is the lower bound of the range that we're searching. 774 // We know that the offset corresponding to the FileID is is less than 775 // SLocOffset. 776 unsigned LessIndex = 0; 777 NumProbes = 0; 778 while (1) { 779 bool Invalid = false; 780 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex; 781 unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset(); 782 if (Invalid) 783 return FileID::get(0); 784 785 ++NumProbes; 786 787 // If the offset of the midpoint is too large, chop the high side of the 788 // range to the midpoint. 789 if (MidOffset > SLocOffset) { 790 GreaterIndex = MiddleIndex; 791 continue; 792 } 793 794 // If the middle index contains the value, succeed and return. 795 // FIXME: This could be made faster by using a function that's aware of 796 // being in the local area. 797 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) { 798 FileID Res = FileID::get(MiddleIndex); 799 800 // If this isn't a macro expansion, remember it. We have good locality 801 // across FileID lookups. 802 if (!LocalSLocEntryTable[MiddleIndex].isExpansion()) 803 LastFileIDLookup = Res; 804 NumBinaryProbes += NumProbes; 805 return Res; 806 } 807 808 // Otherwise, move the low-side up to the middle index. 809 LessIndex = MiddleIndex; 810 } 811 } 812 813 /// \brief Return the FileID for a SourceLocation with a high offset. 814 /// 815 /// This function knows that the SourceLocation is in a loaded buffer, not a 816 /// local one. 817 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const { 818 // Sanity checking, otherwise a bug may lead to hanging in release build. 819 if (SLocOffset < CurrentLoadedOffset) { 820 assert(0 && "Invalid SLocOffset or bad function choice"); 821 return FileID(); 822 } 823 824 // Essentially the same as the local case, but the loaded array is sorted 825 // in the other direction. 826 827 // First do a linear scan from the last lookup position, if possible. 828 unsigned I; 829 int LastID = LastFileIDLookup.ID; 830 if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset) 831 I = 0; 832 else 833 I = (-LastID - 2) + 1; 834 835 unsigned NumProbes; 836 for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) { 837 // Make sure the entry is loaded! 838 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I); 839 if (E.getOffset() <= SLocOffset) { 840 FileID Res = FileID::get(-int(I) - 2); 841 842 if (!E.isExpansion()) 843 LastFileIDLookup = Res; 844 NumLinearScans += NumProbes + 1; 845 return Res; 846 } 847 } 848 849 // Linear scan failed. Do the binary search. Note the reverse sorting of the 850 // table: GreaterIndex is the one where the offset is greater, which is 851 // actually a lower index! 852 unsigned GreaterIndex = I; 853 unsigned LessIndex = LoadedSLocEntryTable.size(); 854 NumProbes = 0; 855 while (1) { 856 ++NumProbes; 857 unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex; 858 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex); 859 if (E.getOffset() == 0) 860 return FileID(); // invalid entry. 861 862 ++NumProbes; 863 864 if (E.getOffset() > SLocOffset) { 865 // Sanity checking, otherwise a bug may lead to hanging in release build. 866 if (GreaterIndex == MiddleIndex) { 867 assert(0 && "binary search missed the entry"); 868 return FileID(); 869 } 870 GreaterIndex = MiddleIndex; 871 continue; 872 } 873 874 if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) { 875 FileID Res = FileID::get(-int(MiddleIndex) - 2); 876 if (!E.isExpansion()) 877 LastFileIDLookup = Res; 878 NumBinaryProbes += NumProbes; 879 return Res; 880 } 881 882 // Sanity checking, otherwise a bug may lead to hanging in release build. 883 if (LessIndex == MiddleIndex) { 884 assert(0 && "binary search missed the entry"); 885 return FileID(); 886 } 887 LessIndex = MiddleIndex; 888 } 889 } 890 891 SourceLocation SourceManager:: 892 getExpansionLocSlowCase(SourceLocation Loc) const { 893 do { 894 // Note: If Loc indicates an offset into a token that came from a macro 895 // expansion (e.g. the 5th character of the token) we do not want to add 896 // this offset when going to the expansion location. The expansion 897 // location is the macro invocation, which the offset has nothing to do 898 // with. This is unlike when we get the spelling loc, because the offset 899 // directly correspond to the token whose spelling we're inspecting. 900 Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart(); 901 } while (!Loc.isFileID()); 902 903 return Loc; 904 } 905 906 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const { 907 do { 908 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 909 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 910 Loc = Loc.getLocWithOffset(LocInfo.second); 911 } while (!Loc.isFileID()); 912 return Loc; 913 } 914 915 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const { 916 do { 917 if (isMacroArgExpansion(Loc)) 918 Loc = getImmediateSpellingLoc(Loc); 919 else 920 Loc = getImmediateExpansionRange(Loc).first; 921 } while (!Loc.isFileID()); 922 return Loc; 923 } 924 925 926 std::pair<FileID, unsigned> 927 SourceManager::getDecomposedExpansionLocSlowCase( 928 const SrcMgr::SLocEntry *E) const { 929 // If this is an expansion record, walk through all the expansion points. 930 FileID FID; 931 SourceLocation Loc; 932 unsigned Offset; 933 do { 934 Loc = E->getExpansion().getExpansionLocStart(); 935 936 FID = getFileID(Loc); 937 E = &getSLocEntry(FID); 938 Offset = Loc.getOffset()-E->getOffset(); 939 } while (!Loc.isFileID()); 940 941 return std::make_pair(FID, Offset); 942 } 943 944 std::pair<FileID, unsigned> 945 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E, 946 unsigned Offset) const { 947 // If this is an expansion record, walk through all the expansion points. 948 FileID FID; 949 SourceLocation Loc; 950 do { 951 Loc = E->getExpansion().getSpellingLoc(); 952 Loc = Loc.getLocWithOffset(Offset); 953 954 FID = getFileID(Loc); 955 E = &getSLocEntry(FID); 956 Offset = Loc.getOffset()-E->getOffset(); 957 } while (!Loc.isFileID()); 958 959 return std::make_pair(FID, Offset); 960 } 961 962 /// getImmediateSpellingLoc - Given a SourceLocation object, return the 963 /// spelling location referenced by the ID. This is the first level down 964 /// towards the place where the characters that make up the lexed token can be 965 /// found. This should not generally be used by clients. 966 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{ 967 if (Loc.isFileID()) return Loc; 968 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 969 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 970 return Loc.getLocWithOffset(LocInfo.second); 971 } 972 973 974 /// getImmediateExpansionRange - Loc is required to be an expansion location. 975 /// Return the start/end of the expansion information. 