1 //===- SourceManager.cpp - Track and cache source files -------------------===// 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 the SourceManager interface. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "clang/Basic/SourceManager.h" 14 #include "clang/Basic/Diagnostic.h" 15 #include "clang/Basic/FileManager.h" 16 #include "clang/Basic/LLVM.h" 17 #include "clang/Basic/SourceLocation.h" 18 #include "clang/Basic/SourceManagerInternals.h" 19 #include "llvm/ADT/DenseMap.h" 20 #include "llvm/ADT/Optional.h" 21 #include "llvm/ADT/None.h" 22 #include "llvm/ADT/STLExtras.h" 23 #include "llvm/ADT/SmallVector.h" 24 #include "llvm/ADT/StringSwitch.h" 25 #include "llvm/ADT/StringRef.h" 26 #include "llvm/Support/Allocator.h" 27 #include "llvm/Support/Capacity.h" 28 #include "llvm/Support/Compiler.h" 29 #include "llvm/Support/ErrorHandling.h" 30 #include "llvm/Support/FileSystem.h" 31 #include "llvm/Support/MathExtras.h" 32 #include "llvm/Support/MemoryBuffer.h" 33 #include "llvm/Support/Path.h" 34 #include "llvm/Support/raw_ostream.h" 35 #include <algorithm> 36 #include <cassert> 37 #include <cstddef> 38 #include <cstdint> 39 #include <memory> 40 #include <tuple> 41 #include <utility> 42 #include <vector> 43 44 using namespace clang; 45 using namespace SrcMgr; 46 using llvm::MemoryBuffer; 47 48 //===----------------------------------------------------------------------===// 49 // SourceManager Helper Classes 50 //===----------------------------------------------------------------------===// 51 52 ContentCache::~ContentCache() { 53 if (shouldFreeBuffer()) 54 delete Buffer.getPointer(); 55 } 56 57 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this 58 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded. 59 unsigned ContentCache::getSizeBytesMapped() const { 60 return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0; 61 } 62 63 /// Returns the kind of memory used to back the memory buffer for 64 /// this content cache. This is used for performance analysis. 65 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const { 66 assert(Buffer.getPointer()); 67 68 // Should be unreachable, but keep for sanity. 69 if (!Buffer.getPointer()) 70 return llvm::MemoryBuffer::MemoryBuffer_Malloc; 71 72 const llvm::MemoryBuffer *buf = Buffer.getPointer(); 73 return buf->getBufferKind(); 74 } 75 76 /// getSize - Returns the size of the content encapsulated by this ContentCache. 77 /// This can be the size of the source file or the size of an arbitrary 78 /// scratch buffer. If the ContentCache encapsulates a source file, that 79 /// file is not lazily brought in from disk to satisfy this query. 80 unsigned ContentCache::getSize() const { 81 return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize() 82 : (unsigned) ContentsEntry->getSize(); 83 } 84 85 void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B, bool DoNotFree) { 86 if (B && B == Buffer.getPointer()) { 87 assert(0 && "Replacing with the same buffer"); 88 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0); 89 return; 90 } 91 92 if (shouldFreeBuffer()) 93 delete Buffer.getPointer(); 94 Buffer.setPointer(B); 95 Buffer.setInt((B && DoNotFree) ? DoNotFreeFlag : 0); 96 } 97 98 const llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag, 99 const SourceManager &SM, 100 SourceLocation Loc, 101 bool *Invalid) const { 102 // Lazily create the Buffer for ContentCaches that wrap files. If we already 103 // computed it, just return what we have. 104 if (Buffer.getPointer() || !ContentsEntry) { 105 if (Invalid) 106 *Invalid = isBufferInvalid(); 107 108 return Buffer.getPointer(); 109 } 110 111 // Check that the file's size fits in an 'unsigned' (with room for a 112 // past-the-end value). This is deeply regrettable, but various parts of 113 // Clang (including elsewhere in this file!) use 'unsigned' to represent file 114 // offsets, line numbers, string literal lengths, and so on, and fail 115 // miserably on large source files. 116 if ((uint64_t)ContentsEntry->getSize() >= 117 std::numeric_limits<unsigned>::max()) { 118 // We can't make a memory buffer of the required size, so just make a small 119 // one. We should never hit a situation where we've already parsed to a 120 // later offset of the file, so it shouldn't matter that the buffer is 121 // smaller than the file. 122 Buffer.setPointer( 123 llvm::MemoryBuffer::getMemBuffer("", ContentsEntry->getName()) 124 .release()); 125 if (Diag.isDiagnosticInFlight()) 126 Diag.SetDelayedDiagnostic(diag::err_file_too_large, 127 ContentsEntry->getName()); 128 else 129 Diag.Report(Loc, diag::err_file_too_large) 130 << ContentsEntry->getName(); 131 132 Buffer.setInt(Buffer.getInt() | InvalidFlag); 133 if (Invalid) *Invalid = true; 134 return Buffer.getPointer(); 135 } 136 137 bool isVolatile = SM.userFilesAreVolatile() && !IsSystemFile; 138 auto BufferOrError = 139 SM.getFileManager().getBufferForFile(ContentsEntry, isVolatile); 140 141 // If we were unable to open the file, then we are in an inconsistent 142 // situation where the content cache referenced a file which no longer 143 // exists. Most likely, we were using a stat cache with an invalid entry but 144 // the file could also have been removed during processing. Since we can't 145 // really deal with this situation, just create an empty buffer. 146 // 147 // FIXME: This is definitely not ideal, but our immediate clients can't 148 // currently handle returning a null entry here. Ideally we should detect 149 // that we are in an inconsistent situation and error out as quickly as 150 // possible. 151 if (!BufferOrError) { 152 StringRef FillStr("<<<MISSING SOURCE FILE>>>\n"); 153 auto BackupBuffer = llvm::WritableMemoryBuffer::getNewUninitMemBuffer( 154 ContentsEntry->getSize(), "<invalid>"); 155 char *Ptr = BackupBuffer->getBufferStart(); 156 for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i) 157 Ptr[i] = FillStr[i % FillStr.size()]; 158 Buffer.setPointer(BackupBuffer.release()); 159 160 if (Diag.isDiagnosticInFlight()) 161 Diag.SetDelayedDiagnostic(diag::err_cannot_open_file, 162 ContentsEntry->getName(), 163 BufferOrError.getError().message()); 164 else 165 Diag.Report(Loc, diag::err_cannot_open_file) 166 << ContentsEntry->getName() << BufferOrError.getError().message(); 167 168 Buffer.setInt(Buffer.getInt() | InvalidFlag); 169 170 if (Invalid) *Invalid = true; 171 return Buffer.getPointer(); 172 } 173 174 Buffer.setPointer(BufferOrError->release()); 175 176 // Check that the file's size is the same as in the file entry (which may 177 // have come from a stat cache). 178 if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) { 179 if (Diag.isDiagnosticInFlight()) 180 Diag.SetDelayedDiagnostic(diag::err_file_modified, 181 ContentsEntry->getName()); 182 else 183 Diag.Report(Loc, diag::err_file_modified) 184 << ContentsEntry->getName(); 185 186 Buffer.setInt(Buffer.getInt() | InvalidFlag); 187 if (Invalid) *Invalid = true; 188 return Buffer.getPointer(); 189 } 190 191 // If the buffer is valid, check to see if it has a UTF Byte Order Mark 192 // (BOM). We only support UTF-8 with and without a BOM right now. See 193 // http://en.wikipedia.org/wiki/Byte_order_mark for more information. 194 StringRef BufStr = Buffer.getPointer()->getBuffer(); 195 const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr) 196 .StartsWith(llvm::StringLiteral::withInnerNUL("\x00\x00\xFE\xFF"), 197 "UTF-32 (BE)") 198 .StartsWith(llvm::StringLiteral::withInnerNUL("\xFF\xFE\x00\x00"), 199 "UTF-32 (LE)") 200 .StartsWith("\xFE\xFF", "UTF-16 (BE)") 201 .StartsWith("\xFF\xFE", "UTF-16 (LE)") 202 .StartsWith("\x2B\x2F\x76", "UTF-7") 203 .StartsWith("\xF7\x64\x4C", "UTF-1") 204 .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC") 205 .StartsWith("\x0E\xFE\xFF", "SCSU") 206 .StartsWith("\xFB\xEE\x28", "BOCU-1") 207 .StartsWith("\x84\x31\x95\x33", "GB-18030") 208 .Default(nullptr); 209 210 if (InvalidBOM) { 211 Diag.Report(Loc, diag::err_unsupported_bom) 212 << InvalidBOM << ContentsEntry->getName(); 213 Buffer.setInt(Buffer.getInt() | InvalidFlag); 214 } 215 216 if (Invalid) 217 *Invalid = isBufferInvalid(); 218 219 return Buffer.getPointer(); 220 } 221 222 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) { 223 auto IterBool = FilenameIDs.try_emplace(Name, FilenamesByID.size()); 224 if (IterBool.second) 225 FilenamesByID.push_back(&*IterBool.first); 226 return IterBool.first->second; 227 } 228 229 /// Add a line note to the line table that indicates that there is a \#line or 230 /// GNU line marker at the specified FID/Offset location which changes the 231 /// presumed location to LineNo/FilenameID. If EntryExit is 0, then this doesn't 232 /// change the presumed \#include stack. If it is 1, this is a file entry, if 233 /// it is 2 then this is a file exit. FileKind specifies whether this is a 234 /// system header or extern C system header. 235 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset, unsigned LineNo, 236 int FilenameID, unsigned EntryExit, 237 SrcMgr::CharacteristicKind FileKind) { 238 std::vector<LineEntry> &Entries = LineEntries[FID]; 239 240 // An unspecified FilenameID means use the last filename if available, or the 241 // main source file otherwise. 242 if (FilenameID == -1 && !Entries.empty()) 243 FilenameID = Entries.back().FilenameID; 244 245 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 246 "Adding line entries out of order!"); 247 248 unsigned IncludeOffset = 0; 249 if (EntryExit == 0) { // No #include stack change. 