SourceManager.cpp revision b088cd3138b1fdb286c51e982ddda1e8af9d7e86
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/SourceManagerInternals.h" 16#include "clang/Basic/Diagnostic.h" 17#include "clang/Basic/FileManager.h" 18#include "llvm/ADT/StringSwitch.h" 19#include "llvm/Support/Compiler.h" 20#include "llvm/Support/MemoryBuffer.h" 21#include "llvm/Support/raw_ostream.h" 22#include "llvm/System/Path.h" 23#include <algorithm> 24#include <string> 25#include <cstring> 26 27using namespace clang; 28using namespace SrcMgr; 29using llvm::MemoryBuffer; 30 31//===----------------------------------------------------------------------===// 32// SourceManager Helper Classes 33//===----------------------------------------------------------------------===// 34 35ContentCache::~ContentCache() { 36 if (shouldFreeBuffer()) 37 delete Buffer.getPointer(); 38} 39 40/// getSizeBytesMapped - Returns the number of bytes actually mapped for 41/// this ContentCache. This can be 0 if the MemBuffer was not actually 42/// instantiated. 43unsigned ContentCache::getSizeBytesMapped() const { 44 return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0; 45} 46 47/// getSize - Returns the size of the content encapsulated by this ContentCache. 48/// This can be the size of the source file or the size of an arbitrary 49/// scratch buffer. If the ContentCache encapsulates a source file, that 50/// file is not lazily brought in from disk to satisfy this query. 51unsigned ContentCache::getSize() const { 52 return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize() 53 : (unsigned) Entry->getSize(); 54} 55 56void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B, 57 bool DoNotFree) { 58 assert(B != Buffer.getPointer()); 59 60 if (shouldFreeBuffer()) 61 delete Buffer.getPointer(); 62 Buffer.setPointer(B); 63 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0); 64} 65 66const llvm::MemoryBuffer *ContentCache::getBuffer(Diagnostic &Diag, 67 const SourceManager &SM, 68 SourceLocation Loc, 69 bool *Invalid) const { 70 // Lazily create the Buffer for ContentCaches that wrap files. If we already 71 // computed it, jsut return what we have. 72 if (Buffer.getPointer() || Entry == 0) { 73 if (Invalid) 74 *Invalid = isBufferInvalid(); 75 76 return Buffer.getPointer(); 77 } 78 79 std::string ErrorStr; 80 Buffer.setPointer(SM.getFileManager().getBufferForFile(Entry, &ErrorStr)); 81 82 // If we were unable to open the file, then we are in an inconsistent 83 // situation where the content cache referenced a file which no longer 84 // exists. Most likely, we were using a stat cache with an invalid entry but 85 // the file could also have been removed during processing. Since we can't 86 // really deal with this situation, just create an empty buffer. 87 // 88 // FIXME: This is definitely not ideal, but our immediate clients can't 89 // currently handle returning a null entry here. Ideally we should detect 90 // that we are in an inconsistent situation and error out as quickly as 91 // possible. 92 if (!Buffer.getPointer()) { 93 const llvm::StringRef FillStr("<<<MISSING SOURCE FILE>>>\n"); 94 Buffer.setPointer(MemoryBuffer::getNewMemBuffer(Entry->getSize(), 95 "<invalid>")); 96 char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart()); 97 for (unsigned i = 0, e = Entry->getSize(); i != e; ++i) 98 Ptr[i] = FillStr[i % FillStr.size()]; 99 100 if (Diag.isDiagnosticInFlight()) 101 Diag.SetDelayedDiagnostic(diag::err_cannot_open_file, 102 Entry->getName(), ErrorStr); 103 else 104 Diag.Report(Loc, diag::err_cannot_open_file) 105 << Entry->getName() << ErrorStr; 106 107 Buffer.setInt(Buffer.getInt() | InvalidFlag); 108 109 if (Invalid) *Invalid = true; 110 return Buffer.getPointer(); 111 } 112 113 // Check that the file's size is the same as in the file entry (which may 114 // have come from a stat cache). 115 if (getRawBuffer()->getBufferSize() != (size_t)Entry->getSize()) { 116 if (Diag.isDiagnosticInFlight()) 117 Diag.SetDelayedDiagnostic(diag::err_file_modified, 118 Entry->getName()); 119 else 120 Diag.Report(Loc, diag::err_file_modified) 121 << Entry->getName(); 122 123 Buffer.setInt(Buffer.getInt() | InvalidFlag); 124 if (Invalid) *Invalid = true; 125 return Buffer.getPointer(); 126 } 127 128 // If the buffer is valid, check to see if it has a UTF Byte Order Mark 129 // (BOM). We only support UTF-8 without a BOM right now. See 130 // http://en.wikipedia.org/wiki/Byte_order_mark for more information. 131 llvm::StringRef BufStr = Buffer.getPointer()->getBuffer(); 132 const char *BOM = llvm::StringSwitch<const char *>(BufStr) 133 .StartsWith("\xEF\xBB\xBF", "UTF-8") 134 .StartsWith("\xFE\xFF", "UTF-16 (BE)") 135 .StartsWith("\xFF\xFE", "UTF-16 (LE)") 136 .StartsWith("\x00\x00\xFE\xFF", "UTF-32 (BE)") 137 .StartsWith("\xFF\xFE\x00\x00", "UTF-32 (LE)") 138 .StartsWith("\x2B\x2F\x76", "UTF-7") 139 .StartsWith("\xF7\x64\x4C", "UTF-1") 140 .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC") 141 .StartsWith("\x0E\xFE\xFF", "SDSU") 142 .StartsWith("\xFB\xEE\x28", "BOCU-1") 143 .StartsWith("\x84\x31\x95\x33", "GB-18030") 144 .Default(0); 145 146 if (BOM) { 147 Diag.Report(Loc, diag::err_unsupported_bom) 148 << BOM << Entry->getName(); 149 Buffer.setInt(Buffer.getInt() | InvalidFlag); 150 } 151 152 if (Invalid) 153 *Invalid = isBufferInvalid(); 154 155 return Buffer.getPointer(); 156} 157 158unsigned LineTableInfo::getLineTableFilenameID(const char *Ptr, unsigned Len) { 159 // Look up the filename in the string table, returning the pre-existing value 160 // if it exists. 161 llvm::StringMapEntry<unsigned> &Entry = 162 FilenameIDs.GetOrCreateValue(Ptr, Ptr+Len, ~0U); 163 if (Entry.getValue() != ~0U) 164 return Entry.getValue(); 165 166 // Otherwise, assign this the next available ID. 167 Entry.setValue(FilenamesByID.size()); 168 FilenamesByID.push_back(&Entry); 169 return FilenamesByID.size()-1; 170} 171 172/// AddLineNote - Add a line note to the line table that indicates that there 173/// is a #line at the specified FID/Offset location which changes the presumed 174/// location to LineNo/FilenameID. 