976 std::pair<SourceLocation,SourceLocation> 977 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const { 978 assert(Loc.isMacroID() && "Not a macro expansion loc!"); 979 const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion(); 980 return Expansion.getExpansionLocRange(); 981 } 982 983 /// getExpansionRange - Given a SourceLocation object, return the range of 984 /// tokens covered by the expansion in the ultimate file. 985 std::pair<SourceLocation,SourceLocation> 986 SourceManager::getExpansionRange(SourceLocation Loc) const { 987 if (Loc.isFileID()) return std::make_pair(Loc, Loc); 988 989 std::pair<SourceLocation,SourceLocation> Res = 990 getImmediateExpansionRange(Loc); 991 992 // Fully resolve the start and end locations to their ultimate expansion 993 // points. 994 while (!Res.first.isFileID()) 995 Res.first = getImmediateExpansionRange(Res.first).first; 996 while (!Res.second.isFileID()) 997 Res.second = getImmediateExpansionRange(Res.second).second; 998 return Res; 999 } 1000 1001 bool SourceManager::isMacroArgExpansion(SourceLocation Loc, 1002 SourceLocation *StartLoc) const { 1003 if (!Loc.isMacroID()) return false; 1004 1005 FileID FID = getFileID(Loc); 1006 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1007 if (!Expansion.isMacroArgExpansion()) return false; 1008 1009 if (StartLoc) 1010 *StartLoc = Expansion.getExpansionLocStart(); 1011 return true; 1012 } 1013 1014 bool SourceManager::isMacroBodyExpansion(SourceLocation Loc) const { 1015 if (!Loc.isMacroID()) return false; 1016 1017 FileID FID = getFileID(Loc); 1018 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1019 return Expansion.isMacroBodyExpansion(); 1020 } 1021 1022 bool SourceManager::isAtStartOfImmediateMacroExpansion(SourceLocation Loc, 1023 SourceLocation *MacroBegin) const { 1024 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1025 1026 std::pair<FileID, unsigned> DecompLoc = getDecomposedLoc(Loc); 1027 if (DecompLoc.second > 0) 1028 return false; // Does not point at the start of expansion range. 1029 1030 bool Invalid = false; 1031 const SrcMgr::ExpansionInfo &ExpInfo = 1032 getSLocEntry(DecompLoc.first, &Invalid).getExpansion(); 1033 if (Invalid) 1034 return false; 1035 SourceLocation ExpLoc = ExpInfo.getExpansionLocStart(); 1036 1037 if (ExpInfo.isMacroArgExpansion()) { 1038 // For macro argument expansions, check if the previous FileID is part of 1039 // the same argument expansion, in which case this Loc is not at the 1040 // beginning of the expansion. 1041 FileID PrevFID = getPreviousFileID(DecompLoc.first); 1042 if (!PrevFID.isInvalid()) { 1043 const SrcMgr::SLocEntry &PrevEntry = getSLocEntry(PrevFID, &Invalid); 1044 if (Invalid) 1045 return false; 1046 if (PrevEntry.isExpansion() && 1047 PrevEntry.getExpansion().getExpansionLocStart() == ExpLoc) 1048 return false; 1049 } 1050 } 1051 1052 if (MacroBegin) 1053 *MacroBegin = ExpLoc; 1054 return true; 1055 } 1056 1057 bool SourceManager::isAtEndOfImmediateMacroExpansion(SourceLocation Loc, 1058 SourceLocation *MacroEnd) const { 1059 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1060 1061 FileID FID = getFileID(Loc); 1062 SourceLocation NextLoc = Loc.getLocWithOffset(1); 1063 if (isInFileID(NextLoc, FID)) 1064 return false; // Does not point at the end of expansion range. 1065 1066 bool Invalid = false; 1067 const SrcMgr::ExpansionInfo &ExpInfo = 1068 getSLocEntry(FID, &Invalid).getExpansion(); 1069 if (Invalid) 1070 return false; 1071 1072 if (ExpInfo.isMacroArgExpansion()) { 1073 // For macro argument expansions, check if the next FileID is part of the 1074 // same argument expansion, in which case this Loc is not at the end of the 1075 // expansion. 1076 FileID NextFID = getNextFileID(FID); 1077 if (!NextFID.isInvalid()) { 1078 const SrcMgr::SLocEntry &NextEntry = getSLocEntry(NextFID, &Invalid); 1079 if (Invalid) 1080 return false; 1081 if (NextEntry.isExpansion() && 1082 NextEntry.getExpansion().getExpansionLocStart() == 1083 ExpInfo.getExpansionLocStart()) 1084 return false; 1085 } 1086 } 1087 1088 if (MacroEnd) 1089 *MacroEnd = ExpInfo.getExpansionLocEnd(); 1090 return true; 1091 } 1092 1093 1094 //===----------------------------------------------------------------------===// 1095 // Queries about the code at a SourceLocation. 1096 //===----------------------------------------------------------------------===// 1097 1098 /// getCharacterData - Return a pointer to the start of the specified location 1099 /// in the appropriate MemoryBuffer. 1100 const char *SourceManager::getCharacterData(SourceLocation SL, 1101 bool *Invalid) const { 1102 // Note that this is a hot function in the getSpelling() path, which is 1103 // heavily used by -E mode. 1104 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL); 1105 1106 // Note that calling 'getBuffer()' may lazily page in a source file. 1107 bool CharDataInvalid = false; 1108 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid); 1109 if (CharDataInvalid || !Entry.isFile()) { 1110 if (Invalid) 1111 *Invalid = true; 1112 1113 return "<<<<INVALID BUFFER>>>>"; 1114 } 1115 llvm::MemoryBuffer *Buffer = Entry.getFile().getContentCache()->getBuffer( 1116 Diag, *this, SourceLocation(), &CharDataInvalid); 1117 if (Invalid) 1118 *Invalid = CharDataInvalid; 1119 return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second); 1120 } 1121 1122 1123 /// getColumnNumber - Return the column # for the specified file position. 1124 /// this is significantly cheaper to compute than the line number. 1125 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos, 1126 bool *Invalid) const { 1127 bool MyInvalid = false; 1128 llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid); 1129 if (Invalid) 1130 *Invalid = MyInvalid; 1131 1132 if (MyInvalid) 1133 return 1; 1134 1135 // It is okay to request a position just past the end of the buffer. 