1 //===--- Replacement.cpp - Framework for clang refactoring tools ----------===// 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 // Implements classes to support/store refactorings. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/Tooling/Core/Replacement.h" 15 16 #include "clang/Basic/Diagnostic.h" 17 #include "clang/Basic/DiagnosticIDs.h" 18 #include "clang/Basic/DiagnosticOptions.h" 19 #include "clang/Basic/FileManager.h" 20 #include "clang/Basic/SourceManager.h" 21 #include "clang/Lex/Lexer.h" 22 #include "clang/Rewrite/Core/Rewriter.h" 23 #include "llvm/Support/FileSystem.h" 24 #include "llvm/Support/Path.h" 25 #include "llvm/Support/raw_os_ostream.h" 26 27 namespace clang { 28 namespace tooling { 29 30 static const char * const InvalidLocation = ""; 31 32 Replacement::Replacement() 33 : FilePath(InvalidLocation) {} 34 35 Replacement::Replacement(StringRef FilePath, unsigned Offset, unsigned Length, 36 StringRef ReplacementText) 37 : FilePath(FilePath), ReplacementRange(Offset, Length), 38 ReplacementText(ReplacementText) {} 39 40 Replacement::Replacement(const SourceManager &Sources, SourceLocation Start, 41 unsigned Length, StringRef ReplacementText) { 42 setFromSourceLocation(Sources, Start, Length, ReplacementText); 43 } 44 45 Replacement::Replacement(const SourceManager &Sources, 46 const CharSourceRange &Range, 47 StringRef ReplacementText, 48 const LangOptions &LangOpts) { 49 setFromSourceRange(Sources, Range, ReplacementText, LangOpts); 50 } 51 52 bool Replacement::isApplicable() const { 53 return FilePath != InvalidLocation; 54 } 55 56 bool Replacement::apply(Rewriter &Rewrite) const { 57 SourceManager &SM = Rewrite.getSourceMgr(); 58 const FileEntry *Entry = SM.getFileManager().getFile(FilePath); 59 if (!Entry) 60 return false; 61 62 FileID ID = SM.getOrCreateFileID(Entry, SrcMgr::C_User); 63 const SourceLocation Start = 64 SM.getLocForStartOfFile(ID). 65 getLocWithOffset(ReplacementRange.getOffset()); 66 // ReplaceText returns false on success. 67 // ReplaceText only fails if the source location is not a file location, in 68 // which case we already returned false earlier. 69 bool RewriteSucceeded = !Rewrite.ReplaceText( 70 Start, ReplacementRange.getLength(), ReplacementText); 71 assert(RewriteSucceeded); 72 return RewriteSucceeded; 73 } 74 75 std::string Replacement::toString() const { 76 std::string Result; 77 llvm::raw_string_ostream Stream(Result); 78 Stream << FilePath << ": " << ReplacementRange.getOffset() << ":+" 79 << ReplacementRange.getLength() << ":\"" << ReplacementText << "\""; 80 return Stream.str(); 81 } 82 83 bool operator<(const Replacement &LHS, const Replacement &RHS) { 84 if (LHS.getOffset() != RHS.getOffset()) 85 return LHS.getOffset() < RHS.getOffset(); 86 87 // Apply longer replacements first, specifically so that deletions are 88 // executed before insertions. It is (hopefully) never the intention to 89 // delete parts of newly inserted code. 90 if (LHS.getLength() != RHS.getLength()) 91 return LHS.getLength() > RHS.getLength(); 92 93 if (LHS.getFilePath() != RHS.getFilePath()) 94 return LHS.getFilePath() < RHS.getFilePath(); 95 return LHS.getReplacementText() < RHS.getReplacementText(); 96 } 97 98 bool operator==(const Replacement &LHS, const Replacement &RHS) { 99 return LHS.getOffset() == RHS.getOffset() && 100 LHS.getLength() == RHS.getLength() && 101 LHS.getFilePath() == RHS.getFilePath() && 102 LHS.getReplacementText() == RHS.getReplacementText(); 103 } 104 105 void Replacement::setFromSourceLocation(const