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/FileManager.h" 16 #include "llvm/Support/Compiler.h" 17 #include "llvm/Support/MemoryBuffer.h" 18 #include "llvm/System/Path.h" 19 #include "llvm/Bitcode/Serialize.h" 20 #include "llvm/Bitcode/Deserialize.h" 21 #include "llvm/Support/Streams.h" 22 #include <algorithm> 23 using namespace clang; 24 using namespace SrcMgr; 25 using llvm::MemoryBuffer; 26 27 //===----------------------------------------------------------------------===// 28 // SourceManager Helper Classes 29 //===----------------------------------------------------------------------===// 30 31 ContentCache::~ContentCache() { 32 delete Buffer; 33 } 34 35 /// getSizeBytesMapped - Returns the number of bytes actually mapped for 36 /// this ContentCache. This can be 0 if the MemBuffer was not actually 37 /// instantiated. 38 unsigned ContentCache::getSizeBytesMapped() const { 39 return Buffer ? Buffer->getBufferSize() : 0; 40 } 41 42 /// getSize - Returns the size of the content encapsulated by this ContentCache. 43 /// This can be the size of the source file or the size of an arbitrary 44 /// scratch buffer. If the ContentCache encapsulates a source file, that 45 /// file is not lazily brought in from disk to satisfy this query. 46 unsigned ContentCache::getSize() const { 47 return Entry ? Entry->getSize() : Buffer->getBufferSize(); 48 } 49 50 const llvm::MemoryBuffer *ContentCache::getBuffer() const { 51 // Lazily create the Buffer for ContentCaches that wrap files. 52 if (!Buffer && Entry) { 53 // FIXME: Should we support a way to not have to do this check over 54 // and over if we cannot open the file? 55 Buffer = MemoryBuffer::getFile(Entry->getName(), 0, Entry->getSize()); 56 } 57 return Buffer; 58 } 59 60 //===----------------------------------------------------------------------===// 61 // Line Table Implementation 62 //===----------------------------------------------------------------------===// 63 64 namespace clang { 65 struct LineEntry { 66 /// FileOffset - The offset in this file that the line entry occurs at. 67 unsigned FileOffset; 68 /// LineNo - The presumed line number of this line entry: #line 4. 69 unsigned LineNo; 70 /// FilenameID - The ID of the filename identified by this line entry: 71 /// #line 4 "foo.c". This is -1 if not specified. 72 int FilenameID; 73 74 static LineEntry get(unsigned Offs, unsigned Line, int Filename) { 75 LineEntry E; 76 E.FileOffset = Offs; 77 E.LineNo = Line; 78 E.FilenameID = Filename; 79 return E; 80 } 81 }; 82 83 inline bool operator<(const LineEntry &E, unsigned Offset) { 84 return E.FileOffset < Offset; 85 } 86 87 inline bool operator<(unsigned Offset, const LineEntry &E) { 88 return Offset < E.FileOffset; 89 } 90 91 /// LineTableInfo - This class is used to hold and unique data used to 92 /// represent #line information. 93 class LineTableInfo { 94 /// FilenameIDs - This map is used to assign unique IDs to filenames in 95 /// #line directives. This allows us to unique the filenames that 96 /// frequently reoccur and reference them with indices. FilenameIDs holds 97 /// the mapping from string -> ID, and FilenamesByID holds the mapping of ID 98 /// to string. 99 llvm::StringMap<unsigned, llvm::BumpPtrAllocator> FilenameIDs; 100 std::vector<llvm::StringMapEntry<unsigned>*> FilenamesByID; 101 102 /// LineEntries - This is a map from FileIDs to a list of line entries (sorted 103 /// by the offset they occur in the file. 104 std::map<unsigned, std::vector<LineEntry> > LineEntries; 105 public: 106 LineTableInfo() { 107 } 108 109 void clear() { 110 FilenameIDs.clear(); 111 FilenamesByID.clear(); 112 } 113 114 ~LineTableInfo() {} 115 116 unsigned getLineTableFilenameID(const char *Ptr, unsigned Len); 117 const char *getFilename(unsigned ID) const { 118 assert(ID < FilenamesByID.size() && "Invalid FilenameID"); 119 return FilenamesByID[ID]->getKeyData(); 120 } 121 122 void AddLineNote(unsigned FID, unsigned Offset, 123 unsigned LineNo, int FilenameID); 124 125 /// FindNearestLineEntry - Find the line entry nearest to FID that is before 126 /// it. If there is no line entry before Offset in FID, return null. 127 const LineEntry *FindNearestLineEntry(unsigned FID, unsigned Offset); 128 }; 129 } // namespace clang 130 131 unsigned LineTableInfo::getLineTableFilenameID(const char *Ptr, unsigned Len) { 132 // Look up the filename in the string table, returning the pre-existing value 133 // if it exists. 