SourceManager.h revision 1036b68525f39cb69ac22c679ed440acd8392a16
1//===--- SourceManager.h - Track and cache source files ---------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file was developed by Chris Lattner and is distributed under 6// the University of Illinois Open Source License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file defines the SourceManager interface. 11// 12//===----------------------------------------------------------------------===// 13 14#ifndef LLVM_CLANG_SOURCEMANAGER_H 15#define LLVM_CLANG_SOURCEMANAGER_H 16 17#include "clang/Basic/SourceLocation.h" 18#include "llvm/Bitcode/SerializationFwd.h" 19#include <vector> 20#include <set> 21#include <list> 22#include <cassert> 23 24namespace llvm { 25class MemoryBuffer; 26} 27 28namespace clang { 29 30class SourceManager; 31class FileManager; 32class FileEntry; 33class IdentifierTokenInfo; 34 35/// SrcMgr - Private classes that are part of the SourceManager implementation. 36/// 37namespace SrcMgr { 38 /// ContentCache - Once instance of this struct is kept for every file 39 /// loaded or used. This object owns the MemoryBuffer object. 40 struct ContentCache { 41 /// Reference to the file entry. This reference does not own 42 /// the FileEntry object. It is possible for this to be NULL if 43 /// the ContentCache encapsulates an imaginary text buffer. 44 const FileEntry* Entry; 45 46 /// Buffer - The actual buffer containing the characters from the input 47 /// file. This is owned by the ContentCache object. 48 const llvm::MemoryBuffer* Buffer; 49 50 /// SourceLineCache - A new[]'d array of offsets for each source line. This 51 /// is lazily computed. This is owned by the ContentCache object. 52 unsigned* SourceLineCache; 53 54 /// NumLines - The number of lines in this ContentCache. This is only valid 55 /// if SourceLineCache is non-null. 56 unsigned NumLines; 57 58 ContentCache(const FileEntry* e = NULL) 59 : Entry(e), Buffer(NULL), SourceLineCache(NULL), NumLines(0) {} 60 61 ~ContentCache(); 62 63 /// The copy ctor does not allow copies where source object has either 64 /// a non-NULL Buffer or SourceLineCache. Ownership of allocated memory 65 /// is not transfered, so this is a logical error. 66 ContentCache(const ContentCache& RHS) : Buffer(NULL),SourceLineCache(NULL) { 67 Entry = RHS.Entry; 68 69 assert (RHS.Buffer == NULL && RHS.SourceLineCache == NULL 70 && "Passed ContentCache object cannot own a buffer."); 71 72 NumLines = RHS.NumLines; 73 } 74 75 /// Emit - Emit this ContentCache to Bitcode. 76 void Emit(llvm::Serializer& S) const; 77 78 /// ReadToSourceManager - Reconstitute a ContentCache from Bitcode 79 // and store it in the specified SourceManager. 80 static void ReadToSourceManager(llvm::Deserializer& D, SourceManager& SMgr, 81 FileManager* FMgr, std::vector<char>& Buf); 82 83 private: 84 // Disable assignments. 85 ContentCache& operator=(const ContentCache& RHS); 86 }; 87 88 /// FileIDInfo - Information about a FileID, basically just the logical file 89 /// that it represents and include stack information. A File SourceLocation 90 /// is a byte offset from the start of this. 91 /// 92 /// FileID's are used to compute the location of a character in memory as well 93 /// as the logical source location, which can be differ from the physical 94 /// location. It is different when #line's are active or when macros have 95 /// been expanded. 96 /// 97 /// Each FileID has include stack information, indicating where it came from. 98 /// For the primary translation unit, it comes from SourceLocation() aka 0. 99 /// This information encodes the #include chain that a token was instantiated 100 /// from. 101 /// 102 /// FileIDInfos contain a "ContentCache *", describing the source file, 103 /// and a Chunk number, which allows a SourceLocation to index into very 104 /// large files (those which there are not enough FilePosBits to address). 105 /// 106 struct FileIDInfo { 107 private: 108 /// IncludeLoc - The location of the #include that brought in this file. 109 /// This SourceLocation object has an invalid SLOC for the main file. 110 SourceLocation IncludeLoc; 111 112 /// ChunkNo - Really large buffers are broken up into chunks that are 113 /// each (1 << SourceLocation::FilePosBits) in size. This specifies the 114 /// chunk number of this FileID. 115 unsigned ChunkNo; 116 117 /// Content - Information about the source buffer itself. 118 const ContentCache* Content; 119 120 public: 121 /// get - Return a FileIDInfo object. 