1 //===--- WhitespaceManager.cpp - Format C++ code --------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// 9 /// \file 10 /// This file implements WhitespaceManager class. 11 /// 12 //===----------------------------------------------------------------------===// 13 14 #include "WhitespaceManager.h" 15 #include "llvm/ADT/STLExtras.h" 16 #include "llvm/ADT/SmallVector.h" 17 #include <algorithm> 18 19 namespace clang { 20 namespace format { 21 22 bool WhitespaceManager::Change::IsBeforeInFile::operator()( 23 const Change &C1, const Change &C2) const { 24 return SourceMgr.isBeforeInTranslationUnit( 25 C1.OriginalWhitespaceRange.getBegin(), 26 C2.OriginalWhitespaceRange.getBegin()); 27 } 28 29 WhitespaceManager::Change::Change(const FormatToken &Tok, 30 bool CreateReplacement, 31 SourceRange OriginalWhitespaceRange, 32 int Spaces, unsigned StartOfTokenColumn, 33 unsigned NewlinesBefore, 34 StringRef PreviousLinePostfix, 35 StringRef CurrentLinePrefix, bool IsAligned, 36 bool ContinuesPPDirective, bool IsInsideToken) 37 : Tok(&Tok), CreateReplacement(CreateReplacement), 38 OriginalWhitespaceRange(OriginalWhitespaceRange), 39 StartOfTokenColumn(StartOfTokenColumn), NewlinesBefore(NewlinesBefore), 40 PreviousLinePostfix(PreviousLinePostfix), 41 CurrentLinePrefix(CurrentLinePrefix), IsAligned(IsAligned), 42 ContinuesPPDirective(ContinuesPPDirective), Spaces(Spaces), 43 IsInsideToken(IsInsideToken), IsTrailingComment(false), TokenLength(0), 44 PreviousEndOfTokenColumn(0), EscapedNewlineColumn(0), 45 StartOfBlockComment(nullptr), IndentationOffset(0), ConditionalsLevel(0) { 46 } 47 48 void WhitespaceManager::replaceWhitespace(FormatToken &Tok, unsigned Newlines, 49 unsigned Spaces, 50 unsigned StartOfTokenColumn, 51 bool IsAligned, bool InPPDirective) { 52 if (Tok.Finalized) 53 return; 54 Tok.setDecision((Newlines > 0) ? FD_Break : FD_Continue); 55 Changes.push_back(Change(Tok, /*CreateReplacement=*/true, Tok.WhitespaceRange, 56 Spaces, StartOfTokenColumn, Newlines, "", "", 57 IsAligned, InPPDirective && !Tok.IsFirst, 58 /*IsInsideToken=*/false)); 59 } 60 61 void WhitespaceManager::addUntouchableToken(const FormatToken &Tok, 62 bool InPPDirective) { 63 if (Tok.Finalized) 64 return; 65 Changes.push_back(Change(Tok, /*CreateReplacement=*/false, 66 Tok.WhitespaceRange, /*Spaces=*/0, 67 Tok.OriginalColumn, Tok.NewlinesBefore, "", "", 68 /*IsAligned=*/false, InPPDirective && !Tok.IsFirst, 69 /*IsInsideToken=*/false)); 70 } 71 72 llvm::Error 73 WhitespaceManager::addReplacement(const tooling::Replacement &Replacement) { 74 return Replaces.add(Replacement); 75 } 76 77 bool WhitespaceManager::inputUsesCRLF(StringRef Text, bool DefaultToCRLF) { 78 size_t LF = Text.count('\n'); 79 size_t CR = Text.count('\r') * 2; 80 return LF == CR ? DefaultToCRLF : CR > LF; 81 } 82 83 void WhitespaceManager::replaceWhitespaceInToken( 84 const FormatToken &Tok, unsigned Offset, unsigned ReplaceChars, 85 StringRef PreviousPostfix, StringRef CurrentPrefix, bool InPPDirective, 86 unsigned Newlines, int Spaces) { 87 if (Tok.Finalized) 88 return; 89 SourceLocation Start = Tok.getStartOfNonWhitespace().getLocWithOffset(Offset); 90 Changes.push_back( 91 Change(Tok, /*CreateReplacement=*/true, 92 SourceRange(Start, Start.getLocWithOffset(ReplaceChars)), Spaces, 93 std::max(0, Spaces), Newlines, PreviousPostfix, CurrentPrefix, 94 /*IsAligned=*/true, InPPDirective && !Tok.IsFirst, 95 /*IsInsideToken=*/true)); 96 } 97 98 const tooling::Replacements &WhitespaceManager::generateReplacements() { 99 if (Changes.empty()) 100 return Replaces; 101 102 llvm::sort(Changes, Change::IsBeforeInFile(SourceMgr)); 103 calculateLineBreakInformation(); 104 alignConsecutiveMacros(); 105 alignConsecutiveDeclarations(); 106 alignConsecutiveBitFields(); 107 alignConsecutiveAssignments(); 108 alignChainedConditionals(); 109 alignTrailingComments(); 110 alignEscapedNewlines(); 111 alignArrayInitializers(); 112 generateChanges(); 113 114 return Replaces; 115 } 116 117 void WhitespaceManager::calculateLineBreakInformation() { 118 Changes[0].PreviousEndOfTokenColumn = 0; 119 Change *LastOutsideTokenChange = &Changes[0]; 120 for (unsigned i = 1, e = Changes.size(); i != e; ++i) { 121 SourceLocation OriginalWhitespaceStart = 122 Changes[i].OriginalWhitespaceRange.getBegin(); 123 SourceLocation PreviousOriginalWhitespaceEnd = 124 Changes[i - 1].OriginalWhitespaceRange.getEnd(); 125 unsigned OriginalWhitespaceStartOffset = 126 SourceMgr.getFileOffset(OriginalWhitespaceStart); 127 unsigned PreviousOriginalWhitespaceEndOffset = 128 SourceMgr.getFileOffset(PreviousOriginalWhitespaceEnd); 129 assert(PreviousOriginalWhitespaceEndOffset <= 130 OriginalWhitespaceStartOffset); 131 const char *const PreviousOriginalWhitespaceEndData = 132 SourceMgr.getCharacterData(PreviousOriginalWhitespaceEnd); 133 StringRef Text(PreviousOriginalWhitespaceEndData, 134 SourceMgr.getCharacterData(OriginalWhitespaceStart) - 135 PreviousOriginalWhitespaceEndData); 136 // Usually consecutive changes would occur in consecutive tokens. This is 137 // not the case however when analyzing some preprocessor runs of the 138 // annotated lines. For example, in this code: 139 // 140 // #if A // line 1 141 // int i = 1; 142 // #else B // line 2 143 // int i = 2; 144 // #endif // line 3 145 // 146 // one of the runs will produce the sequence of lines marked with line 1, 2 147 // and 3. So the two consecutive whitespace changes just before '// line 2' 148 // and before '#endif // line 3' span multiple lines and tokens: 149 // 150 // #else B{change X}[// line 2 151 // int i = 2; 152 // ]{change Y}#endif // line 3 153 // 154 // For this reason, if the text between consecutive changes spans multiple 155 // newlines, the token length must be adjusted to the end of the original 156 // line of the token. 