250 IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset; 251 } else if (EntryExit == 1) { 252 IncludeOffset = Offset-1; 253 } else if (EntryExit == 2) { 254 assert(!Entries.empty() && Entries.back().IncludeOffset && 255 "PPDirectives should have caught case when popping empty include stack"); 256 257 // Get the include loc of the last entries' include loc as our include loc. 258 IncludeOffset = 0; 259 if (const LineEntry *PrevEntry = 260 FindNearestLineEntry(FID, Entries.back().IncludeOffset)) 261 IncludeOffset = PrevEntry->IncludeOffset; 262 } 263 264 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind, 265 IncludeOffset)); 266 } 267 268 /// FindNearestLineEntry - Find the line entry nearest to FID that is before 269 /// it. If there is no line entry before Offset in FID, return null. 270 const LineEntry *LineTableInfo::FindNearestLineEntry(FileID FID, 271 unsigned Offset) { 272 const std::vector<LineEntry> &Entries = LineEntries[FID]; 273 assert(!Entries.empty() && "No #line entries for this FID after all!"); 274 275 // It is very common for the query to be after the last #line, check this 276 // first. 277 if (Entries.back().FileOffset <= Offset) 278 return &Entries.back(); 279 280 // Do a binary search to find the maximal element that is still before Offset. 281 std::vector<LineEntry>::const_iterator I = llvm::upper_bound(Entries, Offset); 282 if (I == Entries.begin()) 283 return nullptr; 284 return &*--I; 285 } 286 287 /// Add a new line entry that has already been encoded into 288 /// the internal representation of the line table. 289 void LineTableInfo::AddEntry(FileID FID, 290 const std::vector<LineEntry> &Entries) { 291 LineEntries[FID] = Entries; 292 } 293 294 /// getLineTableFilenameID - Return the uniqued ID for the specified filename. 295 unsigned SourceManager::getLineTableFilenameID(StringRef Name) { 296 return getLineTable().getLineTableFilenameID(Name); 297 } 298 299 /// AddLineNote - Add a line note to the line table for the FileID and offset 300 /// specified by Loc. If FilenameID is -1, it is considered to be 301 /// unspecified. 302 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 303 int FilenameID, bool IsFileEntry, 304 bool IsFileExit, 305 SrcMgr::CharacteristicKind FileKind) { 306 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 307 308 bool Invalid = false; 309 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 310 if (!Entry.isFile() || Invalid) 311 return; 312 313 const SrcMgr::FileInfo &FileInfo = Entry.getFile(); 314 315 // Remember that this file has #line directives now if it doesn't already. 316 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 317 318 (void) getLineTable(); 319 320 unsigned EntryExit = 0; 321 if (IsFileEntry) 322 EntryExit = 1; 323 else if (IsFileExit) 324 EntryExit = 2; 325 326 LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID, 327 EntryExit, FileKind); 328 } 329 330 LineTableInfo &SourceManager::getLineTable() { 331 if (!LineTable) 332 LineTable.reset(new LineTableInfo()); 333 return *LineTable; 334 } 335 336 //===----------------------------------------------------------------------===// 337 // Private 'Create' methods. 338 //===----------------------------------------------------------------------===// 339 340 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr, 341 bool UserFilesAreVolatile) 342 : Diag(Diag), FileMgr(FileMgr), UserFilesAreVolatile(UserFilesAreVolatile) { 343 clearIDTables(); 344 Diag.setSourceManager(this); 345 } 346 347 SourceManager::~SourceManager() { 348 // Delete FileEntry objects corresponding to content caches. Since the actual 349 // content cache objects are bump pointer allocated, we just have to run the 350 // dtors, but we call the deallocate method for completeness. 351 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) { 352 if (MemBufferInfos[i]) { 353 MemBufferInfos[i]->~ContentCache(); 354 ContentCacheAlloc.Deallocate(MemBufferInfos[i]); 355 } 356 } 357 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator 358 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) { 359 if (I->second) { 360 I->second->~ContentCache(); 361 ContentCacheAlloc.Deallocate(I->second); 362 } 363 } 364 } 365 366 void SourceManager::clearIDTables() { 367 MainFileID = FileID(); 368 LocalSLocEntryTable.clear(); 369 LoadedSLocEntryTable.clear(); 370 SLocEntryLoaded.clear(); 371 LastLineNoFileIDQuery = FileID(); 372 LastLineNoContentCache = nullptr; 373 LastFileIDLookup = FileID(); 374 375 if (LineTable) 376 LineTable->clear(); 377 378 // Use up FileID #0 as an invalid expansion. 379 NextLocalOffset = 0; 380 CurrentLoadedOffset = MaxLoadedOffset; 381 createExpansionLoc(SourceLocation(), SourceLocation(), SourceLocation(), 1); 382 } 383 384 void SourceManager::initializeForReplay(const SourceManager &Old) { 385 assert(MainFileID.isInvalid() && "expected uninitialized SourceManager"); 386 387 auto CloneContentCache = [&](const ContentCache *Cache) -> ContentCache * { 388 auto *Clone = new (ContentCacheAlloc.Allocate<ContentCache>()) ContentCache; 389 Clone->OrigEntry = Cache->OrigEntry; 390 Clone->ContentsEntry = Cache->ContentsEntry; 391 Clone->BufferOverridden = Cache->BufferOverridden; 392 Clone->IsSystemFile = Cache->IsSystemFile; 393 Clone->IsTransient = Cache->IsTransient; 394 Clone->replaceBuffer(Cache->getRawBuffer(), /*DoNotFree*/true); 395 return Clone; 396 }; 397 398 // Ensure all SLocEntries are loaded from the external source. 399 for (unsigned I = 0, N = Old.LoadedSLocEntryTable.size(); I != N; ++I) 400 if (!Old.SLocEntryLoaded[I]) 401 Old.loadSLocEntry(I, nullptr); 402 403 // Inherit any content cache data from the old source manager. 404 for (auto &FileInfo : Old.FileInfos) { 405 SrcMgr::ContentCache *&Slot = FileInfos[FileInfo.first]; 406 if (Slot) 407 continue; 408 Slot = CloneContentCache(FileInfo.second); 409 } 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 /// Create a new ContentCache for the specified memory buffer. 448 /// This does no caching. 449 const ContentCache * 450 SourceManager::createMemBufferContentCache(const llvm::MemoryBuffer *Buffer, 451 bool DoNotFree) { 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->replaceBuffer(Buffer, DoNotFree); 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 /// 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 /// 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 /// 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 /// 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 bool ExpansionIsTokenRange, 605 int LoadedID, 606 unsigned LoadedOffset) { 607 ExpansionInfo Info = ExpansionInfo::create( 608 SpellingLoc, ExpansionLocStart, ExpansionLocEnd, ExpansionIsTokenRange); 609 return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset); 610 } 611 612 SourceLocation SourceManager::createTokenSplitLoc(SourceLocation Spelling, 613 SourceLocation TokenStart, 614 SourceLocation TokenEnd) { 615 assert(getFileID(TokenStart) == getFileID(TokenEnd) && 616 "token spans multiple files"); 617 return createExpansionLocImpl( 618 ExpansionInfo::createForTokenSplit(Spelling, TokenStart, TokenEnd), 619 TokenEnd.getOffset() - TokenStart.getOffset()); 620 } 621 622 SourceLocation 623 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info, 624 unsigned TokLength, 625 int LoadedID, 626 unsigned LoadedOffset) { 627 if (LoadedID < 0) { 628 assert(LoadedID != -1 && "Loading sentinel FileID"); 629 unsigned Index = unsigned(-LoadedID) - 2; 630 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range"); 631 assert(!SLocEntryLoaded[Index] && "FileID already loaded"); 632 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info); 633 SLocEntryLoaded[Index] = true; 634 return SourceLocation::getMacroLoc(LoadedOffset); 635 } 636 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info)); 637 assert(NextLocalOffset + TokLength + 1 > NextLocalOffset && 638 NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset && 639 "Ran out of source locations!"); 640 // See createFileID for that +1. 641 NextLocalOffset += TokLength + 1; 642 return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1)); 643 } 644 645 const llvm::MemoryBuffer * 646 SourceManager::getMemoryBufferForFile(const FileEntry *File, bool *Invalid) { 647 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 648 assert(IR && "getOrCreateContentCache() cannot return NULL"); 649 return IR->getBuffer(Diag, *this, SourceLocation(), Invalid); 650 } 651 652 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 653 llvm::MemoryBuffer *Buffer, 654 bool DoNotFree) { 655 const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile); 656 assert(IR && "getOrCreateContentCache() cannot return NULL"); 657 658 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree); 659 const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true; 660 661 getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile); 662 } 663 664 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 665 const FileEntry *NewFile) { 666 assert(SourceFile->getSize() == NewFile->getSize() && 667 "Different sizes, use the FileManager to create a virtual file with " 668 "the correct size"); 669 assert(FileInfos.count(SourceFile) == 0 && 670 "This function should be called at the initialization stage, before " 671 "any parsing occurs."); 672 getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile; 673 } 674 675 void SourceManager::disableFileContentsOverride(const FileEntry *File) { 676 if (!isFileOverridden(File)) 677 return; 678 679 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 680 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(nullptr); 681 const_cast<SrcMgr::ContentCache *>(IR)->ContentsEntry = IR->OrigEntry; 682 683 assert(OverriddenFilesInfo); 684 OverriddenFilesInfo->OverriddenFiles.erase(File); 685 OverriddenFilesInfo->OverriddenFilesWithBuffer.erase(File); 686 } 687 688 void SourceManager::setFileIsTransient(const FileEntry *File) { 689 const SrcMgr::ContentCache *CC = getOrCreateContentCache(File); 690 const_cast<SrcMgr::ContentCache *>(CC)->IsTransient = true; 691 } 692 693 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const { 694 bool MyInvalid = false; 695 const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid); 696 if (!SLoc.isFile() || MyInvalid) { 697 if (Invalid) 698 *Invalid = true; 699 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 700 } 701 702 const llvm::MemoryBuffer *Buf = SLoc.getFile().getContentCache()->getBuffer( 703 Diag, *this, SourceLocation(), &MyInvalid); 704 if (Invalid) 705 *Invalid = MyInvalid; 706 707 if (MyInvalid) 708 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 709 710 return Buf->getBuffer(); 711 } 712 713 //===----------------------------------------------------------------------===// 714 // SourceLocation manipulation methods. 