SourceManager &Sources, 106 SourceLocation Start, unsigned Length, 107 StringRef ReplacementText) { 108 const std::pair<FileID, unsigned> DecomposedLocation = 109 Sources.getDecomposedLoc(Start); 110 const FileEntry *Entry = Sources.getFileEntryForID(DecomposedLocation.first); 111 this->FilePath = Entry ? Entry->getName() : InvalidLocation; 112 this->ReplacementRange = Range(DecomposedLocation.second, Length); 113 this->ReplacementText = ReplacementText; 114 } 115 116 // FIXME: This should go into the Lexer, but we need to figure out how 117 // to handle ranges for refactoring in general first - there is no obvious 118 // good way how to integrate this into the Lexer yet. 119 static int getRangeSize(const SourceManager &Sources, 120 const CharSourceRange &Range, 121 const LangOptions &LangOpts) { 122 SourceLocation SpellingBegin = Sources.getSpellingLoc(Range.getBegin()); 123 SourceLocation SpellingEnd = Sources.getSpellingLoc(Range.getEnd()); 124 std::pair<FileID, unsigned> Start = Sources.getDecomposedLoc(SpellingBegin); 125 std::pair<FileID, unsigned> End = Sources.getDecomposedLoc(SpellingEnd); 126 if (Start.first != End.first) return -1; 127 if (Range.isTokenRange()) 128 End.second += Lexer::MeasureTokenLength(SpellingEnd, Sources, LangOpts); 129 return End.second - Start.second; 130 } 131 132 void Replacement::setFromSourceRange(const SourceManager &Sources, 133 const CharSourceRange &Range, 134 StringRef ReplacementText, 135 const LangOptions &LangOpts) { 136 setFromSourceLocation(Sources, Sources.getSpellingLoc(Range.getBegin()), 137 getRangeSize(Sources, Range, LangOpts), 138 ReplacementText); 139 } 140 141 template <typename T> 142 unsigned shiftedCodePositionInternal(const T &Replaces, unsigned Position) { 143 unsigned Offset = 0; 144 for (const auto& R : Replaces) { 145 if (R.getOffset() + R.getLength() <= Position) { 146 Offset += R.getReplacementText().size() - R.getLength(); 147 continue; 148 } 149 if (R.getOffset() < Position && 150 R.getOffset() + R.getReplacementText().size() <= Position) { 151 Position = R.getOffset() + R.getReplacementText().size() - 1; 152 } 153 break; 154 } 155 return Position + Offset; 156 } 157 158 unsigned shiftedCodePosition(const Replacements &Replaces, unsigned Position) { 159 return shiftedCodePositionInternal(Replaces, Position); 160 } 161 162 // FIXME: Remove this function when Replacements is implemented as std::vector 163 // instead of std::set. 164 unsigned shiftedCodePosition(const std::vector<Replacement> &Replaces, 165 unsigned Position) { 166 return shiftedCodePositionInternal(Replaces, Position); 167 } 168 169 void deduplicate(std::vector<Replacement> &Replaces, 170 std::vector<Range> &Conflicts) { 171 if (Replaces.empty()) 172 return; 173 174 auto LessNoPath = [](const Replacement &LHS, const Replacement &RHS) { 175 if (LHS.getOffset() != RHS.getOffset()) 176 return LHS.getOffset() < RHS.getOffset(); 177 if (LHS.getLength() != RHS.getLength()) 178 return LHS.getLength() < RHS.getLength(); 179 return LHS.getReplacementText() < RHS.getReplacementText(); 180 }; 181 182 auto EqualNoPath = [](const Replacement &LHS, const Replacement &RHS) { 183 return LHS.getOffset() == RHS.getOffset() && 184 LHS.getLength() == RHS.getLength() && 185 LHS.getReplacementText() == RHS.getReplacementText(); 186 }; 187 188 // Deduplicate. We don't want to deduplicate based on the path as we assume 189 // that all replacements refer to the same file (or are symlinks). 