157 auto NewlinePos = Text.find_first_of('\n'); 158 if (NewlinePos == StringRef::npos) { 159 Changes[i - 1].TokenLength = OriginalWhitespaceStartOffset - 160 PreviousOriginalWhitespaceEndOffset + 161 Changes[i].PreviousLinePostfix.size() + 162 Changes[i - 1].CurrentLinePrefix.size(); 163 } else { 164 Changes[i - 1].TokenLength = 165 NewlinePos + Changes[i - 1].CurrentLinePrefix.size(); 166 } 167 168 // If there are multiple changes in this token, sum up all the changes until 169 // the end of the line. 170 if (Changes[i - 1].IsInsideToken && Changes[i - 1].NewlinesBefore == 0) 171 LastOutsideTokenChange->TokenLength += 172 Changes[i - 1].TokenLength + Changes[i - 1].Spaces; 173 else 174 LastOutsideTokenChange = &Changes[i - 1]; 175 176 Changes[i].PreviousEndOfTokenColumn = 177 Changes[i - 1].StartOfTokenColumn + Changes[i - 1].TokenLength; 178 179 Changes[i - 1].IsTrailingComment = 180 (Changes[i].NewlinesBefore > 0 || Changes[i].Tok->is(tok::eof) || 181 (Changes[i].IsInsideToken && Changes[i].Tok->is(tok::comment))) && 182 Changes[i - 1].Tok->is(tok::comment) && 183 // FIXME: This is a dirty hack. The problem is that 184 // BreakableLineCommentSection does comment reflow changes and here is 185 // the aligning of trailing comments. Consider the case where we reflow 186 // the second line up in this example: 187 // 188 // // line 1 189 // // line 2 190 // 191 // That amounts to 2 changes by BreakableLineCommentSection: 192 // - the first, delimited by (), for the whitespace between the tokens, 193 // - and second, delimited by [], for the whitespace at the beginning 194 // of the second token: 195 // 196 // // line 1( 197 // )[// ]line 2 198 // 199 // So in the end we have two changes like this: 200 // 201 // // line1()[ ]line 2 202 // 203 // Note that the OriginalWhitespaceStart of the second change is the 204 // same as the PreviousOriginalWhitespaceEnd of the first change. 205 // In this case, the below check ensures that the second change doesn't 206 // get treated as a trailing comment change here, since this might 207 // trigger additional whitespace to be wrongly inserted before "line 2" 208 // by the comment aligner here. 209 // 210 // For a proper solution we need a mechanism to say to WhitespaceManager 211 // that a particular change breaks the current sequence of trailing 212 // comments. 213 OriginalWhitespaceStart != PreviousOriginalWhitespaceEnd; 214 } 215 // FIXME: The last token is currently not always an eof token; in those 216 // cases, setting TokenLength of the last token to 0 is wrong. 217 Changes.back().TokenLength = 0; 218 Changes.back().IsTrailingComment = Changes.back().Tok->is(tok::comment); 219 220 const WhitespaceManager::Change *LastBlockComment = nullptr; 221 for (auto &Change : Changes) { 222 // Reset the IsTrailingComment flag for changes inside of trailing comments 223 // so they don't get realigned later. Comment line breaks however still need 224 // to be aligned. 225 if (Change.IsInsideToken && Change.NewlinesBefore == 0) 226 Change.IsTrailingComment = false; 227 Change.StartOfBlockComment = nullptr; 228 Change.IndentationOffset = 0; 229 if (Change.Tok->is(tok::comment)) { 230 if (Change.Tok->is(TT_LineComment) || !Change.IsInsideToken) 231 LastBlockComment = &Change; 232 else { 233 if ((Change.StartOfBlockComment = LastBlockComment)) 234 Change.IndentationOffset = 235 Change.StartOfTokenColumn - 236 Change.StartOfBlockComment->StartOfTokenColumn; 237 } 238 } else { 239 LastBlockComment = nullptr; 240 } 241 } 242 243 // Compute conditional nesting level 244 // Level is increased for each conditional, unless this conditional continues 245 // a chain of conditional, i.e. starts immediately after the colon of another 246 // conditional. 247 SmallVector<bool, 16> ScopeStack; 248 int ConditionalsLevel = 0; 249 for (auto &Change : Changes) { 250 for (unsigned i = 0, e = Change.Tok->FakeLParens.size(); i != e; ++i) { 251 bool isNestedConditional = 252 Change.Tok->FakeLParens[e - 1 - i] == prec::Conditional && 253 !(i == 0 && Change.Tok->Previous && 254 Change.Tok->Previous->is(TT_ConditionalExpr) && 255 Change.Tok->Previous->is(tok::colon)); 256 if (isNestedConditional) 257 ++ConditionalsLevel; 258 ScopeStack.push_back(isNestedConditional); 259 } 260 261 Change.ConditionalsLevel = ConditionalsLevel; 262 263 for (unsigned i = Change.Tok->FakeRParens; i > 0 && ScopeStack.size(); 264 --i) { 265 if (ScopeStack.pop_back_val()) 266 --ConditionalsLevel; 267 } 268 } 269 } 270 271 // Align a single sequence of tokens, see AlignTokens below. 272 template <typename F> 273 static void 274 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End, 275 unsigned Column, F &&Matches, 276 SmallVector<WhitespaceManager::Change, 16> &Changes) { 277 bool FoundMatchOnLine = false; 278 int Shift = 0; 279 280 // ScopeStack keeps track of the current scope depth. It contains indices of 281 // the first token on each scope. 282 // We only run the "Matches" function on tokens from the outer-most scope. 283 // However, we do need to pay special attention to one class of tokens 284 // that are not in the outer-most scope, and that is function parameters 285 // which are split across multiple lines, as illustrated by this example: 286 // double a(int x); 287 // int b(int y, 288 // double z); 289 // In the above example, we need to take special care to ensure that 290 // 'double z' is indented along with it's owning function 'b'. 291 // The same holds for calling a function: 292 // double a = foo(x); 293 // int b = bar(foo(y), 294 // foor(z)); 295 // Similar for broken string literals: 296 // double x = 3.14; 297 // auto s = "Hello" 298 // "World"; 299 // Special handling is required for 'nested' ternary operators. 300 SmallVector<unsigned, 16> ScopeStack; 301 302 for (unsigned i = Start; i != End; ++i) { 303 if (ScopeStack.size() != 0 && 304 Changes[i].indentAndNestingLevel() < 305 Changes[ScopeStack.back()].indentAndNestingLevel()) 306 ScopeStack.pop_back(); 307 308 // Compare current token to previous non-comment token to ensure whether 309 // it is in a deeper scope or not. 310 unsigned PreviousNonComment = i - 1; 311 while (PreviousNonComment > Start && 312 Changes[PreviousNonComment].Tok->is(tok::comment)) 313 PreviousNonComment--; 314 if (i != Start && Changes[i].indentAndNestingLevel() > 315 Changes[PreviousNonComment].indentAndNestingLevel()) 316 ScopeStack.push_back(i); 317 318 bool InsideNestedScope = ScopeStack.size() != 0; 319 bool ContinuedStringLiteral = i > Start && 320 Changes[i].Tok->is(tok::string_literal) && 321 Changes[i - 1].Tok->is(tok::string_literal); 322 bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral; 323 324 if (Changes[i].NewlinesBefore > 0 && !SkipMatchCheck) { 325 Shift = 0; 326 FoundMatchOnLine = false; 327 } 328 329 // If this is the first matching token to be aligned, remember by how many 330 // spaces it has to be shifted, so the rest of the changes on the line are 331 // shifted by the same amount 332 if (!FoundMatchOnLine && !SkipMatchCheck && Matches(Changes[i])) { 333 FoundMatchOnLine = true; 334 Shift = Column - Changes[i].StartOfTokenColumn; 335 Changes[i].Spaces += Shift; 336 } 337 338 // This is for function parameters that are split across multiple lines, 339 // as mentioned in the ScopeStack comment. 