122 static FileIDInfo get(SourceLocation IL, unsigned CN, 123 const ContentCache *Con) { 124 FileIDInfo X; 125 X.IncludeLoc = IL; 126 X.ChunkNo = CN; 127 X.Content = Con; 128 return X; 129 } 130 131 SourceLocation getIncludeLoc() const { return IncludeLoc; } 132 unsigned getChunkNo() const { return ChunkNo; } 133 const ContentCache* getContentCache() const { return Content; } 134 135 /// Emit - Emit this FileIDInfo to Bitcode. 136 void Emit(llvm::Serializer& S) const; 137 138 /// ReadVal - Reconstitute a FileIDInfo from Bitcode. 139 static FileIDInfo ReadVal(llvm::Deserializer& S); 140 }; 141 142 /// MacroIDInfo - Macro SourceLocations refer to these records by their ID. 143 /// Each MacroIDInfo encodes the Instantiation location - where the macro was 144 /// instantiated, and the PhysicalLoc - where the actual character data for 145 /// the token came from. An actual macro SourceLocation stores deltas from 146 /// these positions. 147 class MacroIDInfo { 148 SourceLocation VirtualLoc, PhysicalLoc; 149 public: 150 SourceLocation getVirtualLoc() const { return VirtualLoc; } 151 SourceLocation getPhysicalLoc() const { return PhysicalLoc; } 152 153 /// get - Return a MacroID for a macro expansion. VL specifies 154 /// the instantiation location (where the macro is expanded), and PL 155 /// specifies the physical location (where the characters from the token 156 /// come from). Both VL and PL refer to normal File SLocs. 157 static MacroIDInfo get(SourceLocation VL, SourceLocation PL) { 158 MacroIDInfo X; 159 X.VirtualLoc = VL; 160 X.PhysicalLoc = PL; 161 return X; 162 } 163 164 /// Emit - Emit this MacroIDInfo to Bitcode. 165 void Emit(llvm::Serializer& S) const; 166 167 /// ReadVal - Reconstitute a MacroIDInfo from Bitcode. 168 static MacroIDInfo ReadVal(llvm::Deserializer& S); 169 }; 170} // end SrcMgr namespace. 171} // end clang namespace 172 173namespace std { 174template <> struct less<clang::SrcMgr::ContentCache> { 175 inline bool operator()(const clang::SrcMgr::ContentCache& L, 176 const clang::SrcMgr::ContentCache& R) const { 177 return L.Entry < R.Entry; 178 } 179}; 180} // end std namespace 181 182namespace clang { 183 184/// SourceManager - This file handles loading and caching of source files into 185/// memory. This object owns the MemoryBuffer objects for all of the loaded 186/// files and assigns unique FileID's for each unique #include chain. 187/// 188/// The SourceManager can be queried for information about SourceLocation 189/// objects, turning them into either physical or logical locations. Physical 190/// locations represent where the bytes corresponding to a token came from and 191/// logical locations represent where the location is in the user's view. In 192/// the case of a macro expansion, for example, the physical location indicates 193/// where the expanded token came from and the logical location specifies where 194/// it was expanded. Logical locations are also influenced by #line directives, 195/// etc. 196class SourceManager { 197 /// FileInfos - Memoized information about all of the files tracked by this 198 /// SourceManager. This set allows us to merge ContentCache entries based 199 /// on their FileEntry*. All ContentCache objects will thus have unique, 200 /// non-null, FileEntry pointers. 201 std::set<SrcMgr::ContentCache> FileInfos; 202 203 /// MemBufferInfos - Information about various memory buffers that we have 204 /// read in. This is a list, instead of a vector, because we need pointers to 205 /// the ContentCache objects to be stable. All FileEntry* within the 206 /// stored ContentCache objects are NULL, as they do not refer to a file. 207 std::list<SrcMgr::ContentCache> MemBufferInfos; 208 209 /// FileIDs - Information about each FileID. FileID #0 is not valid, so all 210 /// entries are off by one. 211 std::vector<SrcMgr::FileIDInfo> FileIDs; 212 213 /// MacroIDs - Information about each MacroID. 214 std::vector<SrcMgr::MacroIDInfo> MacroIDs; 215 216 /// LastLineNo - These ivars serve as a cache used in the getLineNumber 217 /// method which is used to speedup getLineNumber calls to nearby locations. 