1 //===--- UnwrappedLineFormatter.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 #include "UnwrappedLineFormatter.h" 10 #include "NamespaceEndCommentsFixer.h" 11 #include "WhitespaceManager.h" 12 #include "llvm/Support/Debug.h" 13 #include <queue> 14 15 #define DEBUG_TYPE "format-formatter" 16 17 namespace clang { 18 namespace format { 19 20 namespace { 21 22 bool startsExternCBlock(const AnnotatedLine &Line) { 23 const FormatToken *Next = Line.First->getNextNonComment(); 24 const FormatToken *NextNext = Next ? Next->getNextNonComment() : nullptr; 25 return Line.startsWith(tok::kw_extern) && Next && Next->isStringLiteral() && 26 NextNext && NextNext->is(tok::l_brace); 27 } 28 29 /// Tracks the indent level of \c AnnotatedLines across levels. 30 /// 31 /// \c nextLine must be called for each \c AnnotatedLine, after which \c 32 /// getIndent() will return the indent for the last line \c nextLine was called 33 /// with. 34 /// If the line is not formatted (and thus the indent does not change), calling 35 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause 36 /// subsequent lines on the same level to be indented at the same level as the 37 /// given line. 38 class LevelIndentTracker { 39 public: 40 LevelIndentTracker(const FormatStyle &Style, 41 const AdditionalKeywords &Keywords, unsigned StartLevel, 42 int AdditionalIndent) 43 : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) { 44 for (unsigned i = 0; i != StartLevel; ++i) 45 IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent); 46 } 47 48 /// Returns the indent for the current line. 49 unsigned getIndent() const { return Indent; } 50 51 /// Update the indent state given that \p Line is going to be formatted 52 /// next. 53 void nextLine(const AnnotatedLine &Line) { 54 Offset = getIndentOffset(*Line.First); 55 // Update the indent level cache size so that we can rely on it 56 // having the right size in adjustToUnmodifiedline. 57 while (IndentForLevel.size() <= Line.Level) 58 IndentForLevel.push_back(-1); 59 if (Line.InPPDirective) { 60 unsigned IndentWidth = 61 (Style.PPIndentWidth >= 0) ? Style.PPIndentWidth : Style.IndentWidth; 62 Indent = Line.Level * IndentWidth + AdditionalIndent; 63 } else { 64 IndentForLevel.resize(Line.Level + 1); 65 Indent = getIndent(Line.Level); 66 } 67 if (static_cast<int>(Indent) + Offset >= 0) 68 Indent += Offset; 69 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 70 Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth; 71 } 72 73 /// Update the indent state given that \p Line indent should be 74 /// skipped. 75 void skipLine(const AnnotatedLine &Line) { 76 while (IndentForLevel.size() <= Line.Level) 77 IndentForLevel.push_back(Indent); 78 } 79 80 /// Update the level indent to adapt to the given \p Line. 81 /// 82 /// When a line is not formatted, we move the subsequent lines on the same 83 /// level to the same indent. 84 /// Note that \c nextLine must have been called before this method. 85 void adjustToUnmodifiedLine(const AnnotatedLine &Line) { 86 unsigned LevelIndent = Line.First->OriginalColumn; 87 if (static_cast<int>(LevelIndent) - Offset >= 0) 88 LevelIndent -= Offset; 89 if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) && 90 !Line.InPPDirective) 91 IndentForLevel[Line.Level] = LevelIndent; 92 } 93 94 private: 95 /// Get the offset of the line relatively to the level. 96 /// 97 /// For example, 'public:' labels in classes are offset by 1 or 2 98 /// characters to the left from their level. 99 int getIndentOffset(const FormatToken &RootToken) { 100 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 101 Style.isCSharp()) 102 return 0; 103 104 auto IsAccessModifier = [this, &RootToken]() { 105 if (RootToken.isAccessSpecifier(Style.isCpp())) 106 return true; 107 else if (RootToken.isObjCAccessSpecifier()) 108 return true; 109 // Handle Qt signals. 110 else if ((RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) && 111 RootToken.Next && RootToken.Next->is(tok::colon))) 112 return true; 113 else if (RootToken.Next && 114 RootToken.Next->isOneOf(Keywords.kw_slots, Keywords.kw_qslots) && 115 RootToken.Next->Next && RootToken.Next->Next->is(tok::colon)) 116 return true; 117 // Handle malformed access specifier e.g. 'private' without trailing ':'. 118 else if (!RootToken.Next && RootToken.isAccessSpecifier(false)) 119 return true; 120 return false; 121 }; 122 123 if (IsAccessModifier()) { 124 // The AccessModifierOffset may be overridden by IndentAccessModifiers, 125 // in which case we take a negative value of the IndentWidth to simulate 126 // the upper indent level. 127 return Style.IndentAccessModifiers ? -Style.IndentWidth 128 : Style.AccessModifierOffset; 129 } 130 return 0; 131 } 132 133 /// Get the indent of \p Level from \p IndentForLevel. 134 /// 135 /// \p IndentForLevel must contain the indent for the level \c l 136 /// at \p IndentForLevel[l], or a value < 0 if the indent for 137 /// that level is unknown. 138 unsigned getIndent(unsigned Level) const { 139 if (IndentForLevel[Level] != -1) 140 return IndentForLevel[Level]; 141 if (Level == 0) 142 return 0; 143 return getIndent(Level - 1) + Style.IndentWidth; 144 } 145 146 const FormatStyle &Style; 147 const AdditionalKeywords &Keywords; 148 const unsigned AdditionalIndent; 149 150 /// The indent in characters for each level. 151 std::vector<int> IndentForLevel; 152 153 /// Offset of the current line relative to the indent level. 154 /// 155 /// For example, the 'public' keywords is often indented with a negative 156 /// offset. 157 int Offset = 0; 158 159 /// The current line's indent. 160 unsigned Indent = 0; 161 }; 162 163 const FormatToken *getMatchingNamespaceToken( 164 const AnnotatedLine *Line, 165 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) { 166 if (!Line->startsWith(tok::r_brace)) 167 return nullptr; 168 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex; 169 if (StartLineIndex == UnwrappedLine::kInvalidIndex) 170 return nullptr; 171 assert(StartLineIndex < AnnotatedLines.size()); 172 return AnnotatedLines[StartLineIndex]->First->getNamespaceToken(); 173 } 174 175 StringRef getNamespaceTokenText(const AnnotatedLine *Line) { 176 const FormatToken *NamespaceToken = Line->First->getNamespaceToken(); 177 return NamespaceToken ? NamespaceToken->TokenText : StringRef(); 178 } 179 180 StringRef getMatchingNamespaceTokenText( 181 const AnnotatedLine *Line, 182 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) { 183 const FormatToken *NamespaceToken = 184 getMatchingNamespaceToken(Line, AnnotatedLines); 185 return NamespaceToken ? NamespaceToken->TokenText : StringRef(); 186 } 187 188 class LineJoiner { 189 public: 190 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords, 191 const SmallVectorImpl<AnnotatedLine *> &Lines) 192 : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()), 193 AnnotatedLines(Lines) {} 194 195 /// Returns the next line, merging multiple lines into one if possible. 