175void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset, 176 unsigned LineNo, int FilenameID) { 177 std::vector<LineEntry> &Entries = LineEntries[FID]; 178 179 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 180 "Adding line entries out of order!"); 181 182 SrcMgr::CharacteristicKind Kind = SrcMgr::C_User; 183 unsigned IncludeOffset = 0; 184 185 if (!Entries.empty()) { 186 // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember 187 // that we are still in "foo.h". 188 if (FilenameID == -1) 189 FilenameID = Entries.back().FilenameID; 190 191 // If we are after a line marker that switched us to system header mode, or 192 // that set #include information, preserve it. 193 Kind = Entries.back().FileKind; 194 IncludeOffset = Entries.back().IncludeOffset; 195 } 196 197 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, Kind, 198 IncludeOffset)); 199} 200 201/// AddLineNote This is the same as the previous version of AddLineNote, but is 202/// used for GNU line markers. If EntryExit is 0, then this doesn't change the 203/// presumed #include stack. If it is 1, this is a file entry, if it is 2 then 204/// this is a file exit. FileKind specifies whether this is a system header or 205/// extern C system header. 206void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset, 207 unsigned LineNo, int FilenameID, 208 unsigned EntryExit, 209 SrcMgr::CharacteristicKind FileKind) { 210 assert(FilenameID != -1 && "Unspecified filename should use other accessor"); 211 212 std::vector<LineEntry> &Entries = LineEntries[FID]; 213 214 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 215 "Adding line entries out of order!"); 216 217 unsigned IncludeOffset = 0; 218 if (EntryExit == 0) { // No #include stack change. 219 IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset; 220 } else if (EntryExit == 1) { 221 IncludeOffset = Offset-1; 222 } else if (EntryExit == 2) { 223 assert(!Entries.empty() && Entries.back().IncludeOffset && 224 "PPDirectives should have caught case when popping empty include stack"); 225 226 // Get the include loc of the last entries' include loc as our include loc. 227 IncludeOffset = 0; 228 if (const LineEntry *PrevEntry = 229 FindNearestLineEntry(FID, Entries.back().IncludeOffset)) 230 IncludeOffset = PrevEntry->IncludeOffset; 231 } 232 233 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind, 234 IncludeOffset)); 235} 236 237 238/// FindNearestLineEntry - Find the line entry nearest to FID that is before 239/// it. If there is no line entry before Offset in FID, return null. 240const LineEntry *LineTableInfo::FindNearestLineEntry(unsigned FID, 241 unsigned Offset) { 242 const std::vector<LineEntry> &Entries = LineEntries[FID]; 243 assert(!Entries.empty() && "No #line entries for this FID after all!"); 244 245 // It is very common for the query to be after the last #line, check this 246 // first. 247 if (Entries.back().FileOffset <= Offset) 248 return &Entries.back(); 249 250 // Do a binary search to find the maximal element that is still before Offset. 251 std::vector<LineEntry>::const_iterator I = 252 std::upper_bound(Entries.begin(), Entries.end(), Offset); 253 if (I == Entries.begin()) return 0; 254 return &*--I; 255} 256 257/// \brief Add a new line entry that has already been encoded into 258/// the internal representation of the line table. 259void LineTableInfo::AddEntry(unsigned FID, 260 const std::vector<LineEntry> &Entries) { 261 LineEntries[FID] = Entries; 262} 263 264/// getLineTableFilenameID - Return the uniqued ID for the specified filename. 265/// 266unsigned SourceManager::getLineTableFilenameID(const char *Ptr, unsigned Len) { 267 if (LineTable == 0) 268 LineTable = new LineTableInfo(); 269 return LineTable->getLineTableFilenameID(Ptr, Len); 270} 271 272 273/// AddLineNote - Add a line note to the line table for the FileID and offset 274/// specified by Loc. If FilenameID is -1, it is considered to be 275/// unspecified. 276void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 277 int FilenameID) { 278 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 279 280 const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile(); 281 282 // Remember that this file has #line directives now if it doesn't already. 283 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 284 285 if (LineTable == 0) 286 LineTable = new LineTableInfo(); 287 LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID); 288} 289 290/// AddLineNote - Add a GNU line marker to the line table. 291void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 292 int FilenameID, bool IsFileEntry, 293 bool IsFileExit, bool IsSystemHeader, 294 bool IsExternCHeader) { 295 // If there is no filename and no flags, this is treated just like a #line, 296 // which does not change the flags of the previous line marker. 297 if (FilenameID == -1) { 298 assert(!IsFileEntry && !IsFileExit && !IsSystemHeader && !IsExternCHeader && 299 "Can't set flags without setting the filename!"); 300 return AddLineNote(Loc, LineNo, FilenameID); 301 } 302 303 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 304 const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile(); 305 306 // Remember that this file has #line directives now if it doesn't already. 307 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 308 309 if (LineTable == 0) 310 LineTable = new LineTableInfo(); 311 312 SrcMgr::CharacteristicKind FileKind; 313 if (IsExternCHeader) 314 FileKind = SrcMgr::C_ExternCSystem; 315 else if (IsSystemHeader) 316 FileKind = SrcMgr::C_System; 317 else 318 FileKind = SrcMgr::C_User; 319 320 unsigned EntryExit = 0; 321 if (IsFileEntry) 322 EntryExit = 1; 323 else if (IsFileExit) 324 EntryExit = 2; 325 326 LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID, 327 EntryExit, FileKind); 328} 329 330LineTableInfo &SourceManager::getLineTable() { 331 if (LineTable == 0) 332 LineTable = new LineTableInfo(); 333 return *LineTable; 334} 335 336//===----------------------------------------------------------------------===// 337// Private 'Create' methods. 338//===----------------------------------------------------------------------===// 339 340SourceManager::SourceManager(Diagnostic &Diag, FileManager &FileMgr) 341 : Diag(Diag), FileMgr(FileMgr), 342 ExternalSLocEntries(0), LineTable(0), NumLinearScans(0), 343 NumBinaryProbes(0) { 344 clearIDTables(); 345 Diag.setSourceManager(this); 346} 347 348SourceManager::~SourceManager() { 349 delete LineTable; 350 351 // Delete FileEntry objects corresponding to content caches. Since the actual 352 // content cache objects are bump pointer allocated, we just have to run the 353 // dtors, but we call the deallocate method for completeness. 