1136 if (FilePos > MemBuf->getBufferSize()) { 1137 if (Invalid) 1138 *Invalid = true; 1139 return 1; 1140 } 1141 1142 // See if we just calculated the line number for this FilePos and can use 1143 // that to lookup the start of the line instead of searching for it. 1144 if (LastLineNoFileIDQuery == FID && 1145 LastLineNoContentCache->SourceLineCache != nullptr && 1146 LastLineNoResult < LastLineNoContentCache->NumLines) { 1147 unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache; 1148 unsigned LineStart = SourceLineCache[LastLineNoResult - 1]; 1149 unsigned LineEnd = SourceLineCache[LastLineNoResult]; 1150 if (FilePos >= LineStart && FilePos < LineEnd) 1151 return FilePos - LineStart + 1; 1152 } 1153 1154 const char *Buf = MemBuf->getBufferStart(); 1155 unsigned LineStart = FilePos; 1156 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r') 1157 --LineStart; 1158 return FilePos-LineStart+1; 1159 } 1160 1161 // isInvalid - Return the result of calling loc.isInvalid(), and 1162 // if Invalid is not null, set its value to same. 1163 template<typename LocType> 1164 static bool isInvalid(LocType Loc, bool *Invalid) { 1165 bool MyInvalid = Loc.isInvalid(); 1166 if (Invalid) 1167 *Invalid = MyInvalid; 1168 return MyInvalid; 1169 } 1170 1171 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc, 1172 bool *Invalid) const { 1173 if (isInvalid(Loc, Invalid)) return 0; 1174 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1175 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1176 } 1177 1178 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc, 1179 bool *Invalid) const { 1180 if (isInvalid(Loc, Invalid)) return 0; 1181 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1182 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1183 } 1184 1185 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc, 1186 bool *Invalid) const { 1187 PresumedLoc PLoc = getPresumedLoc(Loc); 1188 if (isInvalid(PLoc, Invalid)) return 0; 1189 return PLoc.getColumn(); 1190 } 1191 1192 #ifdef __SSE2__ 1193 #include <emmintrin.h> 1194 #endif 1195 1196 static LLVM_ATTRIBUTE_NOINLINE void 1197 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1198 llvm::BumpPtrAllocator &Alloc, 1199 const SourceManager &SM, bool &Invalid); 1200 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1201 llvm::BumpPtrAllocator &Alloc, 1202 const SourceManager &SM, bool &Invalid) { 1203 // Note that calling 'getBuffer()' may lazily page in the file. 1204 MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(), &Invalid); 1205 if (Invalid) 1206 return; 1207 1208 // Find the file offsets of all of the *physical* source lines. This does 1209 // not look at trigraphs, escaped newlines, or anything else tricky. 1210 SmallVector<unsigned, 256> LineOffsets; 1211 1212 // Line #1 starts at char 0. 1213 LineOffsets.push_back(0); 1214 1215 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart(); 1216 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd(); 1217 unsigned Offs = 0; 1218 while (1) { 1219 // Skip over the contents of the line. 1220 const unsigned char *NextBuf = (const unsigned char *)Buf; 1221 1222 #ifdef __SSE2__ 1223 // Try to skip to the next newline using SSE instructions. This is very 1224 // performance sensitive for programs with lots of diagnostics and in -E 1225 // mode. 1226 __m128i CRs = _mm_set1_epi8('\r'); 1227 __m128i LFs = _mm_set1_epi8('\n'); 1228 1229 // First fix up the alignment to 16 bytes. 1230 while (((uintptr_t)NextBuf & 0xF) != 0) { 1231 if (*NextBuf == '\n' || *NextBuf == '\r' || *NextBuf == '\0') 1232 goto FoundSpecialChar; 1233 ++NextBuf; 1234 } 1235 1236 // Scan 16 byte chunks for '\r' and '\n'. Ignore '\0'. 1237 while (NextBuf+16 <= End) { 1238 const __m128i Chunk = *(const __m128i*)NextBuf; 1239 __m128i Cmp = _mm_or_si128(_mm_cmpeq_epi8(Chunk, CRs), 1240 _mm_cmpeq_epi8(Chunk, LFs)); 1241 unsigned Mask = _mm_movemask_epi8(Cmp); 1242 1243 // If we found a newline, adjust the pointer and jump to the handling code. 1244 if (Mask != 0) { 1245 NextBuf += llvm::countTrailingZeros(Mask); 1246 goto FoundSpecialChar; 1247 } 1248 NextBuf += 16; 1249 } 1250 #endif 1251 1252 while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0') 1253 ++NextBuf; 1254 1255 #ifdef __SSE2__ 1256 FoundSpecialChar: 1257 #endif 1258 Offs += NextBuf-Buf; 1259 Buf = NextBuf; 1260 1261 if (Buf[0] == '\n' || Buf[0] == '\r') { 1262 // If this is \n\r or \r\n, skip both characters. 1263 if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1]) { 1264 ++Offs; 1265 ++Buf; 1266 } 1267 ++Offs; 1268 ++Buf; 1269 LineOffsets.push_back(Offs); 1270 } else { 1271 // Otherwise, this is a null. If end of file, exit. 1272 if (Buf == End) break; 1273 // Otherwise, skip the null. 1274 ++Offs; 1275 ++Buf; 1276 } 1277 } 1278 1279 // Copy the offsets into the FileInfo structure. 1280 FI->NumLines = LineOffsets.size(); 1281 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size()); 1282 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache); 1283 } 1284 1285 /// getLineNumber - Given a SourceLocation, return the spelling line number 1286 /// for the position indicated. This requires building and caching a table of 1287 /// line offsets for the MemoryBuffer, so this is not cheap: use only when 1288 /// about to emit a diagnostic. 1289 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos, 1290 bool *Invalid) const { 1291 if (FID.isInvalid()) { 1292 if (Invalid) 1293 *Invalid = true; 1294 return 1; 1295 } 1296 1297 ContentCache *Content; 1298 if (LastLineNoFileIDQuery == FID) 1299 Content = LastLineNoContentCache; 1300 else { 1301 bool MyInvalid = false; 1302 const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid); 1303 if (MyInvalid || !Entry.isFile()) { 1304 if (Invalid) 1305 *Invalid = true; 1306 return 1; 1307 } 1308 1309 Content = const_cast<ContentCache*>(Entry.getFile().getContentCache()); 1310 } 1311 1312 // If this is the first use of line information for this buffer, compute the 1313 /// SourceLineCache for it on demand. 1314 if (!Content->SourceLineCache) { 1315 bool MyInvalid = false; 1316 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1317 if (Invalid) 1318 *Invalid = MyInvalid; 1319 if (MyInvalid) 1320 return 1; 1321 } else if (Invalid) 1322 *Invalid = false; 1323 1324 // Okay, we know we have a line number table. Do a binary search to find the 1325 // line number that this character position lands on. 1326 unsigned *SourceLineCache = Content->SourceLineCache; 1327 unsigned *SourceLineCacheStart = SourceLineCache; 1328 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines; 1329 1330 unsigned QueriedFilePos = FilePos+1; 1331 1332 // FIXME: I would like to be convinced that this code is worth being as 1333 // complicated as it is, binary search isn't that slow. 1334 // 1335 // If it is worth being optimized, then in my opinion it could be more 1336 // performant, simpler, and more obviously correct by just "galloping" outward 1337 // from the queried file position. In fact, this could be incorporated into a 1338 // generic algorithm such as lower_bound_with_hint. 1339 // 1340 // If someone gives me a test case where this matters, and I will do it! - DWD 1341 1342 // If the previous query was to the same file, we know both the file pos from 1343 // that query and the line number returned. This allows us to narrow the 1344 // search space from the entire file to something near the match. 1345 if (LastLineNoFileIDQuery == FID) { 1346 if (QueriedFilePos >= LastLineNoFilePos) { 1347 // FIXME: Potential overflow? 