134 llvm::StringMapEntry<unsigned> &Entry = 135 FilenameIDs.GetOrCreateValue(Ptr, Ptr+Len, ~0U); 136 if (Entry.getValue() != ~0U) 137 return Entry.getValue(); 138 139 // Otherwise, assign this the next available ID. 140 Entry.setValue(FilenamesByID.size()); 141 FilenamesByID.push_back(&Entry); 142 return FilenamesByID.size()-1; 143 } 144 145 /// AddLineNote - Add a line note to the line table that indicates that there 146 /// is a #line at the specified FID/Offset location which changes the presumed 147 /// location to LineNo/FilenameID. 148 void LineTableInfo::AddLineNote(unsigned FID, unsigned Offset, 149 unsigned LineNo, int FilenameID) { 150 std::vector<LineEntry> &Entries = LineEntries[FID]; 151 152 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 153 "Adding line entries out of order!"); 154 155 // If this is a '#line 4' after '#line 42 "foo.h"', make sure to remember that 156 // we are still in "foo.h". 157 if (FilenameID == -1 && !Entries.empty()) 158 FilenameID = Entries.back().FilenameID; 159 160 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID)); 161 } 162 163 /// FindNearestLineEntry - Find the line entry nearest to FID that is before 164 /// it. If there is no line entry before Offset in FID, return null. 165 const LineEntry *LineTableInfo::FindNearestLineEntry(unsigned FID, 166 unsigned Offset) { 167 const std::vector<LineEntry> &Entries = LineEntries[FID]; 168 assert(!Entries.empty() && "No #line entries for this FID after all!"); 169 170 // It is very common for the query to be after the last #line, check this 171 // first. 172 if (Entries.back().FileOffset <= Offset) 173 return &Entries.back(); 174 175 // Do a binary search to find the maximal element that is still before Offset. 176 std::vector<LineEntry>::const_iterator I = 177 std::upper_bound(Entries.begin(), Entries.end(), Offset); 178 if (I == Entries.begin()) return 0; 179 return &*--I; 180 } 181 182 183 /// getLineTableFilenameID - Return the uniqued ID for the specified filename. 184 /// 185 unsigned SourceManager::getLineTableFilenameID(const char *Ptr, unsigned Len) { 186 if (LineTable == 0) 187 LineTable = new LineTableInfo(); 188 return LineTable->getLineTableFilenameID(Ptr, Len); 189 } 190 191 192 /// AddLineNote - Add a line note to the line table for the FileID and offset 193 /// specified by Loc. If FilenameID is -1, it is considered to be 194 /// unspecified. 195 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 196 int FilenameID) { 197 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 198 199 const SrcMgr::FileInfo &FileInfo = getSLocEntry(LocInfo.first).getFile(); 200 201 // Remember that this file has #line directives now if it doesn't already. 202 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 203 204 if (LineTable == 0) 205 LineTable = new LineTableInfo(); 206 LineTable->AddLineNote(LocInfo.first.ID, LocInfo.second, LineNo, FilenameID); 207 } 208 209 210 //===----------------------------------------------------------------------===// 211 // Private 'Create' methods. 212 //===----------------------------------------------------------------------===// 213 214 SourceManager::~SourceManager() { 215 delete LineTable; 216 217 // Delete FileEntry objects corresponding to content caches. Since the actual 218 // content cache objects are bump pointer allocated, we just have to run the 219 // dtors, but we call the deallocate method for completeness. 