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 // Column - The token for which Matches returns true is moved to this column. 271 // RightJustify - Whether it is the token's right end or left end that gets 272 // moved to that column. 273 template <typename F> 274 static void 275 AlignTokenSequence(const FormatStyle &Style, unsigned Start, unsigned End, 276 unsigned Column, bool RightJustify, F &&Matches, 277 SmallVector<WhitespaceManager::Change, 16> &Changes) { 278 bool FoundMatchOnLine = false; 279 int Shift = 0; 280 281 // ScopeStack keeps track of the current scope depth. It contains indices of 282 // the first token on each scope. 283 // We only run the "Matches" function on tokens from the outer-most scope. 284 // However, we do need to pay special attention to one class of tokens 285 // that are not in the outer-most scope, and that is function parameters 286 // which are split across multiple lines, as illustrated by this example: 287 // double a(int x); 288 // int b(int y, 289 // double z); 290 // In the above example, we need to take special care to ensure that 291 // 'double z' is indented along with it's owning function 'b'. 292 // The same holds for calling a function: 293 // double a = foo(x); 294 // int b = bar(foo(y), 295 // foor(z)); 296 // Similar for broken string literals: 297 // double x = 3.14; 298 // auto s = "Hello" 299 // "World"; 300 // Special handling is required for 'nested' ternary operators. 301 SmallVector<unsigned, 16> ScopeStack; 302 303 for (unsigned i = Start; i != End; ++i) { 304 if (ScopeStack.size() != 0 && 305 Changes[i].indentAndNestingLevel() < 306 Changes[ScopeStack.back()].indentAndNestingLevel()) 307 ScopeStack.pop_back(); 308 309 // Compare current token to previous non-comment token to ensure whether 310 // it is in a deeper scope or not. 311 unsigned PreviousNonComment = i - 1; 312 while (PreviousNonComment > Start && 313 Changes[PreviousNonComment].Tok->is(tok::comment)) 314 --PreviousNonComment; 315 if (i != Start && Changes[i].indentAndNestingLevel() > 316 Changes[PreviousNonComment].indentAndNestingLevel()) 317 ScopeStack.push_back(i); 318 319 bool InsideNestedScope = ScopeStack.size() != 0; 320 bool ContinuedStringLiteral = i > Start && 321 Changes[i].Tok->is(tok::string_literal) && 322 Changes[i - 1].Tok->is(tok::string_literal); 323 bool SkipMatchCheck = InsideNestedScope || ContinuedStringLiteral; 324 325 if (Changes[i].NewlinesBefore > 0 && !SkipMatchCheck) { 326 Shift = 0; 327 FoundMatchOnLine = false; 328 } 329 330 // If this is the first matching token to be aligned, remember by how many 331 // spaces it has to be shifted, so the rest of the changes on the line are 332 // shifted by the same amount 333 if (!FoundMatchOnLine && !SkipMatchCheck && Matches(Changes[i])) { 334 FoundMatchOnLine = true; 335 Shift = Column - (RightJustify ? Changes[i].TokenLength : 0) - 336 Changes[i].StartOfTokenColumn; 337 Changes[i].Spaces += Shift; 338 // FIXME: This is a workaround that should be removed when we fix 339 // http://llvm.org/PR53699. An assertion later below verifies this. 340 if (Changes[i].NewlinesBefore == 0) 341 Changes[i].Spaces = 342 std::max(Changes[i].Spaces, 343 static_cast<int>(Changes[i].Tok->SpacesRequiredBefore)); 344 } 345 346 // This is for function parameters that are split across multiple lines, 347 // as mentioned in the ScopeStack comment. 348 if (InsideNestedScope && Changes[i].NewlinesBefore > 0) { 349 unsigned ScopeStart = ScopeStack.back(); 350 auto ShouldShiftBeAdded = [&] { 351 // Function declaration 352 if (Changes[ScopeStart - 1].Tok->is(TT_FunctionDeclarationName)) 353 return true; 354 355 // Lambda. 356 if (Changes[ScopeStart - 1].Tok->is(TT_LambdaLBrace)) 357 return false; 358 359 // Continued function declaration 360 if (ScopeStart > Start + 1 && 361 Changes[ScopeStart - 2].Tok->is(TT_FunctionDeclarationName)) 362 return true; 363 364 // Continued function call 365 if (ScopeStart > Start + 1 && 366 Changes[ScopeStart - 2].Tok->is(tok::identifier) && 367 Changes[ScopeStart - 1].Tok->is(tok::l_paren) && 368 Changes[ScopeStart].Tok->isNot(TT_LambdaLSquare)) { 369 if (Changes[i].Tok->MatchingParen && 370 Changes[i].Tok->MatchingParen->is(TT_LambdaLBrace)) 371 return false; 372 if (Changes[ScopeStart].NewlinesBefore > 0) 373 return false; 374 return Style.BinPackArguments; 375 } 376 377 // Ternary operator 378 if (Changes[i].Tok->is(TT_ConditionalExpr)) 379 return true; 380 381 // Period Initializer .XXX = 1. 382 if (Changes[i].Tok->is(TT_DesignatedInitializerPeriod)) 383 return true; 384 385 // Continued ternary operator 386 if (Changes[i].Tok->Previous && 387 Changes[i].Tok->Previous->is(TT_ConditionalExpr)) 388 return true; 389 390 // Continued braced list. 391 if (ScopeStart > Start + 1 && 392 Changes[ScopeStart - 2].Tok->isNot(tok::identifier) && 393 Changes[ScopeStart - 1].Tok->is(tok::l_brace) && 394 Changes[i].Tok->isNot(tok::r_brace)) { 395 for (unsigned OuterScopeStart : llvm::reverse(ScopeStack)) { 396 // Lambda. 397 if (OuterScopeStart > Start && 398 Changes[OuterScopeStart - 1].Tok->is(TT_LambdaLBrace)) 399 return false; 400 } 401 if (Changes[ScopeStart].NewlinesBefore > 0) 402 return false; 403 return true; 404 } 405 406 return false; 407 }; 408 409 if (ShouldShiftBeAdded()) 410 Changes[i].Spaces += Shift; 411 } 412 413 if (ContinuedStringLiteral) 414 Changes[i].Spaces += Shift; 415 416 // We should not remove required spaces unless we break the line before. 