1348 SourceLineCache = SourceLineCache+LastLineNoResult-1; 1349 1350 // The query is likely to be nearby the previous one. Here we check to 1351 // see if it is within 5, 10 or 20 lines. It can be far away in cases 1352 // where big comment blocks and vertical whitespace eat up lines but 1353 // contribute no tokens. 1354 if (SourceLineCache+5 < SourceLineCacheEnd) { 1355 if (SourceLineCache[5] > QueriedFilePos) 1356 SourceLineCacheEnd = SourceLineCache+5; 1357 else if (SourceLineCache+10 < SourceLineCacheEnd) { 1358 if (SourceLineCache[10] > QueriedFilePos) 1359 SourceLineCacheEnd = SourceLineCache+10; 1360 else if (SourceLineCache+20 < SourceLineCacheEnd) { 1361 if (SourceLineCache[20] > QueriedFilePos) 1362 SourceLineCacheEnd = SourceLineCache+20; 1363 } 1364 } 1365 } 1366 } else { 1367 if (LastLineNoResult < Content->NumLines) 1368 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1; 1369 } 1370 } 1371 1372 unsigned *Pos 1373 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos); 1374 unsigned LineNo = Pos-SourceLineCacheStart; 1375 1376 LastLineNoFileIDQuery = FID; 1377 LastLineNoContentCache = Content; 1378 LastLineNoFilePos = QueriedFilePos; 1379 LastLineNoResult = LineNo; 1380 return LineNo; 1381 } 1382 1383 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc, 1384 bool *Invalid) const { 1385 if (isInvalid(Loc, Invalid)) return 0; 1386 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1387 return getLineNumber(LocInfo.first, LocInfo.second); 1388 } 1389 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc, 1390 bool *Invalid) const { 1391 if (isInvalid(Loc, Invalid)) return 0; 1392 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1393 return getLineNumber(LocInfo.first, LocInfo.second); 1394 } 1395 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc, 1396 bool *Invalid) const { 1397 PresumedLoc PLoc = getPresumedLoc(Loc); 1398 if (isInvalid(PLoc, Invalid)) return 0; 1399 return PLoc.getLine(); 1400 } 1401 1402 /// getFileCharacteristic - return the file characteristic of the specified 1403 /// source location, indicating whether this is a normal file, a system 1404 /// header, or an "implicit extern C" system header. 1405 /// 1406 /// This state can be modified with flags on GNU linemarker directives like: 1407 /// # 4 "foo.h" 3 1408 /// which changes all source locations in the current file after that to be 1409 /// considered to be from a system header. 1410 SrcMgr::CharacteristicKind 1411 SourceManager::getFileCharacteristic(SourceLocation Loc) const { 1412 assert(Loc.isValid() && "Can't get file characteristic of invalid loc!"); 1413 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1414 bool Invalid = false; 1415 const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid); 1416 if (Invalid || !SEntry.isFile()) 1417 return C_User; 1418 1419 const SrcMgr::FileInfo &FI = SEntry.getFile(); 1420 1421 // If there are no #line directives in this file, just return the whole-file 1422 // state. 1423 if (!FI.hasLineDirectives()) 1424 return FI.getFileCharacteristic(); 1425 1426 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1427 // See if there is a #line directive before the location. 1428 const LineEntry *Entry = 1429 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second); 1430 1431 // If this is before the first line marker, use the file characteristic. 1432 if (!Entry) 1433 return FI.getFileCharacteristic(); 1434 1435 return Entry->FileKind; 1436 } 1437 1438 /// Return the filename or buffer identifier of the buffer the location is in. 1439 /// Note that this name does not respect \#line directives. Use getPresumedLoc 1440 /// for normal clients. 1441 const char *SourceManager::getBufferName(SourceLocation Loc, 1442 bool *Invalid) const { 1443 if (isInvalid(Loc, Invalid)) return "<invalid loc>"; 1444 1445 return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier(); 1446 } 1447 1448 1449 /// getPresumedLoc - This method returns the "presumed" location of a 1450 /// SourceLocation specifies. A "presumed location" can be modified by \#line 1451 /// or GNU line marker directives. This provides a view on the data that a 1452 /// user should see in diagnostics, for example. 1453 /// 1454 /// Note that a presumed location is always given as the expansion point of an 1455 /// expansion location, not at the spelling location. 1456 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc, 1457 bool UseLineDirectives) const { 1458 if (Loc.isInvalid()) return PresumedLoc(); 1459 1460 // Presumed locations are always for expansion points. 1461 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1462 1463 bool Invalid = false; 1464 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1465 if (Invalid || !Entry.isFile()) 1466 return PresumedLoc(); 1467 1468 const SrcMgr::FileInfo &FI = Entry.getFile(); 1469 const SrcMgr::ContentCache *C = FI.getContentCache(); 1470 1471 // To get the source name, first consult the FileEntry (if one exists) 1472 // before the MemBuffer as this will avoid unnecessarily paging in the 1473 // MemBuffer. 1474 const char *Filename; 1475 if (C->OrigEntry) 1476 Filename = C->OrigEntry->getName(); 1477 else 1478 Filename = C->getBuffer(Diag, *this)->getBufferIdentifier(); 1479 1480 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid); 1481 if (Invalid) 1482 return PresumedLoc(); 1483 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid); 1484 if (Invalid) 1485 return PresumedLoc(); 1486 1487 SourceLocation IncludeLoc = FI.getIncludeLoc(); 1488 1489 // If we have #line directives in this file, update and overwrite the physical 1490 // location info if appropriate. 1491 if (UseLineDirectives && FI.hasLineDirectives()) { 1492 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1493 // See if there is a #line directive before this. If so, get it. 1494 if (const LineEntry *Entry = 1495 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) { 1496 // If the LineEntry indicates a filename, use it. 1497 if (Entry->FilenameID != -1) 1498 Filename = LineTable->getFilename(Entry->FilenameID); 1499 1500 // Use the line number specified by the LineEntry. This line number may 1501 // be multiple lines down from the line entry. Add the difference in 1502 // physical line numbers from the query point and the line marker to the 1503 // total. 1504 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset); 1505 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1); 1506 1507 // Note that column numbers are not molested by line markers. 1508 1509 // Handle virtual #include manipulation. 1510 if (Entry->IncludeOffset) { 1511 IncludeLoc = getLocForStartOfFile(LocInfo.first); 1512 IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset); 1513 } 1514 } 1515 } 1516 1517 return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc); 1518 } 1519 1520 /// \brief Returns whether the PresumedLoc for a given SourceLocation is 1521 /// in the main file. 1522 /// 1523 /// This computes the "presumed" location for a SourceLocation, then checks 1524 /// whether it came from a file other than the main file. This is different 1525 /// from isWrittenInMainFile() because it takes line marker directives into 1526 /// account. 