340 if (InsideNestedScope && Changes[i].NewlinesBefore > 0) { 341 unsigned ScopeStart = ScopeStack.back(); 342 auto ShouldShiftBeAdded = [&] { 343 // Function declaration 344 if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName)) 345 return true; 346 347 // Continued function declaration 348 if (ScopeStart > Start + 1 && 349 Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) 350 return true; 351 352 // Continued function call 353 if (ScopeStart > Start + 1 && 354 Changes[ScopeStart - 2].Tok->is(tok::identifier) && 355 Changes[ScopeStart - 1].Tok->is(tok::l_paren)) 356 return Style.BinPackArguments; 357 358 // Ternary operator 359 if (Changes[i].Tok->is(TT_ConditionalExpr)) 360 return true; 361 362 // Period Initializer .XXX = 1. 363 if (Changes[i].Tok->is(TT_DesignatedInitializerPeriod)) 364 return true; 365 366 // Continued ternary operator 367 if (Changes[i].Tok->Previous && 368 Changes[i].Tok->Previous->is(TT_ConditionalExpr)) 369 return true; 370 371 return false; 372 }; 373 374 if (ShouldShiftBeAdded()) 375 Changes[i].Spaces += Shift; 376 } 377 378 if (ContinuedStringLiteral) 379 Changes[i].Spaces += Shift; 380 381 Changes[i].StartOfTokenColumn += Shift; 382 if (i + 1 != Changes.size()) 383 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 384 385 // If PointerAlignment is PAS_Right, keep *s or &s next to the token 386 if (Style.PointerAlignment == FormatStyle::PAS_Right && 387 Changes[i].Spaces != 0) { 388 for (int Previous = i - 1; 389 Previous >= 0 && 390 Changes[Previous].Tok->getType() == TT_PointerOrReference; 391 --Previous) { 392 Changes[Previous + 1].Spaces -= Shift; 393 Changes[Previous].Spaces += Shift; 394 } 395 } 396 } 397 } 398 399 // Walk through a subset of the changes, starting at StartAt, and find 400 // sequences of matching tokens to align. To do so, keep track of the lines and 401 // whether or not a matching token was found on a line. If a matching token is 402 // found, extend the current sequence. If the current line cannot be part of a 403 // sequence, e.g. because there is an empty line before it or it contains only 404 // non-matching tokens, finalize the previous sequence. 405 // The value returned is the token on which we stopped, either because we 406 // exhausted all items inside Changes, or because we hit a scope level higher 407 // than our initial scope. 408 // This function is recursive. Each invocation processes only the scope level 409 // equal to the initial level, which is the level of Changes[StartAt]. 410 // If we encounter a scope level greater than the initial level, then we call 411 // ourselves recursively, thereby avoiding the pollution of the current state 412 // with the alignment requirements of the nested sub-level. This recursive 413 // behavior is necessary for aligning function prototypes that have one or more 414 // arguments. 415 // If this function encounters a scope level less than the initial level, 416 // it returns the current position. 417 // There is a non-obvious subtlety in the recursive behavior: Even though we 418 // defer processing of nested levels to recursive invocations of this 419 // function, when it comes time to align a sequence of tokens, we run the 420 // alignment on the entire sequence, including the nested levels. 421 // When doing so, most of the nested tokens are skipped, because their 422 // alignment was already handled by the recursive invocations of this function. 423 // However, the special exception is that we do NOT skip function parameters 424 // that are split across multiple lines. See the test case in FormatTest.cpp 425 // that mentions "split function parameter alignment" for an example of this. 426 template <typename F> 427 static unsigned AlignTokens( 428 const FormatStyle &Style, F &&Matches, 429 SmallVector<WhitespaceManager::Change, 16> &Changes, unsigned StartAt, 430 const FormatStyle::AlignConsecutiveStyle &ACS = FormatStyle::ACS_None) { 431 unsigned MinColumn = 0; 432 unsigned MaxColumn = UINT_MAX; 433 434 // Line number of the start and the end of the current token sequence. 435 unsigned StartOfSequence = 0; 436 unsigned EndOfSequence = 0; 437 438 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and 439 // abort when we hit any token in a higher scope than the starting one. 440 auto IndentAndNestingLevel = StartAt < Changes.size() 441 ? Changes[StartAt].indentAndNestingLevel() 442 : std::tuple<unsigned, unsigned, unsigned>(); 443 444 // Keep track of the number of commas before the matching tokens, we will only 445 // align a sequence of matching tokens if they are preceded by the same number 446 // of commas. 447 unsigned CommasBeforeLastMatch = 0; 448 unsigned CommasBeforeMatch = 0; 449 450 // Whether a matching token has been found on the current line. 451 bool FoundMatchOnLine = false; 452 453 // Whether the current line consists purely of comments. 454 bool LineIsComment = true; 455 456 // Aligns a sequence of matching tokens, on the MinColumn column. 457 // 458 // Sequences start from the first matching token to align, and end at the 459 // first token of the first line that doesn't need to be aligned. 460 // 461 // We need to adjust the StartOfTokenColumn of each Change that is on a line 462 // containing any matching token to be aligned and located after such token. 463 auto AlignCurrentSequence = [&] { 464 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) 465 AlignTokenSequence(Style, StartOfSequence, EndOfSequence, MinColumn, 466 Matches, Changes); 467 MinColumn = 0; 468 MaxColumn = UINT_MAX; 469 StartOfSequence = 0; 470 EndOfSequence = 0; 471 }; 472 473 unsigned i = StartAt; 474 for (unsigned e = Changes.size(); i != e; ++i) { 475 if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel) 476 break; 477 478 if (Changes[i].NewlinesBefore != 0) { 479 CommasBeforeMatch = 0; 480 EndOfSequence = i; 481 482 // Whether to break the alignment sequence because of an empty line. 483 bool EmptyLineBreak = 484 (Changes[i].NewlinesBefore > 1) && 485 (ACS != FormatStyle::ACS_AcrossEmptyLines) && 486 (ACS != FormatStyle::ACS_AcrossEmptyLinesAndComments); 487 488 // Whether to break the alignment sequence because of a line without a 489 // match. 