417 assert(Shift >= 0 || Changes[i].NewlinesBefore > 0 || 418 Changes[i].Spaces >= 419 static_cast<int>(Changes[i].Tok->SpacesRequiredBefore) || 420 Changes[i].Tok->is(tok::eof)); 421 422 Changes[i].StartOfTokenColumn += Shift; 423 if (i + 1 != Changes.size()) 424 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 425 426 // If PointerAlignment is PAS_Right, keep *s or &s next to the token 427 if (Style.PointerAlignment == FormatStyle::PAS_Right && 428 Changes[i].Spaces != 0) { 429 for (int Previous = i - 1; 430 Previous >= 0 && 431 Changes[Previous].Tok->getType() == TT_PointerOrReference; 432 --Previous) { 433 Changes[Previous + 1].Spaces -= Shift; 434 Changes[Previous].Spaces += Shift; 435 } 436 } 437 } 438 } 439 440 // Walk through a subset of the changes, starting at StartAt, and find 441 // sequences of matching tokens to align. To do so, keep track of the lines and 442 // whether or not a matching token was found on a line. If a matching token is 443 // found, extend the current sequence. If the current line cannot be part of a 444 // sequence, e.g. because there is an empty line before it or it contains only 445 // non-matching tokens, finalize the previous sequence. 446 // The value returned is the token on which we stopped, either because we 447 // exhausted all items inside Changes, or because we hit a scope level higher 448 // than our initial scope. 449 // This function is recursive. Each invocation processes only the scope level 450 // equal to the initial level, which is the level of Changes[StartAt]. 451 // If we encounter a scope level greater than the initial level, then we call 452 // ourselves recursively, thereby avoiding the pollution of the current state 453 // with the alignment requirements of the nested sub-level. This recursive 454 // behavior is necessary for aligning function prototypes that have one or more 455 // arguments. 456 // If this function encounters a scope level less than the initial level, 457 // it returns the current position. 458 // There is a non-obvious subtlety in the recursive behavior: Even though we 459 // defer processing of nested levels to recursive invocations of this 460 // function, when it comes time to align a sequence of tokens, we run the 461 // alignment on the entire sequence, including the nested levels. 462 // When doing so, most of the nested tokens are skipped, because their 463 // alignment was already handled by the recursive invocations of this function. 464 // However, the special exception is that we do NOT skip function parameters 465 // that are split across multiple lines. See the test case in FormatTest.cpp 466 // that mentions "split function parameter alignment" for an example of this. 467 // When the parameter RightJustify is true, the operator will be 468 // right-justified. It is used to align compound assignments like `+=` and `=`. 469 // When RightJustify and ACS.PadOperators are true, operators in each block to 470 // be aligned will be padded on the left to the same length before aligning. 471 template <typename F> 472 static unsigned AlignTokens(const FormatStyle &Style, F &&Matches, 473 SmallVector<WhitespaceManager::Change, 16> &Changes, 474 unsigned StartAt, 475 const FormatStyle::AlignConsecutiveStyle &ACS = {}, 476 bool RightJustify = false) { 477 // We arrange each line in 3 parts. The operator to be aligned (the anchor), 478 // and text to its left and right. In the aligned text the width of each part 479 // will be the maximum of that over the block that has been aligned. Maximum 480 // widths of each part so far. When RightJustify is true and ACS.PadOperators 481 // is false, the part from start of line to the right end of the anchor. 482 // Otherwise, only the part to the left of the anchor. Including the space 483 // that exists on its left from the start. Not including the padding added on 484 // the left to right-justify the anchor. 485 unsigned WidthLeft = 0; 486 // The operator to be aligned when RightJustify is true and ACS.PadOperators 487 // is false. 0 otherwise. 488 unsigned WidthAnchor = 0; 489 // Width to the right of the anchor. Plus width of the anchor when 490 // RightJustify is false. 491 unsigned WidthRight = 0; 492 493 // Line number of the start and the end of the current token sequence. 494 unsigned StartOfSequence = 0; 495 unsigned EndOfSequence = 0; 496 497 // Measure the scope level (i.e. depth of (), [], {}) of the first token, and 498 // abort when we hit any token in a higher scope than the starting one. 499 auto IndentAndNestingLevel = StartAt < Changes.size() 500 ? Changes[StartAt].indentAndNestingLevel() 501 : std::tuple<unsigned, unsigned, unsigned>(); 502 503 // Keep track of the number of commas before the matching tokens, we will only 504 // align a sequence of matching tokens if they are preceded by the same number 505 // of commas. 506 unsigned CommasBeforeLastMatch = 0; 507 unsigned CommasBeforeMatch = 0; 508 509 // Whether a matching token has been found on the current line. 510 bool FoundMatchOnLine = false; 511 512 // Whether the current line consists purely of comments. 513 bool LineIsComment = true; 514 515 // Aligns a sequence of matching tokens, on the MinColumn column. 516 // 517 // Sequences start from the first matching token to align, and end at the 518 // first token of the first line that doesn't need to be aligned. 519 // 520 // We need to adjust the StartOfTokenColumn of each Change that is on a line 521 // containing any matching token to be aligned and located after such token. 