1527 bool SourceManager::isInMainFile(SourceLocation Loc) const { 1528 if (Loc.isInvalid()) return false; 1529 1530 // Presumed locations are always for expansion points. 1531 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1532 1533 bool Invalid = false; 1534 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1535 if (Invalid || !Entry.isFile()) 1536 return false; 1537 1538 const SrcMgr::FileInfo &FI = Entry.getFile(); 1539 1540 // Check if there is a line directive for this location. 1541 if (FI.hasLineDirectives()) 1542 if (const LineEntry *Entry = 1543 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) 1544 if (Entry->IncludeOffset) 1545 return false; 1546 1547 return FI.getIncludeLoc().isInvalid(); 1548 } 1549 1550 /// \brief The size of the SLocEntry that \p FID represents. 1551 unsigned SourceManager::getFileIDSize(FileID FID) const { 1552 bool Invalid = false; 1553 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1554 if (Invalid) 1555 return 0; 1556 1557 int ID = FID.ID; 1558 unsigned NextOffset; 1559 if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size())) 1560 NextOffset = getNextLocalOffset(); 1561 else if (ID+1 == -1) 1562 NextOffset = MaxLoadedOffset; 1563 else 1564 NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset(); 1565 1566 return NextOffset - Entry.getOffset() - 1; 1567 } 1568 1569 //===----------------------------------------------------------------------===// 1570 // Other miscellaneous methods. 1571 //===----------------------------------------------------------------------===// 1572 1573 /// \brief Retrieve the inode for the given file entry, if possible. 1574 /// 1575 /// This routine involves a system call, and therefore should only be used 1576 /// in non-performance-critical code. 1577 static Optional<llvm::sys::fs::UniqueID> 1578 getActualFileUID(const FileEntry *File) { 1579 if (!File) 1580 return None; 1581 1582 llvm::sys::fs::UniqueID ID; 1583 if (llvm::sys::fs::getUniqueID(File->getName(), ID)) 1584 return None; 1585 1586 return ID; 1587 } 1588 1589 /// \brief Get the source location for the given file:line:col triplet. 1590 /// 1591 /// If the source file is included multiple times, the source location will 1592 /// be based upon an arbitrary inclusion. 1593 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile, 1594 unsigned Line, 1595 unsigned Col) const { 1596 assert(SourceFile && "Null source file!"); 1597 assert(Line && Col && "Line and column should start from 1!"); 1598 1599 FileID FirstFID = translateFile(SourceFile); 1600 return translateLineCol(FirstFID, Line, Col); 1601 } 1602 1603 /// \brief Get the FileID for the given file. 1604 /// 1605 /// If the source file is included multiple times, the FileID will be the 1606 /// first inclusion. 1607 FileID SourceManager::translateFile(const FileEntry *SourceFile) const { 1608 assert(SourceFile && "Null source file!"); 1609 1610 // Find the first file ID that corresponds to the given file. 1611 FileID FirstFID; 1612 1613 // First, check the main file ID, since it is common to look for a 1614 // location in the main file. 1615 Optional<llvm::sys::fs::UniqueID> SourceFileUID; 1616 Optional<StringRef> SourceFileName; 1617 if (MainFileID.isValid()) { 1618 bool Invalid = false; 1619 const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid); 1620 if (Invalid) 1621 return FileID(); 1622 1623 if (MainSLoc.isFile()) { 1624 const ContentCache *MainContentCache 1625 = MainSLoc.getFile().getContentCache(); 1626 if (!MainContentCache) { 1627 // Can't do anything 1628 } else if (MainContentCache->OrigEntry == SourceFile) { 1629 FirstFID = MainFileID; 1630 } else { 1631 // Fall back: check whether we have the same base name and inode 1632 // as the main file. 1633 const FileEntry *MainFile = MainContentCache->OrigEntry; 1634 SourceFileName = llvm::sys::path::filename(SourceFile->getName()); 1635 if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) { 1636 SourceFileUID = getActualFileUID(SourceFile); 1637 if (SourceFileUID) { 1638 if (Optional<llvm::sys::fs::UniqueID> MainFileUID = 1639 getActualFileUID(MainFile)) { 1640 if (*SourceFileUID == *MainFileUID) { 1641 FirstFID = MainFileID; 1642 SourceFile = MainFile; 1643 } 1644 } 1645 } 1646 } 1647 } 1648 } 1649 } 1650 1651 if (FirstFID.isInvalid()) { 1652 // The location we're looking for isn't in the main file; look 1653 // through all of the local source locations. 1654 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) { 1655 bool Invalid = false; 1656 const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid); 1657 if (Invalid) 1658 return FileID(); 1659 1660 if (SLoc.isFile() && 1661 SLoc.getFile().getContentCache() && 1662 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) { 1663 FirstFID = FileID::get(I); 1664 break; 1665 } 1666 } 1667 // If that still didn't help, try the modules. 1668 if (FirstFID.isInvalid()) { 1669 for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) { 1670 const SLocEntry &SLoc = getLoadedSLocEntry(I); 1671 if (SLoc.isFile() && 1672 SLoc.getFile().getContentCache() && 1673 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) { 1674 FirstFID = FileID::get(-int(I) - 2); 1675 break; 1676 } 1677 } 1678 } 1679 } 1680 1681 // If we haven't found what we want yet, try again, but this time stat() 1682 // each of the files in case the files have changed since we originally 1683 // parsed the file. 1684 if (FirstFID.isInvalid() && 1685 (SourceFileName || 1686 (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) && 1687 (SourceFileUID || (SourceFileUID = getActualFileUID(SourceFile)))) { 1688 bool Invalid = false; 1689 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) { 1690 FileID IFileID; 1691 IFileID.ID = I; 1692 const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid); 1693 if (Invalid) 1694 return FileID(); 1695 1696 if (SLoc.isFile()) { 1697 const ContentCache *FileContentCache 1698 = SLoc.getFile().getContentCache(); 1699 const FileEntry *Entry = FileContentCache ? FileContentCache->OrigEntry 1700 : nullptr; 1701 if (Entry && 1702 *SourceFileName == llvm::sys::path::filename(Entry->getName())) { 1703 if (Optional<llvm::sys::fs::UniqueID> EntryUID = 1704 getActualFileUID(Entry)) { 1705 if (*SourceFileUID == *EntryUID) { 1706 FirstFID = FileID::get(I); 1707 SourceFile = Entry; 1708 break; 1709 } 1710 } 1711 } 1712 } 1713 } 1714 } 1715 1716 (void) SourceFile; 1717 return FirstFID; 1718 } 1719 1720 /// \brief Get the source location in \arg FID for the given line:col. 1721 /// Returns null location if \arg FID is not a file SLocEntry. 