196 const AnnotatedLine *getNextMergedLine(bool DryRun, 197 LevelIndentTracker &IndentTracker) { 198 if (Next == End) 199 return nullptr; 200 const AnnotatedLine *Current = *Next; 201 IndentTracker.nextLine(*Current); 202 unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End); 203 if (MergedLines > 0 && Style.ColumnLimit == 0) 204 // Disallow line merging if there is a break at the start of one of the 205 // input lines. 206 for (unsigned i = 0; i < MergedLines; ++i) 207 if (Next[i + 1]->First->NewlinesBefore > 0) 208 MergedLines = 0; 209 if (!DryRun) 210 for (unsigned i = 0; i < MergedLines; ++i) 211 join(*Next[0], *Next[i + 1]); 212 Next = Next + MergedLines + 1; 213 return Current; 214 } 215 216 private: 217 /// Calculates how many lines can be merged into 1 starting at \p I. 218 unsigned 219 tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker, 220 SmallVectorImpl<AnnotatedLine *>::const_iterator I, 221 SmallVectorImpl<AnnotatedLine *>::const_iterator E) { 222 const unsigned Indent = IndentTracker.getIndent(); 223 224 // Can't join the last line with anything. 225 if (I + 1 == E) 226 return 0; 227 // We can never merge stuff if there are trailing line comments. 228 const AnnotatedLine *TheLine = *I; 229 if (TheLine->Last->is(TT_LineComment)) 230 return 0; 231 const auto &NextLine = *I[1]; 232 if (NextLine.Type == LT_Invalid || NextLine.First->MustBreakBefore) 233 return 0; 234 if (TheLine->InPPDirective && 235 (!NextLine.InPPDirective || NextLine.First->HasUnescapedNewline)) 236 return 0; 237 238 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit) 239 return 0; 240 241 unsigned Limit = 242 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent; 243 // If we already exceed the column limit, we set 'Limit' to 0. The different 244 // tryMerge..() functions can then decide whether to still do merging. 245 Limit = TheLine->Last->TotalLength > Limit 246 ? 0 247 : Limit - TheLine->Last->TotalLength; 248 249 if (TheLine->Last->is(TT_FunctionLBrace) && 250 TheLine->First == TheLine->Last && 251 !Style.BraceWrapping.SplitEmptyFunction && 252 NextLine.First->is(tok::r_brace)) 253 return tryMergeSimpleBlock(I, E, Limit); 254 255 const auto *PreviousLine = I != AnnotatedLines.begin() ? I[-1] : nullptr; 256 // Handle empty record blocks where the brace has already been wrapped. 257 if (PreviousLine && TheLine->Last->is(tok::l_brace) && 258 TheLine->First == TheLine->Last) { 259 bool EmptyBlock = NextLine.First->is(tok::r_brace); 260 261 const FormatToken *Tok = PreviousLine->First; 262 if (Tok && Tok->is(tok::comment)) 263 Tok = Tok->getNextNonComment(); 264 265 if (Tok && Tok->getNamespaceToken()) 266 return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock 267 ? tryMergeSimpleBlock(I, E, Limit) 268 : 0; 269 270 if (Tok && Tok->is(tok::kw_typedef)) 271 Tok = Tok->getNextNonComment(); 272 if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union, 273 tok::kw_extern, Keywords.kw_interface)) 274 return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock 275 ? tryMergeSimpleBlock(I, E, Limit) 276 : 0; 277 278 if (Tok && Tok->is(tok::kw_template) && 279 Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) 280 return 0; 281 } 282 283 auto ShouldMergeShortFunctions = [this, &I, &NextLine, PreviousLine, 284 TheLine]() { 285 if (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All) 286 return true; 287 if (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty && 288 NextLine.First->is(tok::r_brace)) 289 return true; 290 291 if (Style.AllowShortFunctionsOnASingleLine & 292 FormatStyle::SFS_InlineOnly) { 293 // Just checking TheLine->Level != 0 is not enough, because it 294 // provokes treating functions inside indented namespaces as short. 295 if (Style.isJavaScript() && TheLine->Last->is(TT_FunctionLBrace)) 296 return true; 297 298 if (TheLine->Level != 0) { 299 if (!PreviousLine) 300 return false; 301 302 // TODO: Use IndentTracker to avoid loop? 303 // Find the last line with lower level. 304 auto J = I - 1; 305 for (; J != AnnotatedLines.begin(); --J) 306 if ((*J)->Level < TheLine->Level) 307 break; 308 309 // Check if the found line starts a record. 310 for (const FormatToken *RecordTok = (*J)->Last; RecordTok; 311 RecordTok = RecordTok->Previous) 312 if (RecordTok->is(tok::l_brace)) 313 return RecordTok->is(TT_RecordLBrace); 314 } 315 } 316 317 return false; 318 }; 319 320 bool MergeShortFunctions = ShouldMergeShortFunctions(); 321 322 const FormatToken *FirstNonComment = TheLine->First; 323 if (FirstNonComment->is(tok::comment)) { 324 FirstNonComment = FirstNonComment->getNextNonComment(); 325 if (!FirstNonComment) 326 return 0; 327 } 328 // FIXME: There are probably cases where we should use FirstNonComment 329 // instead of TheLine->First. 330 331 if (Style.CompactNamespaces) { 332 if (auto nsToken = TheLine->First->getNamespaceToken()) { 333 int i = 0; 334 unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1; 335 for (; I + 1 + i != E && 336 nsToken->TokenText == getNamespaceTokenText(I[i + 1]) && 337 closingLine == I[i + 1]->MatchingClosingBlockLineIndex && 338 I[i + 1]->Last->TotalLength < Limit; 339 i++, --closingLine) { 340 // No extra indent for compacted namespaces. 341 IndentTracker.skipLine(*I[i + 1]); 342 343 Limit -= I[i + 1]->Last->TotalLength; 344 } 345 return i; 346 } 347 348 if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) { 349 int i = 0; 350 unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1; 351 for (; I + 1 + i != E && 352 nsToken->TokenText == 353 getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) && 354 openingLine == I[i + 1]->MatchingOpeningBlockLineIndex; 355 i++, --openingLine) { 356 // No space between consecutive braces. 357 I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace); 358 359 // Indent like the outer-most namespace. 360 IndentTracker.nextLine(*I[i + 1]); 361 } 362 return i; 363 } 364 } 365 366 // Try to merge a function block with left brace unwrapped. 367 if (TheLine->Last->is(TT_FunctionLBrace) && TheLine->First != TheLine->Last) 368 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0; 369 // Try to merge a control statement block with left brace unwrapped. 