522 auto AlignCurrentSequence = [&] { 523 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) 524 AlignTokenSequence(Style, StartOfSequence, EndOfSequence, 525 WidthLeft + WidthAnchor, RightJustify, Matches, 526 Changes); 527 WidthLeft = 0; 528 WidthAnchor = 0; 529 WidthRight = 0; 530 StartOfSequence = 0; 531 EndOfSequence = 0; 532 }; 533 534 unsigned i = StartAt; 535 for (unsigned e = Changes.size(); i != e; ++i) { 536 if (Changes[i].indentAndNestingLevel() < IndentAndNestingLevel) 537 break; 538 539 if (Changes[i].NewlinesBefore != 0) { 540 CommasBeforeMatch = 0; 541 EndOfSequence = i; 542 543 // Whether to break the alignment sequence because of an empty line. 544 bool EmptyLineBreak = 545 (Changes[i].NewlinesBefore > 1) && !ACS.AcrossEmptyLines; 546 547 // Whether to break the alignment sequence because of a line without a 548 // match. 549 bool NoMatchBreak = 550 !FoundMatchOnLine && !(LineIsComment && ACS.AcrossComments); 551 552 if (EmptyLineBreak || NoMatchBreak) 553 AlignCurrentSequence(); 554 555 // A new line starts, re-initialize line status tracking bools. 556 // Keep the match state if a string literal is continued on this line. 557 if (i == 0 || !Changes[i].Tok->is(tok::string_literal) || 558 !Changes[i - 1].Tok->is(tok::string_literal)) 559 FoundMatchOnLine = false; 560 LineIsComment = true; 561 } 562 563 if (!Changes[i].Tok->is(tok::comment)) 564 LineIsComment = false; 565 566 if (Changes[i].Tok->is(tok::comma)) { 567 ++CommasBeforeMatch; 568 } else if (Changes[i].indentAndNestingLevel() > IndentAndNestingLevel) { 569 // Call AlignTokens recursively, skipping over this scope block. 570 unsigned StoppedAt = AlignTokens(Style, Matches, Changes, i, ACS); 571 i = StoppedAt - 1; 572 continue; 573 } 574 575 if (!Matches(Changes[i])) 576 continue; 577 578 // If there is more than one matching token per line, or if the number of 579 // preceding commas, do not match anymore, end the sequence. 580 if (FoundMatchOnLine || CommasBeforeMatch != CommasBeforeLastMatch) 581 AlignCurrentSequence(); 582 583 CommasBeforeLastMatch = CommasBeforeMatch; 584 FoundMatchOnLine = true; 585 586 if (StartOfSequence == 0) 587 StartOfSequence = i; 588 589 unsigned ChangeWidthLeft = Changes[i].StartOfTokenColumn; 590 unsigned ChangeWidthAnchor = 0; 591 unsigned ChangeWidthRight = 0; 592 if (RightJustify) { 593 if (ACS.PadOperators) 594 ChangeWidthAnchor = Changes[i].TokenLength; 595 else 596 ChangeWidthLeft += Changes[i].TokenLength; 597 } else 598 ChangeWidthRight = Changes[i].TokenLength; 599 for (unsigned j = i + 1; j != e && Changes[j].NewlinesBefore == 0; ++j) { 600 ChangeWidthRight += Changes[j].Spaces; 601 // Changes are generally 1:1 with the tokens, but a change could also be 602 // inside of a token, in which case it's counted more than once: once for 603 // the whitespace surrounding the token (!IsInsideToken) and once for 604 // each whitespace change within it (IsInsideToken). 605 // Therefore, changes inside of a token should only count the space. 606 if (!Changes[j].IsInsideToken) 607 ChangeWidthRight += Changes[j].TokenLength; 608 } 609 610 // If we are restricted by the maximum column width, end the sequence. 611 unsigned NewLeft = std::max(ChangeWidthLeft, WidthLeft); 612 unsigned NewAnchor = std::max(ChangeWidthAnchor, WidthAnchor); 613 unsigned NewRight = std::max(ChangeWidthRight, WidthRight); 614 // `ColumnLimit == 0` means there is no column limit. 615 if (Style.ColumnLimit != 0 && 616 Style.ColumnLimit < NewLeft + NewAnchor + NewRight) { 617 AlignCurrentSequence(); 618 StartOfSequence = i; 619 WidthLeft = ChangeWidthLeft; 620 WidthAnchor = ChangeWidthAnchor; 621 WidthRight = ChangeWidthRight; 622 } else { 623 WidthLeft = NewLeft; 624 WidthAnchor = NewAnchor; 625 WidthRight = NewRight; 626 } 627 } 628 629 EndOfSequence = i; 630 AlignCurrentSequence(); 631 return i; 632 } 633 634 // Aligns a sequence of matching tokens, on the MinColumn column. 635 // 636 // Sequences start from the first matching token to align, and end at the 637 // first token of the first line that doesn't need to be aligned. 638 // 639 // We need to adjust the StartOfTokenColumn of each Change that is on a line 640 // containing any matching token to be aligned and located after such token. 641 static void AlignMacroSequence( 642 unsigned &StartOfSequence, unsigned &EndOfSequence, unsigned &MinColumn, 643 unsigned &MaxColumn, bool &FoundMatchOnLine, 644 std::function<bool(const WhitespaceManager::Change &C)> AlignMacrosMatches, 645 SmallVector<WhitespaceManager::Change, 16> &Changes) { 646 if (StartOfSequence > 0 && StartOfSequence < EndOfSequence) { 647 648 FoundMatchOnLine = false; 649 int Shift = 0; 650 651 for (unsigned I = StartOfSequence; I != EndOfSequence; ++I) { 652 if (Changes[I].NewlinesBefore > 0) { 653 Shift = 0; 654 FoundMatchOnLine = false; 655 } 656 657 // If this is the first matching token to be aligned, remember by how many 658 // spaces it has to be shifted, so the rest of the changes on the line are 659 // shifted by the same amount 660 if (!FoundMatchOnLine && AlignMacrosMatches(Changes[I])) { 661 FoundMatchOnLine = true; 662 Shift = MinColumn - Changes[I].StartOfTokenColumn; 663 Changes[I].Spaces += Shift; 664 } 665 666 assert(Shift >= 0); 667 Changes[I].StartOfTokenColumn += Shift; 668 if (I + 1 != Changes.size()) 669 Changes[I + 1].PreviousEndOfTokenColumn += Shift; 670 } 671 } 672 673 MinColumn = 0; 674 MaxColumn = UINT_MAX; 675 StartOfSequence = 0; 676 EndOfSequence = 0; 677 } 678 679 void WhitespaceManager::alignConsecutiveMacros() { 680 if (!Style.AlignConsecutiveMacros.Enabled) 681 return; 682 683 auto AlignMacrosMatches = [](const Change &C) { 684 const FormatToken *Current = C.Tok; 685 unsigned SpacesRequiredBefore = 1; 686 687 if (Current->SpacesRequiredBefore == 0 || !Current->Previous) 688 return false; 689 690 Current = Current->Previous; 691 692 // If token is a ")", skip over the parameter list, to the 693 // token that precedes the "(" 694 if (Current->is(tok::r_paren) && Current->MatchingParen) { 695 Current = Current->MatchingParen->Previous; 696 SpacesRequiredBefore = 0; 697 } 698 699 if (!Current || !Current->is(tok::identifier)) 700 return false; 701 702 if (!Current->Previous || !Current->Previous->is(tok::pp_define)) 703 return false; 704 705 // For a macro function, 0 spaces are required between the 706 // identifier and the lparen that opens the parameter list. 707 // For a simple macro, 1 space is required between the 708 // identifier and the first token of the defined value. 