715 //===----------------------------------------------------------------------===// 716 717 /// Return the FileID for a SourceLocation. 718 /// 719 /// This is the cache-miss path of getFileID. Not as hot as that function, but 720 /// still very important. It is responsible for finding the entry in the 721 /// SLocEntry tables that contains the specified location. 722 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const { 723 if (!SLocOffset) 724 return FileID::get(0); 725 726 // Now it is time to search for the correct file. See where the SLocOffset 727 // sits in the global view and consult local or loaded buffers for it. 728 if (SLocOffset < NextLocalOffset) 729 return getFileIDLocal(SLocOffset); 730 return getFileIDLoaded(SLocOffset); 731 } 732 733 /// Return the FileID for a SourceLocation with a low offset. 734 /// 735 /// This function knows that the SourceLocation is in a local buffer, not a 736 /// loaded one. 737 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const { 738 assert(SLocOffset < NextLocalOffset && "Bad function choice"); 739 740 // After the first and second level caches, I see two common sorts of 741 // behavior: 1) a lot of searched FileID's are "near" the cached file 742 // location or are "near" the cached expansion location. 2) others are just 743 // completely random and may be a very long way away. 744 // 745 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly 746 // then we fall back to a less cache efficient, but more scalable, binary 747 // search to find the location. 748 749 // See if this is near the file point - worst case we start scanning from the 750 // most newly created FileID. 751 const SrcMgr::SLocEntry *I; 752 753 if (LastFileIDLookup.ID < 0 || 754 LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) { 755 // Neither loc prunes our search. 756 I = LocalSLocEntryTable.end(); 757 } else { 758 // Perhaps it is near the file point. 759 I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID; 760 } 761 762 // Find the FileID that contains this. "I" is an iterator that points to a 763 // FileID whose offset is known to be larger than SLocOffset. 764 unsigned NumProbes = 0; 765 while (true) { 766 --I; 767 if (I->getOffset() <= SLocOffset) { 768 FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin())); 769 770 // If this isn't an expansion, remember it. We have good locality across 771 // FileID lookups. 772 if (!I->isExpansion()) 773 LastFileIDLookup = Res; 774 NumLinearScans += NumProbes+1; 775 return Res; 776 } 777 if (++NumProbes == 8) 778 break; 779 } 780 781 // Convert "I" back into an index. We know that it is an entry whose index is 782 // larger than the offset we are looking for. 783 unsigned GreaterIndex = I - LocalSLocEntryTable.begin(); 784 // LessIndex - This is the lower bound of the range that we're searching. 785 // We know that the offset corresponding to the FileID is is less than 786 // SLocOffset. 787 unsigned LessIndex = 0; 788 NumProbes = 0; 789 while (true) { 790 bool Invalid = false; 791 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex; 792 unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset(); 793 if (Invalid) 794 return FileID::get(0); 795 796 ++NumProbes; 797 798 // If the offset of the midpoint is too large, chop the high side of the 799 // range to the midpoint. 800 if (MidOffset > SLocOffset) { 801 GreaterIndex = MiddleIndex; 802 continue; 803 } 804 805 // If the middle index contains the value, succeed and return. 806 // FIXME: This could be made faster by using a function that's aware of 807 // being in the local area. 808 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) { 809 FileID Res = FileID::get(MiddleIndex); 810 811 // If this isn't a macro expansion, remember it. We have good locality 812 // across FileID lookups. 813 if (!LocalSLocEntryTable[MiddleIndex].isExpansion()) 814 LastFileIDLookup = Res; 815 NumBinaryProbes += NumProbes; 816 return Res; 817 } 818 819 // Otherwise, move the low-side up to the middle index. 820 LessIndex = MiddleIndex; 821 } 822 } 823 824 /// Return the FileID for a SourceLocation with a high offset. 825 /// 826 /// This function knows that the SourceLocation is in a loaded buffer, not a 827 /// local one. 828 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const { 829 // Sanity checking, otherwise a bug may lead to hanging in release build. 830 if (SLocOffset < CurrentLoadedOffset) { 831 assert(0 && "Invalid SLocOffset or bad function choice"); 832 return FileID(); 833 } 834 835 // Essentially the same as the local case, but the loaded array is sorted 836 // in the other direction. 837 838 // First do a linear scan from the last lookup position, if possible. 839 unsigned I; 840 int LastID = LastFileIDLookup.ID; 841 if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset) 842 I = 0; 843 else 844 I = (-LastID - 2) + 1; 845 846 unsigned NumProbes; 847 for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) { 848 // Make sure the entry is loaded! 849 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I); 850 if (E.getOffset() <= SLocOffset) { 851 FileID Res = FileID::get(-int(I) - 2); 852 853 if (!E.isExpansion()) 854 LastFileIDLookup = Res; 855 NumLinearScans += NumProbes + 1; 856 return Res; 857 } 858 } 859 860 // Linear scan failed. Do the binary search. Note the reverse sorting of the 861 // table: GreaterIndex is the one where the offset is greater, which is 862 // actually a lower index! 863 unsigned GreaterIndex = I; 864 unsigned LessIndex = LoadedSLocEntryTable.size(); 865 NumProbes = 0; 866 while (true) { 867 ++NumProbes; 868 unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex; 869 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex); 870 if (E.getOffset() == 0) 871 return FileID(); // invalid entry. 872 873 ++NumProbes; 874 875 if (E.getOffset() > SLocOffset) { 876 // Sanity checking, otherwise a bug may lead to hanging in release build. 877 if (GreaterIndex == MiddleIndex) { 878 assert(0 && "binary search missed the entry"); 879 return FileID(); 880 } 881 GreaterIndex = MiddleIndex; 882 continue; 883 } 884 885 if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) { 886 FileID Res = FileID::get(-int(MiddleIndex) - 2); 887 if (!E.isExpansion()) 888 LastFileIDLookup = Res; 889 NumBinaryProbes += NumProbes; 890 return Res; 891 } 892 893 // Sanity checking, otherwise a bug may lead to hanging in release build. 894 if (LessIndex == MiddleIndex) { 895 assert(0 && "binary search missed the entry"); 896 return FileID(); 897 } 898 LessIndex = MiddleIndex; 899 } 900 } 901 902 SourceLocation SourceManager:: 903 getExpansionLocSlowCase(SourceLocation Loc) const { 904 do { 905 // Note: If Loc indicates an offset into a token that came from a macro 906 // expansion (e.g. the 5th character of the token) we do not want to add 907 // this offset when going to the expansion location. The expansion 908 // location is the macro invocation, which the offset has nothing to do 909 // with. This is unlike when we get the spelling loc, because the offset 910 // directly correspond to the token whose spelling we're inspecting. 911 Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart(); 912 } while (!Loc.isFileID()); 913 914 return Loc; 915 } 916 917 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const { 918 do { 919 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 920 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 921 Loc = Loc.getLocWithOffset(LocInfo.second); 922 } while (!Loc.isFileID()); 923 return Loc; 924 } 925 926 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const { 927 do { 928 if (isMacroArgExpansion(Loc)) 929 Loc = getImmediateSpellingLoc(Loc); 930 else 931 Loc = getImmediateExpansionRange(Loc).getBegin(); 932 } while (!Loc.isFileID()); 933 return Loc; 934 } 935 936 937 std::pair<FileID, unsigned> 938 SourceManager::getDecomposedExpansionLocSlowCase( 939 const SrcMgr::SLocEntry *E) const { 940 // If this is an expansion record, walk through all the expansion points. 941 FileID FID; 942 SourceLocation Loc; 943 unsigned Offset; 944 do { 945 Loc = E->getExpansion().getExpansionLocStart(); 946 947 FID = getFileID(Loc); 948 E = &getSLocEntry(FID); 949 Offset = Loc.getOffset()-E->getOffset(); 950 } while (!Loc.isFileID()); 951 952 return std::make_pair(FID, Offset); 953 } 954 955 std::pair<FileID, unsigned> 956 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E, 957 unsigned Offset) const { 958 // If this is an expansion record, walk through all the expansion points. 959 FileID FID; 960 SourceLocation Loc; 961 do { 962 Loc = E->getExpansion().getSpellingLoc(); 963 Loc = Loc.getLocWithOffset(Offset); 964 965 FID = getFileID(Loc); 966 E = &getSLocEntry(FID); 967 Offset = Loc.getOffset()-E->getOffset(); 968 } while (!Loc.isFileID()); 969 970 return std::make_pair(FID, Offset); 971 } 972 973 /// getImmediateSpellingLoc - Given a SourceLocation object, return the 974 /// spelling location referenced by the ID. This is the first level down 975 /// towards the place where the characters that make up the lexed token can be 976 /// found. This should not generally be used by clients. 