490 bool NoMatchBreak = 491 !FoundMatchOnLine && 492 !(LineIsComment && 493 ((ACS == FormatStyle::ACS_AcrossComments) || 494 (ACS == FormatStyle::ACS_AcrossEmptyLinesAndComments))); 495 496 if (EmptyLineBreak || NoMatchBreak) 497 AlignCurrentSequence(); 498 499 // A new line starts, re-initialize line status tracking bools. 500 // Keep the match state if a string literal is continued on this line. 501 if (i == 0 || !Changes[i].Tok->is(tok::string_literal) || 502 !Changes[i - 1].Tok->is(tok::string_literal)) 503 FoundMatchOnLine = false; 504 LineIsComment = true; 505 } 506 507 if (!Changes[i].Tok->is(tok::comment)) { 508 LineIsComment = false; 509 } 510 511 if (Changes[i].Tok->is(tok::comma)) { 512 ++CommasBeforeMatch; 513 } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) { 514 // Call AlignTokens recursively, skipping over this scope block. 515 unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i, ACS); 516 i = StoppedAt - 1; 517 continue; 518 } 519 520 if (!Matches(Changes[i])) 521 continue; 522 523 // If there is more than one matching token per line, or if the number of 524 // preceding commas, do not match anymore, end the sequence. 525 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch) 526 AlignCurrentSequence(); 527 528 CommasBeforeLastMatch = CommasBeforeMatch; 529 FoundMatchOnLine = true; 530 531 if (StartOfSequence == 0) 532 StartOfSequence = i; 533 534 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn; 535 int LineLengthAfter = Changes[i].TokenLength; 536 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) { 537 LineLengthAfter += Changes[j].Spaces; 538 // Changes are generally 1:1 with the tokens, but a change could also be 539 // inside of a token, in which case it's counted more than once: once for 540 // the whitespace surrounding the token (!IsInsideToken) and once for 541 // each whitespace change within it (IsInsideToken). 542 // Therefore, changes inside of a token should only count the space. 543 if (!Changes[j].IsInsideToken) 544 LineLengthAfter += Changes[j].TokenLength; 545 } 546 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter; 547 548 // If we are restricted by the maximum column width, end the sequence. 549 if (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn || 550 CommasBeforeLastMatch != CommasBeforeMatch) { 551 AlignCurrentSequence(); 552 StartOfSequence = i; 553 } 554 555 MinColumn = std::max(MinColumn, ChangeMinColumn); 556 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 557 } 558 559 EndOfSequence = i; 560 AlignCurrentSequence(); 561 return i; 562 } 563 564 // Aligns a sequence of matching tokens, on the MinColumn column. 565 // 566 // Sequences start from the first matching token to align, and end at the 567 // first token of the first line that doesn't need to be aligned. 568 // 569 // We need to adjust the StartOfTokenColumn of each Change that is on a line 570 // containing any matching token to be aligned and located after such token. 571 static void AlignMacroSequence( 572 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn, 573 unsigned &MaxColumn, bool &FoundMatchOnLine, 574 std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches, 575 SmallVector<WhitespaceManager::Change, 16> &Changes) { 576 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) { 577 578 FoundMatchOnLine = false; 579 int Shift = 0; 580 581 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) { 582 if (Changes[I].NewlinesBefore > 0) { 583 Shift = 0; 584 FoundMatchOnLine = false; 585 } 586 587 // If this is the first matching token to be aligned, remember by how many 588 // spaces it has to be shifted, so the rest of the changes on the line are 589 // shifted by the same amount 590 if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) { 591 FoundMatchOnLine = true; 592 Shift = MinColumn - Changes[I].StartOfTokenColumn; 593 Changes[I].Spaces += Shift; 594 } 595 596 assert(Shift >= 0); 597 Changes[I].StartOfTokenColumn += Shift; 598 if (I + 1 != Changes.size()) 599 Changes[I + 1].PreviousEndOfTokenColumn += Shift; 600 } 601 } 602 603 MinColumn = 0; 604 MaxColumn = UINT_MAX; 605 StartOfSequence = 0; 606 EndOfSequence = 0; 607 } 608 609 void WhitespaceManager::alignConsecutiveMacros() { 610 if (Style.AlignConsecutiveMacros == FormatStyle::ACS_None) 611 return; 612 613 auto AlignMacrosMatches = [](const Change &C) { 614 const FormatToken *Current = C.Tok; 615 unsigned SpacesRequiredBefore = 1; 616 617 if (Current->SpacesRequiredBefore == 0 || !Current->Previous) 618 return false; 619 620 Current = Current->Previous; 621 622 // If token is a ")", skip over the parameter list, to the 623 // token that precedes the "(" 624 if (Current->is(tok::r_paren) && Current->MatchingParen) { 625 Current = Current->MatchingParen->Previous; 626 SpacesRequiredBefore = 0; 627 } 628 629 if (!Current || !Current->is(tok::identifier)) 630 return false; 631 632 if (!Current->Previous || !Current->Previous->is(tok::pp_define)) 633 return false; 634 635 // For a macro function, 0 spaces are required between the 636 // identifier and the lparen that opens the parameter list. 637 // For a simple macro, 1 space is required between the 638 // identifier and the first token of the defined value. 639 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore; 640 }; 641 642 unsigned MinColumn = 0; 643 unsigned MaxColumn = UINT_MAX; 644 645 // Start and end of the token sequence we're processing. 646 unsigned StartOfSequence = 0; 647 unsigned EndOfSequence = 0; 648 649 // Whether a matching token has been found on the current line. 650 bool FoundMatchOnLine = false; 651 652 // Whether the current line consists only of comments 653 bool LineIsComment = true; 654 655 unsigned I = 0; 656 for (unsigned E = Changes.size(); I != E; ++I) { 657 if (Changes[I].NewlinesBefore != 0) { 658 EndOfSequence = I; 659 660 // Whether to break the alignment sequence because of an empty line. 661 bool EmptyLineBreak = 662 (Changes[I].NewlinesBefore > 1) && 663 (Style.AlignConsecutiveMacros != FormatStyle::ACS_AcrossEmptyLines) && 664 (Style.AlignConsecutiveMacros != 665 FormatStyle::ACS_AcrossEmptyLinesAndComments); 666 667 // Whether to break the alignment sequence because of a line without a 668 // match. 669 bool NoMatchBreak = 670 !FoundMatchOnLine && 671 !