220 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) { 221 MemBufferInfos[i]->~ContentCache(); 222 ContentCacheAlloc.Deallocate(MemBufferInfos[i]); 223 } 224 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator 225 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) { 226 I->second->~ContentCache(); 227 ContentCacheAlloc.Deallocate(I->second); 228 } 229 } 230 231 void SourceManager::clearIDTables() { 232 MainFileID = FileID(); 233 SLocEntryTable.clear(); 234 LastLineNoFileIDQuery = FileID(); 235 LastLineNoContentCache = 0; 236 LastFileIDLookup = FileID(); 237 238 if (LineTable) 239 LineTable->clear(); 240 241 // Use up FileID #0 as an invalid instantiation. 242 NextOffset = 0; 243 createInstantiationLoc(SourceLocation(), SourceLocation(), 1); 244 } 245 246 /// getOrCreateContentCache - Create or return a cached ContentCache for the 247 /// specified file. 248 const ContentCache * 249 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt) { 250 assert(FileEnt && "Didn't specify a file entry to use?"); 251 252 // Do we already have information about this file? 253 ContentCache *&Entry = FileInfos[FileEnt]; 254 if (Entry) return Entry; 255 256 // Nope, create a new Cache entry. Make sure it is at least 8-byte aligned 257 // so that FileInfo can use the low 3 bits of the pointer for its own 258 // nefarious purposes. 259 unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment; 260 EntryAlign = std::max(8U, EntryAlign); 261 Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign); 262 new (Entry) ContentCache(FileEnt); 263 return Entry; 264 } 265 266 267 /// createMemBufferContentCache - Create a new ContentCache for the specified 268 /// memory buffer. This does no caching. 269 const ContentCache* 270 SourceManager::createMemBufferContentCache(const MemoryBuffer *Buffer) { 271 // Add a new ContentCache to the MemBufferInfos list and return it. Make sure 272 // it is at least 8-byte aligned so that FileInfo can use the low 3 bits of 273 // the pointer for its own nefarious purposes. 274 unsigned EntryAlign = llvm::AlignOf<ContentCache>::Alignment; 275 EntryAlign = std::max(8U, EntryAlign); 276 ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(1, EntryAlign); 277 new (Entry) ContentCache(); 278 MemBufferInfos.push_back(Entry); 279 Entry->setBuffer(Buffer); 280 return Entry; 281 } 282 283 //===----------------------------------------------------------------------===// 284 // Methods to create new FileID's and instantiations. 285 //===----------------------------------------------------------------------===// 286 287 /// createFileID - Create a new fileID for the specified ContentCache and 288 /// include position. This works regardless of whether the ContentCache 289 /// corresponds to a file or some other input source. 290 FileID SourceManager::createFileID(const ContentCache *File, 291 SourceLocation IncludePos, 292 SrcMgr::CharacteristicKind FileCharacter) { 293 SLocEntryTable.push_back(SLocEntry::get(NextOffset, 294 FileInfo::get(IncludePos, File, 295 FileCharacter))); 296 unsigned FileSize = File->getSize(); 297 assert(NextOffset+FileSize+1 > NextOffset && "Ran out of source locations!"); 298 NextOffset += FileSize+1; 299 300 // Set LastFileIDLookup to the newly created file. The next getFileID call is 301 // almost guaranteed to be from that file. 302 return LastFileIDLookup = FileID::get(SLocEntryTable.size()-1); 303 } 304 305 /// createInstantiationLoc - Return a new SourceLocation that encodes the fact 306 /// that a token from SpellingLoc should actually be referenced from 307 /// InstantiationLoc. 308 SourceLocation SourceManager::createInstantiationLoc(SourceLocation SpellingLoc, 309 SourceLocation InstantLoc, 310 unsigned TokLength) { 311 SLocEntryTable.push_back(SLocEntry::get(NextOffset, 312 InstantiationInfo::get(InstantLoc, 313 SpellingLoc))); 314 assert(NextOffset+TokLength+1 > NextOffset && "Ran out of source locations!"); 315 NextOffset += TokLength+1; 316 return SourceLocation::getMacroLoc(NextOffset-(TokLength+1)); 317 } 318 319 /// getBufferData - Return a pointer to the start and end of the source buffer 320 /// data for the specified FileID. 321 std::pair<const char*, const char*> 322 SourceManager::getBufferData(FileID FID) const { 323 const llvm::MemoryBuffer *Buf = getBuffer(FID); 324 return std::make_pair(Buf->getBufferStart(), Buf->getBufferEnd()); 325 } 326 327 328 //===----------------------------------------------------------------------===// 329 // SourceLocation manipulation methods. 