370 if (TheLine->Last->is(tok::l_brace) && FirstNonComment != TheLine->Last && 371 FirstNonComment->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for, 372 TT_ForEachMacro)) { 373 return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never 374 ? tryMergeSimpleBlock(I, E, Limit) 375 : 0; 376 } 377 // Try to merge a control statement block with left brace wrapped. 378 if (NextLine.First->is(tok::l_brace)) { 379 if ((TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, 380 tok::kw_for, tok::kw_switch, tok::kw_try, 381 tok::kw_do, TT_ForEachMacro) || 382 (TheLine->First->is(tok::r_brace) && TheLine->First->Next && 383 TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) && 384 Style.BraceWrapping.AfterControlStatement == 385 FormatStyle::BWACS_MultiLine) { 386 // If possible, merge the next line's wrapped left brace with the 387 // current line. Otherwise, leave it on the next line, as this is a 388 // multi-line control statement. 389 return (Style.ColumnLimit == 0 || 390 TheLine->Last->TotalLength <= Style.ColumnLimit) 391 ? 1 392 : 0; 393 } 394 if (TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, 395 tok::kw_for, TT_ForEachMacro)) { 396 return (Style.BraceWrapping.AfterControlStatement == 397 FormatStyle::BWACS_Always) 398 ? tryMergeSimpleBlock(I, E, Limit) 399 : 0; 400 } 401 if (TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) && 402 Style.BraceWrapping.AfterControlStatement == 403 FormatStyle::BWACS_MultiLine) { 404 // This case if different from the upper BWACS_MultiLine processing 405 // in that a preceding r_brace is not on the same line as else/catch 406 // most likely because of BeforeElse/BeforeCatch set to true. 407 // If the line length doesn't fit ColumnLimit, leave l_brace on the 408 // next line to respect the BWACS_MultiLine. 409 return (Style.ColumnLimit == 0 || 410 TheLine->Last->TotalLength <= Style.ColumnLimit) 411 ? 1 412 : 0; 413 } 414 } 415 if (PreviousLine && TheLine->First->is(tok::l_brace)) { 416 switch (PreviousLine->First->Tok.getKind()) { 417 case tok::at: 418 // Don't merge block with left brace wrapped after ObjC special blocks. 419 if (PreviousLine->First->Next) { 420 tok::ObjCKeywordKind kwId = 421 PreviousLine->First->Next->Tok.getObjCKeywordID(); 422 if (kwId == tok::objc_autoreleasepool || 423 kwId == tok::objc_synchronized) 424 return 0; 425 } 426 break; 427 428 case tok::kw_case: 429 case tok::kw_default: 430 // Don't merge block with left brace wrapped after case labels. 431 return 0; 432 433 default: 434 break; 435 } 436 } 437 438 // Don't merge an empty template class or struct if SplitEmptyRecords 439 // is defined. 440 if (PreviousLine && Style.BraceWrapping.SplitEmptyRecord && 441 TheLine->Last->is(tok::l_brace) && PreviousLine->Last) { 442 const FormatToken *Previous = PreviousLine->Last; 443 if (Previous) { 444 if (Previous->is(tok::comment)) 445 Previous = Previous->getPreviousNonComment(); 446 if (Previous) { 447 if (Previous->is(tok::greater) && !PreviousLine->InPPDirective) 448 return 0; 449 if (Previous->is(tok::identifier)) { 450 const FormatToken *PreviousPrevious = 451 Previous->getPreviousNonComment(); 452 if (PreviousPrevious && 453 PreviousPrevious->isOneOf(tok::kw_class, tok::kw_struct)) 454 return 0; 455 } 456 } 457 } 458 } 459 460 // Try to merge a block with left brace unwrapped that wasn't yet covered. 461 if (TheLine->Last->is(tok::l_brace)) { 462 const FormatToken *Tok = TheLine->First; 463 bool ShouldMerge = false; 464 if (Tok->is(tok::kw_typedef)) { 465 Tok = Tok->getNextNonComment(); 466 assert(Tok); 467 } 468 if (Tok->isOneOf(tok::kw_class, tok::kw_struct)) { 469 ShouldMerge = !Style.BraceWrapping.AfterClass || 470 (NextLine.First->is(tok::r_brace) && 471 !Style.BraceWrapping.SplitEmptyRecord); 472 } else if (Tok->is(tok::kw_enum)) { 473 ShouldMerge = Style.AllowShortEnumsOnASingleLine; 474 } else { 475 ShouldMerge = !Style.BraceWrapping.AfterFunction || 476 (NextLine.First->is(tok::r_brace) && 477 !Style.BraceWrapping.SplitEmptyFunction); 478 } 479 return ShouldMerge ? tryMergeSimpleBlock(I, E, Limit) : 0; 480 } 481 // Try to merge a function block with left brace wrapped. 482 if (NextLine.First->is(TT_FunctionLBrace) && 483 Style.BraceWrapping.AfterFunction) { 484 if (NextLine.Last->is(TT_LineComment)) 485 return 0; 486 487 // Check for Limit <= 2 to account for the " {". 488 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine))) 489 return 0; 490 Limit -= 2; 491 492 unsigned MergedLines = 0; 493 if (MergeShortFunctions || 494 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty && 495 NextLine.First == NextLine.Last && I + 2 != E && 496 I[2]->First->is(tok::r_brace))) { 497 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); 498 // If we managed to merge the block, count the function header, which is 499 // on a separate line. 500 if (MergedLines > 0) 501 ++MergedLines; 502 } 503 return MergedLines; 504 } 505 auto IsElseLine = [&TheLine]() -> bool { 506 const FormatToken *First = TheLine->First; 507 if (First->is(tok::kw_else)) 508 return true; 509 510 return First->is(tok::r_brace) && First->Next && 511 First->Next->is(tok::kw_else); 512 }; 513 if (TheLine->First->is(tok::kw_if) || 514 (IsElseLine() && (Style.AllowShortIfStatementsOnASingleLine == 515 FormatStyle::SIS_AllIfsAndElse))) { 516 return Style.AllowShortIfStatementsOnASingleLine 517 ? tryMergeSimpleControlStatement(I, E, Limit) 518 : 0; 519 } 520 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do, 521 TT_ForEachMacro)) { 522 return Style.AllowShortLoopsOnASingleLine 523 ? tryMergeSimpleControlStatement(I, E, Limit) 524 : 0; 525 } 526 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) { 527 return Style.AllowShortCaseLabelsOnASingleLine 528 ? tryMergeShortCaseLabels(I, E, Limit) 529 : 0; 530 } 531 if (TheLine->InPPDirective && 532 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) 533 return tryMergeSimplePPDirective(I, E, Limit); 534 return 0; 535 } 536 537 unsigned 538 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 539 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 540 unsigned Limit) { 541 if (Limit == 0) 542 return 0; 543 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline) 544 return 0; 545 if (1 + I[1]->Last->TotalLength > Limit) 546 return 0; 547 return 1; 548 } 549 550 unsigned tryMergeSimpleControlStatement( 551 SmallVectorImpl<AnnotatedLine *>::const_iterator I, 552 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) { 553 if (Limit == 0) 554 return 0; 555 if (Style.BraceWrapping.AfterControlStatement == 556 FormatStyle::BWACS_Always && 557 I[1]->First->is(tok::l_brace) && 558 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 559 return 0; 560 if (I[1]->InPPDirective != (*I)->InPPDirective || 561 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline)) 562 return 0; 563 Limit = limitConsideringMacros(I + 1, E, Limit); 564 AnnotatedLine &Line = **I; 565 if (!Line.First->is(tok::kw_do) && !Line.First->is(tok::kw_else) && 566 !Line.Last->is(tok::kw_else) && Line.Last->isNot(tok::r_paren)) 567 return 0; 568 // Only merge `do while` if `do` is the only statement on the line. 569 if (Line.First->is(tok::kw_do) && !Line.Last->is(tok::kw_do)) 570 return 0; 571 if (1 + I[1]->Last->TotalLength > Limit) 572 return 0; 573 // Don't merge with loops, ifs, a single semicolon or a line comment. 