218 unsigned LastLineNoFileIDQuery; 219 SrcMgr::ContentCache *LastLineNoContentCache; 220 unsigned LastLineNoFilePos; 221 unsigned LastLineNoResult; 222 223 /// MainFileID - The file ID for the main source file of the translation unit. 224 unsigned MainFileID; 225 226public: 227 SourceManager() : LastLineNoFileIDQuery(~0U), MainFileID(0) {} 228 ~SourceManager() {} 229 230 void clearIDTables() { 231 FileIDs.clear(); 232 MacroIDs.clear(); 233 LastLineNoFileIDQuery = ~0U; 234 LastLineNoContentCache = 0; 235 } 236 237 /// getMainFileID - Returns the FileID of the main source file. 238 unsigned getMainFileID() const { return MainFileID; } 239 240 /// createFileID - Create a new FileID that represents the specified file 241 /// being #included from the specified IncludePosition. This returns 0 on 242 /// error and translates NULL into standard input. 243 unsigned createFileID(const FileEntry *SourceFile, SourceLocation IncludePos){ 244 const SrcMgr::ContentCache *IR = getContentCache(SourceFile); 245 if (IR == 0) return 0; // Error opening file? 246 return createFileID(IR, IncludePos); 247 } 248 249 /// createMainFileID - Create the FileID for the main source file. 250 unsigned createMainFileID(const FileEntry *SourceFile, 251 SourceLocation IncludePos) { 252 253 assert (MainFileID == 0 && "MainFileID already set!"); 254 MainFileID = createFileID(SourceFile,IncludePos); 255 return MainFileID; 256 } 257 258 /// createFileIDForMemBuffer - Create a new FileID that represents the 259 /// specified memory buffer. This does no caching of the buffer and takes 260 /// ownership of the MemoryBuffer, so only pass a MemoryBuffer to this once. 261 unsigned createFileIDForMemBuffer(const llvm::MemoryBuffer *Buffer) { 262 return createFileID(createMemBufferContentCache(Buffer), SourceLocation()); 263 } 264 265 /// createMainFileIDForMembuffer - Create the FileID for a memory buffer 266 /// that will represent the FileID for the main source. One example 267 /// of when this would be used is when the main source is read from STDIN. 268 unsigned createMainFileIDForMemBuffer(const llvm::MemoryBuffer *Buffer) { 269 assert (MainFileID == 0 && "MainFileID already set!"); 270 MainFileID = createMainFileIDForMemBuffer(Buffer); 271 return MainFileID; 272 } 273 274 /// getInstantiationLoc - Return a new SourceLocation that encodes the fact 275 /// that a token at Loc should actually be referenced from InstantiationLoc. 276 SourceLocation getInstantiationLoc(SourceLocation Loc, 277 SourceLocation InstantiationLoc); 278 279 /// getBuffer - Return the buffer for the specified FileID. 280 /// 281 const llvm::MemoryBuffer *getBuffer(unsigned FileID) const { 282 return getContentCache(FileID)->Buffer; 283 } 284 285 /// getBufferData - Return a pointer to the start and end of the character 286 /// data for the specified FileID. 287 std::pair<const char*, const char*> getBufferData(unsigned FileID) const; 288 289 /// getIncludeLoc - Return the location of the #include for the specified 290 /// SourceLocation. If this is a macro expansion, this transparently figures 291 /// out which file includes the file being expanded into. 292 SourceLocation getIncludeLoc(SourceLocation ID) const { 293 return getFIDInfo(getLogicalLoc(ID).getFileID())->getIncludeLoc(); 294 } 295 296 /// getCharacterData - Return a pointer to the start of the specified location 297 /// in the appropriate MemoryBuffer. 298 const char *getCharacterData(SourceLocation SL) const; 299 300 /// getColumnNumber - Return the column # for the specified file position. 301 /// This is significantly cheaper to compute than the line number. This 302 /// returns zero if the column number isn't known. This may only be called on 303 /// a file sloc, so you must choose a physical or logical location before 304 /// calling this method. 305 unsigned getColumnNumber(SourceLocation Loc) const; 306 307 unsigned getPhysicalColumnNumber(SourceLocation Loc) const { 308 return getColumnNumber(getPhysicalLoc(Loc)); 309 } 310 unsigned getLogicalColumnNumber(SourceLocation Loc) const { 311 return getColumnNumber(getLogicalLoc(Loc)); 312 } 313 314 315 /// getLineNumber - Given a SourceLocation, return the physical line number 316 /// for the position indicated. This requires building and caching a table of 317 /// line offsets for the MemoryBuffer, so this is not cheap: use only when 318 /// about to emit a diagnostic. 319 unsigned getLineNumber(SourceLocation Loc); 320 321 unsigned getLogicalLineNumber(SourceLocation Loc) { 322 return getLineNumber(getLogicalLoc(Loc)); 323 } 324 unsigned getPhysicalLineNumber(SourceLocation Loc) { 325 return getLineNumber(getPhysicalLoc(Loc)); 326 } 327 328 /// getSourceName - This method returns the name of the file or buffer that 329 /// the SourceLocation specifies. This can be modified with #line directives, 330 /// etc. 331 const char *getSourceName(SourceLocation Loc) const; 332 333 /// Given a SourceLocation object, return the logical location referenced by 334 /// the ID. This logical location is subject to #line directives, etc. 335 SourceLocation getLogicalLoc(SourceLocation Loc) const { 336 // File locations are both physical and logical. 