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(); --i) 264 if (ScopeStack.pop_back_val()) 265 --ConditionalsLevel; 266 } 267 } 268 269 // Align a single sequence of tokens, see AlignTokens below. 270 template <typename F> 271 static void 272 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End, 273 unsigned Column, F &&Matches, 274 SmallVector<WhitespaceManager::Change, 16> &Changes) { 275 bool FoundMatchOnLine = false; 276 int Shift = 0; 277 278 // ScopeStack keeps track of the current scope depth. It contains indices of 279 // the first token on each scope. 280 // We only run the "Matches" function on tokens from the outer-most scope. 281 // However, we do need to pay special attention to one class of tokens 282 // that are not in the outer-most scope, and that is function parameters 283 // which are split across multiple lines, as illustrated by this example: 284 // double a(int x); 285 // int b(int y, 286 // double z); 287 // In the above example, we need to take special care to ensure that 288 // 'double z' is indented along with it's owning function 'b'. 289 // The same holds for calling a function: 290 // double a = foo(x); 291 // int b = bar(foo(y), 292 // foor(z)); 293 // Similar for broken string literals: 294 // double x = 3.14; 295 // auto s = "Hello" 296 // "World"; 297 // Special handling is required for 'nested' ternary operators. 298 SmallVector<unsigned, 16> ScopeStack; 299 300 for (unsigned i = Start; i != End; ++i) { 301 if (ScopeStack.size() != 0 && 302 Changes[i].indentAndNestingLevel() < 303 Changes[ScopeStack.back()].indentAndNestingLevel()) 304 ScopeStack.pop_back(); 305 306 // Compare current token to previous non-comment token to ensure whether 307 // it is in a deeper scope or not. 308 unsigned PreviousNonComment = i - 1; 309 while (PreviousNonComment > Start && 310 Changes[PreviousNonComment].Tok->is(tok::comment)) 311 --PreviousNonComment; 312 if (i != Start && Changes[i].indentAndNestingLevel() > 313 Changes[PreviousNonComment].indentAndNestingLevel()) 314 ScopeStack.push_back(i); 315 316 bool InsideNestedScope = ScopeStack.size() != 0; 317 bool ContinuedStringLiteral = i > Start && 318 Changes[i].Tok->is(tok::string_literal) && 319 Changes[i - 1].Tok->is(tok::string_literal); 320 bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral; 321 322 if (Changes[i].NewlinesBefore > 0 && !SkipMatchCheck) { 323 Shift = 0; 324 FoundMatchOnLine = false; 325 } 326 327 // If this is the first matching token to be aligned, remember by how many 328 // spaces it has to be shifted, so the rest of the changes on the line are 329 // shifted by the same amount 330 if (!FoundMatchOnLine && !SkipMatchCheck && Matches(Changes[i])) { 331 FoundMatchOnLine = true; 332 Shift = Column - Changes[i].StartOfTokenColumn; 333 Changes[i].Spaces += Shift; 334 // FIXME: This is a workaround that should be removed when we fix 335 // http://llvm.org/PR53699. An assertion later below verifies this. 336 if (Changes[i].NewlinesBefore == 0) 337 Changes[i].Spaces = 338 std::max(Changes[i].Spaces, 339 static_cast<int>(Changes[i].Tok->SpacesRequiredBefore)); 340 } 341 342 // This is for function parameters that are split across multiple lines, 343 // as mentioned in the ScopeStack comment. 344 if (InsideNestedScope && Changes[i].NewlinesBefore > 0) { 345 unsigned ScopeStart = ScopeStack.back(); 346 auto ShouldShiftBeAdded = [&] { 347 // Function declaration 348 if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName)) 349 return true; 350 351 // Lambda. 352 if (Changes[ScopeStart - 1].Tok->is(TT_LambdaLBrace)) 353 return false; 354 355 // Continued function declaration 356 if (ScopeStart > Start + 1 && 357 Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) 358 return true; 359 360 // Continued function call 361 if (ScopeStart > Start + 1 && 362 Changes[ScopeStart - 2].Tok->is(tok::identifier) && 363 Changes[ScopeStart - 1].Tok->is(tok::l_paren) && 364 Changes[ScopeStart].Tok->isNot(TT_LambdaLSquare)) { 365 if (Changes[i].Tok->MatchingParen && 366 Changes[i].Tok->MatchingParen->is(TT_LambdaLBrace)) 367 return false; 368 return Style.BinPackArguments; 369 } 370 371 // Ternary operator 372 if (Changes[i].Tok->is(TT_ConditionalExpr)) 373 return true; 374 375 // Period Initializer .XXX = 1. 376 if (Changes[i].Tok->is(TT_DesignatedInitializerPeriod)) 377 return true; 378 379 // Continued ternary operator 380 if (Changes[i].Tok->Previous && 381 Changes[i].Tok->Previous->is(TT_ConditionalExpr)) 382 return true; 383 384 // Continued braced list. 385 if (ScopeStart > Start + 1 && 386 Changes[ScopeStart - 2].Tok->isNot(tok::identifier) && 387 Changes[ScopeStart - 1].Tok->is(tok::l_brace) && 388 Changes[i].Tok->isNot(tok::r_brace)) { 389 for (unsigned OuterScopeStart : llvm::reverse(ScopeStack)) { 390 // Lambda. 391 if (OuterScopeStart > Start && 392 Changes[OuterScopeStart - 1].Tok->is(TT_LambdaLBrace)) 393 return false; 394 } 395 return true; 396 } 397 398 return false; 399 }; 400 401 if (ShouldShiftBeAdded()) 402 Changes[i].Spaces += Shift; 403 } 404 405 if (ContinuedStringLiteral) 406 Changes[i].Spaces += Shift; 407 408 // We should not remove required spaces unless we break the line before. 