354 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) { 355 MemBufferInfos[i]->~ContentCache(); 356 ContentCacheAlloc.Deallocate(MemBufferInfos[i]); 357 } 358 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator 359 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) { 360 I->second->~ContentCache(); 361 ContentCacheAlloc.Deallocate(I->second); 362 } 363} 364 365void SourceManager::clearIDTables() { 366 MainFileID = FileID(); 367 SLocEntryTable.clear(); 368 LastLineNoFileIDQuery = FileID(); 369 LastLineNoContentCache = 0; 370 LastFileIDLookup = FileID(); 371 372 if (LineTable) 373 LineTable->clear(); 374 375 // Use up FileID #0 as an invalid instantiation. 376 NextOffset = 0; 377 createInstantiationLoc(SourceLocation(),SourceLocation(),SourceLocation(), 1); 378} 379 380/// getOrCreateContentCache - Create or return a cached ContentCache for the 381/// specified file. 382const ContentCache * 383SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) { 384 assert(FileEnt && "Didn't specify a file entry to use?"); 385 386 // Do we already have information about this file? 387 ContentCache *&Entry = FileInfos[FileEnt]; 388 if (Entry) return Entry; 389 390 // Nope, create a new Cache entry. Make sure it is at least 8-byte aligned 391 // so that FileInfo can use the low 3 bits of the pointer for its own 392 // nefarious purposes. 393 unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment; 394 EntryAlign = std::max(8U, EntryAlign); 395 Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign); 396 new (Entry) ContentCache(FileEnt); 397 return Entry; 398} 399 400 401/// createMemBufferContentCache - Create a new ContentCache for the specified 402/// memory buffer. This does no caching. 403const ContentCache* 404SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) { 405 // Add a new ContentCache to the MemBufferInfos list and return it. Make sure 406 // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of 407 // the pointer for its own nefarious purposes. 408 unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment; 409 EntryAlign = std::max(8U, EntryAlign); 410 ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign); 411 new (Entry) ContentCache(); 412 MemBufferInfos.push_back(Entry); 413 Entry->setBuffer(Buffer); 414 return Entry; 415} 416 417void SourceManager::PreallocateSLocEntries(ExternalSLocEntrySource *Source, 418 unsigned NumSLocEntries, 419 unsigned NextOffset) { 420 ExternalSLocEntries = Source; 421 this->NextOffset = NextOffset; 422 unsigned CurPrealloc = SLocEntryLoaded.size(); 423 // If we've ever preallocated, we must not count the dummy entry. 424 if (CurPrealloc) --CurPrealloc; 425 SLocEntryLoaded.resize(NumSLocEntries + 1); 426 SLocEntryLoaded[0] = true; 427 SLocEntryTable.resize(SLocEntryTable.size() + NumSLocEntries - CurPrealloc); 428} 429 430void SourceManager::ClearPreallocatedSLocEntries() { 431 unsigned I = 0; 432 for (unsigned N = SLocEntryLoaded.size(); I != N; ++I) 433 if (!SLocEntryLoaded[I]) 434 break; 435 436 // We've already loaded all preallocated source location entries. 437 if (I == SLocEntryLoaded.size()) 438 return; 439 440 // Remove everything from location I onward. 441 SLocEntryTable.resize(I); 442 SLocEntryLoaded.clear(); 443 ExternalSLocEntries = 0; 444} 445 446 447//===----------------------------------------------------------------------===// 448// Methods to create new FileID's and instantiations. 449//===----------------------------------------------------------------------===// 450 451/// createFileID - Create a new FileID for the specified ContentCache and 452/// include position. This works regardless of whether the ContentCache 453/// corresponds to a file or some other input source. 454FileID SourceManager::createFileID(const ContentCache *File, 455 SourceLocation IncludePos, 456 SrcMgr::CharacteristicKind FileCharacter, 457 unsigned PreallocatedID, 458 unsigned Offset) { 459 if (PreallocatedID) { 460 // If we're filling in a preallocated ID, just load in the file 461 // entry and return. 462 assert(PreallocatedID < SLocEntryLoaded.size() && 463 "Preallocate ID out-of-range"); 464 assert(!SLocEntryLoaded[PreallocatedID] && 465 "Source location entry already loaded"); 466 assert(Offset && "Preallocate source location cannot have zero offset"); 467 SLocEntryTable[PreallocatedID] 468 = SLocEntry::get(Offset, FileInfo::get(IncludePos, File, FileCharacter)); 469 SLocEntryLoaded[PreallocatedID] = true; 470 FileID FID = FileID::get(PreallocatedID); 471 return FID; 472 } 473 474 SLocEntryTable.push_back(SLocEntry::get(NextOffset, 475 FileInfo::get(IncludePos, File, 476 FileCharacter))); 477 unsigned FileSize = File->getSize(); 478 assert(NextOffset+FileSize+1 > NextOffset && "Ran out of source locations!"); 479 NextOffset += FileSize+1; 480 481 // Set LastFileIDLookup to the newly created file. The next getFileID call is 482 // almost guaranteed to be from that file. 483 FileID FID = FileID::get(SLocEntryTable.size()-1); 484 return LastFileIDLookup = FID; 485} 486 487/// createInstantiationLoc - Return a new SourceLocation that encodes the fact 488/// that a token from SpellingLoc should actually be referenced from 489/// InstantiationLoc. 490SourceLocation SourceManager::createInstantiationLoc(SourceLocation SpellingLoc, 491 SourceLocation ILocStart, 492 SourceLocation ILocEnd, 493 unsigned TokLength, 494 unsigned PreallocatedID, 495 unsigned Offset) { 496 InstantiationInfo II = InstantiationInfo::get(ILocStart,ILocEnd, SpellingLoc); 497 if (PreallocatedID) { 498 // If we're filling in a preallocated ID, just load in the 499 // instantiation entry and return. 