1722 SourceLocation SourceManager::translateLineCol(FileID FID, 1723 unsigned Line, 1724 unsigned Col) const { 1725 // Lines are used as a one-based index into a zero-based array. This assert 1726 // checks for possible buffer underruns. 1727 assert(Line && Col && "Line and column should start from 1!"); 1728 1729 if (FID.isInvalid()) 1730 return SourceLocation(); 1731 1732 bool Invalid = false; 1733 const SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1734 if (Invalid) 1735 return SourceLocation(); 1736 1737 if (!Entry.isFile()) 1738 return SourceLocation(); 1739 1740 SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset()); 1741 1742 if (Line == 1 && Col == 1) 1743 return FileLoc; 1744 1745 ContentCache *Content 1746 = const_cast<ContentCache *>(Entry.getFile().getContentCache()); 1747 if (!Content) 1748 return SourceLocation(); 1749 1750 // If this is the first use of line information for this buffer, compute the 1751 // SourceLineCache for it on demand. 1752 if (!Content->SourceLineCache) { 1753 bool MyInvalid = false; 1754 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1755 if (MyInvalid) 1756 return SourceLocation(); 1757 } 1758 1759 if (Line > Content->NumLines) { 1760 unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize(); 1761 if (Size > 0) 1762 --Size; 1763 return FileLoc.getLocWithOffset(Size); 1764 } 1765 1766 llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this); 1767 unsigned FilePos = Content->SourceLineCache[Line - 1]; 1768 const char *Buf = Buffer->getBufferStart() + FilePos; 1769 unsigned BufLength = Buffer->getBufferSize() - FilePos; 1770 if (BufLength == 0) 1771 return FileLoc.getLocWithOffset(FilePos); 1772 1773 unsigned i = 0; 1774 1775 // Check that the given column is valid. 1776 while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r') 1777 ++i; 1778 return FileLoc.getLocWithOffset(FilePos + i); 1779 } 1780 1781 /// \brief Compute a map of macro argument chunks to their expanded source 1782 /// location. Chunks that are not part of a macro argument will map to an 1783 /// invalid source location. e.g. if a file contains one macro argument at 1784 /// offset 100 with length 10, this is how the map will be formed: 1785 /// 0 -> SourceLocation() 1786 /// 100 -> Expanded macro arg location 1787 /// 110 -> SourceLocation() 1788 void SourceManager::computeMacroArgsCache(MacroArgsMap *&CachePtr, 1789 FileID FID) const { 1790 assert(FID.isValid()); 1791 assert(!CachePtr); 1792 1793 CachePtr = new MacroArgsMap(); 1794 MacroArgsMap &MacroArgsCache = *CachePtr; 1795 // Initially no macro argument chunk is present. 1796 MacroArgsCache.insert(std::make_pair(0, SourceLocation())); 1797 1798 int ID = FID.ID; 1799 while (1) { 1800 ++ID; 1801 // Stop if there are no more FileIDs to check. 1802 if (ID > 0) { 1803 if (unsigned(ID) >= local_sloc_entry_size()) 1804 return; 1805 } else if (ID == -1) { 1806 return; 1807 } 1808 1809 bool Invalid = false; 1810 const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID, &Invalid); 1811 if (Invalid) 1812 return; 1813 if (Entry.isFile()) { 1814 SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc(); 1815 if (IncludeLoc.isInvalid()) 1816 continue; 1817 if (!isInFileID(IncludeLoc, FID)) 1818 return; // No more files/macros that may be "contained" in this file. 1819 1820 // Skip the files/macros of the #include'd file, we only care about macros 1821 // that lexed macro arguments from our file. 1822 if (Entry.getFile().NumCreatedFIDs) 1823 ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/; 1824 continue; 1825 } 1826 1827 const ExpansionInfo &ExpInfo = Entry.getExpansion(); 1828 1829 if (ExpInfo.getExpansionLocStart().isFileID()) { 1830 if (!isInFileID(ExpInfo.getExpansionLocStart(), FID)) 1831 return; // No more files/macros that may be "contained" in this file. 1832 } 1833 1834 if (!ExpInfo.isMacroArgExpansion()) 1835 continue; 1836 1837 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1838 ExpInfo.getSpellingLoc(), 1839 SourceLocation::getMacroLoc(Entry.getOffset()), 1840 getFileIDSize(FileID::get(ID))); 1841 } 1842 } 1843 1844 void SourceManager::associateFileChunkWithMacroArgExp( 1845 MacroArgsMap &MacroArgsCache, 1846 FileID FID, 1847 SourceLocation SpellLoc, 1848 SourceLocation ExpansionLoc, 1849 unsigned ExpansionLength) const { 1850 if (!SpellLoc.isFileID()) { 1851 unsigned SpellBeginOffs = SpellLoc.getOffset(); 1852 unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength; 1853 1854 // The spelling range for this macro argument expansion can span multiple 1855 // consecutive FileID entries. Go through each entry contained in the 1856 // spelling range and if one is itself a macro argument expansion, recurse 1857 // and associate the file chunk that it represents. 1858 1859 FileID SpellFID; // Current FileID in the spelling range. 1860 unsigned SpellRelativeOffs; 1861 std::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc); 1862 while (1) { 1863 const SLocEntry &Entry = getSLocEntry(SpellFID); 1864 unsigned SpellFIDBeginOffs = Entry.getOffset(); 1865 unsigned SpellFIDSize = getFileIDSize(SpellFID); 1866 unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize; 1867 const ExpansionInfo &Info = Entry.getExpansion(); 1868 if (Info.isMacroArgExpansion()) { 1869 unsigned CurrSpellLength; 1870 if (SpellFIDEndOffs < SpellEndOffs) 1871 CurrSpellLength = SpellFIDSize - SpellRelativeOffs; 1872 else 1873 CurrSpellLength = ExpansionLength; 1874 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1875 Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs), 1876 ExpansionLoc, CurrSpellLength); 1877 } 1878 1879 if (SpellFIDEndOffs >= SpellEndOffs) 1880 return; // we covered all FileID entries in the spelling range. 1881 1882 // Move to the next FileID entry in the spelling range. 1883 unsigned advance = SpellFIDSize - SpellRelativeOffs + 1; 1884 ExpansionLoc = ExpansionLoc.getLocWithOffset(advance); 1885 ExpansionLength -= advance; 1886 ++SpellFID.ID; 1887 SpellRelativeOffs = 0; 1888 } 1889 1890 } 1891 1892 assert(SpellLoc.isFileID()); 1893 1894 unsigned BeginOffs; 1895 if (!isInFileID(SpellLoc, FID, &BeginOffs)) 1896 return; 1897 1898 unsigned EndOffs = BeginOffs + ExpansionLength; 1899 1900 // Add a new chunk for this macro argument. A previous macro argument chunk 1901 // may have been lexed again, so e.g. if the map is 1902 // 0 -> SourceLocation() 1903 // 100 -> Expanded loc #1 1904 // 110 -> SourceLocation() 1905 // and we found a new macro FileID that lexed from offet 105 with length 3, 1906 // the new map will be: 1907 // 0 -> SourceLocation() 1908 // 100 -> Expanded loc #1 1909 // 105 -> Expanded loc #2 1910 // 108 -> Expanded loc #1 1911 // 110 -> SourceLocation() 1912 // 1913 // Since re-lexed macro chunks will always be the same size or less of 1914 // previous chunks, we only need to find where the ending of the new macro 1915 // chunk is mapped to and update the map with new begin/end mappings. 