574 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while, 575 TT_ForEachMacro, TT_LineComment)) 576 return 0; 577 // Only inline simple if's (no nested if or else), unless specified 578 if (Style.AllowShortIfStatementsOnASingleLine == 579 FormatStyle::SIS_WithoutElse) { 580 if (I + 2 != E && Line.startsWith(tok::kw_if) && 581 I[2]->First->is(tok::kw_else)) 582 return 0; 583 } 584 return 1; 585 } 586 587 unsigned 588 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 589 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 590 unsigned Limit) { 591 if (Limit == 0 || I + 1 == E || 592 I[1]->First->isOneOf(tok::kw_case, tok::kw_default)) 593 return 0; 594 if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace)) 595 return 0; 596 unsigned NumStmts = 0; 597 unsigned Length = 0; 598 bool EndsWithComment = false; 599 bool InPPDirective = I[0]->InPPDirective; 600 const unsigned Level = I[0]->Level; 601 for (; NumStmts < 3; ++NumStmts) { 602 if (I + 1 + NumStmts == E) 603 break; 604 const AnnotatedLine *Line = I[1 + NumStmts]; 605 if (Line->InPPDirective != InPPDirective) 606 break; 607 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace)) 608 break; 609 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch, 610 tok::kw_while) || 611 EndsWithComment) 612 return 0; 613 if (Line->First->is(tok::comment)) { 614 if (Level != Line->Level) 615 return 0; 616 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts; 617 for (; J != E; ++J) { 618 Line = *J; 619 if (Line->InPPDirective != InPPDirective) 620 break; 621 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace)) 622 break; 623 if (Line->First->isNot(tok::comment) || Level != Line->Level) 624 return 0; 625 } 626 break; 627 } 628 if (Line->Last->is(tok::comment)) 629 EndsWithComment = true; 630 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space. 631 } 632 if (NumStmts == 0 || NumStmts == 3 || Length > Limit) 633 return 0; 634 return NumStmts; 635 } 636 637 unsigned 638 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 639 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 640 unsigned Limit) { 641 // Don't merge with a preprocessor directive. 642 if (I[1]->Type == LT_PreprocessorDirective) 643 return 0; 644 645 AnnotatedLine &Line = **I; 646 647 // Don't merge ObjC @ keywords and methods. 648 // FIXME: If an option to allow short exception handling clauses on a single 649 // line is added, change this to not return for @try and friends. 650 if (Style.Language != FormatStyle::LK_Java && 651 Line.First->isOneOf(tok::at, tok::minus, tok::plus)) 652 return 0; 653 654 // Check that the current line allows merging. This depends on whether we 655 // are in a control flow statements as well as several style flags. 656 if (Line.First->is(tok::kw_case) || 657 (Line.First->Next && Line.First->Next->is(tok::kw_else))) 658 return 0; 659 // default: in switch statement 660 if (Line.First->is(tok::kw_default)) { 661 const FormatToken *Tok = Line.First->getNextNonComment(); 662 if (Tok && Tok->is(tok::colon)) 663 return 0; 664 } 665 if (Line.First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, tok::kw_do, 666 tok::kw_try, tok::kw___try, tok::kw_catch, 667 tok::kw___finally, tok::kw_for, TT_ForEachMacro, 668 tok::r_brace, Keywords.kw___except)) { 669 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 670 return 0; 671 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Empty && 672 !I[1]->First->is(tok::r_brace)) 673 return 0; 674 // Don't merge when we can't except the case when 675 // the control statement block is empty 676 if (!Style.AllowShortIfStatementsOnASingleLine && 677 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 678 !Style.BraceWrapping.AfterControlStatement && 679 !I[1]->First->is(tok::r_brace)) 680 return 0; 681 if (!Style.AllowShortIfStatementsOnASingleLine && 682 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 683 Style.BraceWrapping.AfterControlStatement == 684 FormatStyle::BWACS_Always && 685 I + 2 != E && !I[2]->First->is(tok::r_brace)) 686 return 0; 687 if (!Style.AllowShortLoopsOnASingleLine && 688 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for, 689 TT_ForEachMacro) && 690 !Style.BraceWrapping.AfterControlStatement && 691 !I[1]->First->is(tok::r_brace)) 692 return 0; 693 if (!Style.AllowShortLoopsOnASingleLine && 694 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for, 695 TT_ForEachMacro) && 696 Style.BraceWrapping.AfterControlStatement == 697 FormatStyle::BWACS_Always && 698 I + 2 != E && !I[2]->First->is(tok::r_brace)) 699 return 0; 700 // FIXME: Consider an option to allow short exception handling clauses on 701 // a single line. 702 // FIXME: This isn't covered by tests. 703 // FIXME: For catch, __except, __finally the first token on the line 704 // is '}', so this isn't correct here. 705 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch, 706 Keywords.kw___except, tok::kw___finally)) 707 return 0; 708 } 709 710 if (Line.Last->is(tok::l_brace)) { 711 FormatToken *Tok = I[1]->First; 712 auto ShouldMerge = [Tok]() { 713 if (Tok->isNot(tok::r_brace) || Tok->MustBreakBefore) 714 return false; 715 const FormatToken *Next = Tok->getNextNonComment(); 716 return !Next || Next->is(tok::semi); 717 }; 718 if (ShouldMerge()) { 719 // We merge empty blocks even if the line exceeds the column limit. 720 Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0; 721 Tok->CanBreakBefore = true; 722 return 1; 723 } else if (Limit != 0 && !Line.startsWithNamespace() && 724 !startsExternCBlock(Line)) { 725 // We don't merge short records. 726 FormatToken *RecordTok = Line.First; 727 // Skip record modifiers. 728 while (RecordTok->Next && 729 RecordTok->isOneOf(tok::kw_typedef, tok::kw_export, 730 Keywords.kw_declare, Keywords.kw_abstract, 731 tok::kw_default, Keywords.kw_override, 732 tok::kw_public, tok::kw_private, 733 tok::kw_protected, Keywords.kw_internal)) 734 RecordTok = RecordTok->Next; 735 if (RecordTok && 736 RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct, 737 Keywords.kw_interface)) 738 return 0; 739 740 // Check that we still have three lines and they fit into the limit. 