330 //===----------------------------------------------------------------------===// 331 332 /// getFileIDSlow - Return the FileID for a SourceLocation. This is a very hot 333 /// method that is used for all SourceManager queries that start with a 334 /// SourceLocation object. It is responsible for finding the entry in 335 /// SLocEntryTable which contains the specified location. 336 /// 337 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const { 338 assert(SLocOffset && "Invalid FileID"); 339 340 // After the first and second level caches, I see two common sorts of 341 // behavior: 1) a lot of searched FileID's are "near" the cached file location 342 // or are "near" the cached instantiation location. 2) others are just 343 // completely random and may be a very long way away. 344 // 345 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly 346 // then we fall back to a less cache efficient, but more scalable, binary 347 // search to find the location. 348 349 // See if this is near the file point - worst case we start scanning from the 350 // most newly created FileID. 351 std::vector<SrcMgr::SLocEntry>::const_iterator I; 352 353 if (SLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) { 354 // Neither loc prunes our search. 355 I = SLocEntryTable.end(); 356 } else { 357 // Perhaps it is near the file point. 358 I = SLocEntryTable.begin()+LastFileIDLookup.ID; 359 } 360 361 // Find the FileID that contains this. "I" is an iterator that points to a 362 // FileID whose offset is known to be larger than SLocOffset. 363 unsigned NumProbes = 0; 364 while (1) { 365 --I; 366 if (I->getOffset() <= SLocOffset) { 367 #if 0 368 printf("lin %d -> %d [%s] %d %d\n", SLocOffset, 369 I-SLocEntryTable.begin(), 370 I->isInstantiation() ? "inst" : "file", 371 LastFileIDLookup.ID, int(SLocEntryTable.end()-I)); 372 #endif 373 FileID Res = FileID::get(I-SLocEntryTable.begin()); 374 375 // If this isn't an instantiation, remember it. We have good locality 376 // across FileID lookups. 377 if (!I->isInstantiation()) 378 LastFileIDLookup = Res; 379 NumLinearScans += NumProbes+1; 380 return Res; 381 } 382 if (++NumProbes == 8) 383 break; 384 } 385 386 // Convert "I" back into an index. We know that it is an entry whose index is 387 // larger than the offset we are looking for. 388 unsigned GreaterIndex = I-SLocEntryTable.begin(); 389 // LessIndex - This is the lower bound of the range that we're searching. 390 // We know that the offset corresponding to the FileID is is less than 391 // SLocOffset. 392 unsigned LessIndex = 0; 393 NumProbes = 0; 394 while (1) { 395 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex; 396 unsigned MidOffset = SLocEntryTable[MiddleIndex].getOffset(); 397 398 ++NumProbes; 399 400 // If the offset of the midpoint is too large, chop the high side of the 401 // range to the midpoint. 402 if (MidOffset > SLocOffset) { 403 GreaterIndex = MiddleIndex; 404 continue; 405 } 406 407 // If the middle index contains the value, succeed and return. 408 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) { 409 #if 0 410 printf("bin %d -> %d [%s] %d %d\n", SLocOffset, 411 I-SLocEntryTable.begin(), 412 I->isInstantiation() ? "inst" : "file", 413 LastFileIDLookup.ID, int(SLocEntryTable.end()-I)); 414 #endif 415 FileID Res = FileID::get(MiddleIndex); 416 417 // If this isn't an instantiation, remember it. We have good locality 418 // across FileID lookups. 419 if (!I->isInstantiation()) 420 LastFileIDLookup = Res; 421 NumBinaryProbes += NumProbes; 422 return Res; 423 } 424 425 // Otherwise, move the low-side up to the middle index. 