500 assert(PreallocatedID < SLocEntryLoaded.size() && 501 "Preallocate ID out-of-range"); 502 assert(!SLocEntryLoaded[PreallocatedID] && 503 "Source location entry already loaded"); 504 assert(Offset && "Preallocate source location cannot have zero offset"); 505 SLocEntryTable[PreallocatedID] = SLocEntry::get(Offset, II); 506 SLocEntryLoaded[PreallocatedID] = true; 507 return SourceLocation::getMacroLoc(Offset); 508 } 509 SLocEntryTable.push_back(SLocEntry::get(NextOffset, II)); 510 assert(NextOffset+TokLength+1 > NextOffset && "Ran out of source locations!"); 511 NextOffset += TokLength+1; 512 return SourceLocation::getMacroLoc(NextOffset-(TokLength+1)); 513} 514 515const llvm::MemoryBuffer * 516SourceManager::getMemoryBufferForFile(const FileEntry *File, 517 bool *Invalid) { 518 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 519 assert(IR && "getOrCreateContentCache() cannot return NULL"); 520 return IR->getBuffer(Diag, *this, SourceLocation(), Invalid); 521} 522 523void SourceManager::overrideFileContents(const FileEntry *SourceFile, 524 const llvm::MemoryBuffer *Buffer, 525 bool DoNotFree) { 526 const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile); 527 assert(IR && "getOrCreateContentCache() cannot return NULL"); 528 529 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree); 530} 531 532llvm::StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const { 533 bool MyInvalid = false; 534 const llvm::MemoryBuffer *Buf = getBuffer(FID, &MyInvalid); 535 if (Invalid) 536 *Invalid = MyInvalid; 537 538 if (MyInvalid) 539 return ""; 540 541 return Buf->getBuffer(); 542} 543 544//===----------------------------------------------------------------------===// 545// SourceLocation manipulation methods. 546//===----------------------------------------------------------------------===// 547 548/// getFileIDSlow - Return the FileID for a SourceLocation. This is a very hot 549/// method that is used for all SourceManager queries that start with a 550/// SourceLocation object. It is responsible for finding the entry in 551/// SLocEntryTable which contains the specified location. 552/// 553FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const { 554 assert(SLocOffset && "Invalid FileID"); 555 556 // After the first and second level caches, I see two common sorts of 557 // behavior: 1) a lot of searched FileID's are "near" the cached file location 558 // or are "near" the cached instantiation location. 2) others are just 559 // completely random and may be a very long way away. 560 // 561 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly 562 // then we fall back to a less cache efficient, but more scalable, binary 563 // search to find the location. 564 565 // See if this is near the file point - worst case we start scanning from the 566 // most newly created FileID. 567 std::vector<SrcMgr::SLocEntry>::const_iterator I; 568 569 if (SLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) { 570 // Neither loc prunes our search. 571 I = SLocEntryTable.end(); 572 } else { 573 // Perhaps it is near the file point. 574 I = SLocEntryTable.begin()+LastFileIDLookup.ID; 575 } 576 577 // Find the FileID that contains this. "I" is an iterator that points to a 578 // FileID whose offset is known to be larger than SLocOffset. 579 unsigned NumProbes = 0; 580 while (1) { 581 --I; 582 if (ExternalSLocEntries) 583 getSLocEntry(FileID::get(I - SLocEntryTable.begin())); 584 if (I->getOffset() <= SLocOffset) { 585#if 0 586 printf("lin %d -> %d [%s] %d %d\n", SLocOffset, 587 I-SLocEntryTable.begin(), 588 I->isInstantiation() ? "inst" : "file", 589 LastFileIDLookup.ID, int(SLocEntryTable.end()-I)); 590#endif 591 FileID Res = FileID::get(I-SLocEntryTable.begin()); 592 593 // If this isn't an instantiation, remember it. We have good locality 594 // across FileID lookups. 595 if (!I->isInstantiation()) 596 LastFileIDLookup = Res; 597 NumLinearScans += NumProbes+1; 598 return Res; 599 } 600 if (++NumProbes == 8) 601 break; 602 } 603 604 // Convert "I" back into an index. We know that it is an entry whose index is 605 // larger than the offset we are looking for. 606 unsigned GreaterIndex = I-SLocEntryTable.begin(); 607 // LessIndex - This is the lower bound of the range that we're searching. 608 // We know that the offset corresponding to the FileID is is less than 609 // SLocOffset. 610 unsigned LessIndex = 0; 611 NumProbes = 0; 612 while (1) { 613 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex; 614 unsigned MidOffset = getSLocEntry(FileID::get(MiddleIndex)).getOffset(); 615 616 ++NumProbes; 617 618 // If the offset of the midpoint is too large, chop the high side of the 619 // range to the midpoint. 620 if (MidOffset > SLocOffset) { 621 GreaterIndex = MiddleIndex; 622 continue; 623 } 624 625 // If the middle index contains the value, succeed and return. 626 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) { 627#if 0 628 printf("bin %d -> %d [%s] %d %d\n", SLocOffset, 629 I-SLocEntryTable.begin(), 630 I->isInstantiation() ? "inst" : "file", 631 LastFileIDLookup.ID, int(SLocEntryTable.end()-I)); 632#endif 633 FileID Res = FileID::get(MiddleIndex); 634 635 // If this isn't an instantiation, remember it. We have good locality 636 // across FileID lookups. 637 if (!I->isInstantiation()) 638 LastFileIDLookup = Res; 639 NumBinaryProbes += NumProbes; 640 return Res; 641 } 642 643 // Otherwise, move the low-side up to the middle index. 644 LessIndex = MiddleIndex; 645 } 646} 647 648SourceLocation SourceManager:: 649getInstantiationLocSlowCase(SourceLocation Loc) const { 650 do { 651 // Note: If Loc indicates an offset into a token that came from a macro 652 // expansion (e.g. the 5th character of the token) we do not want to add 653 // this offset when going to the instantiation location. The instatiation 654 // location is the macro invocation, which the offset has nothing to do 655 // with. This is unlike when we get the spelling loc, because the offset 656 // directly correspond to the token whose spelling we're inspecting. 657 Loc = getSLocEntry(getFileID(Loc)).getInstantiation() 658 .getInstantiationLocStart(); 659 } while (!Loc.isFileID()); 660 661 return Loc; 662} 663 664SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const { 665 do { 666 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 667 Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc(); 668 Loc = Loc.getFileLocWithOffset(LocInfo.second); 669 } while (!Loc.isFileID()); 670 return Loc; 671} 672 673 674std::pair<FileID, unsigned> 675SourceManager::getDecomposedInstantiationLocSlowCase(const SrcMgr::SLocEntry *E, 676 unsigned Offset) const { 677 // If this is an instantiation record, walk through all the instantiation 678 // points. 