741 if (I + 2 == E || I[2]->Type == LT_Invalid) 742 return 0; 743 Limit = limitConsideringMacros(I + 2, E, Limit); 744 745 if (!nextTwoLinesFitInto(I, Limit)) 746 return 0; 747 748 // Second, check that the next line does not contain any braces - if it 749 // does, readability declines when putting it into a single line. 750 if (I[1]->Last->is(TT_LineComment)) 751 return 0; 752 do { 753 if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit)) 754 return 0; 755 Tok = Tok->Next; 756 } while (Tok); 757 758 // Last, check that the third line starts with a closing brace. 759 Tok = I[2]->First; 760 if (Tok->isNot(tok::r_brace)) 761 return 0; 762 763 // Don't merge "if (a) { .. } else {". 764 if (Tok->Next && Tok->Next->is(tok::kw_else)) 765 return 0; 766 767 // Don't merge a trailing multi-line control statement block like: 768 // } else if (foo && 769 // bar) 770 // { <-- current Line 771 // baz(); 772 // } 773 if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) && 774 Style.BraceWrapping.AfterControlStatement == 775 FormatStyle::BWACS_MultiLine) 776 return 0; 777 778 return 2; 779 } 780 } else if (I[1]->First->is(tok::l_brace)) { 781 if (I[1]->Last->is(TT_LineComment)) 782 return 0; 783 784 // Check for Limit <= 2 to account for the " {". 785 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I))) 786 return 0; 787 Limit -= 2; 788 unsigned MergedLines = 0; 789 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never || 790 (I[1]->First == I[1]->Last && I + 2 != E && 791 I[2]->First->is(tok::r_brace))) { 792 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); 793 // If we managed to merge the block, count the statement header, which 794 // is on a separate line. 795 if (MergedLines > 0) 796 ++MergedLines; 797 } 798 return MergedLines; 799 } 800 return 0; 801 } 802 803 /// Returns the modified column limit for \p I if it is inside a macro and 804 /// needs a trailing '\'. 805 unsigned 806 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 807 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 808 unsigned Limit) { 809 if (I[0]->InPPDirective && I + 1 != E && 810 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) 811 return Limit < 2 ? 0 : Limit - 2; 812 return Limit; 813 } 814 815 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 816 unsigned Limit) { 817 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore) 818 return false; 819 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit; 820 } 821 822 bool containsMustBreak(const AnnotatedLine *Line) { 823 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) 824 if (Tok->MustBreakBefore) 825 return true; 826 return false; 827 } 828 829 void join(AnnotatedLine &A, const AnnotatedLine &B) { 830 assert(!A.Last->Next); 831 assert(!B.First->Previous); 832 if (B.Affected) 833 A.Affected = true; 834 A.Last->Next = B.First; 835 B.First->Previous = A.Last; 836 B.First->CanBreakBefore = true; 837 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore; 838 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) { 839 Tok->TotalLength += LengthA; 840 A.Last = Tok; 841 } 842 } 843 844 const FormatStyle &Style; 845 const AdditionalKeywords &Keywords; 846 const SmallVectorImpl<AnnotatedLine *>::const_iterator End; 847 848 SmallVectorImpl<AnnotatedLine *>::const_iterator Next; 849 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines; 850 }; 851 852 static void markFinalized(FormatToken *Tok) { 853 for (; Tok; Tok = Tok->Next) { 854 Tok->Finalized = true; 855 for (AnnotatedLine *Child : Tok->Children) 856 markFinalized(Child->First); 857 } 858 } 859 860 #ifndef NDEBUG 861 static void printLineState(const LineState &State) { 862 llvm::dbgs() << "State: "; 863 for (const ParenState &P : State.Stack) { 864 llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|" 865 << P.LastSpace << "|" << P.NestedBlockIndent << " "; 866 } 867 llvm::dbgs() << State.NextToken->TokenText << "\n"; 868 } 869 #endif 870 871 /// Base class for classes that format one \c AnnotatedLine. 872 class LineFormatter { 873 public: 874 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces, 875 const FormatStyle &Style, 876 UnwrappedLineFormatter *BlockFormatter) 877 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style), 878 BlockFormatter(BlockFormatter) {} 879 virtual ~LineFormatter() {} 880 881 /// Formats an \c AnnotatedLine and returns the penalty. 882 /// 883 /// If \p DryRun is \c false, directly applies the changes. 884 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 885 unsigned FirstStartColumn, bool DryRun) = 0; 886 887 protected: 888 /// If the \p State's next token is an r_brace closing a nested block, 889 /// format the nested block before it. 890 /// 891 /// Returns \c true if all children could be placed successfully and adapts 892 /// \p Penalty as well as \p State. If \p DryRun is false, also directly 893 /// creates changes using \c Whitespaces. 894 /// 895 /// The crucial idea here is that children always get formatted upon 896 /// encountering the closing brace right after the nested block. Now, if we 897 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is 898 /// \c false), the entire block has to be kept on the same line (which is only 899 /// possible if it fits on the line, only contains a single statement, etc. 900 /// 901 /// If \p NewLine is true, we format the nested block on separate lines, i.e. 902 /// break after the "{", format all lines with correct indentation and the put 903 /// the closing "}" on yet another new line. 904 /// 905 /// This enables us to keep the simple structure of the 906 /// \c UnwrappedLineFormatter, where we only have two options for each token: 907 /// break or don't break. 908 bool formatChildren(LineState &State, bool NewLine, bool DryRun, 909 unsigned &Penalty) { 910 const FormatToken *LBrace = State.NextToken->getPreviousNonComment(); 911 FormatToken &Previous = *State.NextToken->Previous; 912 if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) || 913 Previous.Children.size() == 0) 914 // The previous token does not open a block. Nothing to do. We don't 915 // assert so that we can simply call this function for all tokens. 916 return true; 917 918 if (NewLine) { 919 const ParenState &P = State.Stack.back(); 920 921 int AdditionalIndent = 922 P.Indent - Previous.Children[0]->Level * Style.IndentWidth; 923 924 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && 925 P.NestedBlockIndent == P.LastSpace) { 926 if (State.NextToken->MatchingParen && 927 State.NextToken->MatchingParen->is(TT_LambdaLBrace)) 928 State.Stack.pop_back(); 929 if (LBrace->is(TT_LambdaLBrace)) 930 AdditionalIndent = 0; 931 } 932 933 Penalty += 934 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent, 935 /*FixBadIndentation=*/true); 936 return true; 937 } 938 939 if (Previous.Children[0]->First->MustBreakBefore) 940 return false; 941 942 // Cannot merge into one line if this line ends on a comment. 