426 LessIndex = MiddleIndex; 427 } 428 } 429 430 SourceLocation SourceManager:: 431 getInstantiationLocSlowCase(SourceLocation Loc) const { 432 do { 433 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 434 Loc =getSLocEntry(LocInfo.first).getInstantiation().getInstantiationLoc(); 435 Loc = Loc.getFileLocWithOffset(LocInfo.second); 436 } while (!Loc.isFileID()); 437 438 return Loc; 439 } 440 441 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const { 442 do { 443 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 444 Loc = getSLocEntry(LocInfo.first).getInstantiation().getSpellingLoc(); 445 Loc = Loc.getFileLocWithOffset(LocInfo.second); 446 } while (!Loc.isFileID()); 447 return Loc; 448 } 449 450 451 std::pair<FileID, unsigned> 452 SourceManager::getDecomposedInstantiationLocSlowCase(const SrcMgr::SLocEntry *E, 453 unsigned Offset) const { 454 // If this is an instantiation record, walk through all the instantiation 455 // points. 456 FileID FID; 457 SourceLocation Loc; 458 do { 459 Loc = E->getInstantiation().getInstantiationLoc(); 460 461 FID = getFileID(Loc); 462 E = &getSLocEntry(FID); 463 Offset += Loc.getOffset()-E->getOffset(); 464 } while (!Loc.isFileID()); 465 466 return std::make_pair(FID, Offset); 467 } 468 469 std::pair<FileID, unsigned> 470 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E, 471 unsigned Offset) const { 472 // If this is an instantiation record, walk through all the instantiation 473 // points. 474 FileID FID; 475 SourceLocation Loc; 476 do { 477 Loc = E->getInstantiation().getSpellingLoc(); 478 479 FID = getFileID(Loc); 480 E = &getSLocEntry(FID); 481 Offset += Loc.getOffset()-E->getOffset(); 482 } while (!Loc.isFileID()); 483 484 return std::make_pair(FID, Offset); 485 } 486 487 488 //===----------------------------------------------------------------------===// 489 // Queries about the code at a SourceLocation. 490 //===----------------------------------------------------------------------===// 491 492 /// getCharacterData - Return a pointer to the start of the specified location 493 /// in the appropriate MemoryBuffer. 494 const char *SourceManager::getCharacterData(SourceLocation SL) const { 495 // Note that this is a hot function in the getSpelling() path, which is 496 // heavily used by -E mode. 497 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL); 498 499 // Note that calling 'getBuffer()' may lazily page in a source file. 500 return getSLocEntry(LocInfo.first).getFile().getContentCache() 501 ->getBuffer()->getBufferStart() + LocInfo.second; 502 } 503 504 505 /// getColumnNumber - Return the column # for the specified file position. 506 /// this is significantly cheaper to compute than the line number. 507 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos) const { 508 const char *Buf = getBuffer(FID)->getBufferStart(); 509 510 unsigned LineStart = FilePos; 511 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r') 512 --LineStart; 513 return FilePos-LineStart+1; 514 } 515 516 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc) const { 517 if (Loc.isInvalid()) return 0; 518 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 519 return getColumnNumber(LocInfo.first, LocInfo.second); 520 } 521 522 unsigned SourceManager::getInstantiationColumnNumber(SourceLocation Loc) const { 523 if (Loc.isInvalid()) return 0; 524 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 525 return getColumnNumber(LocInfo.first, LocInfo.second); 526 } 527 528 529 530 static void ComputeLineNumbers(ContentCache* FI, 531 llvm::BumpPtrAllocator &Alloc) DISABLE_INLINE; 532 static void ComputeLineNumbers(ContentCache* FI, llvm::BumpPtrAllocator &Alloc){ 533 // Note that calling 'getBuffer()' may lazily page in the file. 534 const MemoryBuffer *Buffer = FI->getBuffer(); 535 536 // Find the file offsets of all of the *physical* source lines. This does 537 // not look at trigraphs, escaped newlines, or anything else tricky. 538 std::vector<unsigned> LineOffsets; 539 540 // Line #1 starts at char 0. 541 LineOffsets.push_back(0); 542 543 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart(); 544 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd(); 545 unsigned Offs = 0; 546 while (1) { 547 // Skip over the contents of the line. 548 // TODO: Vectorize this? This is very performance sensitive for programs 549 // with lots of diagnostics and in -E mode. 550 const unsigned char *NextBuf = (const unsigned char *)Buf; 551 while (*NextBuf != '\n' && *NextBuf != '\r' && *NextBuf != '\0') 552 ++NextBuf; 553 Offs += NextBuf-Buf; 554 Buf = NextBuf; 555 556 if (Buf[0] == '\n' || Buf[0] == '\r') { 557 // If this is \n\r or \r\n, skip both characters. 