977 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{ 978 if (Loc.isFileID()) return Loc; 979 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 980 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 981 return Loc.getLocWithOffset(LocInfo.second); 982 } 983 984 /// getImmediateExpansionRange - Loc is required to be an expansion location. 985 /// Return the start/end of the expansion information. 986 CharSourceRange 987 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const { 988 assert(Loc.isMacroID() && "Not a macro expansion loc!"); 989 const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion(); 990 return Expansion.getExpansionLocRange(); 991 } 992 993 SourceLocation SourceManager::getTopMacroCallerLoc(SourceLocation Loc) const { 994 while (isMacroArgExpansion(Loc)) 995 Loc = getImmediateSpellingLoc(Loc); 996 return Loc; 997 } 998 999 /// getExpansionRange - Given a SourceLocation object, return the range of 1000 /// tokens covered by the expansion in the ultimate file. 1001 CharSourceRange SourceManager::getExpansionRange(SourceLocation Loc) const { 1002 if (Loc.isFileID()) 1003 return CharSourceRange(SourceRange(Loc, Loc), true); 1004 1005 CharSourceRange Res = getImmediateExpansionRange(Loc); 1006 1007 // Fully resolve the start and end locations to their ultimate expansion 1008 // points. 1009 while (!Res.getBegin().isFileID()) 1010 Res.setBegin(getImmediateExpansionRange(Res.getBegin()).getBegin()); 1011 while (!Res.getEnd().isFileID()) { 1012 CharSourceRange EndRange = getImmediateExpansionRange(Res.getEnd()); 1013 Res.setEnd(EndRange.getEnd()); 1014 Res.setTokenRange(EndRange.isTokenRange()); 1015 } 1016 return Res; 1017 } 1018 1019 bool SourceManager::isMacroArgExpansion(SourceLocation Loc, 1020 SourceLocation *StartLoc) const { 1021 if (!Loc.isMacroID()) return false; 1022 1023 FileID FID = getFileID(Loc); 1024 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1025 if (!Expansion.isMacroArgExpansion()) return false; 1026 1027 if (StartLoc) 1028 *StartLoc = Expansion.getExpansionLocStart(); 1029 return true; 1030 } 1031 1032 bool SourceManager::isMacroBodyExpansion(SourceLocation Loc) const { 1033 if (!Loc.isMacroID()) return false; 1034 1035 FileID FID = getFileID(Loc); 1036 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1037 return Expansion.isMacroBodyExpansion(); 1038 } 1039 1040 bool SourceManager::isAtStartOfImmediateMacroExpansion(SourceLocation Loc, 1041 SourceLocation *MacroBegin) const { 1042 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1043 1044 std::pair<FileID, unsigned> DecompLoc = getDecomposedLoc(Loc); 1045 if (DecompLoc.second > 0) 1046 return false; // Does not point at the start of expansion range. 1047 1048 bool Invalid = false; 1049 const SrcMgr::ExpansionInfo &ExpInfo = 1050 getSLocEntry(DecompLoc.first, &Invalid).getExpansion(); 1051 if (Invalid) 1052 return false; 1053 SourceLocation ExpLoc = ExpInfo.getExpansionLocStart(); 1054 1055 if (ExpInfo.isMacroArgExpansion()) { 1056 // For macro argument expansions, check if the previous FileID is part of 1057 // the same argument expansion, in which case this Loc is not at the 1058 // beginning of the expansion. 1059 FileID PrevFID = getPreviousFileID(DecompLoc.first); 1060 if (!PrevFID.isInvalid()) { 1061 const SrcMgr::SLocEntry &PrevEntry = getSLocEntry(PrevFID, &Invalid); 1062 if (Invalid) 1063 return false; 1064 if (PrevEntry.isExpansion() && 1065 PrevEntry.getExpansion().getExpansionLocStart() == ExpLoc) 1066 return false; 1067 } 1068 } 1069 1070 if (MacroBegin) 1071 *MacroBegin = ExpLoc; 1072 return true; 1073 } 1074 1075 bool SourceManager::isAtEndOfImmediateMacroExpansion(SourceLocation Loc, 1076 SourceLocation *MacroEnd) const { 1077 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1078 1079 FileID FID = getFileID(Loc); 1080 SourceLocation NextLoc = Loc.getLocWithOffset(1); 1081 if (isInFileID(NextLoc, FID)) 1082 return false; // Does not point at the end of expansion range. 1083 1084 bool Invalid = false; 1085 const SrcMgr::ExpansionInfo &ExpInfo = 1086 getSLocEntry(FID, &Invalid).getExpansion(); 1087 if (Invalid) 1088 return false; 1089 1090 if (ExpInfo.isMacroArgExpansion()) { 1091 // For macro argument expansions, check if the next FileID is part of the 1092 // same argument expansion, in which case this Loc is not at the end of the 1093 // expansion. 1094 FileID NextFID = getNextFileID(FID); 1095 if (!NextFID.isInvalid()) { 1096 const SrcMgr::SLocEntry &NextEntry = getSLocEntry(NextFID, &Invalid); 1097 if (Invalid) 1098 return false; 1099 if (NextEntry.isExpansion() && 1100 NextEntry.getExpansion().getExpansionLocStart() == 1101 ExpInfo.getExpansionLocStart()) 1102 return false; 1103 } 1104 } 1105 1106 if (MacroEnd) 1107 *MacroEnd = ExpInfo.getExpansionLocEnd(); 1108 return true; 1109 } 1110 1111 //===----------------------------------------------------------------------===// 1112 // Queries about the code at a SourceLocation. 1113 //===----------------------------------------------------------------------===// 1114 1115 /// getCharacterData - Return a pointer to the start of the specified location 1116 /// in the appropriate MemoryBuffer. 1117 const char *SourceManager::getCharacterData(SourceLocation SL, 1118 bool *Invalid) const { 1119 // Note that this is a hot function in the getSpelling() path, which is 1120 // heavily used by -E mode. 1121 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL); 1122 1123 // Note that calling 'getBuffer()' may lazily page in a source file. 1124 bool CharDataInvalid = false; 1125 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid); 1126 if (CharDataInvalid || !Entry.isFile()) { 1127 if (Invalid) 1128 *Invalid = true; 1129 1130 return "<<<<INVALID BUFFER>>>>"; 1131 } 1132 const llvm::MemoryBuffer *Buffer = 1133 Entry.getFile().getContentCache()->getBuffer( 1134 Diag, *this, SourceLocation(), &CharDataInvalid); 1135 if (Invalid) 1136 *Invalid = CharDataInvalid; 1137 return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second); 1138 } 1139 1140 /// getColumnNumber - Return the column # for the specified file position. 1141 /// this is significantly cheaper to compute than the line number. 1142 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos, 1143 bool *Invalid) const { 1144 bool MyInvalid = false; 1145 const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid); 1146 if (Invalid) 1147 *Invalid = MyInvalid; 1148 1149 if (MyInvalid) 1150 return 1; 1151 1152 // It is okay to request a position just past the end of the buffer. 1153 if (FilePos > MemBuf->getBufferSize()) { 1154 if (Invalid) 1155 *Invalid = true; 1156 return 1; 1157 } 1158 1159 const char *Buf = MemBuf->getBufferStart(); 1160 // See if we just calculated the line number for this FilePos and can use 1161 // that to lookup the start of the line instead of searching for it. 1162 if (LastLineNoFileIDQuery == FID && 1163 LastLineNoContentCache->SourceLineCache != nullptr && 1164 LastLineNoResult < LastLineNoContentCache->NumLines) { 1165 unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache; 1166 unsigned LineStart = SourceLineCache[LastLineNoResult - 1]; 1167 unsigned LineEnd = SourceLineCache[LastLineNoResult]; 1168 if (FilePos >= LineStart && FilePos < LineEnd) { 1169 // LineEnd is the LineStart of the next line. 1170 // A line ends with separator LF or CR+LF on Windows. 1171 // FilePos might point to the last separator, 1172 // but we need a column number at most 1 + the last column. 1173 if (FilePos + 1 == LineEnd && FilePos > LineStart) { 1174 if (Buf[FilePos - 1] == '\r' || Buf[FilePos - 1] == '\n') 1175 --FilePos; 1176 } 1177 return FilePos - LineStart + 1; 1178 } 1179 } 1180 1181 unsigned LineStart = FilePos; 1182 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r') 1183 --LineStart; 1184 return FilePos-LineStart+1; 1185 } 1186 1187 // isInvalid - Return the result of calling loc.isInvalid(), and 1188 // if Invalid is not null, set its value to same. 1189 template<typename LocType> 1190 static bool isInvalid(LocType Loc, bool *Invalid) { 1191 bool MyInvalid = Loc.isInvalid(); 1192 if (Invalid) 1193 *Invalid = MyInvalid; 1194 return MyInvalid; 1195 } 1196 1197 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc, 1198 bool *Invalid) const { 1199 if (isInvalid(Loc, Invalid)) return 0; 1200 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1201 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1202 } 1203 1204 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc, 1205 bool *Invalid) const { 1206 if (isInvalid(Loc, Invalid)) return 0; 1207 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1208 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1209 } 1210 1211 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc, 1212 bool *Invalid) const { 1213 PresumedLoc PLoc = getPresumedLoc(Loc); 1214 if (isInvalid(PLoc, Invalid)) return 0; 1215 return PLoc.getColumn(); 1216 } 1217 1218 #ifdef __SSE2__ 1219 #include <emmintrin.h> 1220 #endif 1221 1222 static LLVM_ATTRIBUTE_NOINLINE void 1223 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1224 llvm::BumpPtrAllocator &Alloc, 1225 const SourceManager &SM, bool &Invalid); 1226 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1227 llvm::BumpPtrAllocator &Alloc, 1228 const SourceManager &SM, bool &Invalid) { 1229 // Note that calling 'getBuffer()' may lazily page in the file. 1230 const MemoryBuffer *Buffer = 1231 FI->getBuffer(Diag, SM, SourceLocation(), &Invalid); 1232 if (Invalid) 1233 return; 1234 1235 // Find the file offsets of all of the *physical* source lines. This does 1236 // not look at trigraphs, escaped newlines, or anything else tricky. 