943 if (Previous.is(tok::comment)) 944 return false; 945 946 // Cannot merge multiple statements into a single line. 947 if (Previous.Children.size() > 1) 948 return false; 949 950 const AnnotatedLine *Child = Previous.Children[0]; 951 // We can't put the closing "}" on a line with a trailing comment. 952 if (Child->Last->isTrailingComment()) 953 return false; 954 955 // If the child line exceeds the column limit, we wouldn't want to merge it. 956 // We add +2 for the trailing " }". 957 if (Style.ColumnLimit > 0 && 958 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit) 959 return false; 960 961 if (!DryRun) { 962 Whitespaces->replaceWhitespace( 963 *Child->First, /*Newlines=*/0, /*Spaces=*/1, 964 /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false, 965 State.Line->InPPDirective); 966 } 967 Penalty += 968 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun); 969 970 State.Column += 1 + Child->Last->TotalLength; 971 return true; 972 } 973 974 ContinuationIndenter *Indenter; 975 976 private: 977 WhitespaceManager *Whitespaces; 978 const FormatStyle &Style; 979 UnwrappedLineFormatter *BlockFormatter; 980 }; 981 982 /// Formatter that keeps the existing line breaks. 983 class NoColumnLimitLineFormatter : public LineFormatter { 984 public: 985 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter, 986 WhitespaceManager *Whitespaces, 987 const FormatStyle &Style, 988 UnwrappedLineFormatter *BlockFormatter) 989 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 990 991 /// Formats the line, simply keeping all of the input's line breaking 992 /// decisions. 993 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 994 unsigned FirstStartColumn, bool DryRun) override { 995 assert(!DryRun); 996 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn, 997 &Line, /*DryRun=*/false); 998 while (State.NextToken) { 999 bool Newline = 1000 Indenter->mustBreak(State) || 1001 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0); 1002 unsigned Penalty = 0; 1003 formatChildren(State, Newline, /*DryRun=*/false, Penalty); 1004 Indenter->addTokenToState(State, Newline, /*DryRun=*/false); 1005 } 1006 return 0; 1007 } 1008 }; 1009 1010 /// Formatter that puts all tokens into a single line without breaks. 1011 class NoLineBreakFormatter : public LineFormatter { 1012 public: 1013 NoLineBreakFormatter(ContinuationIndenter *Indenter, 1014 WhitespaceManager *Whitespaces, const FormatStyle &Style, 1015 UnwrappedLineFormatter *BlockFormatter) 1016 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 1017 1018 /// Puts all tokens into a single line. 1019 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 1020 unsigned FirstStartColumn, bool DryRun) override { 1021 unsigned Penalty = 0; 1022 LineState State = 1023 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 1024 while (State.NextToken) { 1025 formatChildren(State, /*NewLine=*/false, DryRun, Penalty); 1026 Indenter->addTokenToState( 1027 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun); 1028 } 1029 return Penalty; 1030 } 1031 }; 1032 1033 /// Finds the best way to break lines. 1034 class OptimizingLineFormatter : public LineFormatter { 1035 public: 1036 OptimizingLineFormatter(ContinuationIndenter *Indenter, 1037 WhitespaceManager *Whitespaces, 1038 const FormatStyle &Style, 1039 UnwrappedLineFormatter *BlockFormatter) 1040 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 1041 1042 /// Formats the line by finding the best line breaks with line lengths 1043 /// below the column limit. 1044 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 1045 unsigned FirstStartColumn, bool DryRun) override { 1046 LineState State = 1047 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 1048 1049 // If the ObjC method declaration does not fit on a line, we should format 1050 // it with one arg per line. 1051 if (State.Line->Type == LT_ObjCMethodDecl) 1052 State.Stack.back().BreakBeforeParameter = true; 1053 1054 // Find best solution in solution space. 1055 return analyzeSolutionSpace(State, DryRun); 1056 } 1057 1058 private: 1059 struct CompareLineStatePointers { 1060 bool operator()(LineState *obj1, LineState *obj2) const { 1061 return *obj1 < *obj2; 1062 } 1063 }; 1064 1065 /// A pair of <penalty, count> that is used to prioritize the BFS on. 1066 /// 1067 /// In case of equal penalties, we want to prefer states that were inserted 1068 /// first. During state generation we make sure that we insert states first 1069 /// that break the line as late as possible. 1070 typedef std::pair<unsigned, unsigned> OrderedPenalty; 1071 1072 /// An edge in the solution space from \c Previous->State to \c State, 1073 /// inserting a newline dependent on the \c NewLine. 1074 struct StateNode { 1075 StateNode(const LineState &State, bool NewLine, StateNode *Previous) 1076 : State(State), NewLine(NewLine), Previous(Previous) {} 1077 LineState State; 1078 bool NewLine; 1079 StateNode *Previous; 1080 }; 1081 1082 /// An item in the prioritized BFS search queue. The \c StateNode's 1083 /// \c State has the given \c OrderedPenalty. 1084 typedef std::pair<OrderedPenalty, StateNode *> QueueItem; 1085 1086 /// The BFS queue type. 1087 typedef std::priority_queue<QueueItem, std::vector<QueueItem>, 1088 std::greater<QueueItem>> 1089 QueueType; 1090 1091 /// Analyze the entire solution space starting from \p InitialState. 1092 /// 1093 /// This implements a variant of Dijkstra's algorithm on the graph that spans 1094 /// the solution space (\c LineStates are the nodes). The algorithm tries to 1095 /// find the shortest path (the one with lowest penalty) from \p InitialState 1096 /// to a state where all tokens are placed. Returns the penalty. 1097 /// 1098 /// If \p DryRun is \c false, directly applies the changes. 1099 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) { 1100 std::set<LineState *, CompareLineStatePointers> Seen; 1101 1102 // Increasing count of \c StateNode items we have created. This is used to 1103 // create a deterministic order independent of the container. 1104 unsigned Count = 0; 1105 QueueType Queue; 1106 1107 // Insert start element into queue. 1108 StateNode *RootNode = 1109 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr); 1110 Queue.push(QueueItem(OrderedPenalty(0, Count), RootNode)); 1111 ++Count; 1112 1113 unsigned Penalty = 0; 1114 1115 // While not empty, take first element and follow edges. 