679 FileID FID; 680 SourceLocation Loc; 681 do { 682 Loc = E->getInstantiation().getInstantiationLocStart(); 683 684 FID = getFileID(Loc); 685 E = &getSLocEntry(FID); 686 Offset += Loc.getOffset()-E->getOffset(); 687 } while (!Loc.isFileID()); 688 689 return std::make_pair(FID, Offset); 690} 691 692std::pair<FileID, unsigned> 693SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E, 694 unsigned Offset) const { 695 // If this is an instantiation record, walk through all the instantiation 696 // points. 697 FileID FID; 698 SourceLocation Loc; 699 do { 700 Loc = E->getInstantiation().getSpellingLoc(); 701 702 FID = getFileID(Loc); 703 E = &getSLocEntry(FID); 704 Offset += Loc.getOffset()-E->getOffset(); 705 } while (!Loc.isFileID()); 706 707 return std::make_pair(FID, Offset); 708} 709 710/// getImmediateSpellingLoc - Given a SourceLocation object, return the 711/// spelling location referenced by the ID. This is the first level down 712/// towards the place where the characters that make up the lexed token can be 713/// found. This should not generally be used by clients. 714SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{ 715 if (Loc.isFileID()) return Loc; 716 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 717 Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc(); 718 return Loc.getFileLocWithOffset(LocInfo.second); 719} 720 721 722/// getImmediateInstantiationRange - Loc is required to be an instantiation 723/// location. Return the start/end of the instantiation information. 724std::pair<SourceLocation,SourceLocation> 725SourceManager::getImmediateInstantiationRange(SourceLocation Loc) const { 726 assert(Loc.isMacroID() && "Not an instantiation loc!"); 727 const InstantiationInfo &II = getSLocEntry(getFileID(Loc)).getInstantiation(); 728 return II.getInstantiationLocRange(); 729} 730 731/// getInstantiationRange - Given a SourceLocation object, return the 732/// range of tokens covered by the instantiation in the ultimate file. 733std::pair<SourceLocation,SourceLocation> 734SourceManager::getInstantiationRange(SourceLocation Loc) const { 735 if (Loc.isFileID()) return std::make_pair(Loc, Loc); 736 737 std::pair<SourceLocation,SourceLocation> Res = 738 getImmediateInstantiationRange(Loc); 739 740 // Fully resolve the start and end locations to their ultimate instantiation 741 // points. 742 while (!Res.first.isFileID()) 743 Res.first = getImmediateInstantiationRange(Res.first).first; 744 while (!Res.second.isFileID()) 745 Res.second = getImmediateInstantiationRange(Res.second).second; 746 return Res; 747} 748 749 750 751//===----------------------------------------------------------------------===// 752// Queries about the code at a SourceLocation. 753//===----------------------------------------------------------------------===// 754 755/// getCharacterData - Return a pointer to the start of the specified location 756/// in the appropriate MemoryBuffer. 757const char *SourceManager::getCharacterData(SourceLocation SL, 758 bool *Invalid) const { 759 // Note that this is a hot function in the getSpelling() path, which is 760 // heavily used by -E mode. 761 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL); 762 763 // Note that calling 'getBuffer()' may lazily page in a source file. 764 bool CharDataInvalid = false; 765 const llvm::MemoryBuffer *Buffer 766 = getSLocEntry(LocInfo.first).getFile().getContentCache() 767 ->getBuffer(Diag, *this, SourceLocation(), &CharDataInvalid); 768 if (Invalid) 769 *Invalid = CharDataInvalid; 770 return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second); 771} 772 773 774/// getColumnNumber - Return the column # for the specified file position. 775/// this is significantly cheaper to compute than the line number. 776unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos, 777 bool *Invalid) const { 778 bool MyInvalid = false; 779 const char *Buf = getBuffer(FID, &MyInvalid)->getBufferStart(); 780 if (Invalid) 781 *Invalid = MyInvalid; 782 783 if (MyInvalid) 784 return 1; 785 786 unsigned LineStart = FilePos; 787 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r') 788 --LineStart; 789 return FilePos-LineStart+1; 790} 791 792// isInvalid - Return the result of calling loc.isInvalid(), and 793// if Invalid is not null, set its value to same. 794static bool isInvalid(SourceLocation Loc, bool *Invalid) { 795 bool MyInvalid = Loc.isInvalid(); 796 if (Invalid) 797 *Invalid = MyInvalid; 798 return MyInvalid; 799} 800 801unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc, 802 bool *Invalid) const { 803 if (isInvalid(Loc, Invalid)) return 0; 804 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 805 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 806} 807 808unsigned SourceManager::getInstantiationColumnNumber(SourceLocation Loc, 809 bool *Invalid) const { 810 if (isInvalid(Loc, Invalid)) return 0; 811 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 812 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 813} 814 815static LLVM_ATTRIBUTE_NOINLINE void 816ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI, 817 llvm::BumpPtrAllocator &Alloc, 818 const SourceManager &SM, bool &Invalid); 819static void ComputeLineNumbers(Diagnostic &Diag, ContentCache *FI, 820 llvm::BumpPtrAllocator &Alloc, 821 const SourceManager &SM, bool &Invalid) { 822 // Note that calling 'getBuffer()' may lazily page in the file. 823 const MemoryBuffer *Buffer = FI->getBuffer(Diag, SM, SourceLocation(), 824 &Invalid); 825 if (Invalid) 826 return; 827 828 // Find the file offsets of all of the *physical* source lines. This does 829 // not look at trigraphs, escaped newlines, or anything else tricky. 830 std::vector<unsigned> LineOffsets; 831 832 // Line #1 starts at char 0. 833 LineOffsets.push_back(0); 834 835 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart(); 836 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd(); 837 unsigned Offs = 0; 838 while (1) { 839 // Skip over the contents of the line. 840 // TODO: Vectorize this? This is very performance sensitive for programs 841 // with lots of diagnostics and in -E mode. 842 const unsigned char *NextBuf = (const unsigned char *)Buf; 843 while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0') 844 ++NextBuf; 845 Offs += NextBuf-Buf; 846 Buf = NextBuf; 847 848 if (Buf[0] == '\n' || Buf[0] == '\r') { 849 // If this is \n\r or \r\n, skip both characters. 