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