1916 1917 MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs); 1918 --I; 1919 SourceLocation EndOffsMappedLoc = I->second; 1920 MacroArgsCache[BeginOffs] = ExpansionLoc; 1921 MacroArgsCache[EndOffs] = EndOffsMappedLoc; 1922 } 1923 1924 /// \brief If \arg Loc points inside a function macro argument, the returned 1925 /// location will be the macro location in which the argument was expanded. 1926 /// If a macro argument is used multiple times, the expanded location will 1927 /// be at the first expansion of the argument. 1928 /// e.g. 1929 /// MY_MACRO(foo); 1930 /// ^ 1931 /// Passing a file location pointing at 'foo', will yield a macro location 1932 /// where 'foo' was expanded into. 1933 SourceLocation 1934 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const { 1935 if (Loc.isInvalid() || !Loc.isFileID()) 1936 return Loc; 1937 1938 FileID FID; 1939 unsigned Offset; 1940 std::tie(FID, Offset) = getDecomposedLoc(Loc); 1941 if (FID.isInvalid()) 1942 return Loc; 1943 1944 MacroArgsMap *&MacroArgsCache = MacroArgsCacheMap[FID]; 1945 if (!MacroArgsCache) 1946 computeMacroArgsCache(MacroArgsCache, FID); 1947 1948 assert(!MacroArgsCache->empty()); 1949 MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset); 1950 --I; 1951 1952 unsigned MacroArgBeginOffs = I->first; 1953 SourceLocation MacroArgExpandedLoc = I->second; 1954 if (MacroArgExpandedLoc.isValid()) 1955 return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs); 1956 1957 return Loc; 1958 } 1959 1960 std::pair<FileID, unsigned> 1961 SourceManager::getDecomposedIncludedLoc(FileID FID) const { 1962 if (FID.isInvalid()) 1963 return std::make_pair(FileID(), 0); 1964 1965 // Uses IncludedLocMap to retrieve/cache the decomposed loc. 1966 1967 typedef std::pair<FileID, unsigned> DecompTy; 1968 typedef llvm::DenseMap<FileID, DecompTy> MapTy; 1969 std::pair<MapTy::iterator, bool> 1970 InsertOp = IncludedLocMap.insert(std::make_pair(FID, DecompTy())); 1971 DecompTy &DecompLoc = InsertOp.first->second; 1972 if (!InsertOp.second) 1973 return DecompLoc; // already in map. 1974 1975 SourceLocation UpperLoc; 1976 bool Invalid = false; 1977 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1978 if (!Invalid) { 1979 if (Entry.isExpansion()) 1980 UpperLoc = Entry.getExpansion().getExpansionLocStart(); 1981 else 1982 UpperLoc = Entry.getFile().getIncludeLoc(); 1983 } 1984 1985 if (UpperLoc.isValid()) 1986 DecompLoc = getDecomposedLoc(UpperLoc); 1987 1988 return DecompLoc; 1989 } 1990 1991 /// Given a decomposed source location, move it up the include/expansion stack 1992 /// to the parent source location. If this is possible, return the decomposed 1993 /// version of the parent in Loc and return false. If Loc is the top-level 1994 /// entry, return true and don't modify it. 1995 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc, 1996 const SourceManager &SM) { 1997 std::pair<FileID, unsigned> UpperLoc = SM.getDecomposedIncludedLoc(Loc.first); 1998 if (UpperLoc.first.isInvalid()) 1999 return true; // We reached the top. 2000 2001 Loc = UpperLoc; 2002 return false; 2003 } 2004 2005 /// Return the cache entry for comparing the given file IDs 2006 /// for isBeforeInTranslationUnit. 2007 InBeforeInTUCacheEntry &SourceManager::getInBeforeInTUCache(FileID LFID, 2008 FileID RFID) const { 2009 // This is a magic number for limiting the cache size. It was experimentally 2010 // derived from a small Objective-C project (where the cache filled 2011 // out to ~250 items). We can make it larger if necessary. 2012 enum { MagicCacheSize = 300 }; 2013 IsBeforeInTUCacheKey Key(LFID, RFID); 2014 2015 // If the cache size isn't too large, do a lookup and if necessary default 2016 // construct an entry. We can then return it to the caller for direct 2017 // use. When they update the value, the cache will get automatically 2018 // updated as well. 2019 if (IBTUCache.size() < MagicCacheSize) 2020 return IBTUCache[Key]; 2021 2022 // Otherwise, do a lookup that will not construct a new value. 2023 InBeforeInTUCache::iterator I = IBTUCache.find(Key); 2024 if (I != IBTUCache.end()) 2025 return I->second; 2026 2027 // Fall back to the overflow value. 2028 return IBTUCacheOverflow; 2029 } 2030 2031 /// \brief Determines the order of 2 source locations in the translation unit. 2032 /// 2033 /// \returns true if LHS source location comes before RHS, false otherwise. 2034 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS, 2035 SourceLocation RHS) const { 2036 assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!"); 2037 if (LHS == RHS) 2038 return false; 2039 2040 std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS); 2041 std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS); 2042 2043 // getDecomposedLoc may have failed to return a valid FileID because, e.g. it 2044 // is a serialized one referring to a file that was removed after we loaded 2045 // the PCH. 2046 if (LOffs.first.isInvalid() || ROffs.first.isInvalid()) 2047 return LOffs.first.isInvalid() && !ROffs.first.isInvalid(); 2048 2049 // If the source locations are in the same file, just compare offsets. 2050 if (LOffs.first == ROffs.first) 2051 return LOffs.second < ROffs.second; 2052 2053 // If we are comparing a source location with multiple locations in the same 2054 // file, we get a big win by caching the result. 2055 InBeforeInTUCacheEntry &IsBeforeInTUCache = 2056 getInBeforeInTUCache(LOffs.first, ROffs.first); 2057 2058 // If we are comparing a source location with multiple locations in the same 2059 // file, we get a big win by caching the result. 2060 if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first)) 2061 return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second); 2062 2063 // Okay, we missed in the cache, start updating the cache for this query. 2064 IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first, 2065 /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID); 2066 2067 // We need to find the common ancestor. The only way of doing this is to 2068 // build the complete include chain for one and then walking up the chain 2069 // of the other looking for a match. 2070 // We use a map from FileID to Offset to store the chain. Easier than writing 2071 // a custom set hash info that only depends on the first part of a pair. 2072 typedef llvm::SmallDenseMap<FileID, unsigned, 16> LocSet; 2073 LocSet LChain; 2074 do { 2075 LChain.insert(LOffs); 2076 // We catch the case where LOffs is in a file included by ROffs and 2077 // quit early. The other way round unfortunately remains suboptimal. 