(LineIsComment && ((Style.AlignConsecutiveMacros == 672 FormatStyle::ACS_AcrossComments) || 673 (Style.AlignConsecutiveMacros == 674 FormatStyle::ACS_AcrossEmptyLinesAndComments))); 675 676 if (EmptyLineBreak || NoMatchBreak) 677 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 678 FoundMatchOnLine, AlignMacrosMatches, Changes); 679 680 // A new line starts, re-initialize line status tracking bools. 681 FoundMatchOnLine = false; 682 LineIsComment = true; 683 } 684 685 if (!Changes[I].Tok->is(tok::comment)) { 686 LineIsComment = false; 687 } 688 689 if (!AlignMacrosMatches(Changes[I])) 690 continue; 691 692 FoundMatchOnLine = true; 693 694 if (StartOfSequence == 0) 695 StartOfSequence = I; 696 697 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn; 698 int LineLengthAfter = -Changes[I].Spaces; 699 for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j) 700 LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength; 701 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter; 702 703 MinColumn = std::max(MinColumn, ChangeMinColumn); 704 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 705 } 706 707 EndOfSequence = I; 708 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 709 FoundMatchOnLine, AlignMacrosMatches, Changes); 710 } 711 712 void WhitespaceManager::alignConsecutiveAssignments() { 713 if (Style.AlignConsecutiveAssignments == FormatStyle::ACS_None) 714 return; 715 716 AlignTokens( 717 Style, 718 [&](const Change &C) { 719 // Do not align on equal signs that are first on a line. 720 if (C.NewlinesBefore > 0) 721 return false; 722 723 // Do not align on equal signs that are last on a line. 724 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 725 return false; 726 727 return C.Tok->is(tok::equal); 728 }, 729 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments); 730 } 731 732 void WhitespaceManager::alignConsecutiveBitFields() { 733 if (Style.AlignConsecutiveBitFields == FormatStyle::ACS_None) 734 return; 735 736 AlignTokens( 737 Style, 738 [&](Change const &C) { 739 // Do not align on ':' that is first on a line. 740 if (C.NewlinesBefore > 0) 741 return false; 742 743 // Do not align on ':' that is last on a line. 744 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 745 return false; 746 747 return C.Tok->is(TT_BitFieldColon); 748 }, 749 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields); 750 } 751 752 void WhitespaceManager::alignConsecutiveDeclarations() { 753 if (Style.AlignConsecutiveDeclarations == FormatStyle::ACS_None) 754 return; 755 756 AlignTokens( 757 Style, 758 [](Change const &C) { 759 // tok::kw_operator is necessary for aligning operator overload 760 // definitions. 761 if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator)) 762 return true; 763 if (C.Tok->isNot(TT_StartOfName)) 764 return false; 765 if (C.Tok->Previous && 766 C.Tok->Previous->is(TT_StatementAttributeLikeMacro)) 767 return false; 768 // Check if there is a subsequent name that starts the same declaration. 769 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) { 770 if (Next->is(tok::comment)) 771 continue; 772 if (Next->is(TT_PointerOrReference)) 773 return false; 774 if (!Next->Tok.getIdentifierInfo()) 775 break; 776 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName, 777 tok::kw_operator)) 778 return false; 779 } 780 return true; 781 }, 782 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations); 783 } 784 785 void WhitespaceManager::alignChainedConditionals() { 786 if (Style.BreakBeforeTernaryOperators) { 787 AlignTokens( 788 Style, 789 [](Change const &C) { 790 // Align question operators and last colon 791 return C.Tok->is(TT_ConditionalExpr) && 792 ((C.Tok->is(tok::question) && !C.NewlinesBefore) || 793 (C.Tok->is(tok::colon) && C.Tok->Next && 794 (C.Tok->Next->FakeLParens.size() == 0 || 795 C.Tok->Next->FakeLParens.back() != prec::Conditional))); 796 }, 797 Changes, /*StartAt=*/0); 798 } else { 799 static auto AlignWrappedOperand = [](Change const &C) { 800 FormatToken *Previous = C.Tok->getPreviousNonComment(); 801 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) && 802 (Previous->is(tok::colon) && 803 (C.Tok->FakeLParens.size() == 0 || 804 C.Tok->FakeLParens.back() != prec::Conditional)); 805 }; 806 // Ensure we keep alignment of wrapped operands with non-wrapped operands 807 // Since we actually align the operators, the wrapped operands need the 808 // extra offset to be properly aligned. 809 for (Change &C : Changes) { 810 if (AlignWrappedOperand(C)) 811 C.StartOfTokenColumn -= 2; 812 } 813 AlignTokens( 814 Style, 815 [this](Change const &C) { 816 // Align question operators if next operand is not wrapped, as 817 // well as wrapped operands after question operator or last 818 // colon in conditional sequence 819 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) && 820 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 && 821 !(&C + 1)->IsTrailingComment) || 822 AlignWrappedOperand(C); 823 }, 824 Changes, /*StartAt=*/0); 825 } 826 } 827 828 void WhitespaceManager::alignTrailingComments() { 829 unsigned MinColumn = 0; 830 unsigned MaxColumn = UINT_MAX; 831 unsigned StartOfSequence = 0; 832 bool BreakBeforeNext = false; 833 unsigned Newlines = 0; 834 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 835 if (Changes[i].StartOfBlockComment) 836 continue; 837 Newlines += Changes[i].NewlinesBefore; 838 if (!Changes[i].IsTrailingComment) 839 continue; 840 841 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn; 842 unsigned ChangeMaxColumn; 843 844 if (Style.ColumnLimit == 0) 845 ChangeMaxColumn = UINT_MAX; 846 else if (Style.ColumnLimit >= Changes[i].TokenLength) 847 ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength; 848 else 849 ChangeMaxColumn = ChangeMinColumn; 850 851 // If we don't create a replacement for this change, we have to consider 852 // it to be immovable. 853 if (!Changes[i].CreateReplacement) 854 ChangeMaxColumn = ChangeMinColumn; 855 856 if (i + 1 != e && Changes[i + 1].ContinuesPPDirective) 857 ChangeMaxColumn -= 2; 858 // If this comment follows an } in column 0, it probably documents the 859 // closing of a namespace and we don't want to align it. 860 bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 && 861 Changes[i - 1].Tok->is(tok::r_brace) && 862 Changes[i - 1].StartOfTokenColumn == 0; 863 bool WasAlignedWithStartOfNextLine = false; 864 if (Changes[i].NewlinesBefore == 1) { // A comment on its own line. 865 unsigned CommentColumn = SourceMgr.getSpellingColumnNumber( 866 Changes[i].OriginalWhitespaceRange.getEnd()); 867 for (unsigned j = i + 1; j != e; ++j) { 868 if (Changes[j].Tok->is(tok::comment)) 869 continue; 870 871 unsigned NextColumn = SourceMgr.getSpellingColumnNumber( 872 Changes[j].OriginalWhitespaceRange.getEnd()); 873 // The start of the next token was previously aligned with the 874 // start of this comment. 