190 std::sort(Replaces.begin(), Replaces.end(), LessNoPath); 191 Replaces.erase(std::unique(Replaces.begin(), Replaces.end(), EqualNoPath), 192 Replaces.end()); 193 194 // Detect conflicts 195 Range ConflictRange(Replaces.front().getOffset(), 196 Replaces.front().getLength()); 197 unsigned ConflictStart = 0; 198 unsigned ConflictLength = 1; 199 for (unsigned i = 1; i < Replaces.size(); ++i) { 200 Range Current(Replaces[i].getOffset(), Replaces[i].getLength()); 201 if (ConflictRange.overlapsWith(Current)) { 202 // Extend conflicted range 203 ConflictRange = Range(ConflictRange.getOffset(), 204 std::max(ConflictRange.getLength(), 205 Current.getOffset() + Current.getLength() - 206 ConflictRange.getOffset())); 207 ++ConflictLength; 208 } else { 209 if (ConflictLength > 1) 210 Conflicts.push_back(Range(ConflictStart, ConflictLength)); 211 ConflictRange = Current; 212 ConflictStart = i; 213 ConflictLength = 1; 214 } 215 } 216 217 if (ConflictLength > 1) 218 Conflicts.push_back(Range(ConflictStart, ConflictLength)); 219 } 220 221 bool applyAllReplacements(const Replacements &Replaces, Rewriter &Rewrite) { 222 bool Result = true; 223 for (Replacements::const_iterator I = Replaces.begin(), 224 E = Replaces.end(); 225 I != E; ++I) { 226 if (I->isApplicable()) { 227 Result = I->apply(Rewrite) && Result; 228 } else { 229 Result = false; 230 } 231 } 232 return Result; 233 } 234 235 // FIXME: Remove this function when Replacements is implemented as std::vector 236 // instead of std::set. 237 bool applyAllReplacements(const std::vector<Replacement> &Replaces, 238 Rewriter &Rewrite) { 239 bool Result = true; 240 for (std::vector<Replacement>::const_iterator I = Replaces.begin(), 241 E = Replaces.end(); 242 I != E; ++I) { 243 if (I->isApplicable()) { 244 Result = I->apply(Rewrite) && Result; 245 } else { 246 Result = false; 247 } 248 } 249 return Result; 250 } 251 252 llvm::Expected<std::string> applyAllReplacements(StringRef Code, 253 const Replacements &Replaces) { 254 if (Replaces.empty()) 255 return Code.str(); 256 257 IntrusiveRefCntPtr<vfs::InMemoryFileSystem> InMemoryFileSystem( 258 new vfs::InMemoryFileSystem); 259 FileManager Files(FileSystemOptions(), InMemoryFileSystem); 260 DiagnosticsEngine Diagnostics( 261 IntrusiveRefCntPtr<DiagnosticIDs>(new DiagnosticIDs), 262 new DiagnosticOptions); 263 SourceManager SourceMgr(Diagnostics, Files); 264 Rewriter Rewrite(SourceMgr, LangOptions()); 265 InMemoryFileSystem->addFile( 266 "<stdin>", 0, llvm::MemoryBuffer::getMemBuffer(Code, "<stdin>")); 267 FileID ID = SourceMgr.createFileID(Files.getFile("<stdin>"), SourceLocation(), 268 clang::SrcMgr::C_User); 269 for (Replacements::const_iterator I = Replaces.begin(), E = Replaces.end(); 270 I != E; ++I) { 271 Replacement Replace("<stdin>", I->getOffset(), I->getLength(), 272 I->getReplacementText()); 273 if (!Replace.apply(Rewrite)) 274 return llvm::make_error<llvm::StringError>( 275 "Failed to apply replacement: " + Replace.toString(), 276 llvm::inconvertibleErrorCode()); 277 } 278 std::string Result; 279 llvm::raw_string_ostream OS(Result); 280 Rewrite.getEditBuffer(ID).write(OS); 281 OS.flush(); 282 return Result; 283 } 284 285 // Merge and sort overlapping ranges in \p Ranges. 