1116 while (!Queue.empty()) { 1117 Penalty = Queue.top().first.first; 1118 StateNode *Node = Queue.top().second; 1119 if (!Node->State.NextToken) { 1120 LLVM_DEBUG(llvm::dbgs() 1121 << "\n---\nPenalty for line: " << Penalty << "\n"); 1122 break; 1123 } 1124 Queue.pop(); 1125 1126 // Cut off the analysis of certain solutions if the analysis gets too 1127 // complex. See description of IgnoreStackForComparison. 1128 if (Count > 50000) 1129 Node->State.IgnoreStackForComparison = true; 1130 1131 if (!Seen.insert(&Node->State).second) 1132 // State already examined with lower penalty. 1133 continue; 1134 1135 FormatDecision LastFormat = Node->State.NextToken->getDecision(); 1136 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue) 1137 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue); 1138 if (LastFormat == FD_Unformatted || LastFormat == FD_Break) 1139 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue); 1140 } 1141 1142 if (Queue.empty()) { 1143 // We were unable to find a solution, do nothing. 1144 // FIXME: Add diagnostic? 1145 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n"); 1146 return 0; 1147 } 1148 1149 // Reconstruct the solution. 1150 if (!DryRun) 1151 reconstructPath(InitialState, Queue.top().second); 1152 1153 LLVM_DEBUG(llvm::dbgs() 1154 << "Total number of analyzed states: " << Count << "\n"); 1155 LLVM_DEBUG(llvm::dbgs() << "---\n"); 1156 1157 return Penalty; 1158 } 1159 1160 /// Add the following state to the analysis queue \c Queue. 1161 /// 1162 /// Assume the current state is \p PreviousNode and has been reached with a 1163 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true. 1164 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode, 1165 bool NewLine, unsigned *Count, QueueType *Queue) { 1166 if (NewLine && !Indenter->canBreak(PreviousNode->State)) 1167 return; 1168 if (!NewLine && Indenter->mustBreak(PreviousNode->State)) 1169 return; 1170 1171 StateNode *Node = new (Allocator.Allocate()) 1172 StateNode(PreviousNode->State, NewLine, PreviousNode); 1173 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty)) 1174 return; 1175 1176 Penalty += Indenter->addTokenToState(Node->State, NewLine, true); 1177 1178 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node)); 1179 ++(*Count); 1180 } 1181 1182 /// Applies the best formatting by reconstructing the path in the 1183 /// solution space that leads to \c Best. 1184 void reconstructPath(LineState &State, StateNode *Best) { 1185 llvm::SmallVector<StateNode *> Path; 1186 // We do not need a break before the initial token. 1187 while (Best->Previous) { 1188 Path.push_back(Best); 1189 Best = Best->Previous; 1190 } 1191 for (const auto &Node : llvm::reverse(Path)) { 1192 unsigned Penalty = 0; 1193 formatChildren(State, Node->NewLine, /*DryRun=*/false, Penalty); 1194 Penalty += Indenter->addTokenToState(State, Node->NewLine, false); 1195 1196 LLVM_DEBUG({ 1197 printLineState(Node->Previous->State); 1198 if (Node->NewLine) 1199 llvm::dbgs() << "Penalty for placing " 1200 << Node->Previous->State.NextToken->Tok.getName() 1201 << " on a new line: " << Penalty << "\n"; 1202 }); 1203 } 1204 } 1205 1206 llvm::SpecificBumpPtrAllocator<StateNode> Allocator; 1207 }; 1208 1209 } // anonymous namespace 1210 1211 unsigned UnwrappedLineFormatter::format( 1212 const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun, 1213 int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn, 1214 unsigned NextStartColumn, unsigned LastStartColumn) { 1215 LineJoiner Joiner(Style, Keywords, Lines); 1216 1217 // Try to look up already computed penalty in DryRun-mode. 1218 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey( 1219 &Lines, AdditionalIndent); 1220 auto CacheIt = PenaltyCache.find(CacheKey); 1221 if (DryRun && CacheIt != PenaltyCache.end()) 1222 return CacheIt->second; 1223 1224 assert(!Lines.empty()); 1225 unsigned Penalty = 0; 1226 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level, 1227 AdditionalIndent); 1228 const AnnotatedLine *PrevPrevLine = nullptr; 1229 const AnnotatedLine *PreviousLine = nullptr; 1230 const AnnotatedLine *NextLine = nullptr; 1231 1232 // The minimum level of consecutive lines that have been formatted. 1233 unsigned RangeMinLevel = UINT_MAX; 1234 1235 bool FirstLine = true; 1236 for (const AnnotatedLine *Line = 1237 Joiner.getNextMergedLine(DryRun, IndentTracker); 1238 Line; PrevPrevLine = PreviousLine, PreviousLine = Line, Line = NextLine, 1239 FirstLine = false) { 1240 assert(Line->First); 1241 const AnnotatedLine &TheLine = *Line; 1242 unsigned Indent = IndentTracker.getIndent(); 1243 1244 // We continue formatting unchanged lines to adjust their indent, e.g. if a 1245 // scope was added. However, we need to carefully stop doing this when we 1246 // exit the scope of affected lines to prevent indenting a the entire 1247 // remaining file if it currently missing a closing brace. 1248 bool PreviousRBrace = 1249 PreviousLine && PreviousLine->startsWith(tok::r_brace); 1250 bool ContinueFormatting = 1251 TheLine.Level > RangeMinLevel || 1252 (TheLine.Level == RangeMinLevel && !PreviousRBrace && 1253 !TheLine.startsWith(tok::r_brace)); 1254 1255 bool FixIndentation = (FixBadIndentation || ContinueFormatting) && 1256 Indent != TheLine.First->OriginalColumn; 1257 bool ShouldFormat = TheLine.Affected || FixIndentation; 1258 // We cannot format this line; if the reason is that the line had a 1259 // parsing error, remember that. 1260 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) { 1261 Status->FormatComplete = false; 1262 Status->Line = 1263 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation()); 1264 } 1265 1266 if (ShouldFormat && TheLine.Type != LT_Invalid) { 1267 if (!DryRun) { 1268 bool LastLine = TheLine.First->is(tok::eof); 1269 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, Indent, 1270 LastLine ? LastStartColumn : NextStartColumn + Indent); 1271 } 1272 1273 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1274 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine); 1275 bool FitsIntoOneLine = 1276 TheLine.Last->TotalLength + Indent <= ColumnLimit || 1277 (TheLine.Type == LT_ImportStatement && 1278 (!Style.isJavaScript() || !Style.JavaScriptWrapImports)) || 1279 (Style.isCSharp() && 1280 TheLine.InPPDirective); // don't split #regions in C# 1281 if (Style.ColumnLimit == 0) 1282 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this) 1283 .formatLine(TheLine, NextStartColumn + Indent, 1284 FirstLine ? FirstStartColumn : 0, DryRun); 1285 else if (FitsIntoOneLine) 1286 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this) 1287 .formatLine(TheLine, NextStartColumn + Indent, 1288 FirstLine ? FirstStartColumn : 0, DryRun); 1289 else 1290 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this) 1291 .formatLine(TheLine, NextStartColumn + Indent, 1292 FirstLine ? FirstStartColumn : 0, DryRun); 1293 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level); 1294 } else { 1295 // If no token in the current line is affected, we still need to format 1296 // affected children. 