709 return Current->Next->SpacesRequiredBefore == SpacesRequiredBefore; 710 }; 711 712 unsigned MinColumn = 0; 713 unsigned MaxColumn = UINT_MAX; 714 715 // Start and end of the token sequence we're processing. 716 unsigned StartOfSequence = 0; 717 unsigned EndOfSequence = 0; 718 719 // Whether a matching token has been found on the current line. 720 bool FoundMatchOnLine = false; 721 722 // Whether the current line consists only of comments 723 bool LineIsComment = true; 724 725 unsigned I = 0; 726 for (unsigned E = Changes.size(); I != E; ++I) { 727 if (Changes[I].NewlinesBefore != 0) { 728 EndOfSequence = I; 729 730 // Whether to break the alignment sequence because of an empty line. 731 bool EmptyLineBreak = (Changes[I].NewlinesBefore > 1) && 732 !Style.AlignConsecutiveMacros.AcrossEmptyLines; 733 734 // Whether to break the alignment sequence because of a line without a 735 // match. 736 bool NoMatchBreak = 737 !FoundMatchOnLine && 738 !(LineIsComment && Style.AlignConsecutiveMacros.AcrossComments); 739 740 if (EmptyLineBreak || NoMatchBreak) 741 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 742 FoundMatchOnLine, AlignMacrosMatches, Changes); 743 744 // A new line starts, re-initialize line status tracking bools. 745 FoundMatchOnLine = false; 746 LineIsComment = true; 747 } 748 749 if (!Changes[I].Tok->is(tok::comment)) 750 LineIsComment = false; 751 752 if (!AlignMacrosMatches(Changes[I])) 753 continue; 754 755 FoundMatchOnLine = true; 756 757 if (StartOfSequence == 0) 758 StartOfSequence = I; 759 760 unsigned ChangeMinColumn = Changes[I].StartOfTokenColumn; 761 int LineLengthAfter = -Changes[I].Spaces; 762 for (unsigned j = I; j != E && Changes[j].NewlinesBefore == 0; ++j) 763 LineLengthAfter += Changes[j].Spaces + Changes[j].TokenLength; 764 unsigned ChangeMaxColumn = Style.ColumnLimit - LineLengthAfter; 765 766 MinColumn = std::max(MinColumn, ChangeMinColumn); 767 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 768 } 769 770 EndOfSequence = I; 771 AlignMacroSequence(StartOfSequence, EndOfSequence, MinColumn, MaxColumn, 772 FoundMatchOnLine, AlignMacrosMatches, Changes); 773 } 774 775 void WhitespaceManager::alignConsecutiveAssignments() { 776 if (!Style.AlignConsecutiveAssignments.Enabled) 777 return; 778 779 AlignTokens( 780 Style, 781 [&](const Change &C) { 782 // Do not align on equal signs that are first on a line. 783 if (C.NewlinesBefore > 0) 784 return false; 785 786 // Do not align on equal signs that are last on a line. 787 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 788 return false; 789 790 // Do not align operator= overloads. 791 FormatToken *Previous = C.Tok->getPreviousNonComment(); 792 if (Previous && Previous->is(tok::kw_operator)) 793 return false; 794 795 return Style.AlignConsecutiveAssignments.AlignCompound 796 ? C.Tok->getPrecedence() == prec::Assignment 797 : C.Tok->is(tok::equal); 798 }, 799 Changes, /*StartAt=*/0, Style.AlignConsecutiveAssignments, 800 /*RightJustify=*/true); 801 } 802 803 void WhitespaceManager::alignConsecutiveBitFields() { 804 if (!Style.AlignConsecutiveBitFields.Enabled) 805 return; 806 807 AlignTokens( 808 Style, 809 [&](Change const &C) { 810 // Do not align on ':' that is first on a line. 811 if (C.NewlinesBefore > 0) 812 return false; 813 814 // Do not align on ':' that is last on a line. 815 if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0) 816 return false; 817 818 return C.Tok->is(TT_BitFieldColon); 819 }, 820 Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields); 821 } 822 823 void WhitespaceManager::alignConsecutiveDeclarations() { 824 if (!Style.AlignConsecutiveDeclarations.Enabled) 825 return; 826 827 AlignTokens( 828 Style, 829 [](Change const &C) { 830 // tok::kw_operator is necessary for aligning operator overload 831 // definitions. 832 if (C.Tok->isOneOf(TT_FunctionDeclarationName, tok::kw_operator)) 833 return true; 834 if (C.Tok->isNot(TT_StartOfName)) 835 return false; 836 if (C.Tok->Previous && 837 C.Tok->Previous->is(TT_StatementAttributeLikeMacro)) 838 return false; 839 // Check if there is a subsequent name that starts the same declaration. 840 for (FormatToken *Next = C.Tok->Next; Next; Next = Next->Next) { 841 if (Next->is(tok::comment)) 842 continue; 843 if (Next->is(TT_PointerOrReference)) 844 return false; 845 if (!Next->Tok.getIdentifierInfo()) 846 break; 847 if (Next->isOneOf(TT_StartOfName, TT_FunctionDeclarationName, 848 tok::kw_operator)) 849 return false; 850 } 851 return true; 852 }, 853 Changes, /*StartAt=*/0, Style.AlignConsecutiveDeclarations); 854 } 855 856 void WhitespaceManager::alignChainedConditionals() { 857 if (Style.BreakBeforeTernaryOperators) { 858 AlignTokens( 859 Style, 860 [](Change const &C) { 861 // Align question operators and last colon 862 return C.Tok->is(TT_ConditionalExpr) && 863 ((C.Tok->is(tok::question) && !C.NewlinesBefore) || 864 (C.Tok->is(tok::colon) && C.Tok->Next && 865 (C.Tok->Next->FakeLParens.size() == 0 || 866 C.Tok->Next->FakeLParens.back() != prec::Conditional))); 867 }, 868 Changes, /*StartAt=*/0); 869 } else { 870 static auto AlignWrappedOperand = [](Change const &C) { 871 FormatToken *Previous = C.Tok->getPreviousNonComment(); 872 return C.NewlinesBefore && Previous && Previous->is(TT_ConditionalExpr) && 873 (Previous->is(tok::colon) && 874 (C.Tok->FakeLParens.size() == 0 || 875 C.Tok->FakeLParens.back() != prec::Conditional)); 876 }; 877 // Ensure we keep alignment of wrapped operands with non-wrapped operands 878 // Since we actually align the operators, the wrapped operands need the 879 // extra offset to be properly aligned. 