875 WasAlignedWithStartOfNextLine = 876 CommentColumn == NextColumn || 877 CommentColumn == NextColumn + Style.IndentWidth; 878 break; 879 } 880 } 881 if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) { 882 alignTrailingComments(StartOfSequence, i, MinColumn); 883 MinColumn = ChangeMinColumn; 884 MaxColumn = ChangeMinColumn; 885 StartOfSequence = i; 886 } else if (BreakBeforeNext || Newlines > 1 || 887 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) || 888 // Break the comment sequence if the previous line did not end 889 // in a trailing comment. 890 (Changes[i].NewlinesBefore == 1 && i > 0 && 891 !Changes[i - 1].IsTrailingComment) || 892 WasAlignedWithStartOfNextLine) { 893 alignTrailingComments(StartOfSequence, i, MinColumn); 894 MinColumn = ChangeMinColumn; 895 MaxColumn = ChangeMaxColumn; 896 StartOfSequence = i; 897 } else { 898 MinColumn = std::max(MinColumn, ChangeMinColumn); 899 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 900 } 901 BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) || 902 // Never start a sequence with a comment at the beginning 903 // of the line. 904 (Changes[i].NewlinesBefore == 1 && StartOfSequence == i); 905 Newlines = 0; 906 } 907 alignTrailingComments(StartOfSequence, Changes.size(), MinColumn); 908 } 909 910 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End, 911 unsigned Column) { 912 for (unsigned i = Start; i != End; ++i) { 913 int Shift = 0; 914 if (Changes[i].IsTrailingComment) { 915 Shift = Column - Changes[i].StartOfTokenColumn; 916 } 917 if (Changes[i].StartOfBlockComment) { 918 Shift = Changes[i].IndentationOffset + 919 Changes[i].StartOfBlockComment->StartOfTokenColumn - 920 Changes[i].StartOfTokenColumn; 921 } 922 if (Shift < 0) 923 continue; 924 Changes[i].Spaces += Shift; 925 if (i + 1 != Changes.size()) 926 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 927 Changes[i].StartOfTokenColumn += Shift; 928 } 929 } 930 931 void WhitespaceManager::alignEscapedNewlines() { 932 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign) 933 return; 934 935 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left; 936 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 937 unsigned StartOfMacro = 0; 938 for (unsigned i = 1, e = Changes.size(); i < e; ++i) { 939 Change &C = Changes[i]; 940 if (C.NewlinesBefore > 0) { 941 if (C.ContinuesPPDirective) { 942 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine); 943 } else { 944 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine); 945 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 946 StartOfMacro = i; 947 } 948 } 949 } 950 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine); 951 } 952 953 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End, 954 unsigned Column) { 955 for (unsigned i = Start; i < End; ++i) { 956 Change &C = Changes[i]; 957 if (C.NewlinesBefore > 0) { 958 assert(C.ContinuesPPDirective); 959 if (C.PreviousEndOfTokenColumn + 1 > Column) 960 C.EscapedNewlineColumn = 0; 961 else 962 C.EscapedNewlineColumn = Column; 963 } 964 } 965 } 966 967 void WhitespaceManager::alignArrayInitializers() { 968 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None) 969 return; 970 971 for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size(); 972 ChangeIndex < ChangeEnd; ++ChangeIndex) { 973 auto &C = Changes[ChangeIndex]; 974 if (C.Tok->IsArrayInitializer) { 975 bool FoundComplete = false; 976 for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd; 977 ++InsideIndex) { 978 if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) { 979 alignArrayInitializers(ChangeIndex, InsideIndex + 1); 980 ChangeIndex = InsideIndex + 1; 981 FoundComplete = true; 982 break; 983 } 984 } 985 if (!FoundComplete) 986 ChangeIndex = ChangeEnd; 987 } 988 } 989 } 990 991 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) { 992 993 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right) 994 alignArrayInitializersRightJustified(getCells(Start, End)); 995 else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left) 996 alignArrayInitializersLeftJustified(getCells(Start, End)); 997 } 998 999 void WhitespaceManager::alignArrayInitializersRightJustified( 1000 CellDescriptions &&CellDescs) { 1001 auto &Cells = CellDescs.Cells; 1002 1003 // Now go through and fixup the spaces. 1004 auto *CellIter = Cells.begin(); 1005 for (auto i = 0U; i < CellDescs.CellCount; i++, ++CellIter) { 1006 unsigned NetWidth = 0U; 1007 if (isSplitCell(*CellIter)) 1008 NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1009 auto CellWidth = getMaximumCellWidth(CellIter, NetWidth); 1010 1011 if (Changes[CellIter->Index].Tok->is(tok::r_brace)) { 1012 // So in here we want to see if there is a brace that falls 1013 // on a line that was split. If so on that line we make sure that 1014 // the spaces in front of the brace are enough. 1015 Changes[CellIter->Index].NewlinesBefore = 0; 1016 Changes[CellIter->Index].Spaces = 0; 1017 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1018 Next = Next->NextColumnElement) { 1019 Changes[Next->Index].Spaces = 0; 1020 Changes[Next->Index].NewlinesBefore = 0; 1021 } 1022 // Unless the array is empty, we need the position of all the 1023 // immediately adjacent cells 1024 if (CellIter != Cells.begin()) { 1025 auto ThisNetWidth = 1026 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1027 auto MaxNetWidth = 1028 getMaximumNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces, 1029 CellDescs.CellCount); 1030 if (ThisNetWidth < MaxNetWidth) 1031 Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1032 auto RowCount = 1U; 1033 auto Offset = std::distance(Cells.begin(), CellIter); 1034 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1035 Next = Next->NextColumnElement) { 1036 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCount); 1037 auto *End = Start + Offset; 1038 ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1039 if (ThisNetWidth < MaxNetWidth) 1040 Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1041 ++RowCount; 1042 } 1043 } 1044 } else { 1045 auto ThisWidth = 1046 calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) + 1047 NetWidth; 1048 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1049 Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth)); 1050 Changes[CellIter->Index].Spaces += (i > 0) ? 