409 assert(Shift >= 0 || Changes[i].NewlinesBefore > 0 || 410 Changes[i].Spaces >= 411 static_cast<int>(Changes[i].Tok->SpacesRequiredBefore) || 412 Changes[i].Tok->is(tok::eof)); 413 414 Changes[i].StartOfTokenColumn += Shift; 415 if (i + 1 != Changes.size()) 416 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 417 418 // If PointerAlignment is PAS_Right, keep *s or &s next to the token 419 if (Style.PointerAlignment == FormatStyle::PAS_Right && 420 Changes[i].Spaces != 0) { 421 for (int Previous = i - 1; 422 Previous >= 0 && 423 Changes[Previous].Tok->getType() == TT_PointerOrReference; 424 --Previous) { 425 Changes[Previous + 1].Spaces -= Shift; 426 Changes[Previous].Spaces += Shift; 427 } 428 } 429 } 430 } 431 432 // Walk through a subset of the changes, starting at StartAt, and find 433 // sequences of matching tokens to align. To do so, keep track of the lines and 434 // whether or not a matching token was found on a line. If a matching token is 435 // found, extend the current sequence. If the current line cannot be part of a 436 // sequence, e.g. because there is an empty line before it or it contains only 437 // non-matching tokens, finalize the previous sequence. 438 // The value returned is the token on which we stopped, either because we 439 // exhausted all items inside Changes, or because we hit a scope level higher 440 // than our initial scope. 441 // This function is recursive. Each invocation processes only the scope level 442 // equal to the initial level, which is the level of Changes[StartAt]. 443 // If we encounter a scope level greater than the initial level, then we call 444 // ourselves recursively, thereby avoiding the pollution of the current state 445 // with the alignment requirements of the nested sub-level. This recursive 446 // behavior is necessary for aligning function prototypes that have one or more 447 // arguments. 448 // If this function encounters a scope level less than the initial level, 449 // it returns the current position. 450 // There is a non-obvious subtlety in the recursive behavior: Even though we 451 // defer processing of nested levels to recursive invocations of this 452 // function, when it comes time to align a sequence of tokens, we run the 453 // alignment on the entire sequence, including the nested levels. 454 // When doing so, most of the nested tokens are skipped, because their 455 // alignment was already handled by the recursive invocations of this function. 456 // However, the special exception is that we do NOT skip function parameters 457 // that are split across multiple lines. See the test case in FormatTest.cpp 458 // that mentions "split function parameter alignment" for an example of this. 459 template <typename F> 460 static unsigned AlignTokens( 461 const FormatStyle &Style, F &&Matches, 462 SmallVector<WhitespaceManager::Change, 16> &Changes, unsigned StartAt, 463 const FormatStyle::AlignConsecutiveStyle &ACS = FormatStyle::ACS_None) { 464 unsigned MinColumn = 0; 465 unsigned MaxColumn = UINT_MAX; 466 467 // Line number of the start and the end of the current token sequence. 468 unsigned StartOfSequence = 0; 469 unsigned EndOfSequence = 0; 470 471 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and 472 // abort when we hit any token in a higher scope than the starting one. 473 auto IndentAndNestingLevel = StartAt < Changes.size() 474 ? Changes[StartAt].indentAndNestingLevel() 475 : std::tuple<unsigned, unsigned, unsigned>(); 476 477 // Keep track of the number of commas before the matching tokens, we will only 478 // align a sequence of matching tokens if they are preceded by the same number 479 // of commas. 480 unsigned CommasBeforeLastMatch = 0; 481 unsigned CommasBeforeMatch = 0; 482 483 // Whether a matching token has been found on the current line. 484 bool FoundMatchOnLine = false; 485 486 // Whether the current line consists purely of comments. 487 bool LineIsComment = true; 488 489 // Aligns a sequence of matching tokens, on the MinColumn column. 490 // 491 // Sequences start from the first matching token to align, and end at the 492 // first token of the first line that doesn't need to be aligned. 493 // 494 // We need to adjust the StartOfTokenColumn of each Change that is on a line 495 // containing any matching token to be aligned and located after such token. 496 auto AlignCurrentSequence = [&] { 497 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) 498 AlignTokenSequence(Style, StartOfSequence, EndOfSequence, MinColumn, 499 Matches, Changes); 500 MinColumn = 0; 501 MaxColumn = UINT_MAX; 502 StartOfSequence = 0; 503 EndOfSequence = 0; 504 }; 505 506 unsigned i = StartAt; 507 for (unsigned e = Changes.size(); i != e; ++i) { 508 if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel) 509 break; 510 511 if (Changes[i].NewlinesBefore != 0) { 512 CommasBeforeMatch = 0; 513 EndOfSequence = i; 514 515 // Whether to break the alignment sequence because of an empty line. 516 bool EmptyLineBreak = 517 (Changes[i].NewlinesBefore > 1) && 518 (ACS != FormatStyle::ACS_AcrossEmptyLines) && 519 (ACS != FormatStyle::ACS_AcrossEmptyLinesAndComments); 520 521 // Whether to break the alignment sequence because of a line without a 522 // match. 523 bool NoMatchBreak = 524 !FoundMatchOnLine && 525 !(LineIsComment && 526 ((ACS == FormatStyle::ACS_AcrossComments) || 527 (ACS == FormatStyle::ACS_AcrossEmptyLinesAndComments))); 528 529 if (EmptyLineBreak || NoMatchBreak) 530 AlignCurrentSequence(); 531 532 // A new line starts, re-initialize line status tracking bools. 