880 for (Change &C : Changes) 881 if (AlignWrappedOperand(C)) 882 C.StartOfTokenColumn -= 2; 883 AlignTokens( 884 Style, 885 [this](Change const &C) { 886 // Align question operators if next operand is not wrapped, as 887 // well as wrapped operands after question operator or last 888 // colon in conditional sequence 889 return (C.Tok->is(TT_ConditionalExpr) && C.Tok->is(tok::question) && 890 &C != &Changes.back() && (&C + 1)->NewlinesBefore == 0 && 891 !(&C + 1)->IsTrailingComment) || 892 AlignWrappedOperand(C); 893 }, 894 Changes, /*StartAt=*/0); 895 } 896 } 897 898 void WhitespaceManager::alignTrailingComments() { 899 unsigned MinColumn = 0; 900 unsigned MaxColumn = UINT_MAX; 901 unsigned StartOfSequence = 0; 902 bool BreakBeforeNext = false; 903 unsigned Newlines = 0; 904 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 905 if (Changes[i].StartOfBlockComment) 906 continue; 907 Newlines += Changes[i].NewlinesBefore; 908 if (!Changes[i].IsTrailingComment) 909 continue; 910 911 unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn; 912 unsigned ChangeMaxColumn; 913 914 if (Style.ColumnLimit == 0) 915 ChangeMaxColumn = UINT_MAX; 916 else if (Style.ColumnLimit >= Changes[i].TokenLength) 917 ChangeMaxColumn = Style.ColumnLimit - Changes[i].TokenLength; 918 else 919 ChangeMaxColumn = ChangeMinColumn; 920 921 // If we don't create a replacement for this change, we have to consider 922 // it to be immovable. 923 if (!Changes[i].CreateReplacement) 924 ChangeMaxColumn = ChangeMinColumn; 925 926 if (i + 1 != e && Changes[i + 1].ContinuesPPDirective) 927 ChangeMaxColumn -= 2; 928 // If this comment follows an } in column 0, it probably documents the 929 // closing of a namespace and we don't want to align it. 930 bool FollowsRBraceInColumn0 = i > 0 && Changes[i].NewlinesBefore == 0 && 931 Changes[i - 1].Tok->is(tok::r_brace) && 932 Changes[i - 1].StartOfTokenColumn == 0; 933 bool WasAlignedWithStartOfNextLine = false; 934 if (Changes[i].NewlinesBefore == 1) { // A comment on its own line. 935 unsigned CommentColumn = SourceMgr.getSpellingColumnNumber( 936 Changes[i].OriginalWhitespaceRange.getEnd()); 937 for (unsigned j = i + 1; j != e; ++j) { 938 if (Changes[j].Tok->is(tok::comment)) 939 continue; 940 941 unsigned NextColumn = SourceMgr.getSpellingColumnNumber( 942 Changes[j].OriginalWhitespaceRange.getEnd()); 943 // The start of the next token was previously aligned with the 944 // start of this comment. 945 WasAlignedWithStartOfNextLine = 946 CommentColumn == NextColumn || 947 CommentColumn == NextColumn + Style.IndentWidth; 948 break; 949 } 950 } 951 if (!Style.AlignTrailingComments || FollowsRBraceInColumn0) { 952 alignTrailingComments(StartOfSequence, i, MinColumn); 953 MinColumn = ChangeMinColumn; 954 MaxColumn = ChangeMinColumn; 955 StartOfSequence = i; 956 } else if (BreakBeforeNext || Newlines > 1 || 957 (ChangeMinColumn > MaxColumn || ChangeMaxColumn < MinColumn) || 958 // Break the comment sequence if the previous line did not end 959 // in a trailing comment. 960 (Changes[i].NewlinesBefore == 1 && i > 0 && 961 !Changes[i - 1].IsTrailingComment) || 962 WasAlignedWithStartOfNextLine) { 963 alignTrailingComments(StartOfSequence, i, MinColumn); 964 MinColumn = ChangeMinColumn; 965 MaxColumn = ChangeMaxColumn; 966 StartOfSequence = i; 967 } else { 968 MinColumn = std::max(MinColumn, ChangeMinColumn); 969 MaxColumn = std::min(MaxColumn, ChangeMaxColumn); 970 } 971 BreakBeforeNext = (i == 0) || (Changes[i].NewlinesBefore > 1) || 972 // Never start a sequence with a comment at the beginning 973 // of the line. 974 (Changes[i].NewlinesBefore == 1 && StartOfSequence == i); 975 Newlines = 0; 976 } 977 alignTrailingComments(StartOfSequence, Changes.size(), MinColumn); 978 } 979 980 void WhitespaceManager::alignTrailingComments(unsigned Start, unsigned End, 981 unsigned Column) { 982 for (unsigned i = Start; i != End; ++i) { 983 int Shift = 0; 984 if (Changes[i].IsTrailingComment) 985 Shift = Column - Changes[i].StartOfTokenColumn; 986 if (Changes[i].StartOfBlockComment) { 987 Shift = Changes[i].IndentationOffset + 988 Changes[i].StartOfBlockComment->StartOfTokenColumn - 989 Changes[i].StartOfTokenColumn; 990 } 991 if (Shift < 0) 992 continue; 993 Changes[i].Spaces += Shift; 994 if (i + 1 != Changes.size()) 995 Changes[i + 1].PreviousEndOfTokenColumn += Shift; 996 Changes[i].StartOfTokenColumn += Shift; 997 } 998 } 999 1000 void WhitespaceManager::alignEscapedNewlines() { 1001 if (Style.AlignEscapedNewlines == FormatStyle::ENAS_DontAlign) 1002 return; 1003 1004 bool AlignLeft = Style.AlignEscapedNewlines == FormatStyle::ENAS_Left; 1005 unsigned MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 1006 unsigned StartOfMacro = 0; 1007 for (unsigned i = 1, e = Changes.size(); i < e; ++i) { 1008 Change &C = Changes[i]; 1009 if (C.NewlinesBefore > 0) { 1010 if (C.ContinuesPPDirective) { 1011 MaxEndOfLine = std::max(C.PreviousEndOfTokenColumn + 2, MaxEndOfLine); 1012 } else { 1013 alignEscapedNewlines(StartOfMacro + 1, i, MaxEndOfLine); 1014 MaxEndOfLine = AlignLeft ? 