1237 SmallVector<unsigned, 256> LineOffsets; 1238 1239 // Line #1 starts at char 0. 1240 LineOffsets.push_back(0); 1241 1242 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart(); 1243 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd(); 1244 unsigned I = 0; 1245 while (true) { 1246 // Skip over the contents of the line. 1247 while (Buf[I] != '\n' && Buf[I] != '\r' && Buf[I] != '\0') 1248 ++I; 1249 1250 if (Buf[I] == '\n' || Buf[I] == '\r') { 1251 // If this is \r\n, skip both characters. 1252 if (Buf[I] == '\r' && Buf[I+1] == '\n') 1253 ++I; 1254 ++I; 1255 LineOffsets.push_back(I); 1256 } else { 1257 // Otherwise, this is a NUL. If end of file, exit. 1258 if (Buf+I == End) break; 1259 ++I; 1260 } 1261 } 1262 1263 // Copy the offsets into the FileInfo structure. 1264 FI->NumLines = LineOffsets.size(); 1265 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size()); 1266 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache); 1267 } 1268 1269 /// getLineNumber - Given a SourceLocation, return the spelling line number 1270 /// for the position indicated. This requires building and caching a table of 1271 /// line offsets for the MemoryBuffer, so this is not cheap: use only when 1272 /// about to emit a diagnostic. 1273 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos, 1274 bool *Invalid) const { 1275 if (FID.isInvalid()) { 1276 if (Invalid) 1277 *Invalid = true; 1278 return 1; 1279 } 1280 1281 ContentCache *Content; 1282 if (LastLineNoFileIDQuery == FID) 1283 Content = LastLineNoContentCache; 1284 else { 1285 bool MyInvalid = false; 1286 const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid); 1287 if (MyInvalid || !Entry.isFile()) { 1288 if (Invalid) 1289 *Invalid = true; 1290 return 1; 1291 } 1292 1293 Content = const_cast<ContentCache*>(Entry.getFile().getContentCache()); 1294 } 1295 1296 // If this is the first use of line information for this buffer, compute the 1297 /// SourceLineCache for it on demand. 1298 if (!Content->SourceLineCache) { 1299 bool MyInvalid = false; 1300 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1301 if (Invalid) 1302 *Invalid = MyInvalid; 1303 if (MyInvalid) 1304 return 1; 1305 } else if (Invalid) 1306 *Invalid = false; 1307 1308 // Okay, we know we have a line number table. Do a binary search to find the 1309 // line number that this character position lands on. 1310 unsigned *SourceLineCache = Content->SourceLineCache; 1311 unsigned *SourceLineCacheStart = SourceLineCache; 1312 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines; 1313 1314 unsigned QueriedFilePos = FilePos+1; 1315 1316 // FIXME: I would like to be convinced that this code is worth being as 1317 // complicated as it is, binary search isn't that slow. 1318 // 1319 // If it is worth being optimized, then in my opinion it could be more 1320 // performant, simpler, and more obviously correct by just "galloping" outward 1321 // from the queried file position. In fact, this could be incorporated into a 1322 // generic algorithm such as lower_bound_with_hint. 1323 // 1324 // If someone gives me a test case where this matters, and I will do it! - DWD 1325 1326 // If the previous query was to the same file, we know both the file pos from 1327 // that query and the line number returned. This allows us to narrow the 1328 // search space from the entire file to something near the match. 1329 if (LastLineNoFileIDQuery == FID) { 1330 if (QueriedFilePos >= LastLineNoFilePos) { 1331 // FIXME: Potential overflow? 1332 SourceLineCache = SourceLineCache+LastLineNoResult-1; 1333 1334 // The query is likely to be nearby the previous one. Here we check to 1335 // see if it is within 5, 10 or 20 lines. It can be far away in cases 1336 // where big comment blocks and vertical whitespace eat up lines but 1337 // contribute no tokens. 1338 if (SourceLineCache+5 < SourceLineCacheEnd) { 1339 if (SourceLineCache[5] > QueriedFilePos) 1340 SourceLineCacheEnd = SourceLineCache+5; 1341 else if (SourceLineCache+10 < SourceLineCacheEnd) { 1342 if (SourceLineCache[10] > QueriedFilePos) 1343 SourceLineCacheEnd = SourceLineCache+10; 1344 else if (SourceLineCache+20 < SourceLineCacheEnd) { 1345 if (SourceLineCache[20] > QueriedFilePos) 1346 SourceLineCacheEnd = SourceLineCache+20; 1347 } 1348 } 1349 } 1350 } else { 1351 if (LastLineNoResult < Content->NumLines) 1352 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1; 1353 } 1354 } 1355 1356 unsigned *Pos 1357 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos); 1358 unsigned LineNo = Pos-SourceLineCacheStart; 1359 1360 LastLineNoFileIDQuery = FID; 1361 LastLineNoContentCache = Content; 1362 LastLineNoFilePos = QueriedFilePos; 1363 LastLineNoResult = LineNo; 1364 return LineNo; 1365 } 1366 1367 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc, 1368 bool *Invalid) const { 1369 if (isInvalid(Loc, Invalid)) return 0; 1370 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1371 return getLineNumber(LocInfo.first, LocInfo.second); 1372 } 1373 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc, 1374 bool *Invalid) const { 1375 if (isInvalid(Loc, Invalid)) return 0; 1376 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1377 return getLineNumber(LocInfo.first, LocInfo.second); 1378 } 1379 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc, 1380 bool *Invalid) const { 1381 PresumedLoc PLoc = getPresumedLoc(Loc); 1382 if (isInvalid(PLoc, Invalid)) return 0; 1383 return PLoc.getLine(); 1384 } 1385 1386 /// getFileCharacteristic - return the file characteristic of the specified 1387 /// source location, indicating whether this is a normal file, a system 1388 /// header, or an "implicit extern C" system header. 1389 /// 1390 /// This state can be modified with flags on GNU linemarker directives like: 1391 /// # 4 "foo.h" 3 1392 /// which changes all source locations in the current file after that to be 1393 /// considered to be from a system header. 1394 SrcMgr::CharacteristicKind 1395 SourceManager::getFileCharacteristic(SourceLocation Loc) const { 1396 assert(Loc.isValid() && "Can't get file characteristic of invalid loc!"); 1397 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1398 bool Invalid = false; 1399 const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid); 1400 if (Invalid || !SEntry.isFile()) 1401 return C_User; 1402 1403 const SrcMgr::FileInfo &FI = SEntry.getFile(); 1404 1405 // If there are no #line directives in this file, just return the whole-file 1406 // state. 1407 if (!FI.hasLineDirectives()) 1408 return FI.getFileCharacteristic(); 1409 1410 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1411 // See if there is a #line directive before the location. 1412 const LineEntry *Entry = 1413 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second); 1414 1415 // If this is before the first line marker, use the file characteristic. 1416 if (!Entry) 1417 return FI.getFileCharacteristic(); 1418 1419 return Entry->FileKind; 1420 } 1421 1422 /// Return the filename or buffer identifier of the buffer the location is in. 1423 /// Note that this name does not respect \#line directives. Use getPresumedLoc 1424 /// for normal clients. 1425 StringRef SourceManager::getBufferName(SourceLocation Loc, 1426 bool *Invalid) const { 1427 if (isInvalid(Loc, Invalid)) return "<invalid loc>"; 1428 1429 return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier(); 1430 } 1431 1432 /// getPresumedLoc - This method returns the "presumed" location of a 1433 /// SourceLocation specifies. A "presumed location" can be modified by \#line 1434 /// or GNU line marker directives. This provides a view on the data that a 1435 /// user should see in diagnostics, for example. 1436 /// 1437 /// Note that a presumed location is always given as the expansion point of an 1438 /// expansion location, not at the spelling location. 1439 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc, 1440 bool UseLineDirectives) const { 1441 if (Loc.isInvalid()) return PresumedLoc(); 1442 1443 // Presumed locations are always for expansion points. 1444 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1445 1446 bool Invalid = false; 1447 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1448 if (Invalid || !Entry.isFile()) 1449 return PresumedLoc(); 1450 1451 const SrcMgr::FileInfo &FI = Entry.getFile(); 1452 const SrcMgr::ContentCache *C = FI.getContentCache(); 1453 1454 // To get the source name, first consult the FileEntry (if one exists) 1455 // before the MemBuffer as this will avoid unnecessarily paging in the 1456 // MemBuffer. 1457 FileID FID = LocInfo.first; 1458 StringRef Filename; 1459 if (C->OrigEntry) 1460 Filename = C->OrigEntry->getName(); 1461 else 1462 Filename = C->getBuffer(Diag, *this)->getBufferIdentifier(); 1463 1464 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid); 1465 if (Invalid) 1466 return PresumedLoc(); 1467 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid); 1468 if (Invalid) 1469 return PresumedLoc(); 1470 1471 SourceLocation IncludeLoc = FI.getIncludeLoc(); 1472 1473 // If we have #line directives in this file, update and overwrite the physical 1474 // location info if appropriate. 1475 if (UseLineDirectives && FI.hasLineDirectives()) { 1476 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1477 // See if there is a #line directive before this. If so, get it. 1478 if (const LineEntry *Entry = 1479 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) { 1480 // If the LineEntry indicates a filename, use it. 1481 if (Entry->FilenameID != -1) { 1482 Filename = LineTable->getFilename(Entry->FilenameID); 1483 // The contents of files referenced by #line are not in the 1484 // SourceManager 1485 FID = FileID::get(0); 1486 } 1487 1488 // Use the line number specified by the LineEntry. This line number may 1489 // be multiple lines down from the line entry. Add the difference in 1490 // physical line numbers from the query point and the line marker to the 1491 // total. 