1297 if (TheLine.ChildrenAffected) 1298 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) 1299 if (!Tok->Children.empty()) 1300 format(Tok->Children, DryRun); 1301 1302 // Adapt following lines on the current indent level to the same level 1303 // unless the current \c AnnotatedLine is not at the beginning of a line. 1304 bool StartsNewLine = 1305 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst; 1306 if (StartsNewLine) 1307 IndentTracker.adjustToUnmodifiedLine(TheLine); 1308 if (!DryRun) { 1309 bool ReformatLeadingWhitespace = 1310 StartsNewLine && ((PreviousLine && PreviousLine->Affected) || 1311 TheLine.LeadingEmptyLinesAffected); 1312 // Format the first token. 1313 if (ReformatLeadingWhitespace) 1314 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, 1315 TheLine.First->OriginalColumn, 1316 TheLine.First->OriginalColumn); 1317 else 1318 Whitespaces->addUntouchableToken(*TheLine.First, 1319 TheLine.InPPDirective); 1320 1321 // Notify the WhitespaceManager about the unchanged whitespace. 1322 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next) 1323 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective); 1324 } 1325 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1326 RangeMinLevel = UINT_MAX; 1327 } 1328 if (!DryRun) 1329 markFinalized(TheLine.First); 1330 } 1331 PenaltyCache[CacheKey] = Penalty; 1332 return Penalty; 1333 } 1334 1335 void UnwrappedLineFormatter::formatFirstToken( 1336 const AnnotatedLine &Line, const AnnotatedLine *PreviousLine, 1337 const AnnotatedLine *PrevPrevLine, 1338 const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent, 1339 unsigned NewlineIndent) { 1340 FormatToken &RootToken = *Line.First; 1341 if (RootToken.is(tok::eof)) { 1342 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u); 1343 unsigned TokenIndent = Newlines ? NewlineIndent : 0; 1344 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent, 1345 TokenIndent); 1346 return; 1347 } 1348 unsigned Newlines = 1349 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1); 1350 // Remove empty lines before "}" where applicable. 1351 if (RootToken.is(tok::r_brace) && 1352 (!RootToken.Next || 1353 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) && 1354 // Do not remove empty lines before namespace closing "}". 1355 !getNamespaceToken(&Line, Lines)) 1356 Newlines = std::min(Newlines, 1u); 1357 // Remove empty lines at the start of nested blocks (lambdas/arrow functions) 1358 if (PreviousLine == nullptr && Line.Level > 0) 1359 Newlines = std::min(Newlines, 1u); 1360 if (Newlines == 0 && !RootToken.IsFirst) 1361 Newlines = 1; 1362 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline) 1363 Newlines = 0; 1364 1365 // Remove empty lines after "{". 1366 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine && 1367 PreviousLine->Last->is(tok::l_brace) && 1368 !PreviousLine->startsWithNamespace() && 1369 !(PrevPrevLine && PrevPrevLine->startsWithNamespace() && 1370 PreviousLine->startsWith(tok::l_brace)) && 1371 !startsExternCBlock(*PreviousLine)) 1372 Newlines = 1; 1373 1374 // Insert or remove empty line before access specifiers. 1375 if (PreviousLine && RootToken.isAccessSpecifier()) { 1376 switch (Style.EmptyLineBeforeAccessModifier) { 1377 case FormatStyle::ELBAMS_Never: 1378 if (Newlines > 1) 1379 Newlines = 1; 1380 break; 1381 case FormatStyle::ELBAMS_Leave: 1382 Newlines = std::max(RootToken.NewlinesBefore, 1u); 1383 break; 1384 case FormatStyle::ELBAMS_LogicalBlock: 1385 if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1) 1386 Newlines = 2; 1387 if (PreviousLine->First->isAccessSpecifier()) 1388 Newlines = 1; // Previous is an access modifier remove all new lines. 1389 break; 1390 case FormatStyle::ELBAMS_Always: { 1391 const FormatToken *previousToken; 1392 if (PreviousLine->Last->is(tok::comment)) 1393 previousToken = PreviousLine->Last->getPreviousNonComment(); 1394 else 1395 previousToken = PreviousLine->Last; 1396 if ((!previousToken || !previousToken->is(tok::l_brace)) && Newlines <= 1) 1397 Newlines = 2; 1398 } break; 1399 } 1400 } 1401 1402 // Insert or remove empty line after access specifiers. 1403 if (PreviousLine && PreviousLine->First->isAccessSpecifier() && 1404 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) { 1405 // EmptyLineBeforeAccessModifier is handling the case when two access 1406 // modifiers follow each other. 1407 if (!RootToken.isAccessSpecifier()) { 1408 switch (Style.EmptyLineAfterAccessModifier) { 1409 case FormatStyle::ELAAMS_Never: 1410 Newlines = 1; 1411 break; 1412 case FormatStyle::ELAAMS_Leave: 1413 Newlines = std::max(Newlines, 1u); 1414 break; 1415 case FormatStyle::ELAAMS_Always: 1416 if (RootToken.is(tok::r_brace)) // Do not add at end of class. 1417 Newlines = 1u; 1418 else 1419 Newlines = std::max(Newlines, 2u); 1420 break; 1421 } 1422 } 1423 } 1424 1425 if (Newlines) 1426 Indent = NewlineIndent; 1427 1428 // Preprocessor directives get indented before the hash only if specified 1429 if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 1430 (Line.Type == LT_PreprocessorDirective || 1431 Line.Type == LT_ImportStatement)) 1432 Indent = 0; 1433 1434 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent, 1435 /*IsAligned=*/false, 1436 Line.InPPDirective && 1437 !RootToken.HasUnescapedNewline); 1438 } 1439 1440 unsigned 1441 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective, 1442 const AnnotatedLine *NextLine) const { 1443 // In preprocessor directives reserve two chars for trailing " \" if the 1444 // next line continues the preprocessor directive. 1445 bool ContinuesPPDirective = 1446 InPPDirective && 1447 // If there is no next line, this is likely a child line and the parent 1448 // continues the preprocessor directive. 1449 (!NextLine || 1450 (NextLine->InPPDirective && 1451 // If there is an unescaped newline between this line and the next, the 1452 // next line starts a new preprocessor directive. 1453 !NextLine->First->HasUnescapedNewline)); 1454 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0); 1455 } 1456 1457 } // namespace format 1458 } // namespace clang 1459