533 // Keep the match state if a string literal is continued on this line. 534 if (i == 0 || !Changes[i].Tok->is(tok::string_literal) || 535 !Changes[i - 1].Tok->is(tok::string_literal)) 536 FoundMatchOnLine = false; 537 LineIsComment = true; 538 } 539 540 if (!Changes[i].Tok->is(tok::comment)) 541 LineIsComment = false; 542 543 if (Changes[i].Tok->is(tok::comma)) { 544 ++CommasBeforeMatch; 545 } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) { 546 // Call AlignTokens recursively, skipping over this scope block. 547 unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i, ACS); 548 i = StoppedAt - 1; 549 continue; 550 } 551 552 if (!Matches(Changes[i])) 553 continue; 554 555 // If there is more than one matching token per line, or if the number of 556 // preceding commas, do not match anymore, end the sequence. 557 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch) 558 AlignCurrentSequence(); 559 560 CommasBeforeLastMatch = CommasBeforeMatch; 561 FoundMatchOnLine = true; 562 563 if (StartOfSequence == 0) 564 StartOfSequence = i; 565 566 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn; 567 int LineLengthAfter = Changes[i].TokenLength; 568 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) { 569 LineLengthAfter += Changes[j].Spaces; 570 // Changes are generally 1:1 with the tokens, but a change could also be 571 // inside of a token, in which case it's counted more than once: once for 572 // the whitespace surrounding the token (!IsInsideToken) and once for 573 // each whitespace change within it (IsInsideToken). 574 // Therefore, changes inside of a token should only count the space. 575 if (!Changes[j].IsInsideToken) 576 LineLengthAfter += Changes[j].TokenLength; 577 } 578 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter; 579 580 // If we are restricted by the maximum column width, end the sequence. 581 if (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn || 582 CommasBeforeLastMatch != CommasBeforeMatch) { 583 AlignCurrentSequence(); 584 StartOfSequence = i; 585 } 586 587 MinColumn = std::max(MinColumn, ChangeMinColumn); 588 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 589 } 590 591 EndOfSequence = i; 592 AlignCurrentSequence(); 593 return i; 594 } 595 596 // Aligns a sequence of matching tokens, on the MinColumn column. 597 // 598 // Sequences start from the first matching token to align, and end at the 599 // first token of the first line that doesn't need to be aligned. 600 // 601 // We need to adjust the StartOfTokenColumn of each Change that is on a line 602 // containing any matching token to be aligned and located after such token. 603 static void AlignMacroSequence( 604 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn, 605 unsigned &MaxColumn, bool &FoundMatchOnLine, 606 std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches, 607 SmallVector<WhitespaceManager::Change, 16> &Changes) { 608 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) { 609 610 FoundMatchOnLine = false; 611 int Shift = 0; 612 613 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) { 614 if (Changes[I].NewlinesBefore > 0) { 615 Shift = 0; 616 FoundMatchOnLine = false; 617 } 618 619 // If this is the first matching token to be aligned, remember by how many 620 // spaces it has to be shifted, so the rest of the changes on the line are 621 // shifted by the same amount 622 if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) { 623 FoundMatchOnLine = true; 624 Shift = MinColumn - Changes[I].StartOfTokenColumn; 625 Changes[I].Spaces += Shift; 626 } 627 628 assert(Shift >= 0); 629 Changes[I].StartOfTokenColumn += Shift; 630 if (I + 1 != Changes.size()) 631 Changes[I + 1].PreviousEndOfTokenColumn += Shift; 632 } 633 } 634 635 MinColumn = 0; 636 MaxColumn = UINT_MAX; 637 StartOfSequence = 0; 638 EndOfSequence = 0; 639 } 640 641 void WhitespaceManager::alignConsecutiveMacros() { 642 if (Style.AlignConsecutiveMacros == FormatStyle::ACS_None) 643 return; 644 645 auto AlignMacrosMatches = [](const Change &C) { 646 const FormatToken *Current = C.Tok; 647 unsigned SpacesRequiredBefore = 1; 648 649 if (Current->SpacesRequiredBefore == 0 || !Current->Previous) 650 return false; 651 652 Current = Current->Previous; 653 654 // If token is a ")", skip over the parameter list, to the 655 // token that precedes the "(" 656 if (Current->is(tok::r_paren) && Current->MatchingParen) { 657 Current = Current->MatchingParen->Previous; 658 SpacesRequiredBefore = 0; 659 } 660 661 if (!Current || !Current->is(tok::identifier)) 662 return false; 663 664 if (!Current->Previous || !Current->Previous->is(tok::pp_define)) 665 return false; 666 667 // For a macro function, 0 spaces are required between the 668 // identifier and the lparen that opens the parameter list. 669 // For a simple macro, 1 space is required between the 670 // identifier and the first token of the defined value. 671 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore; 672 }; 673 674 unsigned MinColumn = 0; 675 unsigned MaxColumn = UINT_MAX; 676 677 // Start and end of the token sequence we're processing. 