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 overridden 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_else, tok::kw_while, 324 tok::kw_for, tok::kw_switch, tok::kw_try, 325 tok::kw_do, 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_else, 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->is(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_else, tok::kw_while, tok::kw_do, 582 tok::kw_try, tok::kw___try, tok::kw_catch, 583 tok::kw___finally, tok::kw_for, tok::r_brace, 584 Keywords.kw___except)) { 585 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 586 return 0; 587 // Don't merge when we can't except the case when 588 // the control statement block is empty 589 if (!Style.AllowShortIfStatementsOnASingleLine && 590 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 591 !Style.BraceWrapping.AfterControlStatement && 592 !I[1]->First->is(tok::r_brace)) 593 return 0; 594 if (!Style.AllowShortIfStatementsOnASingleLine && 595 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 596 Style.BraceWrapping.AfterControlStatement == 597 FormatStyle::BWACS_Always && 598 I + 2 != E && !I[2]->First->is(tok::r_brace)) 599 return 0; 600 if (!Style.AllowShortLoopsOnASingleLine && 601 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) && 602 !Style.BraceWrapping.AfterControlStatement && 603 !I[1]->First->is(tok::r_brace)) 604 return 0; 605 if (!Style.AllowShortLoopsOnASingleLine && 606 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for) && 607 Style.BraceWrapping.AfterControlStatement == 608 FormatStyle::BWACS_Always && 609 I + 2 != E && !I[2]->First->is(tok::r_brace)) 610 return 0; 611 // FIXME: Consider an option to allow short exception handling clauses on 612 // a single line. 613 // FIXME: This isn't covered by tests. 614 // FIXME: For catch, __except, __finally the first token on the line 615 // is '}', so this isn't correct here. 616 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch, 617 Keywords.kw___except, tok::kw___finally)) 618 return 0; 619 } 620 621 if (Line.Last->is(tok::l_brace)) { 622 FormatToken *Tok = I[1]->First; 623 if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore && 624 (Tok->getNextNonComment() == nullptr || 625 Tok->getNextNonComment()->is(tok::semi))) { 626 // We merge empty blocks even if the line exceeds the column limit. 627 Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0; 628 Tok->CanBreakBefore = true; 629 return 1; 630 } else if (Limit != 0 && !Line.startsWithNamespace() && 631 !startsExternCBlock(Line)) { 632 // We don't merge short records. 633 FormatToken *RecordTok = Line.First; 634 // Skip record modifiers. 635 while (RecordTok->Next && 636 RecordTok->isOneOf(tok::kw_typedef, tok::kw_export, 637 Keywords.kw_declare, Keywords.kw_abstract, 638 tok::kw_default, Keywords.kw_override, 639 tok::kw_public, tok::kw_private, 640 tok::kw_protected, Keywords.kw_internal)) 641 RecordTok = RecordTok->Next; 642 if (RecordTok && 643 RecordTok->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct, 644 Keywords.kw_interface)) 645 return 0; 646 647 // Check that we still have three lines and they fit into the limit. 648 if (I + 2 == E || I[2]->Type == LT_Invalid) 649 return 0; 650 Limit = limitConsideringMacros(I + 2, E, Limit); 651 652 if (!nextTwoLinesFitInto(I, Limit)) 653 return 0; 654 655 // Second, check that the next line does not contain any braces - if it 656 // does, readability declines when putting it into a single line. 657 if (I[1]->Last->is(TT_LineComment)) 658 return 0; 659 do { 660 if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit)) 661 return 0; 662 Tok = Tok->Next; 663 } while (Tok); 664 665 // Last, check that the third line starts with a closing brace. 666 Tok = I[2]->First; 667 if (Tok->isNot(tok::r_brace)) 668 return 0; 669 670 // Don't merge "if (a) { .. } else {". 671 if (Tok->Next && Tok->Next->is(tok::kw_else)) 672 return 0; 673 674 // Don't merge a trailing multi-line control statement block like: 675 // } else if (foo && 676 // bar) 677 // { <-- current Line 678 // baz(); 679 // } 680 if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) && 681 Style.BraceWrapping.AfterControlStatement == 682 FormatStyle::BWACS_MultiLine) 683 return 0; 684 685 return 2; 686 } 687 } else if (I[1]->First->is(tok::l_brace)) { 688 if (I[1]->Last->is(TT_LineComment)) 689 return 0; 690 691 // Check for Limit <= 2 to account for the " {". 