558 if ((Buf[1] == '\n' || Buf[1] == '\r') && Buf[0] != Buf[1]) 559 ++Offs, ++Buf; 560 ++Offs, ++Buf; 561 LineOffsets.push_back(Offs); 562 } else { 563 // Otherwise, this is a null. If end of file, exit. 564 if (Buf == End) break; 565 // Otherwise, skip the null. 566 ++Offs, ++Buf; 567 } 568 } 569 570 // Copy the offsets into the FileInfo structure. 571 FI->NumLines = LineOffsets.size(); 572 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size()); 573 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache); 574 } 575 576 /// getLineNumber - Given a SourceLocation, return the spelling line number 577 /// for the position indicated. This requires building and caching a table of 578 /// line offsets for the MemoryBuffer, so this is not cheap: use only when 579 /// about to emit a diagnostic. 580 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos) const { 581 ContentCache *Content; 582 if (LastLineNoFileIDQuery == FID) 583 Content = LastLineNoContentCache; 584 else 585 Content = const_cast<ContentCache*>(getSLocEntry(FID) 586 .getFile().getContentCache()); 587 588 // If this is the first use of line information for this buffer, compute the 589 /// SourceLineCache for it on demand. 590 if (Content->SourceLineCache == 0) 591 ComputeLineNumbers(Content, ContentCacheAlloc); 592 593 // Okay, we know we have a line number table. Do a binary search to find the 594 // line number that this character position lands on. 595 unsigned *SourceLineCache = Content->SourceLineCache; 596 unsigned *SourceLineCacheStart = SourceLineCache; 597 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines; 598 599 unsigned QueriedFilePos = FilePos+1; 600 601 // If the previous query was to the same file, we know both the file pos from 602 // that query and the line number returned. This allows us to narrow the 603 // search space from the entire file to something near the match. 604 if (LastLineNoFileIDQuery == FID) { 605 if (QueriedFilePos >= LastLineNoFilePos) { 606 SourceLineCache = SourceLineCache+LastLineNoResult-1; 607 608 // The query is likely to be nearby the previous one. Here we check to 609 // see if it is within 5, 10 or 20 lines. It can be far away in cases 610 // where big comment blocks and vertical whitespace eat up lines but 611 // contribute no tokens. 612 if (SourceLineCache+5 < SourceLineCacheEnd) { 613 if (SourceLineCache[5] > QueriedFilePos) 614 SourceLineCacheEnd = SourceLineCache+5; 615 else if (SourceLineCache+10 < SourceLineCacheEnd) { 616 if (SourceLineCache[10] > QueriedFilePos) 617 SourceLineCacheEnd = SourceLineCache+10; 618 else if (SourceLineCache+20 < SourceLineCacheEnd) { 619 if (SourceLineCache[20] > QueriedFilePos) 620 SourceLineCacheEnd = SourceLineCache+20; 621 } 622 } 623 } 624 } else { 625 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1; 626 } 627 } 628 629 // If the spread is large, do a "radix" test as our initial guess, based on 630 // the assumption that lines average to approximately the same length. 631 // NOTE: This is currently disabled, as it does not appear to be profitable in 632 // initial measurements. 633 if (0 && SourceLineCacheEnd-SourceLineCache > 20) { 634 unsigned FileLen = Content->SourceLineCache[Content->NumLines-1]; 635 636 // Take a stab at guessing where it is. 637 unsigned ApproxPos = Content->NumLines*QueriedFilePos / FileLen; 638 639 // Check for -10 and +10 lines. 