1492 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset); 1493 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1); 1494 1495 // Note that column numbers are not molested by line markers. 1496 1497 // Handle virtual #include manipulation. 1498 if (Entry->IncludeOffset) { 1499 IncludeLoc = getLocForStartOfFile(LocInfo.first); 1500 IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset); 1501 } 1502 } 1503 } 1504 1505 return PresumedLoc(Filename.data(), FID, LineNo, ColNo, IncludeLoc); 1506 } 1507 1508 /// Returns whether the PresumedLoc for a given SourceLocation is 1509 /// in the main file. 1510 /// 1511 /// This computes the "presumed" location for a SourceLocation, then checks 1512 /// whether it came from a file other than the main file. This is different 1513 /// from isWrittenInMainFile() because it takes line marker directives into 1514 /// account. 1515 bool SourceManager::isInMainFile(SourceLocation Loc) const { 1516 if (Loc.isInvalid()) return false; 1517 1518 // Presumed locations are always for expansion points. 1519 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1520 1521 bool Invalid = false; 1522 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1523 if (Invalid || !Entry.isFile()) 1524 return false; 1525 1526 const SrcMgr::FileInfo &FI = Entry.getFile(); 1527 1528 // Check if there is a line directive for this location. 1529 if (FI.hasLineDirectives()) 1530 if (const LineEntry *Entry = 1531 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) 1532 if (Entry->IncludeOffset) 1533 return false; 1534 1535 return FI.getIncludeLoc().isInvalid(); 1536 } 1537 1538 /// The size of the SLocEntry that \p FID represents. 1539 unsigned SourceManager::getFileIDSize(FileID FID) const { 1540 bool Invalid = false; 1541 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1542 if (Invalid) 1543 return 0; 1544 1545 int ID = FID.ID; 1546 unsigned NextOffset; 1547 if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size())) 1548 NextOffset = getNextLocalOffset(); 1549 else if (ID+1 == -1) 1550 NextOffset = MaxLoadedOffset; 1551 else 1552 NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset(); 1553 1554 return NextOffset - Entry.getOffset() - 1; 1555 } 1556 1557 //===----------------------------------------------------------------------===// 1558 // Other miscellaneous methods. 1559 //===----------------------------------------------------------------------===// 1560 1561 /// Retrieve the inode for the given file entry, if possible. 1562 /// 1563 /// This routine involves a system call, and therefore should only be used 1564 /// in non-performance-critical code. 1565 static Optional<llvm::sys::fs::UniqueID> 1566 getActualFileUID(const FileEntry *File) { 1567 if (!File) 1568 return None; 1569 1570 llvm::sys::fs::UniqueID ID; 1571 if (llvm::sys::fs::getUniqueID(File->getName(), ID)) 1572 return None; 1573 1574 return ID; 1575 } 1576 1577 /// Get the source location for the given file:line:col triplet. 1578 /// 1579 /// If the source file is included multiple times, the source location will 1580 /// be based upon an arbitrary inclusion. 1581 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile, 1582 unsigned Line, 1583 unsigned Col) const { 1584 assert(SourceFile && "Null source file!"); 1585 assert(Line && Col && "Line and column should start from 1!"); 1586 1587 FileID FirstFID = translateFile(SourceFile); 1588 return translateLineCol(FirstFID, Line, Col); 1589 } 1590 1591 /// Get the FileID for the given file. 1592 /// 1593 /// If the source file is included multiple times, the FileID will be the 1594 /// first inclusion. 1595 FileID SourceManager::translateFile(const FileEntry *SourceFile) const { 1596 assert(SourceFile && "Null source file!"); 1597 1598 // Find the first file ID that corresponds to the given file. 1599 FileID FirstFID; 1600 1601 // First, check the main file ID, since it is common to look for a 1602 // location in the main file. 1603 Optional<llvm::sys::fs::UniqueID> SourceFileUID; 1604 Optional<StringRef> SourceFileName; 1605 if (MainFileID.isValid()) { 1606 bool Invalid = false; 1607 const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid); 1608 if (Invalid) 1609 return FileID(); 1610 1611 if (MainSLoc.isFile()) { 1612 const ContentCache *MainContentCache 1613 = MainSLoc.getFile().getContentCache(); 1614 if (!MainContentCache || !MainContentCache->OrigEntry) { 1615 // Can't do anything 1616 } else if (MainContentCache->OrigEntry == SourceFile) { 1617 FirstFID = MainFileID; 1618 } else { 1619 // Fall back: check whether we have the same base name and inode 1620 // as the main file. 1621 const FileEntry *MainFile = MainContentCache->OrigEntry; 1622 SourceFileName = llvm::sys::path::filename(SourceFile->getName()); 1623 if (*SourceFileName == llvm::sys::path::filename(MainFile->getName())) { 1624 SourceFileUID = getActualFileUID(SourceFile); 1625 if (SourceFileUID) { 1626 if (Optional<llvm::sys::fs::UniqueID> MainFileUID = 1627 getActualFileUID(MainFile)) { 1628 if (*SourceFileUID == *MainFileUID) { 1629 FirstFID = MainFileID; 1630 SourceFile = MainFile; 1631 } 1632 } 1633 } 1634 } 1635 } 1636 } 1637 } 1638 1639 if (FirstFID.isInvalid()) { 1640 // The location we're looking for isn't in the main file; look 1641 // through all of the local source locations. 1642 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) { 1643 bool Invalid = false; 1644 const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid); 1645 if (Invalid) 1646 return FileID(); 1647 1648 if (SLoc.isFile() && 1649 SLoc.getFile().getContentCache() && 1650 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) { 1651 FirstFID = FileID::get(I); 1652 break; 1653 } 1654 } 1655 // If that still didn't help, try the modules. 1656 if (FirstFID.isInvalid()) { 1657 for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) { 1658 const SLocEntry &SLoc = getLoadedSLocEntry(I); 1659 if (SLoc.isFile() && 1660 SLoc.getFile().getContentCache() && 1661 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) { 1662 FirstFID = FileID::get(-int(I) - 2); 1663 break; 1664 } 1665 } 1666 } 1667 } 1668 1669 // If we haven't found what we want yet, try again, but this time stat() 1670 // each of the files in case the files have changed since we originally 1671 // parsed the file. 1672 if (FirstFID.isInvalid() && 1673 (SourceFileName || 1674 (SourceFileName = llvm::sys::path::filename(SourceFile->getName()))) && 1675 (SourceFileUID || (SourceFileUID = getActualFileUID(SourceFile)))) { 1676 bool Invalid = false; 1677 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) { 1678 FileID IFileID; 1679 IFileID.ID = I; 1680 const SLocEntry &SLoc = getSLocEntry(IFileID, &Invalid); 1681 if (Invalid) 1682 return FileID(); 1683 1684 if (SLoc.isFile()) { 1685 const ContentCache *FileContentCache 1686 = SLoc.getFile().getContentCache(); 1687 const FileEntry *Entry = FileContentCache ? FileContentCache->OrigEntry 1688 : nullptr; 1689 if (Entry && 1690 *SourceFileName == llvm::sys::path::filename(Entry->getName())) { 1691 if (Optional<llvm::sys::fs::UniqueID> EntryUID = 1692 getActualFileUID(Entry)) { 1693 if (*SourceFileUID == *EntryUID) { 1694 FirstFID = FileID::get(I); 1695 SourceFile = Entry; 1696 break; 1697 } 1698 } 1699 } 1700 } 1701 } 1702 } 1703 1704 (void) SourceFile; 1705 return FirstFID; 1706 } 1707 1708 /// Get the source location in \arg FID for the given line:col. 1709 /// Returns null location if \arg FID is not a file SLocEntry. 1710 SourceLocation SourceManager::translateLineCol(FileID FID, 1711 unsigned Line, 1712 unsigned Col) const { 1713 // Lines are used as a one-based index into a zero-based array. This assert 1714 // checks for possible buffer underruns. 1715 assert(Line && Col && "Line and column should start from 1!"); 1716 1717 if (FID.isInvalid()) 1718 return SourceLocation(); 1719 1720 bool Invalid = false; 1721 const SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1722 if (Invalid) 1723 return SourceLocation(); 1724 1725 if (!Entry.isFile()) 1726 return SourceLocation(); 1727 1728 SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset()); 1729 1730 if (Line == 1 && Col == 1) 1731 return FileLoc; 1732 1733 ContentCache *Content 1734 = const_cast<ContentCache *>(Entry.getFile().getContentCache()); 1735 if (!Content) 1736 return SourceLocation(); 1737 1738 // If this is the first use of line information for this buffer, compute the 1739 // SourceLineCache for it on demand. 1740 if (!Content->SourceLineCache) { 1741 bool MyInvalid = false; 1742 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1743 if (MyInvalid) 1744 return SourceLocation(); 1745 } 1746 1747 if (Line > Content->NumLines) { 1748 unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize(); 1749 if (Size > 0) 1750 --Size; 1751 return FileLoc.getLocWithOffset(Size); 1752 } 1753 1754 const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, *this); 1755 unsigned FilePos = Content->SourceLineCache[Line - 1]; 1756 const char *Buf = Buffer->getBufferStart() + FilePos; 1757 unsigned BufLength = Buffer->getBufferSize() - FilePos; 1758 if (BufLength == 0) 1759 return FileLoc.getLocWithOffset(FilePos); 1760 1761 unsigned i = 0; 1762 1763 // Check that the given column is valid. 1764 while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r') 1765 ++i; 1766 return FileLoc.getLocWithOffset(FilePos + i); 1767 } 1768 1769 /// Compute a map of macro argument chunks to their expanded source 1770 /// location. Chunks that are not part of a macro argument will map to an 1771 /// invalid source location. e.g. if a file contains one macro argument at 1772 /// offset 100 with length 10, this is how the map will be formed: 1773 /// 0 -> SourceLocation() 1774 /// 100 -> Expanded macro arg location 1775 /// 110 -> SourceLocation() 1776 void SourceManager::computeMacroArgsCache(MacroArgsMap &MacroArgsCache, 1777 FileID FID) const { 1778 assert(FID.isValid()); 1779 1780 // Initially no macro argument chunk is present. 