850 if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1]) 851 ++Offs, ++Buf; 852 ++Offs, ++Buf; 853 LineOffsets.push_back(Offs); 854 } else { 855 // Otherwise, this is a null. If end of file, exit. 856 if (Buf == End) break; 857 // Otherwise, skip the null. 858 ++Offs, ++Buf; 859 } 860 } 861 862 // Copy the offsets into the FileInfo structure. 863 FI->NumLines = LineOffsets.size(); 864 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size()); 865 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache); 866} 867 868/// getLineNumber - Given a SourceLocation, return the spelling line number 869/// for the position indicated. This requires building and caching a table of 870/// line offsets for the MemoryBuffer, so this is not cheap: use only when 871/// about to emit a diagnostic. 872unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos, 873 bool *Invalid) const { 874 ContentCache *Content; 875 if (LastLineNoFileIDQuery == FID) 876 Content = LastLineNoContentCache; 877 else 878 Content = const_cast<ContentCache*>(getSLocEntry(FID) 879 .getFile().getContentCache()); 880 881 // If this is the first use of line information for this buffer, compute the 882 /// SourceLineCache for it on demand. 883 if (Content->SourceLineCache == 0) { 884 bool MyInvalid = false; 885 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 886 if (Invalid) 887 *Invalid = MyInvalid; 888 if (MyInvalid) 889 return 1; 890 } else if (Invalid) 891 *Invalid = false; 892 893 // Okay, we know we have a line number table. Do a binary search to find the 894 // line number that this character position lands on. 895 unsigned *SourceLineCache = Content->SourceLineCache; 896 unsigned *SourceLineCacheStart = SourceLineCache; 897 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines; 898 899 unsigned QueriedFilePos = FilePos+1; 900 901 // FIXME: I would like to be convinced that this code is worth being as 902 // complicated as it is, binary search isn't that slow. 903 // 904 // If it is worth being optimized, then in my opinion it could be more 905 // performant, simpler, and more obviously correct by just "galloping" outward 906 // from the queried file position. In fact, this could be incorporated into a 907 // generic algorithm such as lower_bound_with_hint. 908 // 909 // If someone gives me a test case where this matters, and I will do it! - DWD 910 911 // If the previous query was to the same file, we know both the file pos from 912 // that query and the line number returned. This allows us to narrow the 913 // search space from the entire file to something near the match. 914 if (LastLineNoFileIDQuery == FID) { 915 if (QueriedFilePos >= LastLineNoFilePos) { 916 // FIXME: Potential overflow? 917 SourceLineCache = SourceLineCache+LastLineNoResult-1; 918 919 // The query is likely to be nearby the previous one. Here we check to 920 // see if it is within 5, 10 or 20 lines. It can be far away in cases 921 // where big comment blocks and vertical whitespace eat up lines but 922 // contribute no tokens. 923 if (SourceLineCache+5 < SourceLineCacheEnd) { 924 if (SourceLineCache[5] > QueriedFilePos) 925 SourceLineCacheEnd = SourceLineCache+5; 926 else if (SourceLineCache+10 < SourceLineCacheEnd) { 927 if (SourceLineCache[10] > QueriedFilePos) 928 SourceLineCacheEnd = SourceLineCache+10; 929 else if (SourceLineCache+20 < SourceLineCacheEnd) { 930 if (SourceLineCache[20] > QueriedFilePos) 931 SourceLineCacheEnd = SourceLineCache+20; 932 } 933 } 934 } 935 } else { 936 if (LastLineNoResult < Content->NumLines) 937 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1; 938 } 939 } 940 941 // If the spread is large, do a "radix" test as our initial guess, based on 942 // the assumption that lines average to approximately the same length. 943 // NOTE: This is currently disabled, as it does not appear to be profitable in 944 // initial measurements. 945 if (0 && SourceLineCacheEnd-SourceLineCache > 20) { 946 unsigned FileLen = Content->SourceLineCache[Content->NumLines-1]; 947 948 // Take a stab at guessing where it is. 949 unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen; 950 951 // Check for -10 and +10 lines. 952 unsigned LowerBound = std::max(int(ApproxPos-10), 0); 953 unsigned UpperBound = std::min(ApproxPos+10, FileLen); 954 955 // If the computed lower bound is less than the query location, move it in. 956 if (SourceLineCache < SourceLineCacheStart+LowerBound && 957 SourceLineCacheStart[LowerBound] < QueriedFilePos) 958 SourceLineCache = SourceLineCacheStart+LowerBound; 959 960 // If the computed upper bound is greater than the query location, move it. 961 if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound && 962 SourceLineCacheStart[UpperBound] >= QueriedFilePos) 963 SourceLineCacheEnd = SourceLineCacheStart+UpperBound; 964 } 965 966 unsigned *Pos 967 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos); 968 unsigned LineNo = Pos-SourceLineCacheStart; 969 970 LastLineNoFileIDQuery = FID; 971 LastLineNoContentCache = Content; 972 LastLineNoFilePos = QueriedFilePos; 973 LastLineNoResult = LineNo; 974 return LineNo; 975} 976 977unsigned SourceManager::getInstantiationLineNumber(SourceLocation Loc, 978 bool *Invalid) const { 979 if (isInvalid(Loc, Invalid)) return 0; 980 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 981 return getLineNumber(LocInfo.first, LocInfo.second); 982} 983unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc, 984 bool *Invalid) const { 985 if (isInvalid(Loc, Invalid)) return 0; 986 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 987 return getLineNumber(LocInfo.first, LocInfo.second); 988} 989 990/// getFileCharacteristic - return the file characteristic of the specified 991/// source location, indicating whether this is a normal file, a system 992/// header, or an "implicit extern C" system header. 993/// 994/// This state can be modified with flags on GNU linemarker directives like: 995/// # 4 "foo.h" 3 996/// which changes all source locations in the current file after that to be 997/// considered to be from a system header. 998SrcMgr::CharacteristicKind 999SourceManager::getFileCharacteristic(SourceLocation Loc) const { 1000 assert(!Loc.isInvalid() && "Can't get file characteristic of invalid loc!"); 1001 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 1002 const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile(); 1003 1004 // If there are no #line directives in this file, just return the whole-file 1005 // state. 