640 unsigned LowerBound = std::max(int(ApproxPos-10), 0); 641 unsigned UpperBound = std::min(ApproxPos+10, FileLen); 642 643 // If the computed lower bound is less than the query location, move it in. 644 if (SourceLineCache < SourceLineCacheStart+LowerBound && 645 SourceLineCacheStart[LowerBound] < QueriedFilePos) 646 SourceLineCache = SourceLineCacheStart+LowerBound; 647 648 // If the computed upper bound is greater than the query location, move it. 649 if (SourceLineCacheEnd > SourceLineCacheStart+UpperBound && 650 SourceLineCacheStart[UpperBound] >= QueriedFilePos) 651 SourceLineCacheEnd = SourceLineCacheStart+UpperBound; 652 } 653 654 unsigned *Pos 655 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos); 656 unsigned LineNo = Pos-SourceLineCacheStart; 657 658 LastLineNoFileIDQuery = FID; 659 LastLineNoContentCache = Content; 660 LastLineNoFilePos = QueriedFilePos; 661 LastLineNoResult = LineNo; 662 return LineNo; 663 } 664 665 unsigned SourceManager::getInstantiationLineNumber(SourceLocation Loc) const { 666 if (Loc.isInvalid()) return 0; 667 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 668 return getLineNumber(LocInfo.first, LocInfo.second); 669 } 670 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc) const { 671 if (Loc.isInvalid()) return 0; 672 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 673 return getLineNumber(LocInfo.first, LocInfo.second); 674 } 675 676 677 /// getPresumedLoc - This method returns the "presumed" location of a 678 /// SourceLocation specifies. A "presumed location" can be modified by #line 679 /// or GNU line marker directives. This provides a view on the data that a 680 /// user should see in diagnostics, for example. 681 /// 682 /// Note that a presumed location is always given as the instantiation point 683 /// of an instantiation location, not at the spelling location. 684 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc) const { 685 if (Loc.isInvalid()) return PresumedLoc(); 686 687 // Presumed locations are always for instantiation points. 688 std::pair<FileID, unsigned> LocInfo = getDecomposedInstantiationLoc(Loc); 689 690 const SrcMgr::FileInfo &FI = getSLocEntry(LocInfo.first).getFile(); 691 const SrcMgr::ContentCache *C = FI.getContentCache(); 692 693 // To get the source name, first consult the FileEntry (if one exists) 694 // before the MemBuffer as this will avoid unnecessarily paging in the 695 // MemBuffer. 696 const char *Filename = 697 C->Entry ? C->Entry->getName() : C->getBuffer()->getBufferIdentifier(); 698 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second); 699 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second); 700 SourceLocation IncludeLoc = FI.getIncludeLoc(); 701 702 // If we have #line directives in this file, update and overwrite the physical 703 // location info if appropriate. 704 if (FI.hasLineDirectives()) { 705 assert(LineTable && "Can't have linetable entries without a LineTable!"); 706 // See if there is a #line directive before this. If so, get it. 707 if (const LineEntry *Entry = 708 LineTable->FindNearestLineEntry(LocInfo.first.ID, LocInfo.second)) { 709 // If the LineEntry indicates a filename, use it. 710 if (Entry->FilenameID != -1) 711 Filename = LineTable->getFilename(Entry->FilenameID); 712 713 // Use the line number specified by the LineEntry. This line number may 714 // be multiple lines down from the line entry. Add the difference in 715 // physical line numbers from the query point and the line marker to the 716 // total. 717 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset); 718 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1); 719 720 // Note that column numbers are not molested by line markers. 721 } 722 } 723 724 return PresumedLoc(Filename, LineNo, ColNo, IncludeLoc); 725 } 726 727 //===----------------------------------------------------------------------===// 728 // Other miscellaneous methods. 729 //===----------------------------------------------------------------------===// 730 731 732 /// PrintStats - Print statistics to stderr. 733 /// 734 void SourceManager::PrintStats() const { 735 llvm::cerr << "\n*** Source Manager Stats:\n"; 736 llvm::cerr << FileInfos.size() << " files mapped, " << MemBufferInfos.size() 737 << " mem buffers mapped.\n"; 738 llvm::cerr << SLocEntryTable.size() << " SLocEntry's allocated, " 739 << NextOffset << "B of Sloc address space used.