1781 MacroArgsCache.insert(std::make_pair(0, SourceLocation())); 1782 1783 int ID = FID.ID; 1784 while (true) { 1785 ++ID; 1786 // Stop if there are no more FileIDs to check. 1787 if (ID > 0) { 1788 if (unsigned(ID) >= local_sloc_entry_size()) 1789 return; 1790 } else if (ID == -1) { 1791 return; 1792 } 1793 1794 bool Invalid = false; 1795 const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID, &Invalid); 1796 if (Invalid) 1797 return; 1798 if (Entry.isFile()) { 1799 SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc(); 1800 if (IncludeLoc.isInvalid()) 1801 continue; 1802 if (!isInFileID(IncludeLoc, FID)) 1803 return; // No more files/macros that may be "contained" in this file. 1804 1805 // Skip the files/macros of the #include'd file, we only care about macros 1806 // that lexed macro arguments from our file. 1807 if (Entry.getFile().NumCreatedFIDs) 1808 ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/; 1809 continue; 1810 } 1811 1812 const ExpansionInfo &ExpInfo = Entry.getExpansion(); 1813 1814 if (ExpInfo.getExpansionLocStart().isFileID()) { 1815 if (!isInFileID(ExpInfo.getExpansionLocStart(), FID)) 1816 return; // No more files/macros that may be "contained" in this file. 1817 } 1818 1819 if (!ExpInfo.isMacroArgExpansion()) 1820 continue; 1821 1822 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1823 ExpInfo.getSpellingLoc(), 1824 SourceLocation::getMacroLoc(Entry.getOffset()), 1825 getFileIDSize(FileID::get(ID))); 1826 } 1827 } 1828 1829 void SourceManager::associateFileChunkWithMacroArgExp( 1830 MacroArgsMap &MacroArgsCache, 1831 FileID FID, 1832 SourceLocation SpellLoc, 1833 SourceLocation ExpansionLoc, 1834 unsigned ExpansionLength) const { 1835 if (!SpellLoc.isFileID()) { 1836 unsigned SpellBeginOffs = SpellLoc.getOffset(); 1837 unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength; 1838 1839 // The spelling range for this macro argument expansion can span multiple 1840 // consecutive FileID entries. Go through each entry contained in the 1841 // spelling range and if one is itself a macro argument expansion, recurse 1842 // and associate the file chunk that it represents. 1843 1844 FileID SpellFID; // Current FileID in the spelling range. 1845 unsigned SpellRelativeOffs; 1846 std::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc); 1847 while (true) { 1848 const SLocEntry &Entry = getSLocEntry(SpellFID); 1849 unsigned SpellFIDBeginOffs = Entry.getOffset(); 1850 unsigned SpellFIDSize = getFileIDSize(SpellFID); 1851 unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize; 1852 const ExpansionInfo &Info = Entry.getExpansion(); 1853 if (Info.isMacroArgExpansion()) { 1854 unsigned CurrSpellLength; 1855 if (SpellFIDEndOffs < SpellEndOffs) 1856 CurrSpellLength = SpellFIDSize - SpellRelativeOffs; 1857 else 1858 CurrSpellLength = ExpansionLength; 1859 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1860 Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs), 1861 ExpansionLoc, CurrSpellLength); 1862 } 1863 1864 if (SpellFIDEndOffs >= SpellEndOffs) 1865 return; // we covered all FileID entries in the spelling range. 1866 1867 // Move to the next FileID entry in the spelling range. 1868 unsigned advance = SpellFIDSize - SpellRelativeOffs + 1; 1869 ExpansionLoc = ExpansionLoc.getLocWithOffset(advance); 1870 ExpansionLength -= advance; 1871 ++SpellFID.ID; 1872 SpellRelativeOffs = 0; 1873 } 1874 } 1875 1876 assert(SpellLoc.isFileID()); 1877 1878 unsigned BeginOffs; 1879 if (!isInFileID(SpellLoc, FID, &BeginOffs)) 1880 return; 1881 1882 unsigned EndOffs = BeginOffs + ExpansionLength; 1883 1884 // Add a new chunk for this macro argument. A previous macro argument chunk 1885 // may have been lexed again, so e.g. if the map is 1886 // 0 -> SourceLocation() 1887 // 100 -> Expanded loc #1 1888 // 110 -> SourceLocation() 1889 // and we found a new macro FileID that lexed from offset 105 with length 3, 1890 // the new map will be: 1891 // 0 -> SourceLocation() 1892 // 100 -> Expanded loc #1 1893 // 105 -> Expanded loc #2 1894 // 108 -> Expanded loc #1 1895 // 110 -> SourceLocation() 1896 // 1897 // Since re-lexed macro chunks will always be the same size or less of 1898 // previous chunks, we only need to find where the ending of the new macro 1899 // chunk is mapped to and update the map with new begin/end mappings. 1900 1901 MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs); 1902 --I; 1903 SourceLocation EndOffsMappedLoc = I->second; 1904 MacroArgsCache[BeginOffs] = ExpansionLoc; 1905 MacroArgsCache[EndOffs] = EndOffsMappedLoc; 1906 } 1907 1908 /// If \arg Loc points inside a function macro argument, the returned 1909 /// location will be the macro location in which the argument was expanded. 1910 /// If a macro argument is used multiple times, the expanded location will 1911 /// be at the first expansion of the argument. 1912 /// e.g. 1913 /// MY_MACRO(foo); 1914 /// ^ 1915 /// Passing a file location pointing at 'foo', will yield a macro location 1916 /// where 'foo' was expanded into. 1917 SourceLocation 1918 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const { 1919 if (Loc.isInvalid() || !Loc.isFileID()) 1920 return Loc; 1921 1922 FileID FID; 1923 unsigned Offset; 1924 std::tie(FID, Offset) = getDecomposedLoc(Loc); 1925 if (FID.isInvalid()) 1926 return Loc; 1927 1928 std::unique_ptr<MacroArgsMap> &MacroArgsCache = MacroArgsCacheMap[FID]; 1929 if (!MacroArgsCache) { 1930 MacroArgsCache = llvm::make_unique<MacroArgsMap>(); 1931 computeMacroArgsCache(*MacroArgsCache, FID); 1932 } 1933 1934 assert(!MacroArgsCache->empty()); 1935 MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset); 1936 --I; 1937 1938 unsigned MacroArgBeginOffs = I->first; 1939 SourceLocation MacroArgExpandedLoc = I->second; 1940 if (MacroArgExpandedLoc.isValid()) 1941 return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs); 1942 1943 return Loc; 1944 } 1945 1946 std::pair<FileID, unsigned> 1947 SourceManager::getDecomposedIncludedLoc(FileID FID) const { 1948 if (FID.isInvalid()) 1949 return std::make_pair(FileID(), 0); 1950 1951 // Uses IncludedLocMap to retrieve/cache the decomposed loc. 1952 1953 using DecompTy = std::pair<FileID, unsigned>; 1954 auto InsertOp = IncludedLocMap.try_emplace(FID); 1955 DecompTy &DecompLoc = InsertOp.first->second; 1956 if (!InsertOp.second) 1957 return DecompLoc; // already in map. 1958 1959 SourceLocation UpperLoc; 1960 bool Invalid = false; 1961 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1962 if (!Invalid) { 1963 if (Entry.isExpansion()) 1964 UpperLoc = Entry.getExpansion().getExpansionLocStart(); 1965 else 1966 UpperLoc = Entry.getFile().getIncludeLoc(); 1967 } 1968 1969 if (UpperLoc.isValid()) 1970 DecompLoc = getDecomposedLoc(UpperLoc); 1971 1972 return DecompLoc; 1973 } 1974 1975 /// Given a decomposed source location, move it up the include/expansion stack 1976 /// to the parent source location. If this is possible, return the decomposed 1977 /// version of the parent in Loc and return false. If Loc is the top-level 1978 /// entry, return true and don't modify it. 1979 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc, 1980 const SourceManager &SM) { 1981 std::pair<FileID, unsigned> UpperLoc = SM.getDecomposedIncludedLoc(Loc.first); 1982 if (UpperLoc.first.isInvalid()) 1983 return true; // We reached the top. 1984 1985 Loc = UpperLoc; 1986 return false; 1987 } 1988 1989 /// Return the cache entry for comparing the given file IDs 1990 /// for isBeforeInTranslationUnit. 1991 InBeforeInTUCacheEntry &SourceManager::getInBeforeInTUCache(FileID LFID, 1992 FileID RFID) const { 1993 // This is a magic number for limiting the cache size. It was experimentally 1994 // derived from a small Objective-C project (where the cache filled 1995 // out to ~250 items). We can make it larger if necessary. 1996 enum { MagicCacheSize = 300 }; 1997 IsBeforeInTUCacheKey Key(LFID, RFID); 1998 1999 // If the cache size isn't too large, do a lookup and if necessary default 2000 // construct an entry. We can then return it to the caller for direct 2001 // use. When they update the value, the cache will get automatically 2002 // updated as well. 2003 if (IBTUCache.size() < MagicCacheSize) 2004 return IBTUCache[Key]; 2005 2006 // Otherwise, do a lookup that will not construct a new value. 2007 InBeforeInTUCache::iterator I = IBTUCache.find(Key); 2008 if (I != IBTUCache.end()) 2009 return I->second; 2010 2011 // Fall back to the overflow value. 2012 return IBTUCacheOverflow; 2013 } 2014 2015 /// Determines the order of 2 source locations in the translation unit. 2016 /// 2017 /// \returns true if LHS source location comes before RHS, false otherwise. 2018 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS, 2019 SourceLocation RHS) const { 2020 assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!"); 2021 if (LHS == RHS) 2022 return false; 2023 2024 std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS); 2025 std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS); 2026 2027 // getDecomposedLoc may have failed to return a valid FileID because, e.g. it 2028 // is a serialized one referring to a file that was removed after we loaded 2029 // the PCH. 2030 if (LOffs.first.isInvalid() || ROffs.first.isInvalid()) 2031 return LOffs.first.isInvalid() && !ROffs.first.isInvalid(); 2032 2033 std::pair<bool, bool> InSameTU = isInTheSameTranslationUnit(LOffs, ROffs); 2034 if (InSameTU.first) 2035 return InSameTU.second; 2036 2037 // If we arrived here, the location is either in a built-ins buffer or 2038 // associated with global inline asm. PR5662 and PR22576 are examples. 