692 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I))) 693 return 0; 694 Limit -= 2; 695 unsigned MergedLines = 0; 696 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never || 697 (I[1]->First == I[1]->Last && I + 2 != E && 698 I[2]->First->is(tok::r_brace))) { 699 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); 700 // If we managed to merge the block, count the statement header, which 701 // is on a separate line. 702 if (MergedLines > 0) 703 ++MergedLines; 704 } 705 return MergedLines; 706 } 707 return 0; 708 } 709 710 /// Returns the modified column limit for \p I if it is inside a macro and 711 /// needs a trailing '\'. 712 unsigned 713 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 714 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 715 unsigned Limit) { 716 if (I[0]->InPPDirective && I + 1 != E && 717 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) { 718 return Limit < 2 ? 0 : Limit - 2; 719 } 720 return Limit; 721 } 722 723 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 724 unsigned Limit) { 725 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore) 726 return false; 727 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit; 728 } 729 730 bool containsMustBreak(const AnnotatedLine *Line) { 731 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) { 732 if (Tok->MustBreakBefore) 733 return true; 734 } 735 return false; 736 } 737 738 void join(AnnotatedLine &A, const AnnotatedLine &B) { 739 assert(!A.Last->Next); 740 assert(!B.First->Previous); 741 if (B.Affected) 742 A.Affected = true; 743 A.Last->Next = B.First; 744 B.First->Previous = A.Last; 745 B.First->CanBreakBefore = true; 746 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore; 747 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) { 748 Tok->TotalLength += LengthA; 749 A.Last = Tok; 750 } 751 } 752 753 const FormatStyle &Style; 754 const AdditionalKeywords &Keywords; 755 const SmallVectorImpl<AnnotatedLine *>::const_iterator End; 756 757 SmallVectorImpl<AnnotatedLine *>::const_iterator Next; 758 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines; 759 }; 760 761 static void markFinalized(FormatToken *Tok) { 762 for (; Tok; Tok = Tok->Next) { 763 Tok->Finalized = true; 764 for (AnnotatedLine *Child : Tok->Children) 765 markFinalized(Child->First); 766 } 767 } 768 769 #ifndef NDEBUG 770 static void printLineState(const LineState &State) { 771 llvm::dbgs() << "State: "; 772 for (const ParenState &P : State.Stack) { 773 llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|" 774 << P.LastSpace << "|" << P.NestedBlockIndent << " "; 775 } 776 llvm::dbgs() << State.NextToken->TokenText << "\n"; 777 } 778 #endif 779 780 /// Base class for classes that format one \c AnnotatedLine. 781 class LineFormatter { 782 public: 783 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces, 784 const FormatStyle &Style, 785 UnwrappedLineFormatter *BlockFormatter) 786 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style), 787 BlockFormatter(BlockFormatter) {} 788 virtual ~LineFormatter() {} 789 790 /// Formats an \c AnnotatedLine and returns the penalty. 791 /// 792 /// If \p DryRun is \c false, directly applies the changes. 793 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 794 unsigned FirstStartColumn, bool DryRun) = 0; 795 796 protected: 797 /// If the \p State's next token is an r_brace closing a nested block, 798 /// format the nested block before it. 799 /// 800 /// Returns \c true if all children could be placed successfully and adapts 801 /// \p Penalty as well as \p State. If \p DryRun is false, also directly 802 /// creates changes using \c Whitespaces. 803 /// 804 /// The crucial idea here is that children always get formatted upon 805 /// encountering the closing brace right after the nested block. Now, if we 806 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is 807 /// \c false), the entire block has to be kept on the same line (which is only 808 /// possible if it fits on the line, only contains a single statement, etc. 809 /// 810 /// If \p NewLine is true, we format the nested block on separate lines, i.e. 811 /// break after the "{", format all lines with correct indentation and the put 812 /// the closing "}" on yet another new line. 813 /// 814 /// This enables us to keep the simple structure of the 815 /// \c UnwrappedLineFormatter, where we only have two options for each token: 816 /// break or don't break. 