1 : 0; 1051 } 1052 alignToStartOfCell(CellIter->Index, CellIter->EndIndex); 1053 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1054 Next = Next->NextColumnElement) { 1055 ThisWidth = 1056 calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth; 1057 if (Changes[Next->Index].NewlinesBefore == 0) { 1058 Changes[Next->Index].Spaces = (CellWidth - ThisWidth); 1059 Changes[Next->Index].Spaces += (i > 0) ? 1 : 0; 1060 } 1061 alignToStartOfCell(Next->Index, Next->EndIndex); 1062 } 1063 } 1064 } 1065 } 1066 1067 void WhitespaceManager::alignArrayInitializersLeftJustified( 1068 CellDescriptions &&CellDescs) { 1069 auto &Cells = CellDescs.Cells; 1070 1071 // Now go through and fixup the spaces. 1072 auto *CellIter = Cells.begin(); 1073 // The first cell needs to be against the left brace. 1074 if (Changes[CellIter->Index].NewlinesBefore == 0) 1075 Changes[CellIter->Index].Spaces = 0; 1076 else 1077 Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces; 1078 ++CellIter; 1079 for (auto i = 1U; i < CellDescs.CellCount; i++, ++CellIter) { 1080 auto MaxNetWidth = getMaximumNetWidth( 1081 Cells.begin(), CellIter, CellDescs.InitialSpaces, CellDescs.CellCount); 1082 auto ThisNetWidth = 1083 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1084 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1085 Changes[CellIter->Index].Spaces = 1086 MaxNetWidth - ThisNetWidth + 1087 (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1088 } 1089 auto RowCount = 1U; 1090 auto Offset = std::distance(Cells.begin(), CellIter); 1091 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1092 Next = Next->NextColumnElement) { 1093 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCount); 1094 auto *End = Start + Offset; 1095 auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1096 if (Changes[Next->Index].NewlinesBefore == 0) { 1097 Changes[Next->Index].Spaces = 1098 MaxNetWidth - ThisNetWidth + 1099 (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1100 } 1101 ++RowCount; 1102 } 1103 } 1104 } 1105 1106 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) { 1107 if (Cell.HasSplit) 1108 return true; 1109 for (const auto *Next = Cell.NextColumnElement; Next != nullptr; 1110 Next = Next->NextColumnElement) { 1111 if (Next->HasSplit) 1112 return true; 1113 } 1114 return false; 1115 } 1116 1117 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start, 1118 unsigned End) { 1119 1120 unsigned Depth = 0; 1121 unsigned Cell = 0; 1122 unsigned CellCount = 0; 1123 unsigned InitialSpaces = 0; 1124 unsigned InitialTokenLength = 0; 1125 unsigned EndSpaces = 0; 1126 SmallVector<CellDescription> Cells; 1127 const FormatToken *MatchingParen = nullptr; 1128 for (unsigned i = Start; i < End; ++i) { 1129 auto &C = Changes[i]; 1130 if (C.Tok->is(tok::l_brace)) 1131 ++Depth; 1132 else if (C.Tok->is(tok::r_brace)) 1133 --Depth; 1134 if (Depth == 2) { 1135 if (C.Tok->is(tok::l_brace)) { 1136 Cell = 0; 1137 MatchingParen = C.Tok->MatchingParen; 1138 if (InitialSpaces == 0) { 1139 InitialSpaces = C.Spaces + C.TokenLength; 1140 InitialTokenLength = C.TokenLength; 1141 auto j = i - 1; 1142 for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) { 1143 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1144 InitialTokenLength += Changes[j].TokenLength; 1145 } 1146 if (C.NewlinesBefore == 0) { 1147 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1148 InitialTokenLength += Changes[j].TokenLength; 1149 } 1150 } 1151 } else if (C.Tok->is(tok::comma)) { 1152 if (!Cells.empty()) 1153 Cells.back().EndIndex = i; 1154 if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma 1155 ++Cell; 1156 } 1157 } else if (Depth == 1) { 1158 if (C.Tok == MatchingParen) { 1159 if (!Cells.empty()) 1160 Cells.back().EndIndex = i; 1161 Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr}); 1162 CellCount = C.Tok->Previous->isNot(tok::comma) ? Cell + 1 : Cell; 1163 // Go to the next non-comment and ensure there is a break in front 1164 const auto *NextNonComment = C.Tok->getNextNonComment(); 1165 while (NextNonComment->is(tok::comma)) 1166 NextNonComment = NextNonComment->getNextNonComment(); 1167 auto j = i; 1168 while (Changes[j].Tok != NextNonComment && j < End) 1169 j++; 1170 if (j < End && Changes[j].NewlinesBefore == 0 && 1171 Changes[j].Tok->isNot(tok::r_brace)) { 1172 Changes[j].NewlinesBefore = 1; 1173 // Account for the added token lengths 1174 Changes[j].Spaces = InitialSpaces - InitialTokenLength; 1175 } 1176 } else if (C.Tok->is(tok::comment)) { 1177 // Trailing comments stay at a space past the last token 1178 C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2; 1179 } else if (C.Tok->is(tok::l_brace)) { 1180 // We need to make sure that the ending braces is aligned to the 1181 // start of our initializer 1182 auto j = i - 1; 1183 for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j) 1184 ; // Nothing the loop does the work 1185 EndSpaces = Changes[j].Spaces; 1186 } 1187 } else if (Depth == 0 && C.Tok->is(tok::r_brace)) { 1188 C.NewlinesBefore = 1; 1189 C.Spaces = EndSpaces; 1190 } 1191 if (C.Tok->StartsColumn) { 1192 // This gets us past tokens that have been split over multiple 1193 // lines 1194 bool HasSplit = false; 1195 if (Changes[i].NewlinesBefore > 0) { 1196 // So if we split a line previously and the tail line + this token is 1197 // less then the column limit we remove the split here and just put 1198 // the column start at a space past the comma 1199 // 1200 // FIXME This if branch covers the cases where the column is not 1201 // the first column. This leads to weird pathologies like the formatting 1202 // auto foo = Items{ 1203 // Section{ 1204 // 0, bar(), 1205 // } 1206 // }; 1207 // Well if it doesn't lead to that it's indicative that the line 1208 // breaking should be revisited. Unfortunately alot of other options 1209 // interact with this 1210 auto j = i - 1; 1211 if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) && 1212 Changes[j - 1].NewlinesBefore > 0) { 1213 --j; 1214 auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength; 1215 if (LineLimit < Style.ColumnLimit) { 1216 Changes[i].NewlinesBefore = 0; 1217 Changes[i].Spaces = 1; 1218 } 1219 } 1220 } 1221 while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) { 1222 Changes[i].Spaces = InitialSpaces; 1223 ++i; 1224 HasSplit = true; 1225 } 1226 if (Changes[i].Tok != C.Tok) 1227 --i; 1228 Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr}); 1229 } 1230 } 1231 1232 return linkCells({Cells, CellCount, InitialSpaces}); 1233 } 1234 1235 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End, 1236 bool WithSpaces) const { 1237 unsigned CellWidth = 0; 1238 for (auto i = Start; i < End; i++) { 1239 if (Changes[i].NewlinesBefore > 0) 1240 CellWidth = 0; 1241 CellWidth += Changes[i].TokenLength; 1242 CellWidth += (WithSpaces ? Changes[i].Spaces : 0); 1243 } 1244 return CellWidth; 1245 } 1246 1247 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) { 1248 if ((End - Start) <= 1) 1249 return; 1250 // If the line is broken anywhere in there make sure everything 1251 // is aligned to the parent 1252 for (auto i = Start + 1; i < End; i++) { 1253 if (Changes[i].NewlinesBefore > 0) 1254 Changes[i].Spaces = Changes[Start].Spaces; 1255 } 1256 } 1257 1258 WhitespaceManager::CellDescriptions 1259 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) { 1260 auto &Cells = CellDesc.Cells; 1261 for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) { 1262 if (CellIter->NextColumnElement == nullptr && 1263 ((CellIter + 1) != Cells.end())) { 1264 for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) { 1265 if (NextIter->Cell == CellIter->Cell) { 1266 CellIter->NextColumnElement = &(*NextIter); 1267 break; 1268 } 1269 } 1270 } 1271 } 1272 return std::move(CellDesc); 1273 } 1274 1275 void WhitespaceManager::generateChanges() { 1276 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 1277 const Change &C = Changes[i]; 1278 if (i > 0 && Changes[i - 1].OriginalWhitespaceRange.getBegin() == 1279 C.OriginalWhitespaceRange.getBegin()) { 1280 // Do not generate two replacements for the same location. 1281 continue; 1282 } 1283 if (C.CreateReplacement) { 1284 std::string ReplacementText = C.PreviousLinePostfix; 1285 if (C.ContinuesPPDirective) 1286 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore, 1287 C.PreviousEndOfTokenColumn, 1288 C.EscapedNewlineColumn); 1289 else 1290 appendNewlineText(ReplacementText, C.NewlinesBefore); 1291 // FIXME: This assert should hold if we computed the column correctly. 1292 // assert((int)C.StartOfTokenColumn >= C.Spaces); 1293 appendIndentText( 1294 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces), 1295 std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces), 1296 C.IsAligned); 1297 ReplacementText.append(C.CurrentLinePrefix); 1298 storeReplacement(C.OriginalWhitespaceRange, ReplacementText); 1299 } 1300 } 1301 } 1302 1303 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) { 1304 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) - 1305 SourceMgr.getFileOffset(Range.getBegin()); 1306 // Don't create a replacement, if it does not change anything. 1307 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()), 1308 WhitespaceLength) == Text) 1309 return; 1310 auto Err = Replaces.add(tooling::Replacement( 1311 SourceMgr, CharSourceRange::getCharRange(Range), Text)); 1312 // FIXME: better error handling. For now, just print an error message in the 1313 // release version. 1314 if (Err) { 1315 llvm::errs() << llvm::toString(std::move(Err)) << "\n"; 1316 assert(false); 1317 } 1318 } 1319 1320 void WhitespaceManager::appendNewlineText(std::string &Text, 1321 unsigned Newlines) { 1322 for (unsigned i = 0; i < Newlines; ++i) 1323 Text.append(UseCRLF ? "\r\n" : "\n"); 1324 } 1325 1326 void WhitespaceManager::appendEscapedNewlineText( 1327 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn, 1328 unsigned EscapedNewlineColumn) { 1329 if (Newlines > 0) { 1330 unsigned Spaces = 1331 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1); 1332 for (unsigned i = 0; i < Newlines; ++i) { 1333 Text.append(Spaces, ' '); 1334 Text.append(UseCRLF ? "\\\r\n" : "\\\n"); 1335 Spaces = std::max<int>(0, EscapedNewlineColumn - 1); 1336 } 1337 } 1338 } 1339 1340 void WhitespaceManager::appendIndentText(std::string &Text, 1341 unsigned IndentLevel, unsigned Spaces, 1342 unsigned WhitespaceStartColumn, 1343 bool IsAligned) { 1344 switch (Style.UseTab) { 1345 case FormatStyle::UT_Never: 1346 Text.append(Spaces, ' '); 1347 break; 1348 case FormatStyle::UT_Always: { 1349 if (Style.TabWidth) { 1350 unsigned FirstTabWidth = 1351 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth; 1352 1353 // Insert only spaces when we want to end up before the next tab. 1354 if (Spaces < FirstTabWidth || Spaces == 1) { 1355 Text.append(Spaces, ' '); 1356 break; 1357 } 1358 // Align to the next tab. 1359 Spaces -= FirstTabWidth; 1360 Text.append("\t"); 1361 1362 Text.append(Spaces / Style.TabWidth, '\t'); 1363 Text.append(Spaces % Style.TabWidth, ' '); 1364 } else if (Spaces == 1) { 1365 Text.append(Spaces, ' '); 1366 } 1367 break; 1368 } 1369 case FormatStyle::UT_ForIndentation: 1370 if (WhitespaceStartColumn == 0) { 1371 unsigned Indentation = IndentLevel * Style.IndentWidth; 1372 Spaces = appendTabIndent(Text, Spaces, Indentation); 1373 } 1374 Text.append(Spaces, ' '); 1375 break; 1376 case FormatStyle::UT_ForContinuationAndIndentation: 1377 if (WhitespaceStartColumn == 0) 1378 Spaces = appendTabIndent(Text, Spaces, Spaces); 1379 Text.append(Spaces, ' '); 1380 break; 1381 case FormatStyle::UT_AlignWithSpaces: 1382 if (WhitespaceStartColumn == 0) { 1383 unsigned Indentation = 1384 IsAligned ? IndentLevel * Style.IndentWidth : Spaces; 1385 Spaces = appendTabIndent(Text, Spaces, Indentation); 1386 } 1387 Text.append(Spaces, ' '); 1388 break; 1389 } 1390 } 1391 1392 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces, 1393 unsigned Indentation) { 1394 // This happens, e.g. when a line in a block comment is indented less than the 1395 // first one. 1396 if (Indentation > Spaces) 1397 Indentation = Spaces; 1398 if (Style.TabWidth) { 1399 unsigned Tabs = Indentation / Style.TabWidth; 1400 Text.append(Tabs, '\t'); 1401 Spaces -= Tabs * Style.TabWidth; 1402 } 1403 return Spaces; 1404 } 1405 1406 } // namespace format 1407 } // namespace clang 1408