678 unsigned StartOfSequence = 0; 679 unsigned EndOfSequence = 0; 680 681 // Whether a matching token has been found on the current line. 682 bool FoundMatchOnLine = false; 683 684 // Whether the current line consists only of comments 685 bool LineIsComment = true; 686 687 unsigned I = 0; 688 for (unsigned E = Changes.size(); I != E; ++I) { 689 if (Changes[I].NewlinesBefore != 0) { 690 EndOfSequence = I; 691 692 // Whether to break the alignment sequence because of an empty line. 693 bool EmptyLineBreak = 694 (Changes[I].NewlinesBefore > 1) && 695 (Style.AlignConsecutiveMacros != FormatStyle::ACS_AcrossEmptyLines) && 696 (Style.AlignConsecutiveMacros != 697 FormatStyle::ACS_AcrossEmptyLinesAndComments); 698 699 // Whether to break the alignment sequence because of a line without a 700 // match. 701 bool NoMatchBreak = 702 !FoundMatchOnLine && 703 !(LineIsComment && ((Style.AlignConsecutiveMacros == 704 FormatStyle::ACS_AcrossComments) || 705 (Style.AlignConsecutiveMacros == 706 FormatStyle::ACS_AcrossEmptyLinesAndComments))); 707 708 if (EmptyLineBreak || NoMatchBreak) 709 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 710 FoundMatchOnLine, AlignMacrosMatches, Changes); 711 712 // A new line starts, re-initialize line status tracking bools. 713 FoundMatchOnLine = false; 714 LineIsComment = true; 715 } 716 717 if (!Changes[I].Tok->is(tok::comment)) 718 LineIsComment = false; 719 720 if (!AlignMacrosMatches(Changes[I])) 721 continue; 722 723 FoundMatchOnLine = true; 724 725 if (StartOfSequence == 0) 726 StartOfSequence = I; 727 728 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn; 729 int LineLengthAfter = -Changes[I].Spaces; 730 for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j) 731 LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength; 732 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter; 733 734 MinColumn = std::max(MinColumn, ChangeMinColumn); 735 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 736 } 737 738 EndOfSequence = I; 739 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 740 FoundMatchOnLine, AlignMacrosMatches, Changes); 741 } 742 743 void WhitespaceManager::alignConsecutiveAssignments() { 744 if (Style.AlignConsecutiveAssignments == FormatStyle::ACS_None) 745 return; 746 747 AlignTokens( 748 Style, 749 [&](const Change &C) { 750 // Do not align on equal signs that are first on a line. 751 if (C.NewlinesBefore > 0) 752 return false; 753 754 // Do not align on equal signs that are last on a line. 755 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 756 return false; 757 758 // Do not align operator= overloads. 759 FormatToken *Previous = C.Tok->getPreviousNonComment(); 760 if (Previous && Previous->is(tok::kw_operator)) 761 return false; 762 763 return C.Tok->is(tok::equal); 764 }, 765 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments); 766 } 767 768 void WhitespaceManager::alignConsecutiveBitFields() { 769 if (Style.AlignConsecutiveBitFields == FormatStyle::ACS_None) 770 return; 771 772 AlignTokens( 773 Style, 774 [&](Change const &C) { 775 // Do not align on ':' that is first on a line. 776 if (C.NewlinesBefore > 0) 777 return false; 778 779 // Do not align on ':' that is last on a line. 780 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 781 return false; 782 783 return C.Tok->is(TT_BitFieldColon); 784 }, 785 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields); 786 } 787 788 void WhitespaceManager::alignConsecutiveDeclarations() { 789 if (Style.AlignConsecutiveDeclarations == FormatStyle::ACS_None) 790 return; 791 792 AlignTokens( 793 Style, 794 [](Change const &C) { 795 // tok::kw_operator is necessary for aligning operator overload 796 // definitions. 797 if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator)) 798 return true; 799 if (C.Tok->isNot(TT_StartOfName)) 800 return false; 801 if (C.Tok->Previous && 802 C.Tok->Previous->is(TT_StatementAttributeLikeMacro)) 803 return false; 804 // Check if there is a subsequent name that starts the same declaration. 805 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) { 806 if (Next->is(tok::comment)) 807 continue; 808 if (Next->is(TT_PointerOrReference)) 809 return false; 810 if (!Next->Tok.getIdentifierInfo()) 811 break; 812 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName, 813 tok::kw_operator)) 814 return false; 815 } 816 return true; 817 }, 818 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations); 819 } 820 821 void WhitespaceManager::alignChainedConditionals() { 822 if (Style.BreakBeforeTernaryOperators) { 823 AlignTokens( 824 Style, 825 [](Change const &C) { 826 // Align question operators and last colon 827 return C.Tok->is(TT_ConditionalExpr) && 828 ((C.Tok->is(tok::question) && !C.NewlinesBefore) || 829 (C.Tok->is(tok::colon) && C.Tok->Next && 830 (C.Tok->Next->FakeLParens.size() == 0 || 831 C.Tok->Next->FakeLParens.back() != prec::Conditional))); 832 }, 833 Changes, /*StartAt=*/0); 834 } else { 835 static auto AlignWrappedOperand = [](Change const &C) { 836 FormatToken *Previous = C.Tok->getPreviousNonComment(); 837 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) && 838 (Previous->is(tok::colon) && 839 (C.Tok->FakeLParens.size() == 0 || 840 C.Tok->FakeLParens.back() != prec::Conditional)); 841 }; 842 // Ensure we keep alignment of wrapped operands with non-wrapped operands 843 // Since we actually align the operators, the wrapped operands need the 844 // extra offset to be properly aligned. 