817 bool formatChildren(LineState &State, bool NewLine, bool DryRun, 818 unsigned &Penalty) { 819 const FormatToken *LBrace = State.NextToken->getPreviousNonComment(); 820 FormatToken &Previous = *State.NextToken->Previous; 821 if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) || 822 Previous.Children.size() == 0) 823 // The previous token does not open a block. Nothing to do. We don't 824 // assert so that we can simply call this function for all tokens. 825 return true; 826 827 if (NewLine) { 828 const ParenState &P = State.Stack.back(); 829 830 int AdditionalIndent = 831 P.Indent - Previous.Children[0]->Level * Style.IndentWidth; 832 833 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && 834 P.NestedBlockIndent == P.LastSpace) { 835 if (State.NextToken->MatchingParen && 836 State.NextToken->MatchingParen->is(TT_LambdaLBrace)) { 837 State.Stack.pop_back(); 838 } 839 if (LBrace->is(TT_LambdaLBrace)) 840 AdditionalIndent = 0; 841 } 842 843 Penalty += 844 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent, 845 /*FixBadIndentation=*/true); 846 return true; 847 } 848 849 if (Previous.Children[0]->First->MustBreakBefore) 850 return false; 851 852 // Cannot merge into one line if this line ends on a comment. 853 if (Previous.is(tok::comment)) 854 return false; 855 856 // Cannot merge multiple statements into a single line. 857 if (Previous.Children.size() > 1) 858 return false; 859 860 const AnnotatedLine *Child = Previous.Children[0]; 861 // We can't put the closing "}" on a line with a trailing comment. 862 if (Child->Last->isTrailingComment()) 863 return false; 864 865 // If the child line exceeds the column limit, we wouldn't want to merge it. 866 // We add +2 for the trailing " }". 867 if (Style.ColumnLimit > 0 && 868 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit) 869 return false; 870 871 if (!DryRun) { 872 Whitespaces->replaceWhitespace( 873 *Child->First, /*Newlines=*/0, /*Spaces=*/1, 874 /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false, 875 State.Line->InPPDirective); 876 } 877 Penalty += 878 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun); 879 880 State.Column += 1 + Child->Last->TotalLength; 881 return true; 882 } 883 884 ContinuationIndenter *Indenter; 885 886 private: 887 WhitespaceManager *Whitespaces; 888 const FormatStyle &Style; 889 UnwrappedLineFormatter *BlockFormatter; 890 }; 891 892 /// Formatter that keeps the existing line breaks. 893 class NoColumnLimitLineFormatter : public LineFormatter { 894 public: 895 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter, 896 WhitespaceManager *Whitespaces, 897 const FormatStyle &Style, 898 UnwrappedLineFormatter *BlockFormatter) 899 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 900 901 /// Formats the line, simply keeping all of the input's line breaking 902 /// decisions. 903 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 904 unsigned FirstStartColumn, bool DryRun) override { 905 assert(!DryRun); 906 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn, 907 &Line, /*DryRun=*/false); 908 while (State.NextToken) { 909 bool Newline = 910 Indenter->mustBreak(State) || 911 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0); 912 unsigned Penalty = 0; 913 formatChildren(State, Newline, /*DryRun=*/false, Penalty); 914 Indenter->addTokenToState(State, Newline, /*DryRun=*/false); 915 } 916 return 0; 917 } 918 }; 919 920 /// Formatter that puts all tokens into a single line without breaks. 