0 : Style.ColumnLimit; 1015 StartOfMacro = i; 1016 } 1017 } 1018 } 1019 alignEscapedNewlines(StartOfMacro + 1, Changes.size(), MaxEndOfLine); 1020 } 1021 1022 void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End, 1023 unsigned Column) { 1024 for (unsigned i = Start; i < End; ++i) { 1025 Change &C = Changes[i]; 1026 if (C.NewlinesBefore > 0) { 1027 assert(C.ContinuesPPDirective); 1028 if (C.PreviousEndOfTokenColumn + 1 > Column) 1029 C.EscapedNewlineColumn = 0; 1030 else 1031 C.EscapedNewlineColumn = Column; 1032 } 1033 } 1034 } 1035 1036 void WhitespaceManager::alignArrayInitializers() { 1037 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_None) 1038 return; 1039 1040 for (unsigned ChangeIndex = 1U, ChangeEnd = Changes.size(); 1041 ChangeIndex < ChangeEnd; ++ChangeIndex) { 1042 auto &C = Changes[ChangeIndex]; 1043 if (C.Tok->IsArrayInitializer) { 1044 bool FoundComplete = false; 1045 for (unsigned InsideIndex = ChangeIndex + 1; InsideIndex < ChangeEnd; 1046 ++InsideIndex) { 1047 if (Changes[InsideIndex].Tok == C.Tok->MatchingParen) { 1048 alignArrayInitializers(ChangeIndex, InsideIndex + 1); 1049 ChangeIndex = InsideIndex + 1; 1050 FoundComplete = true; 1051 break; 1052 } 1053 } 1054 if (!FoundComplete) 1055 ChangeIndex = ChangeEnd; 1056 } 1057 } 1058 } 1059 1060 void WhitespaceManager::alignArrayInitializers(unsigned Start, unsigned End) { 1061 1062 if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Right) 1063 alignArrayInitializersRightJustified(getCells(Start, End)); 1064 else if (Style.AlignArrayOfStructures == FormatStyle::AIAS_Left) 1065 alignArrayInitializersLeftJustified(getCells(Start, End)); 1066 } 1067 1068 void WhitespaceManager::alignArrayInitializersRightJustified( 1069 CellDescriptions &&CellDescs) { 1070 if (!CellDescs.isRectangular()) 1071 return; 1072 1073 auto &Cells = CellDescs.Cells; 1074 // Now go through and fixup the spaces. 1075 auto *CellIter = Cells.begin(); 1076 for (auto i = 0U; i < CellDescs.CellCounts[0]; ++i, ++CellIter) { 1077 unsigned NetWidth = 0U; 1078 if (isSplitCell(*CellIter)) 1079 NetWidth = getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1080 auto CellWidth = getMaximumCellWidth(CellIter, NetWidth); 1081 1082 if (Changes[CellIter->Index].Tok->is(tok::r_brace)) { 1083 // So in here we want to see if there is a brace that falls 1084 // on a line that was split. If so on that line we make sure that 1085 // the spaces in front of the brace are enough. 1086 const auto *Next = CellIter; 1087 do { 1088 const FormatToken *Previous = Changes[Next->Index].Tok->Previous; 1089 if (Previous && Previous->isNot(TT_LineComment)) { 1090 Changes[Next->Index].Spaces = 0; 1091 Changes[Next->Index].NewlinesBefore = 0; 1092 } 1093 Next = Next->NextColumnElement; 1094 } while (Next); 1095 // Unless the array is empty, we need the position of all the 1096 // immediately adjacent cells 1097 if (CellIter != Cells.begin()) { 1098 auto ThisNetWidth = 1099 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1100 auto MaxNetWidth = getMaximumNetWidth( 1101 Cells.begin(), CellIter, CellDescs.InitialSpaces, 1102 CellDescs.CellCounts[0], CellDescs.CellCounts.size()); 1103 if (ThisNetWidth < MaxNetWidth) 1104 Changes[CellIter->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1105 auto RowCount = 1U; 1106 auto Offset = std::distance(Cells.begin(), CellIter); 1107 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1108 Next = Next->NextColumnElement) { 1109 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]); 1110 auto *End = Start + Offset; 1111 ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1112 if (ThisNetWidth < MaxNetWidth) 1113 Changes[Next->Index].Spaces = (MaxNetWidth - ThisNetWidth); 1114 ++RowCount; 1115 } 1116 } 1117 } else { 1118 auto ThisWidth = 1119 calculateCellWidth(CellIter->Index, CellIter->EndIndex, true) + 1120 NetWidth; 1121 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1122 Changes[CellIter->Index].Spaces = (CellWidth - (ThisWidth + NetWidth)); 1123 Changes[CellIter->Index].Spaces += (i > 0) ? 1 : 0; 1124 } 1125 alignToStartOfCell(CellIter->Index, CellIter->EndIndex); 1126 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1127 Next = Next->NextColumnElement) { 1128 ThisWidth = 1129 calculateCellWidth(Next->Index, Next->EndIndex, true) + NetWidth; 1130 if (Changes[Next->Index].NewlinesBefore == 0) { 1131 Changes[Next->Index].Spaces = (CellWidth - ThisWidth); 1132 Changes[Next->Index].Spaces += (i > 0) ? 1 : 0; 1133 } 1134 alignToStartOfCell(Next->Index, Next->EndIndex); 1135 } 1136 } 1137 } 1138 } 1139 1140 void WhitespaceManager::alignArrayInitializersLeftJustified( 1141 CellDescriptions &&CellDescs) { 1142 1143 if (!CellDescs.isRectangular()) 1144 return; 1145 1146 auto &Cells = CellDescs.Cells; 1147 // Now go through and fixup the spaces. 1148 auto *CellIter = Cells.begin(); 1149 // The first cell needs to be against the left brace. 1150 if (Changes[CellIter->Index].NewlinesBefore == 0) 1151 Changes[CellIter->Index].Spaces = 0; 1152 else 1153 Changes[CellIter->Index].Spaces = CellDescs.InitialSpaces; 1154 ++CellIter; 1155 for (auto i = 1U; i < CellDescs.CellCounts[0]; i++, ++CellIter) { 1156 auto MaxNetWidth = getMaximumNetWidth( 1157 Cells.begin(), CellIter, CellDescs.InitialSpaces, 1158 CellDescs.CellCounts[0], CellDescs.CellCounts.size()); 1159 auto ThisNetWidth = 1160 getNetWidth(Cells.begin(), CellIter, CellDescs.InitialSpaces); 1161 if (Changes[CellIter->Index].NewlinesBefore == 0) { 1162 Changes[CellIter->Index].Spaces = 1163 MaxNetWidth - ThisNetWidth + 1164 (Changes[CellIter->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1165 } 1166 auto RowCount = 1U; 1167 auto Offset = std::distance(Cells.begin(), CellIter); 1168 for (const auto *Next = CellIter->NextColumnElement; Next != nullptr; 1169 Next = Next->NextColumnElement) { 1170 if (RowCount > CellDescs.CellCounts.size()) 1171 break; 1172 auto *Start = (Cells.begin() + RowCount * CellDescs.CellCounts[0]); 1173 auto *End = Start + Offset; 1174 auto ThisNetWidth = getNetWidth(Start, End, CellDescs.InitialSpaces); 1175 if (Changes[Next->Index].NewlinesBefore == 0) { 1176 Changes[Next->Index].Spaces = 1177 MaxNetWidth - ThisNetWidth + 1178 (Changes[Next->Index].Tok->isNot(tok::r_brace) ? 