845 for (Change &C : Changes) 846 if (AlignWrappedOperand(C)) 847 C.StartOfTokenColumn -= 2; 848 AlignTokens( 849 Style, 850 [this](Change const &C) { 851 // Align question operators if next operand is not wrapped, as 852 // well as wrapped operands after question operator or last 853 // colon in conditional sequence 854 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) && 855 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 && 856 !(&C + 1)->IsTrailingComment) || 857 AlignWrappedOperand(C); 858 }, 859 Changes, /*StartAt=*/0); 860 } 861 } 862 863 void WhitespaceManager::alignTrailingComments() { 864 unsigned MinColumn = 0; 865 unsigned MaxColumn = UINT_MAX; 866 unsigned StartOfSequence = 0; 867 bool BreakBeforeNext = false; 868 unsigned Newlines = 0; 869 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 870 if (Changes[i].StartOfBlockComment) 871 continue; 872 Newlines += Changes[i].NewlinesBefore; 873 if (!Changes[i].IsTrailingComment) 874 continue; 875 876 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn; 877 unsigned ChangeMaxColumn; 878 879 if (Style.ColumnLimit == 0) 880 ChangeMaxColumn = UINT_MAX; 881 else if (Style.ColumnLimit >= Changes[i].TokenLength) 882 ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength; 883 else 884 ChangeMaxColumn = ChangeMinColumn; 885 886 // If we don't create a replacement for this change, we have to consider 887 // it to be immovable. 888 if (!Changes[i].CreateReplacement) 889 ChangeMaxColumn = ChangeMinColumn; 890 891 if (i + 1 != e && Changes[i + 1].ContinuesPPDirective) 892 ChangeMaxColumn -= 2; 893 // If this comment follows an } in column 0, it probably documents the 894 // closing of a namespace and we don't want to align it. 895 bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 && 896 Changes[i - 1].Tok->is(tok::r_brace) && 897 Changes[i - 1].StartOfTokenColumn == 0; 898 bool WasAlignedWithStartOfNextLine = false; 899 if (Changes[i].NewlinesBefore == 1) { // A comment on its own line. 900 unsigned CommentColumn = SourceMgr.getSpellingColumnNumber( 901 Changes[i].OriginalWhitespaceRange.getEnd()); 902 for (unsigned j = i + 1; j != e; ++j) { 903 if (Changes[j].Tok->is(tok::comment)) 904 continue; 905 906 unsigned NextColumn = SourceMgr.getSpellingColumnNumber( 907 Changes[j].OriginalWhitespaceRange.getEnd()); 908 // The start of the next token was previously aligned with the 909 // start of this comment. 910 WasAlignedWithStartOfNextLine = 911 CommentColumn == NextColumn || 912 CommentColumn == NextColumn + Style.IndentWidth; 913 break; 914 } 915 } 916 if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) { 917 alignTrailingComments(StartOfSequence, i, MinColumn); 918 MinColumn = ChangeMinColumn; 919 MaxColumn = ChangeMinColumn; 920 StartOfSequence = i; 921 } else if (BreakBeforeNext || Newlines > 1 || 922 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) || 923 // Break the comment sequence if the previous line did not end 924 // in a trailing comment. 925 (Changes[i].NewlinesBefore == 1 && i > 0 && 926 !Changes[i - 1].IsTrailingComment) || 927 WasAlignedWithStartOfNextLine) { 928 alignTrailingComments(StartOfSequence, i, MinColumn); 929 MinColumn = ChangeMinColumn; 930 MaxColumn = ChangeMaxColumn; 931 StartOfSequence = i; 932 } else { 933 MinColumn = std::max(MinColumn, ChangeMinColumn); 934 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 935 } 936 BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) || 937 // Never start a sequence with a comment at the beginning 938 // of the line. 939 (Changes[i].NewlinesBefore == 1 && StartOfSequence == i); 940 Newlines = 0; 941 } 942 alignTrailingComments(StartOfSequence, Changes.size(), MinColumn); 943 } 944 945 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End, 946 unsigned Column) { 947 for (unsigned i = Start; i != End; ++i) { 948 int Shift = 0; 949 if (Changes[i].IsTrailingComment) 950 Shift = Column - Changes[i].StartOfTokenColumn; 951 if (Changes[i].StartOfBlockComment) { 952 Shift = Changes[i].IndentationOffset + 953 Changes[i].StartOfBlockComment->StartOfTokenColumn - 954 Changes[i].StartOfTokenColumn; 955 } 956 if (Shift < 0) 957 continue; 958 Changes[i].Spaces += Shift; 959 if (i + 1 != Changes.size()) 960 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 961 Changes[i].StartOfTokenColumn += Shift; 962 } 963 } 964 965 void WhitespaceManager::alignEscapedNewlines() { 966 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign) 967 return; 968 969 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left; 970 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 971 unsigned StartOfMacro = 0; 972 for (unsigned i = 1, e = Changes.size(); i < e; ++i) { 973 Change &C = Changes[i]; 974 if (C.NewlinesBefore > 0) { 975 if (C.ContinuesPPDirective) { 976 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine); 977 } else { 978 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine); 979 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 980 StartOfMacro = i; 981 } 982 } 983 } 984 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine); 985 } 986 987 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End, 988 unsigned Column) { 989 for (unsigned i = Start; i < End; ++i) { 990 Change &C = Changes[i]; 991 if (C.NewlinesBefore > 0) { 992 assert(C.ContinuesPPDirective); 993 if (C.PreviousEndOfTokenColumn + 1 > Column) 994 C.EscapedNewlineColumn = 0; 995 else 996 C.EscapedNewlineColumn = Column; 997 } 998 } 999 } 1000 1001 void WhitespaceManager::alignArrayInitializers() { 1002 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None) 1003 return; 1004 1005 for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size(); 1006 ChangeIndex < ChangeEnd; ++ChangeIndex) { 1007 auto &C = Changes[ChangeIndex]; 1008 if (C.Tok->IsArrayInitializer) { 1009 bool FoundComplete = false; 1010 for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd; 1011 ++InsideIndex) { 1012 if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) { 1013 alignArrayInitializers(ChangeIndex, InsideIndex + 1); 1014 ChangeIndex = InsideIndex + 1; 1015 FoundComplete = true; 1016 break; 1017 } 1018 } 1019 if (!FoundComplete) 1020 ChangeIndex = ChangeEnd; 1021 } 1022 } 1023 } 1024 1025 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) { 1026 1027 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right) 1028 alignArrayInitializersRightJustified(getCells(Start, End)); 1029 else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left) 1030 alignArrayInitializersLeftJustified(getCells(Start, End)); 1031 } 1032 1033 void WhitespaceManager::alignArrayInitializersRightJustified( 1034 CellDescriptions &&CellDescs) { 1035 auto &Cells = CellDescs.Cells; 1036 1037 // Now go through and fixup the spaces. 1038 auto *CellIter = Cells.begin(); 1039 for (auto i = 0U; i < CellDescs.CellCount; ++i, ++CellIter) { 1040 unsigned NetWidth = 0U; 1041 if (isSplitCell(*CellIter)) 1042 NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1043 auto CellWidth = getMaximumCellWidth(CellIter, NetWidth); 1044 1045 if (Changes[CellIter->Index].Tok->is(tok::r_brace)) { 1046 // So in here we want to see if there is a brace that falls 1047 // on a line that was split. If so on that line we make sure that 1048 // the spaces in front of the brace are enough. 1049 Changes[CellIter->Index].NewlinesBefore = 0; 1050 Changes[CellIter->Index].Spaces = 0; 1051 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1052 Next = Next->NextColumnElement) { 1053 Changes[Next->Index].Spaces = 0; 1054 Changes[Next->Index].NewlinesBefore = 0; 1055 } 1056 // Unless the array is empty, we need the position of all the 1057 // immediately adjacent cells 1058 if (CellIter != Cells.begin()) { 1059 auto ThisNetWidth = 1060 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1061 auto MaxNetWidth = 1062 getMaximumNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces, 1063 CellDescs.CellCount); 1064 if (ThisNetWidth < MaxNetWidth) 1065 Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1066 auto RowCount = 1U; 1067 auto Offset = std::distance(Cells.begin(), CellIter); 1068 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1069 Next = Next->NextColumnElement) { 1070 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCount); 1071 auto *End = Start + Offset; 1072 ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1073 if (ThisNetWidth < MaxNetWidth) 1074 Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1075 ++RowCount; 1076 } 1077 } 1078 } else { 1079 auto ThisWidth = 1080 calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) + 1081 NetWidth; 1082 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1083 Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth)); 1084 Changes[CellIter->Index].Spaces += (i > 0) ? 1 : 0; 1085 } 1086 alignToStartOfCell(CellIter->Index, CellIter->EndIndex); 1087 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1088 Next = Next->NextColumnElement) { 1089 ThisWidth = 1090 calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth; 1091 if (Changes[Next->Index].NewlinesBefore == 0) { 1092 Changes[Next->Index].Spaces = (CellWidth - ThisWidth); 1093 Changes[Next->Index].Spaces += (i > 0) ? 1 : 0; 1094 } 1095 alignToStartOfCell(Next->Index, Next->EndIndex); 1096 } 1097 } 1098 } 1099 } 1100 1101 void WhitespaceManager::alignArrayInitializersLeftJustified( 1102 CellDescriptions &&CellDescs) { 1103 auto &Cells = CellDescs.Cells; 1104 1105 // Now go through and fixup the spaces. 1106 auto *CellIter = Cells.begin(); 1107 // The first cell needs to be against the left brace. 1108 if (Changes[CellIter->Index].NewlinesBefore == 0) 1109 Changes[CellIter->Index].Spaces = 0; 1110 else 1111 Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces; 1112 ++CellIter; 1113 for (auto i = 1U; i < CellDescs.CellCount; i++, ++CellIter) { 1114 auto MaxNetWidth = getMaximumNetWidth( 1115 Cells.begin(), CellIter, CellDescs.InitialSpaces, CellDescs.CellCount); 1116 auto ThisNetWidth = 1117 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1118 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1119 Changes[CellIter->Index].Spaces = 1120 MaxNetWidth - ThisNetWidth + 1121 (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1122 } 1123 auto RowCount = 1U; 1124 auto Offset = std::distance(Cells.begin(), CellIter); 1125 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1126 Next = Next->NextColumnElement) { 1127 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCount); 1128 auto *End = Start + Offset; 1129 auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1130 if (Changes[Next->Index].NewlinesBefore == 0) { 1131 Changes[Next->Index].Spaces = 1132 MaxNetWidth - ThisNetWidth + 1133 (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1134 } 1135 ++RowCount; 1136 } 1137 } 1138 } 1139 1140 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) { 1141 if (Cell.HasSplit) 1142 return true; 1143 for (const auto *Next = Cell.NextColumnElement; Next != nullptr; 1144 Next = Next->NextColumnElement) 1145 if (Next->HasSplit) 1146 return true; 1147 return false; 1148 } 1149 1150 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start, 1151 unsigned End) { 1152 1153 unsigned Depth = 0; 1154 unsigned Cell = 0; 1155 unsigned CellCount = 0; 1156 unsigned InitialSpaces = 0; 1157 unsigned InitialTokenLength = 0; 1158 unsigned EndSpaces = 0; 1159 SmallVector<CellDescription> Cells; 1160 const FormatToken *MatchingParen = nullptr; 1161 for (unsigned i = Start; i < End; ++i) { 1162 auto &C = Changes[i]; 1163 if (C.Tok->is(tok::l_brace)) 1164 ++Depth; 1165 else if (C.Tok->is(tok::r_brace)) 1166 --Depth; 1167 if (Depth == 2) { 1168 if (C.Tok->is(tok::l_brace)) { 1169 Cell = 0; 1170 MatchingParen = C.Tok->MatchingParen; 1171 if (InitialSpaces == 0) { 1172 InitialSpaces = C.Spaces + C.TokenLength; 1173 InitialTokenLength = C.TokenLength; 1174 auto j = i - 1; 1175 for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) { 1176 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1177 InitialTokenLength += Changes[j].TokenLength; 1178 } 1179 if (C.NewlinesBefore == 0) { 1180 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1181 InitialTokenLength += Changes[j].TokenLength; 1182 } 1183 } 1184 } else if (C.Tok->is(tok::comma)) { 1185 if (!Cells.empty()) 1186 Cells.back().EndIndex = i; 1187 if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma 1188 ++Cell; 1189 } 1190 } else if (Depth == 1) { 1191 if (C.Tok == MatchingParen) { 1192 if (!Cells.empty()) 1193 Cells.back().EndIndex = i; 1194 Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr}); 1195 CellCount = C.Tok->Previous->isNot(tok::comma) ? Cell + 1 : Cell; 1196 // Go to the next non-comment and ensure there is a break in front 1197 const auto *NextNonComment = C.Tok->getNextNonComment(); 1198 while (NextNonComment->is(tok::comma)) 1199 NextNonComment = NextNonComment->getNextNonComment(); 1200 auto j = i; 1201 while (Changes[j].Tok != NextNonComment && j < End) 1202 ++j; 1203 if (j < End && Changes[j].NewlinesBefore == 0 && 1204 Changes[j].Tok->isNot(tok::r_brace)) { 1205 Changes[j].NewlinesBefore = 1; 1206 // Account for the added token lengths 1207 Changes[j].Spaces = InitialSpaces - InitialTokenLength; 1208 } 1209 } else if (C.Tok->is(tok::comment)) { 1210 // Trailing comments stay at a space past the last token 1211 C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2; 1212 } else if (C.Tok->is(tok::l_brace)) { 1213 // We need to make sure that the ending braces is aligned to the 1214 // start of our initializer 1215 auto j = i - 1; 1216 for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j) 1217 ; // Nothing the loop does the work 1218 EndSpaces = Changes[j].Spaces; 1219 } 1220 } else if (Depth == 0 && C.Tok->is(tok::r_brace)) { 1221 C.NewlinesBefore = 1; 1222 C.Spaces = EndSpaces; 1223 } 1224 if (C.Tok->StartsColumn) { 1225 // This gets us past tokens that have been split over multiple 1226 // lines 1227 bool HasSplit = false; 1228 if (Changes[i].NewlinesBefore > 0) { 1229 // So if we split a line previously and the tail line + this token is 1230 // less then the column limit we remove the split here and just put 1231 // the column start at a space past the comma 1232 // 1233 // FIXME This if branch covers the cases where the column is not 1234 // the first column. This leads to weird pathologies like the formatting 1235 // auto foo = Items{ 1236 // Section{ 1237 // 0, bar(), 1238 // } 1239 // }; 1240 // Well if it doesn't lead to that it's indicative that the line 1241 // breaking should be revisited. Unfortunately alot of other options 1242 // interact with this 1243 auto j = i - 1; 1244 if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) && 1245 Changes[j - 1].NewlinesBefore > 0) { 1246 --j; 1247 auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength; 1248 if (LineLimit < Style.ColumnLimit) { 1249 Changes[i].NewlinesBefore = 0; 1250 Changes[i].Spaces = 1; 1251 } 1252 } 1253 } 1254 while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) { 1255 Changes[i].Spaces = InitialSpaces; 1256 ++i; 1257 HasSplit = true; 1258 } 1259 if (Changes[i].Tok != C.Tok) 1260 --i; 1261 Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr}); 1262 } 1263 } 1264 1265 return linkCells({Cells, CellCount, InitialSpaces}); 1266 } 1267 1268 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End, 1269 bool WithSpaces) const { 1270 unsigned CellWidth = 0; 1271 for (auto i = Start; i < End; i++) { 1272 if (Changes[i].NewlinesBefore > 0) 1273 CellWidth = 0; 1274 CellWidth += Changes[i].TokenLength; 1275 CellWidth += (WithSpaces ? Changes[i].Spaces : 0); 1276 } 1277 return CellWidth; 1278 } 1279 1280 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) { 1281 if ((End - Start) <= 1) 1282 return; 1283 // If the line is broken anywhere in there make sure everything 1284 // is aligned to the parent 1285 for (auto i = Start + 1; i < End; i++) 