921 class NoLineBreakFormatter : public LineFormatter { 922 public: 923 NoLineBreakFormatter(ContinuationIndenter *Indenter, 924 WhitespaceManager *Whitespaces, const FormatStyle &Style, 925 UnwrappedLineFormatter *BlockFormatter) 926 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 927 928 /// Puts all tokens into a single line. 929 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 930 unsigned FirstStartColumn, bool DryRun) override { 931 unsigned Penalty = 0; 932 LineState State = 933 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 934 while (State.NextToken) { 935 formatChildren(State, /*NewLine=*/false, DryRun, Penalty); 936 Indenter->addTokenToState( 937 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun); 938 } 939 return Penalty; 940 } 941 }; 942 943 /// Finds the best way to break lines. 944 class OptimizingLineFormatter : public LineFormatter { 945 public: 946 OptimizingLineFormatter(ContinuationIndenter *Indenter, 947 WhitespaceManager *Whitespaces, 948 const FormatStyle &Style, 949 UnwrappedLineFormatter *BlockFormatter) 950 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 951 952 /// Formats the line by finding the best line breaks with line lengths 953 /// below the column limit. 954 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 955 unsigned FirstStartColumn, bool DryRun) override { 956 LineState State = 957 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 958 959 // If the ObjC method declaration does not fit on a line, we should format 960 // it with one arg per line. 961 if (State.Line->Type == LT_ObjCMethodDecl) 962 State.Stack.back().BreakBeforeParameter = true; 963 964 // Find best solution in solution space. 965 return analyzeSolutionSpace(State, DryRun); 966 } 967 968 private: 969 struct CompareLineStatePointers { 970 bool operator()(LineState *obj1, LineState *obj2) const { 971 return *obj1 < *obj2; 972 } 973 }; 974 975 /// A pair of <penalty, count> that is used to prioritize the BFS on. 976 /// 977 /// In case of equal penalties, we want to prefer states that were inserted 978 /// first. During state generation we make sure that we insert states first 979 /// that break the line as late as possible. 980 typedef std::pair<unsigned, unsigned> OrderedPenalty; 981 982 /// An edge in the solution space from \c Previous->State to \c State, 983 /// inserting a newline dependent on the \c NewLine. 984 struct StateNode { 985 StateNode(const LineState &State, bool NewLine, StateNode *Previous) 986 : State(State), NewLine(NewLine), Previous(Previous) {} 987 LineState State; 988 bool NewLine; 989 StateNode *Previous; 990 }; 991 992 /// An item in the prioritized BFS search queue. The \c StateNode's 993 /// \c State has the given \c OrderedPenalty. 994 typedef std::pair<OrderedPenalty, StateNode *> QueueItem; 995 996 /// The BFS queue type. 997 typedef std::priority_queue<QueueItem, std::vector<QueueItem>, 998 std::greater<QueueItem>> 999 QueueType; 1000 1001 /// Analyze the entire solution space starting from \p InitialState. 1002 /// 1003 /// This implements a variant of Dijkstra's algorithm on the graph that spans 1004 /// the solution space (\c LineStates are the nodes). The algorithm tries to 1005 /// find the shortest path (the one with lowest penalty) from \p InitialState 1006 /// to a state where all tokens are placed. Returns the penalty. 1007 /// 1008 /// If \p DryRun is \c false, directly applies the changes. 1009 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) { 1010 std::set<LineState *, CompareLineStatePointers> Seen; 1011 1012 // Increasing count of \c StateNode items we have created. This is used to 1013 // create a deterministic order independent of the container. 1014 unsigned Count = 0; 1015 QueueType Queue; 1016 1017 // Insert start element into queue. 1018 StateNode *Node = 1019 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr); 1020 Queue.push(QueueItem(OrderedPenalty(0, Count), Node)); 1021 ++Count; 1022 1023 unsigned Penalty = 0; 1024 1025 // While not empty, take first element and follow edges. 1026 while (!Queue.empty()) { 1027 Penalty = Queue.top().first.first; 1028 StateNode *Node = Queue.top().second; 1029 if (!Node->State.NextToken) { 1030 LLVM_DEBUG(llvm::dbgs() 1031 << "\n---\nPenalty for line: " << Penalty << "\n"); 1032 break; 1033 } 1034 Queue.pop(); 1035 1036 // Cut off the analysis of certain solutions if the analysis gets too 1037 // complex. See description of IgnoreStackForComparison. 1038 if (Count > 50000) 1039 Node->State.IgnoreStackForComparison = true; 1040 1041 if (!Seen.insert(&Node->State).second) 1042 // State already examined with lower penalty. 