286 static std::vector<Range> mergeAndSortRanges(std::vector<Range> Ranges) { 287 std::sort(Ranges.begin(), Ranges.end(), 288 [](const Range &LHS, const Range &RHS) { 289 if (LHS.getOffset() != RHS.getOffset()) 290 return LHS.getOffset() < RHS.getOffset(); 291 return LHS.getLength() < RHS.getLength(); 292 }); 293 std::vector<Range> Result; 294 for (const auto &R : Ranges) { 295 if (Result.empty() || 296 Result.back().getOffset() + Result.back().getLength() < R.getOffset()) { 297 Result.push_back(R); 298 } else { 299 unsigned NewEnd = 300 std::max(Result.back().getOffset() + Result.back().getLength(), 301 R.getOffset() + R.getLength()); 302 Result[Result.size() - 1] = 303 Range(Result.back().getOffset(), NewEnd - Result.back().getOffset()); 304 } 305 } 306 return Result; 307 } 308 309 std::vector<Range> calculateChangedRanges(const Replacements &Replaces) { 310 std::vector<Range> ChangedRanges; 311 int Shift = 0; 312 for (const Replacement &R : Replaces) { 313 unsigned Offset = R.getOffset() + Shift; 314 unsigned Length = R.getReplacementText().size(); 315 Shift += Length - R.getLength(); 316 ChangedRanges.push_back(Range(Offset, Length)); 317 } 318 return mergeAndSortRanges(ChangedRanges); 319 } 320 321 std::vector<Range> 322 calculateRangesAfterReplacements(const Replacements &Replaces, 323 const std::vector<Range> &Ranges) { 324 auto MergedRanges = mergeAndSortRanges(Ranges); 325 tooling::Replacements FakeReplaces; 326 for (const auto &R : MergedRanges) 327 FakeReplaces.insert(Replacement(Replaces.begin()->getFilePath(), 328 R.getOffset(), R.getLength(), 329 std::string(R.getLength(), ' '))); 330 tooling::Replacements NewReplaces = mergeReplacements(FakeReplaces, Replaces); 331 return calculateChangedRanges(NewReplaces); 332 } 333 334 namespace { 335 // Represents a merged replacement, i.e. a replacement consisting of multiple 336 // overlapping replacements from 'First' and 'Second' in mergeReplacements. 337 // 338 // Position projection: 339 // Offsets and lengths of the replacements can generally refer to two different 340 // coordinate spaces. Replacements from 'First' refer to the original text 341 // whereas replacements from 'Second' refer to the text after applying 'First'. 342 // 343 // MergedReplacement always operates in the coordinate space of the original 344 // text, i.e. transforms elements from 'Second' to take into account what was 345 // changed based on the elements from 'First'. 346 // 347 // We can correctly calculate this projection as we look at the replacements in 348 // order of strictly increasing offsets. 349 // 350 // Invariants: 351 // * We always merge elements from 'First' into elements from 'Second' and vice 352 // versa. Within each set, the replacements are non-overlapping. 353 // * We only extend to the right, i.e. merge elements with strictly increasing 354 // offsets. 355 class MergedReplacement { 356 public: 357 MergedReplacement(const Replacement &R, bool MergeSecond, int D) 358 : MergeSecond(MergeSecond), Delta(D), FilePath(R.getFilePath()), 359 Offset(R.getOffset() + (MergeSecond ? 0 : Delta)), Length(R.getLength()), 360 Text(R.getReplacementText()) { 361 Delta += MergeSecond ? 0 : Text.size() - Length; 362 DeltaFirst = MergeSecond ? Text.size() - Length : 0; 363 } 364 365 // Merges the next element 'R' into this merged element. As we always merge 366 // from 'First' into 'Second' or vice versa, the MergedReplacement knows what 367 // set the next element is coming from. 368 void merge(const Replacement &R) { 369 if (MergeSecond) { 370 unsigned REnd = R.getOffset() + Delta + R.getLength(); 371 unsigned End = Offset + Text.size(); 372 if (REnd > End) { 373 Length += REnd - End; 374 MergeSecond = false; 375 } 376 StringRef TextRef = Text; 377 StringRef Head = TextRef.substr(0, R.getOffset() + Delta - Offset); 378 StringRef Tail = TextRef.substr(REnd - Offset); 379 Text = (Head + R.getReplacementText() + Tail).str(); 380 Delta += R.getReplacementText().size() - R.getLength(); 381 } else { 382 unsigned End = Offset + Length; 383 StringRef RText = R.getReplacementText(); 384 StringRef Tail = RText.substr(End - R.getOffset()); 385 Text = (Text + Tail).str(); 386 if (R.getOffset() + RText.size() > End) { 387 Length = R.getOffset() + R.getLength() - Offset; 388 MergeSecond = true; 389 } else { 390 Length += R.getLength() - RText.size(); 391 } 392 DeltaFirst += RText.size() - R.getLength(); 393 } 394 } 395 396 // Returns 'true' if 'R' starts strictly after the MergedReplacement and thus 397 // doesn't need to be merged. 398 bool endsBefore(const Replacement &R) const { 399 if (MergeSecond) 400 return Offset + Text.size() < R.getOffset() + Delta; 401 return Offset + Length < R.getOffset(); 402 } 403 404 // Returns 'true' if an element from the second set should be merged next. 405 bool mergeSecond() const { return MergeSecond; } 406 int deltaFirst() const { return DeltaFirst; } 407 Replacement asReplacement() const { return {FilePath, Offset, Length, Text}; } 408 409 private: 410 bool MergeSecond; 411 412 // Amount of characters that elements from 'Second' need to be shifted by in 413 // order to refer to the original text. 414 int Delta; 415 416 // Sum of all deltas (text-length - length) of elements from 'First' merged 417 // into this element. This is used to update 'Delta' once the 418 // MergedReplacement is completed. 419 int DeltaFirst; 420 421 // Data of the actually merged replacement. FilePath and Offset aren't changed 422 // as the element is only extended to the right. 423 const StringRef FilePath; 424 const unsigned Offset; 425 unsigned Length; 426 std::string Text; 427 }; 428 } // namespace 429 430 std::map<std::string, Replacements> 431 groupReplacementsByFile(const Replacements &Replaces) { 432 std::map<std::string, Replacements> FileToReplaces; 433 for (const auto &Replace : Replaces) { 434 FileToReplaces[Replace.getFilePath()].insert(Replace); 435 } 436 return FileToReplaces; 437 } 438 439 Replacements mergeReplacements(const Replacements &First, 440 const Replacements &Second) { 441 if (First.empty() || Second.empty()) 442 return First.empty() ? Second : First; 443 444 // Delta is the amount of characters that replacements from 'Second' need to 445 // be shifted so that their offsets refer to the original text. 446 int Delta = 0; 447 Replacements Result; 448 449 // Iterate over both sets and always add the next element (smallest total 450 // Offset) from either 'First' or 'Second'. Merge that element with 451 // subsequent replacements as long as they overlap. See more details in the 452 // comment on MergedReplacement. 453 for (auto FirstI = First.begin(), SecondI = Second.begin(); 454 FirstI != First.end() || SecondI != Second.end();) { 455 bool NextIsFirst = SecondI == Second.end() || 456 (FirstI != First.end() && 457 FirstI->getOffset() < SecondI->getOffset() + Delta); 458 MergedReplacement Merged(NextIsFirst ? *FirstI : *SecondI, NextIsFirst, 459 Delta); 460 ++(NextIsFirst ? FirstI : SecondI); 461 462 while ((Merged.mergeSecond() && SecondI != Second.end()) || 463 (!Merged.mergeSecond() && FirstI != First.end())) { 464 auto &I = Merged.mergeSecond() ? SecondI : FirstI; 465 if (Merged.endsBefore(*I)) 466 break; 467 Merged.merge(*I); 468 ++I; 469 } 470 Delta -= Merged.deltaFirst(); 471 Result.insert(Merged.asReplacement()); 472 } 473 return Result; 474 } 475 476 } // end namespace tooling 477 } // end namespace clang 478