2039 2040 StringRef LB = getBuffer(LOffs.first)->getBufferIdentifier(); 2041 StringRef RB = getBuffer(ROffs.first)->getBufferIdentifier(); 2042 bool LIsBuiltins = LB == "<built-in>"; 2043 bool RIsBuiltins = RB == "<built-in>"; 2044 // Sort built-in before non-built-in. 2045 if (LIsBuiltins || RIsBuiltins) { 2046 if (LIsBuiltins != RIsBuiltins) 2047 return LIsBuiltins; 2048 // Both are in built-in buffers, but from different files. We just claim that 2049 // lower IDs come first. 2050 return LOffs.first < ROffs.first; 2051 } 2052 bool LIsAsm = LB == "<inline asm>"; 2053 bool RIsAsm = RB == "<inline asm>"; 2054 // Sort assembler after built-ins, but before the rest. 2055 if (LIsAsm || RIsAsm) { 2056 if (LIsAsm != RIsAsm) 2057 return RIsAsm; 2058 assert(LOffs.first == ROffs.first); 2059 return false; 2060 } 2061 bool LIsScratch = LB == "<scratch space>"; 2062 bool RIsScratch = RB == "<scratch space>"; 2063 // Sort scratch after inline asm, but before the rest. 2064 if (LIsScratch || RIsScratch) { 2065 if (LIsScratch != RIsScratch) 2066 return LIsScratch; 2067 return LOffs.second < ROffs.second; 2068 } 2069 llvm_unreachable("Unsortable locations found"); 2070 } 2071 2072 std::pair<bool, bool> SourceManager::isInTheSameTranslationUnit( 2073 std::pair<FileID, unsigned> &LOffs, 2074 std::pair<FileID, unsigned> &ROffs) const { 2075 // If the source locations are in the same file, just compare offsets. 2076 if (LOffs.first == ROffs.first) 2077 return std::make_pair(true, LOffs.second < ROffs.second); 2078 2079 // If we are comparing a source location with multiple locations in the same 2080 // file, we get a big win by caching the result. 2081 InBeforeInTUCacheEntry &IsBeforeInTUCache = 2082 getInBeforeInTUCache(LOffs.first, ROffs.first); 2083 2084 // If we are comparing a source location with multiple locations in the same 2085 // file, we get a big win by caching the result. 2086 if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first)) 2087 return std::make_pair( 2088 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second)); 2089 2090 // Okay, we missed in the cache, start updating the cache for this query. 2091 IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first, 2092 /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID); 2093 2094 // We need to find the common ancestor. The only way of doing this is to 2095 // build the complete include chain for one and then walking up the chain 2096 // of the other looking for a match. 2097 // We use a map from FileID to Offset to store the chain. Easier than writing 2098 // a custom set hash info that only depends on the first part of a pair. 2099 using LocSet = llvm::SmallDenseMap<FileID, unsigned, 16>; 2100 LocSet LChain; 2101 do { 2102 LChain.insert(LOffs); 2103 // We catch the case where LOffs is in a file included by ROffs and 2104 // quit early. The other way round unfortunately remains suboptimal. 2105 } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this)); 2106 LocSet::iterator I; 2107 while((I = LChain.find(ROffs.first)) == LChain.end()) { 2108 if (MoveUpIncludeHierarchy(ROffs, *this)) 2109 break; // Met at topmost file. 2110 } 2111 if (I != LChain.end()) 2112 LOffs = *I; 2113 2114 // If we exited because we found a nearest common ancestor, compare the 2115 // locations within the common file and cache them. 2116 if (LOffs.first == ROffs.first) { 2117 IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second); 2118 return std::make_pair( 2119 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second)); 2120 } 2121 // Clear the lookup cache, it depends on a common location. 2122 IsBeforeInTUCache.clear(); 2123 return std::make_pair(false, false); 2124 } 2125 2126 void SourceManager::PrintStats() const { 2127 llvm::errs() << "\n*** Source Manager Stats:\n"; 2128 llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size() 2129 << " mem buffers mapped.\n"; 2130 llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated (" 2131 << llvm::capacity_in_bytes(LocalSLocEntryTable) 2132 << " bytes of capacity), " 2133 << NextLocalOffset << "B of Sloc address space used.\n"; 2134 llvm::errs() << LoadedSLocEntryTable.size() 2135 << " loaded SLocEntries allocated, " 2136 << MaxLoadedOffset - CurrentLoadedOffset 2137 << "B of Sloc address space used.\n"; 2138 2139 unsigned NumLineNumsComputed = 0; 2140 unsigned NumFileBytesMapped = 0; 2141 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){ 2142 NumLineNumsComputed += I->second->SourceLineCache != nullptr; 2143 NumFileBytesMapped += I->second->getSizeBytesMapped(); 2144 } 2145 unsigned NumMacroArgsComputed = MacroArgsCacheMap.size(); 2146 2147 llvm::errs() << NumFileBytesMapped << " bytes of files mapped, " 2148 << NumLineNumsComputed << " files with line #'s computed, " 2149 << NumMacroArgsComputed << " files with macro args computed.\n"; 2150 llvm::errs() << "FileID scans: " << NumLinearScans << " linear, " 2151 << NumBinaryProbes << " binary.\n"; 2152 } 2153 2154 LLVM_DUMP_METHOD void SourceManager::dump() const { 2155 llvm::raw_ostream &out = llvm::errs(); 2156 2157 auto DumpSLocEntry = [&](int ID, const SrcMgr::SLocEntry &Entry, 2158 llvm::Optional<unsigned> NextStart) { 2159 out << "SLocEntry <FileID " << ID << "> " << (Entry.isFile() ? "file" : "expansion") 2160 << " <SourceLocation " << Entry.getOffset() << ":"; 2161 if (NextStart) 2162 out << *NextStart << ">\n"; 2163 else 2164 out << "???\?>\n"; 2165 if (Entry.isFile()) { 2166 auto &FI = Entry.getFile(); 2167 if (FI.NumCreatedFIDs) 2168 out << " covers <FileID " << ID << ":" << int(ID + FI.NumCreatedFIDs) 2169 << ">\n"; 2170 if (FI.getIncludeLoc().isValid()) 2171 out << " included from " << FI.getIncludeLoc().getOffset() << "\n"; 2172 if (auto *CC = FI.getContentCache()) { 2173 out << " for " << (CC->OrigEntry ? CC->OrigEntry->getName() : "<none>") 2174 << "\n"; 2175 if (CC->BufferOverridden) 2176 out << " contents overridden\n"; 2177 if (CC->ContentsEntry != CC->OrigEntry) { 2178 out << " contents from " 2179 << (CC->ContentsEntry ? CC->ContentsEntry->getName() : "<none>") 2180 << "\n"; 2181 } 2182 } 2183 } else { 2184 auto &EI = Entry.getExpansion(); 2185 out << " spelling from " << EI.getSpellingLoc().getOffset() << "\n"; 2186 out << " macro " << (EI.isMacroArgExpansion() ? "arg" : "body") 2187 << " range <" << EI.getExpansionLocStart().getOffset() << ":" 2188 << EI.getExpansionLocEnd().getOffset() << ">\n"; 2189 } 2190 }; 2191 2192 // Dump local SLocEntries. 2193 for (unsigned ID = 0, NumIDs = LocalSLocEntryTable.size(); ID != NumIDs; ++ID) { 2194 DumpSLocEntry(ID, LocalSLocEntryTable[ID], 2195 ID == NumIDs - 1 ? NextLocalOffset 2196 : LocalSLocEntryTable[ID + 1].getOffset()); 2197 } 2198 // Dump loaded SLocEntries. 2199 llvm::Optional<unsigned> NextStart; 2200 for (unsigned Index = 0; Index != LoadedSLocEntryTable.size(); ++Index) { 2201 int ID = -(int)Index - 2; 2202 if (SLocEntryLoaded[Index]) { 2203 DumpSLocEntry(ID, LoadedSLocEntryTable[Index], NextStart); 2204 NextStart = LoadedSLocEntryTable[Index].getOffset(); 2205 } else { 2206 NextStart = None; 2207 } 2208 } 2209 } 2210 2211 ExternalSLocEntrySource::~ExternalSLocEntrySource() = default; 2212 2213 /// Return the amount of memory used by memory buffers, breaking down 2214 /// by heap-backed versus mmap'ed memory. 2215 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const { 2216 size_t malloc_bytes = 0; 2217 size_t mmap_bytes = 0; 2218 2219 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) 2220 if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped()) 2221 switch (MemBufferInfos[i]->getMemoryBufferKind()) { 2222 case llvm::MemoryBuffer::MemoryBuffer_MMap: 2223 mmap_bytes += sized_mapped; 2224 break; 2225 case llvm::MemoryBuffer::MemoryBuffer_Malloc: 2226 malloc_bytes += sized_mapped; 2227 break; 2228 } 2229 2230 return MemoryBufferSizes(malloc_bytes, mmap_bytes); 2231 } 2232 2233 size_t SourceManager::getDataStructureSizes() const { 2234 size_t size = llvm::capacity_in_bytes(MemBufferInfos) 2235 + llvm::capacity_in_bytes(LocalSLocEntryTable) 2236 + llvm::capacity_in_bytes(LoadedSLocEntryTable) 2237 + llvm::capacity_in_bytes(SLocEntryLoaded) 2238 + llvm::capacity_in_bytes(FileInfos); 2239 2240 if (OverriddenFilesInfo) 2241 size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles); 2242 2243 return size; 2244 } 2245 2246 SourceManagerForFile::SourceManagerForFile(StringRef FileName, 2247 StringRef Content) { 2248 // This is referenced by `FileMgr` and will be released by `FileMgr` when it 2249 // is deleted. 2250 IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> InMemoryFileSystem( 2251 new llvm::vfs::InMemoryFileSystem); 2252 InMemoryFileSystem->addFile( 2253 FileName, 0, 2254 llvm::MemoryBuffer::getMemBuffer(Content, FileName, 2255 /*RequiresNullTerminator=*/false)); 2256 // This is passed to `SM` as reference, so the pointer has to be referenced 2257 // in `Environment` so that `FileMgr` can out-live this function scope. 2258 FileMgr = 2259 llvm::make_unique<FileManager>(FileSystemOptions(), InMemoryFileSystem); 2260 // This is passed to `SM` as reference, so the pointer has to be referenced 2261 // by `Environment` due to the same reason above. 2262 Diagnostics = llvm::make_unique<DiagnosticsEngine>( 2263 IntrusiveRefCntPtr<DiagnosticIDs>(new DiagnosticIDs), 2264 new DiagnosticOptions); 2265 SourceMgr = llvm::make_unique<SourceManager>(*Diagnostics, *FileMgr); 2266 FileID ID = SourceMgr->createFileID(FileMgr->getFile(FileName), 2267 SourceLocation(), clang::SrcMgr::C_User); 2268 assert(ID.isValid()); 2269 SourceMgr->setMainFileID(ID); 2270 } 2271