1043 continue; 1044 1045 FormatDecision LastFormat = Node->State.NextToken->getDecision(); 1046 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue) 1047 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue); 1048 if (LastFormat == FD_Unformatted || LastFormat == FD_Break) 1049 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue); 1050 } 1051 1052 if (Queue.empty()) { 1053 // We were unable to find a solution, do nothing. 1054 // FIXME: Add diagnostic? 1055 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n"); 1056 return 0; 1057 } 1058 1059 // Reconstruct the solution. 1060 if (!DryRun) 1061 reconstructPath(InitialState, Queue.top().second); 1062 1063 LLVM_DEBUG(llvm::dbgs() 1064 << "Total number of analyzed states: " << Count << "\n"); 1065 LLVM_DEBUG(llvm::dbgs() << "---\n"); 1066 1067 return Penalty; 1068 } 1069 1070 /// Add the following state to the analysis queue \c Queue. 1071 /// 1072 /// Assume the current state is \p PreviousNode and has been reached with a 1073 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true. 1074 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode, 1075 bool NewLine, unsigned *Count, QueueType *Queue) { 1076 if (NewLine && !Indenter->canBreak(PreviousNode->State)) 1077 return; 1078 if (!NewLine && Indenter->mustBreak(PreviousNode->State)) 1079 return; 1080 1081 StateNode *Node = new (Allocator.Allocate()) 1082 StateNode(PreviousNode->State, NewLine, PreviousNode); 1083 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty)) 1084 return; 1085 1086 Penalty += Indenter->addTokenToState(Node->State, NewLine, true); 1087 1088 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node)); 1089 ++(*Count); 1090 } 1091 1092 /// Applies the best formatting by reconstructing the path in the 1093 /// solution space that leads to \c Best. 1094 void reconstructPath(LineState &State, StateNode *Best) { 1095 std::deque<StateNode *> Path; 1096 // We do not need a break before the initial token. 1097 while (Best->Previous) { 1098 Path.push_front(Best); 1099 Best = Best->Previous; 1100 } 1101 for (auto I = Path.begin(), E = Path.end(); I != E; ++I) { 1102 unsigned Penalty = 0; 1103 formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty); 1104 Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false); 1105 1106 LLVM_DEBUG({ 1107 printLineState((*I)->Previous->State); 1108 if ((*I)->NewLine) { 1109 llvm::dbgs() << "Penalty for placing " 1110 << (*I)->Previous->State.NextToken->Tok.getName() 1111 << " on a new line: " << Penalty << "\n"; 1112 } 1113 }); 1114 } 1115 } 1116 1117 llvm::SpecificBumpPtrAllocator<StateNode> Allocator; 1118 }; 1119 1120 } // anonymous namespace 1121 1122 unsigned UnwrappedLineFormatter::format( 1123 const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun, 1124 int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn, 1125 unsigned NextStartColumn, unsigned LastStartColumn) { 1126 LineJoiner Joiner(Style, Keywords, Lines); 1127 1128 // Try to look up already computed penalty in DryRun-mode. 1129 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey( 1130 &Lines, AdditionalIndent); 1131 auto CacheIt = PenaltyCache.find(CacheKey); 1132 if (DryRun && CacheIt != PenaltyCache.end()) 1133 return CacheIt->second; 1134 1135 assert(!Lines.empty()); 1136 unsigned Penalty = 0; 1137 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level, 1138 AdditionalIndent); 1139 const AnnotatedLine *PrevPrevLine = nullptr; 1140 const AnnotatedLine *PreviousLine = nullptr; 1141 const AnnotatedLine *NextLine = nullptr; 1142 1143 // The minimum level of consecutive lines that have been formatted. 1144 unsigned RangeMinLevel = UINT_MAX; 1145 1146 bool FirstLine = true; 1147 for (const AnnotatedLine *Line = 1148 Joiner.getNextMergedLine(DryRun, IndentTracker); 1149 Line; Line = NextLine, FirstLine = false) { 1150 const AnnotatedLine &TheLine = *Line; 1151 unsigned Indent = IndentTracker.getIndent(); 1152 1153 // We continue formatting unchanged lines to adjust their indent, e.g. if a 1154 // scope was added. However, we need to carefully stop doing this when we 1155 // exit the scope of affected lines to prevent indenting a the entire 1156 // remaining file if it currently missing a closing brace. 