2078 } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this)); 2079 LocSet::iterator I; 2080 while((I = LChain.find(ROffs.first)) == LChain.end()) { 2081 if (MoveUpIncludeHierarchy(ROffs, *this)) 2082 break; // Met at topmost file. 2083 } 2084 if (I != LChain.end()) 2085 LOffs = *I; 2086 2087 // If we exited because we found a nearest common ancestor, compare the 2088 // locations within the common file and cache them. 2089 if (LOffs.first == ROffs.first) { 2090 IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second); 2091 return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second); 2092 } 2093 2094 // If we arrived here, the location is either in a built-ins buffer or 2095 // associated with global inline asm. PR5662 and PR22576 are examples. 2096 2097 // Clear the lookup cache, it depends on a common location. 2098 IsBeforeInTUCache.clear(); 2099 const char *LB = getBuffer(LOffs.first)->getBufferIdentifier(); 2100 const char *RB = getBuffer(ROffs.first)->getBufferIdentifier(); 2101 bool LIsBuiltins = strcmp("<built-in>", LB) == 0; 2102 bool RIsBuiltins = strcmp("<built-in>", RB) == 0; 2103 // Sort built-in before non-built-in. 2104 if (LIsBuiltins || RIsBuiltins) { 2105 if (LIsBuiltins != RIsBuiltins) 2106 return LIsBuiltins; 2107 // Both are in built-in buffers, but from different files. We just claim that 2108 // lower IDs come first. 2109 return LOffs.first < ROffs.first; 2110 } 2111 bool LIsAsm = strcmp("<inline asm>", LB) == 0; 2112 bool RIsAsm = strcmp("<inline asm>", RB) == 0; 2113 // Sort assembler after built-ins, but before the rest. 2114 if (LIsAsm || RIsAsm) { 2115 if (LIsAsm != RIsAsm) 2116 return RIsAsm; 2117 assert(LOffs.first == ROffs.first); 2118 return false; 2119 } 2120 bool LIsScratch = strcmp("<scratch space>", LB) == 0; 2121 bool RIsScratch = strcmp("<scratch space>", RB) == 0; 2122 // Sort scratch after inline asm, but before the rest. 2123 if (LIsScratch || RIsScratch) { 2124 if (LIsScratch != RIsScratch) 2125 return LIsScratch; 2126 return LOffs.second < ROffs.second; 2127 } 2128 llvm_unreachable("Unsortable locations found"); 2129 } 2130 2131 void SourceManager::PrintStats() const { 2132 llvm::errs() << "\n*** Source Manager Stats:\n"; 2133 llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size() 2134 << " mem buffers mapped.\n"; 2135 llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated (" 2136 << llvm::capacity_in_bytes(LocalSLocEntryTable) 2137 << " bytes of capacity), " 2138 << NextLocalOffset << "B of Sloc address space used.\n"; 2139 llvm::errs() << LoadedSLocEntryTable.size() 2140 << " loaded SLocEntries allocated, " 2141 << MaxLoadedOffset - CurrentLoadedOffset 2142 << "B of Sloc address space used.\n"; 2143 2144 unsigned NumLineNumsComputed = 0; 2145 unsigned NumFileBytesMapped = 0; 2146 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){ 2147 NumLineNumsComputed += I->second->SourceLineCache != nullptr; 2148 NumFileBytesMapped += I->second->getSizeBytesMapped(); 2149 } 2150 unsigned NumMacroArgsComputed = MacroArgsCacheMap.size(); 2151 2152 llvm::errs() << NumFileBytesMapped << " bytes of files mapped, " 2153 << NumLineNumsComputed << " files with line #'s computed, " 2154 << NumMacroArgsComputed << " files with macro args computed.\n"; 2155 llvm::errs() << "FileID scans: " << NumLinearScans << " linear, " 2156 << NumBinaryProbes << " binary.\n"; 2157 } 2158 2159 LLVM_DUMP_METHOD void SourceManager::dump() const { 2160 llvm::raw_ostream &out = llvm::errs(); 2161 2162 auto DumpSLocEntry = [&](int ID, const SrcMgr::SLocEntry &Entry, 2163 llvm::Optional<unsigned> NextStart) { 2164 out << "SLocEntry <FileID " << ID << "> " << (Entry.isFile() ? "file" : "expansion") 2165 << " <SourceLocation " << Entry.getOffset() << ":"; 2166 if (NextStart) 2167 out << *NextStart << ">\n"; 2168 else 2169 out << "???\?>\n"; 2170 if (Entry.isFile()) { 2171 auto &FI = Entry.getFile(); 2172 if (FI.NumCreatedFIDs) 2173 out << " covers <FileID " << ID << ":" << int(ID + FI.NumCreatedFIDs) 2174 << ">\n"; 2175 if (FI.getIncludeLoc().isValid()) 2176 out << " included from " << FI.getIncludeLoc().getOffset() << "\n"; 2177 if (auto *CC = FI.getContentCache()) { 2178 out << " for " << (CC->OrigEntry ? CC->OrigEntry->getName() : "<none>") 2179 << "\n"; 2180 if (CC->BufferOverridden) 2181 out << " contents overridden\n"; 2182 if (CC->ContentsEntry != CC->OrigEntry) { 2183 out << " contents from " 2184 << (CC->ContentsEntry ? CC->ContentsEntry->getName() : "<none>") 2185 << "\n"; 2186 } 2187 } 2188 } else { 2189 auto &EI = Entry.getExpansion(); 2190 out << " spelling from " << EI.getSpellingLoc().getOffset() << "\n"; 2191 out << " macro " << (EI.isMacroArgExpansion() ? "arg" : "body") 2192 << " range <" << EI.getExpansionLocStart().getOffset() << ":" 2193 << EI.getExpansionLocEnd().getOffset() << ">\n"; 2194 } 2195 }; 2196 2197 // Dump local SLocEntries. 2198 for (unsigned ID = 0, NumIDs = LocalSLocEntryTable.size(); ID != NumIDs; ++ID) { 2199 DumpSLocEntry(ID, LocalSLocEntryTable[ID], 2200 ID == NumIDs - 1 ? NextLocalOffset 2201 : LocalSLocEntryTable[ID + 1].getOffset()); 2202 } 2203 // Dump loaded SLocEntries. 2204 llvm::Optional<unsigned> NextStart; 2205 for (unsigned Index = 0; Index != LoadedSLocEntryTable.size(); ++Index) { 2206 int ID = -(int)Index - 2; 2207 if (SLocEntryLoaded[Index]) { 2208 DumpSLocEntry(ID, LoadedSLocEntryTable[Index], NextStart); 2209 NextStart = LoadedSLocEntryTable[Index].getOffset(); 2210 } else { 2211 NextStart = None; 2212 } 2213 } 2214 } 2215 2216 ExternalSLocEntrySource::~ExternalSLocEntrySource() { } 2217 2218 /// Return the amount of memory used by memory buffers, breaking down 2219 /// by heap-backed versus mmap'ed memory. 2220 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const { 2221 size_t malloc_bytes = 0; 2222 size_t mmap_bytes = 0; 2223 2224 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) 2225 if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped()) 2226 switch (MemBufferInfos[i]->getMemoryBufferKind()) { 2227 case llvm::MemoryBuffer::MemoryBuffer_MMap: 2228 mmap_bytes += sized_mapped; 2229 break; 2230 case llvm::MemoryBuffer::MemoryBuffer_Malloc: 2231 malloc_bytes += sized_mapped; 2232 break; 2233 } 2234 2235 return MemoryBufferSizes(malloc_bytes, mmap_bytes); 2236 } 2237 2238 size_t SourceManager::getDataStructureSizes() const { 2239 size_t size = llvm::capacity_in_bytes(MemBufferInfos) 2240 + llvm::capacity_in_bytes(LocalSLocEntryTable) 2241 + llvm::capacity_in_bytes(LoadedSLocEntryTable) 2242 + llvm::capacity_in_bytes(SLocEntryLoaded) 2243 + llvm::capacity_in_bytes(FileInfos); 2244 2245 if (OverriddenFilesInfo) 2246 size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles); 2247 2248 return size; 2249 } 2250