1286 if (Changes[i].NewlinesBefore > 0) 1287 Changes[i].Spaces = Changes[Start].Spaces; 1288 } 1289 1290 WhitespaceManager::CellDescriptions 1291 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) { 1292 auto &Cells = CellDesc.Cells; 1293 for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) { 1294 if (CellIter->NextColumnElement == nullptr && 1295 ((CellIter + 1) != Cells.end())) { 1296 for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) { 1297 if (NextIter->Cell == CellIter->Cell) { 1298 CellIter->NextColumnElement = &(*NextIter); 1299 break; 1300 } 1301 } 1302 } 1303 } 1304 return std::move(CellDesc); 1305 } 1306 1307 void WhitespaceManager::generateChanges() { 1308 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 1309 const Change &C = Changes[i]; 1310 if (i > 0 && Changes[i - 1].OriginalWhitespaceRange.getBegin() == 1311 C.OriginalWhitespaceRange.getBegin()) { 1312 // Do not generate two replacements for the same location. 1313 continue; 1314 } 1315 if (C.CreateReplacement) { 1316 std::string ReplacementText = C.PreviousLinePostfix; 1317 if (C.ContinuesPPDirective) 1318 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore, 1319 C.PreviousEndOfTokenColumn, 1320 C.EscapedNewlineColumn); 1321 else 1322 appendNewlineText(ReplacementText, C.NewlinesBefore); 1323 // FIXME: This assert should hold if we computed the column correctly. 1324 // assert((int)C.StartOfTokenColumn >= C.Spaces); 1325 appendIndentText( 1326 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces), 1327 std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces), 1328 C.IsAligned); 1329 ReplacementText.append(C.CurrentLinePrefix); 1330 storeReplacement(C.OriginalWhitespaceRange, ReplacementText); 1331 } 1332 } 1333 } 1334 1335 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) { 1336 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) - 1337 SourceMgr.getFileOffset(Range.getBegin()); 1338 // Don't create a replacement, if it does not change anything. 1339 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()), 1340 WhitespaceLength) == Text) 1341 return; 1342 auto Err = Replaces.add(tooling::Replacement( 1343 SourceMgr, CharSourceRange::getCharRange(Range), Text)); 1344 // FIXME: better error handling. For now, just print an error message in the 1345 // release version. 1346 if (Err) { 1347 llvm::errs() << llvm::toString(std::move(Err)) << "\n"; 1348 assert(false); 1349 } 1350 } 1351 1352 void WhitespaceManager::appendNewlineText(std::string &Text, 1353 unsigned Newlines) { 1354 if (UseCRLF) { 1355 Text.reserve(Text.size() + 2 * Newlines); 1356 for (unsigned i = 0; i < Newlines; ++i) 1357 Text.append("\r\n"); 1358 } else { 1359 Text.append(Newlines, '\n'); 1360 } 1361 } 1362 1363 void WhitespaceManager::appendEscapedNewlineText( 1364 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn, 1365 unsigned EscapedNewlineColumn) { 1366 if (Newlines > 0) { 1367 unsigned Spaces = 1368 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1); 1369 for (unsigned i = 0; i < Newlines; ++i) { 1370 Text.append(Spaces, ' '); 1371 Text.append(UseCRLF ? "\\\r\n" : "\\\n"); 1372 Spaces = std::max<int>(0, EscapedNewlineColumn - 1); 1373 } 1374 } 1375 } 1376 1377 void WhitespaceManager::appendIndentText(std::string &Text, 1378 unsigned IndentLevel, unsigned Spaces, 1379 unsigned WhitespaceStartColumn, 1380 bool IsAligned) { 1381 switch (Style.UseTab) { 1382 case FormatStyle::UT_Never: 1383 Text.append(Spaces, ' '); 1384 break; 1385 case FormatStyle::UT_Always: { 1386 if (Style.TabWidth) { 1387 unsigned FirstTabWidth = 1388 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth; 1389 1390 // Insert only spaces when we want to end up before the next tab. 1391 if (Spaces < FirstTabWidth || Spaces == 1) { 1392 Text.append(Spaces, ' '); 1393 break; 1394 } 1395 // Align to the next tab. 1396 Spaces -= FirstTabWidth; 1397 Text.append("\t"); 1398 1399 Text.append(Spaces / Style.TabWidth, '\t'); 1400 Text.append(Spaces % Style.TabWidth, ' '); 1401 } else if (Spaces == 1) { 1402 Text.append(Spaces, ' '); 1403 } 1404 break; 1405 } 1406 case FormatStyle::UT_ForIndentation: 1407 if (WhitespaceStartColumn == 0) { 1408 unsigned Indentation = IndentLevel * Style.IndentWidth; 1409 Spaces = appendTabIndent(Text, Spaces, Indentation); 1410 } 1411 Text.append(Spaces, ' '); 1412 break; 1413 case FormatStyle::UT_ForContinuationAndIndentation: 1414 if (WhitespaceStartColumn == 0) 1415 Spaces = appendTabIndent(Text, Spaces, Spaces); 1416 Text.append(Spaces, ' '); 1417 break; 1418 case FormatStyle::UT_AlignWithSpaces: 1419 if (WhitespaceStartColumn == 0) { 1420 unsigned Indentation = 1421 IsAligned ? IndentLevel * Style.IndentWidth : Spaces; 1422 Spaces = appendTabIndent(Text, Spaces, Indentation); 1423 } 1424 Text.append(Spaces, ' '); 1425 break; 1426 } 1427 } 1428 1429 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces, 1430 unsigned Indentation) { 1431 // This happens, e.g. when a line in a block comment is indented less than the 1432 // first one. 1433 if (Indentation > Spaces) 1434 Indentation = Spaces; 1435 if (Style.TabWidth) { 1436 unsigned Tabs = Indentation / Style.TabWidth; 1437 Text.append(Tabs, '\t'); 1438 Spaces -= Tabs * Style.TabWidth; 1439 } 1440 return Spaces; 1441 } 1442 1443 } // namespace format 1444 } // namespace clang 1445