337 if (Loc.isFileID()) return Loc; 338 339 return MacroIDs[Loc.getMacroID()].getVirtualLoc(); 340 } 341 342 /// getPhysicalLoc - Given a SourceLocation object, return the physical 343 /// location referenced by the ID. 344 SourceLocation getPhysicalLoc(SourceLocation Loc) const { 345 // File locations are both physical and logical. 346 if (Loc.isFileID()) return Loc; 347 348 SourceLocation PLoc = MacroIDs[Loc.getMacroID()].getPhysicalLoc(); 349 return PLoc.getFileLocWithOffset(Loc.getMacroPhysOffs()); 350 } 351 352 /// getContentCacheForLoc - Return the ContentCache for the physloc of the 353 /// specified SourceLocation, if one exists. 354 const SrcMgr::ContentCache* getContentCacheForLoc(SourceLocation Loc) const { 355 Loc = getPhysicalLoc(Loc); 356 unsigned FileID = Loc.getFileID(); 357 assert(FileID-1 < FileIDs.size() && "Invalid FileID!"); 358 return FileIDs[FileID-1].getContentCache(); 359 } 360 361 /// getFileEntryForLoc - Return the FileEntry record for the physloc of the 362 /// specified SourceLocation, if one exists. 363 const FileEntry* getFileEntryForLoc(SourceLocation Loc) const { 364 return getContentCacheForLoc(Loc)->Entry; 365 } 366 367 /// getDecomposedFileLoc - Decompose the specified file location into a raw 368 /// FileID + Offset pair. The first element is the FileID, the second is the 369 /// offset from the start of the buffer of the location. 370 std::pair<unsigned, unsigned> getDecomposedFileLoc(SourceLocation Loc) const { 371 assert(Loc.isFileID() && "Isn't a File SourceLocation"); 372 373 // TODO: Add a flag "is first chunk" to SLOC. 374 const SrcMgr::FileIDInfo *FIDInfo = getFIDInfo(Loc.getFileID()); 375 376 // If this file has been split up into chunks, factor in the chunk number 377 // that the FileID references. 378 unsigned ChunkNo = FIDInfo->getChunkNo(); 379 unsigned Offset = Loc.getRawFilePos(); 380 Offset += (ChunkNo << SourceLocation::FilePosBits); 381 382 return std::pair<unsigned,unsigned>(Loc.getFileID()-ChunkNo, Offset); 383 } 384 385 /// PrintStats - Print statistics to stderr. 386 /// 387 void PrintStats() const; 388 389 /// Emit - Emit this SourceManager to Bitcode. 390 void Emit(llvm::Serializer& S) const; 391 392 /// Read - Reconstitute a SourceManager from Bitcode. 393 static SourceManager* CreateAndRegister(llvm::Deserializer& S, 394 FileManager &FMgr); 395 396private: 397 friend class SrcMgr::ContentCache; // Used for deserialization. 398 399 /// createFileID - Create a new fileID for the specified ContentCache and 400 /// include position. This works regardless of whether the ContentCache 401 /// corresponds to a file or some other input source. 402 unsigned createFileID(const SrcMgr::ContentCache* File, 403 SourceLocation IncludePos); 404 405 /// getContentCache - Create or return a cached ContentCache for the specified 406 /// file. This returns null on failure. 407 const SrcMgr::ContentCache* getContentCache(const FileEntry* SourceFile); 408 409 /// createMemBufferContentCache - Create a new ContentCache for the specified 410 /// memory buffer. 411 const SrcMgr::ContentCache* 412 createMemBufferContentCache(const llvm::MemoryBuffer* Buf); 413 414 const SrcMgr::FileIDInfo* getFIDInfo(unsigned FileID) const { 415 assert(FileID-1 < FileIDs.size() && "Invalid FileID!"); 416 return &FileIDs[FileID-1]; 417 } 418 419 const SrcMgr::ContentCache *getContentCache(unsigned FileID) const { 420 return getContentCache(getFIDInfo(FileID)); 421 } 422 423 /// Return the ContentCache structure for the specified FileID. 424 /// This is always the physical reference for the ID. 425 const SrcMgr::ContentCache* 426 getContentCache(const SrcMgr::FileIDInfo* FIDInfo) const { 427 return FIDInfo->getContentCache(); 428 } 429 430 /// getFullFilePos - This (efficient) method returns the offset from the start 431 /// of the file that the specified physical SourceLocation represents. This 432 /// returns the location of the physical character data, not the logical file 433 /// position. 434 unsigned getFullFilePos(SourceLocation PhysLoc) const { 435 return getDecomposedFileLoc(PhysLoc).second; 436 } 437}; 438 439 440} // end namespace clang 441 442#endif 443