1 : 0); 1179 } 1180 ++RowCount; 1181 } 1182 } 1183 } 1184 1185 bool WhitespaceManager::isSplitCell(const CellDescription &Cell) { 1186 if (Cell.HasSplit) 1187 return true; 1188 for (const auto *Next = Cell.NextColumnElement; Next != nullptr; 1189 Next = Next->NextColumnElement) 1190 if (Next->HasSplit) 1191 return true; 1192 return false; 1193 } 1194 1195 WhitespaceManager::CellDescriptions WhitespaceManager::getCells(unsigned Start, 1196 unsigned End) { 1197 1198 unsigned Depth = 0; 1199 unsigned Cell = 0; 1200 SmallVector<unsigned> CellCounts; 1201 unsigned InitialSpaces = 0; 1202 unsigned InitialTokenLength = 0; 1203 unsigned EndSpaces = 0; 1204 SmallVector<CellDescription> Cells; 1205 const FormatToken *MatchingParen = nullptr; 1206 for (unsigned i = Start; i < End; ++i) { 1207 auto &C = Changes[i]; 1208 if (C.Tok->is(tok::l_brace)) 1209 ++Depth; 1210 else if (C.Tok->is(tok::r_brace)) 1211 --Depth; 1212 if (Depth == 2) { 1213 if (C.Tok->is(tok::l_brace)) { 1214 Cell = 0; 1215 MatchingParen = C.Tok->MatchingParen; 1216 if (InitialSpaces == 0) { 1217 InitialSpaces = C.Spaces + C.TokenLength; 1218 InitialTokenLength = C.TokenLength; 1219 auto j = i - 1; 1220 for (; Changes[j].NewlinesBefore == 0 && j > Start; --j) { 1221 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1222 InitialTokenLength += Changes[j].TokenLength; 1223 } 1224 if (C.NewlinesBefore == 0) { 1225 InitialSpaces += Changes[j].Spaces + Changes[j].TokenLength; 1226 InitialTokenLength += Changes[j].TokenLength; 1227 } 1228 } 1229 } else if (C.Tok->is(tok::comma)) { 1230 if (!Cells.empty()) 1231 Cells.back().EndIndex = i; 1232 if (C.Tok->getNextNonComment()->isNot(tok::r_brace)) // dangling comma 1233 ++Cell; 1234 } 1235 } else if (Depth == 1) { 1236 if (C.Tok == MatchingParen) { 1237 if (!Cells.empty()) 1238 Cells.back().EndIndex = i; 1239 Cells.push_back(CellDescription{i, ++Cell, i + 1, false, nullptr}); 1240 CellCounts.push_back(C.Tok->Previous->isNot(tok::comma) ? Cell + 1 1241 : Cell); 1242 // Go to the next non-comment and ensure there is a break in front 1243 const auto *NextNonComment = C.Tok->getNextNonComment(); 1244 while (NextNonComment->is(tok::comma)) 1245 NextNonComment = NextNonComment->getNextNonComment(); 1246 auto j = i; 1247 while (Changes[j].Tok != NextNonComment && j < End) 1248 ++j; 1249 if (j < End && Changes[j].NewlinesBefore == 0 && 1250 Changes[j].Tok->isNot(tok::r_brace)) { 1251 Changes[j].NewlinesBefore = 1; 1252 // Account for the added token lengths 1253 Changes[j].Spaces = InitialSpaces - InitialTokenLength; 1254 } 1255 } else if (C.Tok->is(tok::comment)) { 1256 // Trailing comments stay at a space past the last token 1257 C.Spaces = Changes[i - 1].Tok->is(tok::comma) ? 1 : 2; 1258 } else if (C.Tok->is(tok::l_brace)) { 1259 // We need to make sure that the ending braces is aligned to the 1260 // start of our initializer 1261 auto j = i - 1; 1262 for (; j > 0 && !Changes[j].Tok->ArrayInitializerLineStart; --j) 1263 ; // Nothing the loop does the work 1264 EndSpaces = Changes[j].Spaces; 1265 } 1266 } else if (Depth == 0 && C.Tok->is(tok::r_brace)) { 1267 C.NewlinesBefore = 1; 1268 C.Spaces = EndSpaces; 1269 } 1270 if (C.Tok->StartsColumn) { 1271 // This gets us past tokens that have been split over multiple 1272 // lines 1273 bool HasSplit = false; 1274 if (Changes[i].NewlinesBefore > 0) { 1275 // So if we split a line previously and the tail line + this token is 1276 // less then the column limit we remove the split here and just put 1277 // the column start at a space past the comma 1278 // 1279 // FIXME This if branch covers the cases where the column is not 1280 // the first column. This leads to weird pathologies like the formatting 1281 // auto foo = Items{ 1282 // Section{ 1283 // 0, bar(), 1284 // } 1285 // }; 1286 // Well if it doesn't lead to that it's indicative that the line 1287 // breaking should be revisited. Unfortunately alot of other options 1288 // interact with this 1289 auto j = i - 1; 1290 if ((j - 1) > Start && Changes[j].Tok->is(tok::comma) && 1291 Changes[j - 1].NewlinesBefore > 0) { 1292 --j; 1293 auto LineLimit = Changes[j].Spaces + Changes[j].TokenLength; 1294 if (LineLimit < Style.ColumnLimit) { 1295 Changes[i].NewlinesBefore = 0; 1296 Changes[i].Spaces = 1; 1297 } 1298 } 1299 } 1300 while (Changes[i].NewlinesBefore > 0 && Changes[i].Tok == C.Tok) { 1301 Changes[i].Spaces = InitialSpaces; 1302 ++i; 1303 HasSplit = true; 1304 } 1305 if (Changes[i].Tok != C.Tok) 1306 --i; 1307 Cells.push_back(CellDescription{i, Cell, i, HasSplit, nullptr}); 1308 } 1309 } 1310 1311 return linkCells({Cells, CellCounts, InitialSpaces}); 1312 } 1313 1314 unsigned WhitespaceManager::calculateCellWidth(unsigned Start, unsigned End, 1315 bool WithSpaces) const { 1316 unsigned CellWidth = 0; 1317 for (auto i = Start; i < End; i++) { 1318 if (Changes[i].NewlinesBefore > 0) 1319 CellWidth = 0; 1320 CellWidth += Changes[i].TokenLength; 1321 CellWidth += (WithSpaces ? Changes[i].Spaces : 0); 1322 } 1323 return CellWidth; 1324 } 1325 1326 void WhitespaceManager::alignToStartOfCell(unsigned Start, unsigned