1006 if (!FI.hasLineDirectives()) 1007 return FI.getFileCharacteristic(); 1008 1009 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1010 // See if there is a #line directive before the location. 1011 const LineEntry *Entry = 1012 LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second); 1013 1014 // If this is before the first line marker, use the file characteristic. 1015 if (!Entry) 1016 return FI.getFileCharacteristic(); 1017 1018 return Entry->FileKind; 1019} 1020 1021/// Return the filename or buffer identifier of the buffer the location is in. 1022/// Note that this name does not respect #line directives. Use getPresumedLoc 1023/// for normal clients. 1024const char *SourceManager::getBufferName(SourceLocation Loc, 1025 bool *Invalid) const { 1026 if (isInvalid(Loc, Invalid)) return "<invalid loc>"; 1027 1028 return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier(); 1029} 1030 1031 1032/// getPresumedLoc - This method returns the "presumed" location of a 1033/// SourceLocation specifies. A "presumed location" can be modified by #line 1034/// or GNU line marker directives. This provides a view on the data that a 1035/// user should see in diagnostics, for example. 1036/// 1037/// Note that a presumed location is always given as the instantiation point 1038/// of an instantiation location, not at the spelling location. 1039PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const { 1040 if (Loc.isInvalid()) return PresumedLoc(); 1041 1042 // Presumed locations are always for instantiation points. 1043 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 1044 1045 const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile(); 1046 const SrcMgr::ContentCache *C = FI.getContentCache(); 1047 1048 // To get the source name, first consult the FileEntry (if one exists) 1049 // before the MemBuffer as this will avoid unnecessarily paging in the 1050 // MemBuffer. 1051 const char *Filename; 1052 if (C->Entry) 1053 Filename = C->Entry->getName(); 1054 else 1055 Filename = C->getBuffer(Diag, *this)->getBufferIdentifier(); 1056 bool Invalid = false; 1057 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid); 1058 if (Invalid) 1059 return PresumedLoc(); 1060 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid); 1061 if (Invalid) 1062 return PresumedLoc(); 1063 1064 SourceLocation IncludeLoc = FI.getIncludeLoc(); 1065 1066 // If we have #line directives in this file, update and overwrite the physical 1067 // location info if appropriate. 1068 if (FI.hasLineDirectives()) { 1069 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1070 // See if there is a #line directive before this. If so, get it. 1071 if (const LineEntry *Entry = 1072 LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) { 1073 // If the LineEntry indicates a filename, use it. 1074 if (Entry->FilenameID != -1) 1075 Filename = LineTable->getFilename(Entry->FilenameID); 1076 1077 // Use the line number specified by the LineEntry. This line number may 1078 // be multiple lines down from the line entry. Add the difference in 1079 // physical line numbers from the query point and the line marker to the 1080 // total. 1081 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset); 1082 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1); 1083 1084 // Note that column numbers are not molested by line markers. 1085 1086 // Handle virtual #include manipulation. 1087 if (Entry->IncludeOffset) { 1088 IncludeLoc = getLocForStartOfFile(LocInfo.first); 1089 IncludeLoc = IncludeLoc.getFileLocWithOffset(Entry->IncludeOffset); 1090 } 1091 } 1092 } 1093 1094 return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc); 1095} 1096 1097//===----------------------------------------------------------------------===// 1098// Other miscellaneous methods. 1099//===----------------------------------------------------------------------===// 1100 1101/// \brief Get the source location for the given file:line:col triplet. 1102/// 1103/// If the source file is included multiple times, the source location will 1104/// be based upon the first inclusion. 1105SourceLocation SourceManager::getLocation(const FileEntry *SourceFile, 1106 unsigned Line, unsigned Col) const { 1107 assert(SourceFile && "Null source file!"); 1108 assert(Line && Col && "Line and column should start from 1!"); 1109 1110 fileinfo_iterator FI = FileInfos.find(SourceFile); 1111 if (FI == FileInfos.end()) 1112 return SourceLocation(); 1113 ContentCache *Content = FI->second; 1114 1115 // If this is the first use of line information for this buffer, compute the 1116 /// SourceLineCache for it on demand. 1117 if (Content->SourceLineCache == 0) { 1118 bool MyInvalid = false; 1119 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1120 if (MyInvalid) 1121 return SourceLocation(); 1122 } 1123 1124 // Find the first file ID that corresponds to the given file. 1125 FileID FirstFID; 1126 1127 // First, check the main file ID, since it is common to look for a 1128 // location in the main file. 1129 if (!MainFileID.isInvalid()) { 1130 const SLocEntry &MainSLoc = getSLocEntry(MainFileID); 1131 if (MainSLoc.isFile() && MainSLoc.getFile().getContentCache() == Content) 1132 FirstFID = MainFileID; 1133 } 1134 1135 if (FirstFID.isInvalid()) { 1136 // The location we're looking for isn't in the main file; look 1137 // through all of the source locations. 1138 for (unsigned I = 0, N = sloc_entry_size(); I != N; ++I) { 1139 const SLocEntry &SLoc = getSLocEntry(I); 1140 if (SLoc.isFile() && SLoc.getFile().getContentCache() == Content) { 1141 FirstFID = FileID::get(I); 1142 break; 1143 } 1144 } 1145 } 1146 1147 if (FirstFID.isInvalid()) 1148 return SourceLocation(); 1149 1150 if (Line > Content->NumLines) { 1151 unsigned Size = Content->getBuffer(Diag, *this)->getBufferSize(); 1152 if (Size > 0) 1153 --Size; 1154 return getLocForStartOfFile(FirstFID).getFileLocWithOffset(Size); 1155 } 1156 1157 unsigned FilePos = Content->SourceLineCache[Line - 1]; 1158 const char *Buf = Content->getBuffer(Diag, *this)->getBufferStart() + FilePos; 1159 unsigned BufLength = Content->getBuffer(Diag, *this)->getBufferEnd() - Buf; 1160 unsigned i = 0; 1161 1162 // Check that the given column is valid. 