\n"; 740 741 unsigned NumLineNumsComputed = 0; 742 unsigned NumFileBytesMapped = 0; 743 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){ 744 NumLineNumsComputed += I->second->SourceLineCache != 0; 745 NumFileBytesMapped += I->second->getSizeBytesMapped(); 746 } 747 748 llvm::cerr << NumFileBytesMapped << " bytes of files mapped, " 749 << NumLineNumsComputed << " files with line #'s computed.\n"; 750 llvm::cerr << "FileID scans: " << NumLinearScans << " linear, " 751 << NumBinaryProbes << " binary.\n"; 752 } 753 754 //===----------------------------------------------------------------------===// 755 // Serialization. 756 //===----------------------------------------------------------------------===// 757 758 void ContentCache::Emit(llvm::Serializer& S) const { 759 S.FlushRecord(); 760 S.EmitPtr(this); 761 762 if (Entry) { 763 llvm::sys::Path Fname(Buffer->getBufferIdentifier()); 764 765 if (Fname.isAbsolute()) 766 S.EmitCStr(Fname.c_str()); 767 else { 768 // Create an absolute path. 769 // FIXME: This will potentially contain ".." and "." in the path. 770 llvm::sys::Path path = llvm::sys::Path::GetCurrentDirectory(); 771 path.appendComponent(Fname.c_str()); 772 S.EmitCStr(path.c_str()); 773 } 774 } 775 else { 776 const char* p = Buffer->getBufferStart(); 777 const char* e = Buffer->getBufferEnd(); 778 779 S.EmitInt(e-p); 780 781 for ( ; p != e; ++p) 782 S.EmitInt(*p); 783 } 784 785 S.FlushRecord(); 786 } 787 788 void ContentCache::ReadToSourceManager(llvm::Deserializer& D, 789 SourceManager& SMgr, 790 FileManager* FMgr, 791 std::vector<char>& Buf) { 792 if (FMgr) { 793 llvm::SerializedPtrID PtrID = D.ReadPtrID(); 794 D.ReadCStr(Buf,false); 795 796 // Create/fetch the FileEntry. 797 const char* start = &Buf[0]; 798 const FileEntry* E = FMgr->getFile(start,start+Buf.size()); 799 800 // FIXME: Ideally we want a lazy materialization of the ContentCache 801 // anyway, because we don't want to read in source files unless this 802 // is absolutely needed. 803 if (!E) 804 D.RegisterPtr(PtrID,NULL); 805 else 806 // Get the ContextCache object and register it with the deserializer. 807 D.RegisterPtr(PtrID, SMgr.getOrCreateContentCache(E)); 808 return; 809 } 810 811 // Register the ContextCache object with the deserializer. 812 /* FIXME: 813 ContentCache *Entry 814 SMgr.MemBufferInfos.push_back(ContentCache()); 815 = const_cast<ContentCache&>(SMgr.MemBufferInfos.back()); 816 D.RegisterPtr(&Entry); 817 818 // Create the buffer. 819 unsigned Size = D.ReadInt(); 820 Entry.Buffer = MemoryBuffer::getNewUninitMemBuffer(Size); 821 822 // Read the contents of the buffer. 823 char* p = const_cast<char*>(Entry.Buffer->getBufferStart()); 824 for (unsigned i = 0; i < Size ; ++i) 825 p[i] = D.ReadInt(); 826 */ 827 } 828 829 void SourceManager::Emit(llvm::Serializer& S) const { 830 S.EnterBlock(); 831 S.EmitPtr(this); 832 S.EmitInt(MainFileID.getOpaqueValue()); 833 834 // Emit: FileInfos. Just emit the file name. 835 S.EnterBlock(); 836 837 // FIXME: Emit FileInfos. 838 //std::for_each(FileInfos.begin(), FileInfos.end(), 839 // S.MakeEmitter<ContentCache>()); 840 841 S.ExitBlock(); 842 843 // Emit: MemBufferInfos 844 S.EnterBlock(); 845 846 /* FIXME: EMIT. 847 std::for_each(MemBufferInfos.begin(), MemBufferInfos.end(), 848 S.MakeEmitter<ContentCache>()); 849 */ 850 851 S.ExitBlock(); 852 853 // FIXME: Emit SLocEntryTable. 854 855 S.ExitBlock(); 856 } 857 858 SourceManager* 859 SourceManager::CreateAndRegister(llvm::Deserializer &D, FileManager &FMgr) { 860 SourceManager *M = new SourceManager(); 861 D.RegisterPtr(M); 862 863 // Read: the FileID of the main source file of the translation unit. 864 M->MainFileID = FileID::get(D.ReadInt()); 865 866 std::vector<char> Buf; 867 868 /*{ // FIXME Read: FileInfos. 869 llvm::Deserializer::Location BLoc = D.getCurrentBlockLocation(); 870 while (!D.FinishedBlock(BLoc)) 871 ContentCache::ReadToSourceManager(D,*M,&FMgr,Buf); 872 }*/ 873 874 { // Read: MemBufferInfos. 875 llvm::Deserializer::Location BLoc = D.getCurrentBlockLocation(); 876 while (!D.FinishedBlock(BLoc)) 877 ContentCache::ReadToSourceManager(D,*M,NULL,Buf); 878 } 879 880 // FIXME: Read SLocEntryTable. 881 882 return M; 883 } 884