End) { 1327 if ((End - Start) <= 1) 1328 return; 1329 // If the line is broken anywhere in there make sure everything 1330 // is aligned to the parent 1331 for (auto i = Start + 1; i < End; i++) 1332 if (Changes[i].NewlinesBefore > 0) 1333 Changes[i].Spaces = Changes[Start].Spaces; 1334 } 1335 1336 WhitespaceManager::CellDescriptions 1337 WhitespaceManager::linkCells(CellDescriptions &&CellDesc) { 1338 auto &Cells = CellDesc.Cells; 1339 for (auto *CellIter = Cells.begin(); CellIter != Cells.end(); ++CellIter) { 1340 if (CellIter->NextColumnElement == nullptr && 1341 ((CellIter + 1) != Cells.end())) { 1342 for (auto *NextIter = CellIter + 1; NextIter != Cells.end(); ++NextIter) { 1343 if (NextIter->Cell == CellIter->Cell) { 1344 CellIter->NextColumnElement = &(*NextIter); 1345 break; 1346 } 1347 } 1348 } 1349 } 1350 return std::move(CellDesc); 1351 } 1352 1353 void WhitespaceManager::generateChanges() { 1354 for (unsigned i = 0, e = Changes.size(); i != e; ++i) { 1355 const Change &C = Changes[i]; 1356 if (i > 0 && Changes[i - 1].OriginalWhitespaceRange.getBegin() == 1357 C.OriginalWhitespaceRange.getBegin()) { 1358 // Do not generate two replacements for the same location. 1359 continue; 1360 } 1361 if (C.CreateReplacement) { 1362 std::string ReplacementText = C.PreviousLinePostfix; 1363 if (C.ContinuesPPDirective) 1364 appendEscapedNewlineText(ReplacementText, C.NewlinesBefore, 1365 C.PreviousEndOfTokenColumn, 1366 C.EscapedNewlineColumn); 1367 else 1368 appendNewlineText(ReplacementText, C.NewlinesBefore); 1369 // FIXME: This assert should hold if we computed the column correctly. 1370 // assert((int)C.StartOfTokenColumn >= C.Spaces); 1371 appendIndentText( 1372 ReplacementText, C.Tok->IndentLevel, std::max(0, C.Spaces), 1373 std::max((int)C.StartOfTokenColumn, C.Spaces) - std::max(0, C.Spaces), 1374 C.IsAligned); 1375 ReplacementText.append(C.CurrentLinePrefix); 1376 storeReplacement(C.OriginalWhitespaceRange, ReplacementText); 1377 } 1378 } 1379 } 1380 1381 void WhitespaceManager::storeReplacement(SourceRange Range, StringRef Text) { 1382 unsigned WhitespaceLength = SourceMgr.getFileOffset(Range.getEnd()) - 1383 SourceMgr.getFileOffset(Range.getBegin()); 1384 // Don't create a replacement, if it does not change anything. 1385 if (StringRef(SourceMgr.getCharacterData(Range.getBegin()), 1386 WhitespaceLength) == Text) 1387 return; 1388 auto Err = Replaces.add(tooling::Replacement( 1389 SourceMgr, CharSourceRange::getCharRange(Range), Text)); 1390 // FIXME: better error handling. For now, just print an error message in the 1391 // release version. 1392 if (Err) { 1393 llvm::errs() << llvm::toString(std::move(Err)) << "\n"; 1394 assert(false); 1395 } 1396 } 1397 1398 void WhitespaceManager::appendNewlineText(std::string &Text, 1399 unsigned Newlines) { 1400 if (UseCRLF) { 1401 Text.reserve(Text.size() + 2 * Newlines); 1402 for (unsigned i = 0; i < Newlines; ++i) 1403 Text.append("\r\n"); 1404 } else { 1405 Text.append(Newlines, '\n'); 1406 } 1407 } 1408 1409 void WhitespaceManager::appendEscapedNewlineText( 1410 std::string &Text, unsigned Newlines, unsigned PreviousEndOfTokenColumn, 1411 unsigned EscapedNewlineColumn) { 1412 if (Newlines > 0) { 1413 unsigned Spaces = 1414 std::max<int>(1, EscapedNewlineColumn - PreviousEndOfTokenColumn - 1); 1415 for (unsigned i = 0; i < Newlines; ++i) { 1416 Text.append(Spaces, ' '); 1417 Text.append(UseCRLF ? "\\\r\n" : "\\\n"); 1418 Spaces = std::max<int>(0, EscapedNewlineColumn - 1); 1419 } 1420 } 1421 } 1422 1423 void WhitespaceManager::appendIndentText(std::string &Text, 1424 unsigned IndentLevel, unsigned Spaces, 1425 unsigned WhitespaceStartColumn, 1426 bool IsAligned) { 1427 switch (Style.UseTab) { 1428 case FormatStyle::UT_Never: 1429 Text.append(Spaces, ' '); 1430 break; 1431 case FormatStyle::UT_Always: { 1432 if (Style.TabWidth) { 1433 unsigned FirstTabWidth = 1434 Style.TabWidth - WhitespaceStartColumn % Style.TabWidth; 1435 1436 // Insert only spaces when we want to end up before the next tab. 1437 if (Spaces < FirstTabWidth || Spaces == 1) { 1438 Text.append(Spaces, ' '); 1439 break; 1440 } 1441 // Align to the next tab. 1442 Spaces -= FirstTabWidth; 1443 Text.append("\t"); 1444 1445 Text.append(Spaces / Style.TabWidth, '\t'); 1446 Text.append(Spaces % Style.TabWidth, ' '); 1447 } else if (Spaces == 1) { 1448 Text.append(Spaces, ' '); 1449 } 1450 break; 1451 } 1452 case FormatStyle::UT_ForIndentation: 1453 if (WhitespaceStartColumn == 0) { 1454 unsigned Indentation = IndentLevel * Style.IndentWidth; 1455 Spaces = appendTabIndent(Text, Spaces, Indentation); 1456 } 1457 Text.append(Spaces, ' '); 1458 break; 1459 case FormatStyle::UT_ForContinuationAndIndentation: 1460 if (WhitespaceStartColumn == 0) 1461 Spaces = appendTabIndent(Text, Spaces, Spaces); 1462 Text.append(Spaces, ' '); 1463 break; 1464 case FormatStyle::UT_AlignWithSpaces: 1465 if (WhitespaceStartColumn == 0) { 1466 unsigned Indentation = 1467 IsAligned ? IndentLevel * Style.IndentWidth : Spaces; 1468 Spaces = appendTabIndent(Text, Spaces, Indentation); 1469 } 1470 Text.append(Spaces, ' '); 1471 break; 1472 } 1473 } 1474 1475 unsigned WhitespaceManager::appendTabIndent(std::string &Text, unsigned Spaces, 1476 unsigned Indentation) { 1477 // This happens, e.g. when a line in a block comment is indented less than the 1478 // first one. 1479 if (Indentation > Spaces) 1480 Indentation = Spaces; 1481 if (Style.TabWidth) { 1482 unsigned Tabs = Indentation / Style.TabWidth; 1483 Text.append(Tabs, '\t'); 1484 Spaces -= Tabs * Style.TabWidth; 1485 } 1486 return Spaces; 1487 } 1488 1489 } // namespace format 1490 } // namespace clang 1491