1163 while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r') 1164 ++i; 1165 if (i < Col-1) 1166 return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + i); 1167 1168 return getLocForStartOfFile(FirstFID).getFileLocWithOffset(FilePos + Col - 1); 1169} 1170 1171/// Given a decomposed source location, move it up the include/instantiation 1172/// stack to the parent source location. If this is possible, return the 1173/// decomposed version of the parent in Loc and return false. If Loc is the 1174/// top-level entry, return true and don't modify it. 1175static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc, 1176 const SourceManager &SM) { 1177 SourceLocation UpperLoc; 1178 const SrcMgr::SLocEntry &Entry = SM.getSLocEntry(Loc.first); 1179 if (Entry.isInstantiation()) 1180 UpperLoc = Entry.getInstantiation().getInstantiationLocStart(); 1181 else 1182 UpperLoc = Entry.getFile().getIncludeLoc(); 1183 1184 if (UpperLoc.isInvalid()) 1185 return true; // We reached the top. 1186 1187 Loc = SM.getDecomposedLoc(UpperLoc); 1188 return false; 1189} 1190 1191 1192/// \brief Determines the order of 2 source locations in the translation unit. 1193/// 1194/// \returns true if LHS source location comes before RHS, false otherwise. 1195bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS, 1196 SourceLocation RHS) const { 1197 assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!"); 1198 if (LHS == RHS) 1199 return false; 1200 1201 std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS); 1202 std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS); 1203 1204 // If the source locations are in the same file, just compare offsets. 1205 if (LOffs.first == ROffs.first) 1206 return LOffs.second < ROffs.second; 1207 1208 // If we are comparing a source location with multiple locations in the same 1209 // file, we get a big win by caching the result. 1210 if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first)) 1211 return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second); 1212 1213 // Okay, we missed in the cache, start updating the cache for this query. 1214 IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first); 1215 1216 // "Traverse" the include/instantiation stacks of both locations and try to 1217 // find a common "ancestor". FileIDs build a tree-like structure that 1218 // reflects the #include hierarchy, and this algorithm needs to find the 1219 // nearest common ancestor between the two locations. For example, if you 1220 // have a.c that includes b.h and c.h, and are comparing a location in b.h to 1221 // a location in c.h, we need to find that their nearest common ancestor is 1222 // a.c, and compare the locations of the two #includes to find their relative 1223 // ordering. 1224 // 1225 // SourceManager assigns FileIDs in order of parsing. This means that an 1226 // includee always has a larger FileID than an includer. While you might 1227 // think that we could just compare the FileID's here, that doesn't work to 1228 // compare a point at the end of a.c with a point within c.h. Though c.h has 1229 // a larger FileID, we have to compare the include point of c.h to the 1230 // location in a.c. 1231 // 1232 // Despite not being able to directly compare FileID's, we can tell that a 1233 // larger FileID is necessarily more deeply nested than a lower one and use 1234 // this information to walk up the tree to the nearest common ancestor. 1235 do { 1236 // If LOffs is larger than ROffs, then LOffs must be more deeply nested than 1237 // ROffs, walk up the #include chain. 1238 if (LOffs.first.ID > ROffs.first.ID) { 1239 if (MoveUpIncludeHierarchy(LOffs, *this)) 1240 break; // We reached the top. 1241 1242 } else { 1243 // Otherwise, ROffs is larger than LOffs, so ROffs must be more deeply 1244 // nested than LOffs, walk up the #include chain. 1245 if (MoveUpIncludeHierarchy(ROffs, *this)) 1246 break; // We reached the top. 1247 } 1248 } while (LOffs.first != ROffs.first); 1249 1250 // If we exited because we found a nearest common ancestor, compare the 1251 // locations within the common file and cache them. 1252 if (LOffs.first == ROffs.first) { 1253 IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second); 1254 return IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second); 1255 } 1256 1257 // There is no common ancestor, most probably because one location is in the 1258 // predefines buffer or an AST file. 1259 // FIXME: We should rearrange the external interface so this simply never 1260 // happens; it can't conceptually happen. Also see PR5662. 1261 IsBeforeInTUCache.setQueryFIDs(FileID(), FileID()); // Don't try caching. 1262 1263 // Zip both entries up to the top level record. 1264 while (!MoveUpIncludeHierarchy(LOffs, *this)) /*empty*/; 1265 while (!MoveUpIncludeHierarchy(ROffs, *this)) /*empty*/; 1266 1267 // If exactly one location is a memory buffer, assume it preceeds the other. 1268 1269 // Strip off macro instantation locations, going up to the top-level File 1270 // SLocEntry. 1271 bool LIsMB = getFileEntryForID(LOffs.first) == 0; 1272 bool RIsMB = getFileEntryForID(ROffs.first) == 0; 1273 if (LIsMB != RIsMB) 1274 return LIsMB; 1275 1276 // Otherwise, just assume FileIDs were created in order. 1277 return LOffs.first < ROffs.first; 1278} 1279 1280/// PrintStats - Print statistics to stderr. 1281/// 1282void SourceManager::PrintStats() const { 1283 llvm::errs() << "\n*** Source Manager Stats:\n"; 1284 llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size() 1285 << " mem buffers mapped.\n"; 1286 llvm::errs() << SLocEntryTable.size() << " SLocEntry's allocated, " 1287 << NextOffset << "B of Sloc address space used.\n"; 1288 1289 unsigned NumLineNumsComputed = 0; 1290 unsigned NumFileBytesMapped = 0; 1291 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){ 1292 NumLineNumsComputed += I->second->SourceLineCache != 0; 1293 NumFileBytesMapped += I->second->getSizeBytesMapped(); 1294 } 1295 1296 llvm::errs() << NumFileBytesMapped << " bytes of files mapped, " 1297 << NumLineNumsComputed << " files with line #'s computed.\n"; 1298 llvm::errs() << "FileID scans: " << NumLinearScans << " linear, " 1299 << NumBinaryProbes << " binary.\n"; 1300} 1301 1302ExternalSLocEntrySource::~ExternalSLocEntrySource() { } 1303