1157 bool PreviousRBrace = 1158 PreviousLine && PreviousLine->startsWith(tok::r_brace); 1159 bool ContinueFormatting = 1160 TheLine.Level > RangeMinLevel || 1161 (TheLine.Level == RangeMinLevel && !PreviousRBrace && 1162 !TheLine.startsWith(tok::r_brace)); 1163 1164 bool FixIndentation = (FixBadIndentation || ContinueFormatting) && 1165 Indent != TheLine.First->OriginalColumn; 1166 bool ShouldFormat = TheLine.Affected || FixIndentation; 1167 // We cannot format this line; if the reason is that the line had a 1168 // parsing error, remember that. 1169 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) { 1170 Status->FormatComplete = false; 1171 Status->Line = 1172 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation()); 1173 } 1174 1175 if (ShouldFormat && TheLine.Type != LT_Invalid) { 1176 if (!DryRun) { 1177 bool LastLine = Line->First->is(tok::eof); 1178 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, Indent, 1179 LastLine ? LastStartColumn : NextStartColumn + Indent); 1180 } 1181 1182 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1183 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine); 1184 bool FitsIntoOneLine = 1185 TheLine.Last->TotalLength + Indent <= ColumnLimit || 1186 (TheLine.Type == LT_ImportStatement && 1187 (Style.Language != FormatStyle::LK_JavaScript || 1188 !Style.JavaScriptWrapImports)) || 1189 (Style.isCSharp() && 1190 TheLine.InPPDirective); // don't split #regions in C# 1191 if (Style.ColumnLimit == 0) 1192 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this) 1193 .formatLine(TheLine, NextStartColumn + Indent, 1194 FirstLine ? FirstStartColumn : 0, DryRun); 1195 else if (FitsIntoOneLine) 1196 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this) 1197 .formatLine(TheLine, NextStartColumn + Indent, 1198 FirstLine ? FirstStartColumn : 0, DryRun); 1199 else 1200 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this) 1201 .formatLine(TheLine, NextStartColumn + Indent, 1202 FirstLine ? FirstStartColumn : 0, DryRun); 1203 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level); 1204 } else { 1205 // If no token in the current line is affected, we still need to format 1206 // affected children. 1207 if (TheLine.ChildrenAffected) 1208 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) 1209 if (!Tok->Children.empty()) 1210 format(Tok->Children, DryRun); 1211 1212 // Adapt following lines on the current indent level to the same level 1213 // unless the current \c AnnotatedLine is not at the beginning of a line. 1214 bool StartsNewLine = 1215 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst; 1216 if (StartsNewLine) 1217 IndentTracker.adjustToUnmodifiedLine(TheLine); 1218 if (!DryRun) { 1219 bool ReformatLeadingWhitespace = 1220 StartsNewLine && ((PreviousLine && PreviousLine->Affected) || 1221 TheLine.LeadingEmptyLinesAffected); 1222 // Format the first token. 1223 if (ReformatLeadingWhitespace) 1224 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, 1225 TheLine.First->OriginalColumn, 1226 TheLine.First->OriginalColumn); 1227 else 1228 Whitespaces->addUntouchableToken(*TheLine.First, 1229 TheLine.InPPDirective); 1230 1231 // Notify the WhitespaceManager about the unchanged whitespace. 1232 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next) 1233 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective); 1234 } 1235 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1236 RangeMinLevel = UINT_MAX; 1237 } 1238 if (!DryRun) 1239 markFinalized(TheLine.First); 1240 PrevPrevLine = PreviousLine; 1241 PreviousLine = &TheLine; 1242 } 1243 PenaltyCache[CacheKey] = Penalty; 1244 return Penalty; 1245 } 1246 1247 void UnwrappedLineFormatter::formatFirstToken( 1248 const AnnotatedLine &Line, const AnnotatedLine *PreviousLine, 1249 const AnnotatedLine *PrevPrevLine, 1250 const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent, 1251 unsigned NewlineIndent) { 1252 FormatToken &RootToken = *Line.First; 1253 if (RootToken.is(tok::eof)) { 1254 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u); 1255 unsigned TokenIndent = Newlines ? NewlineIndent : 0; 1256 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent, 1257 TokenIndent); 1258 return; 1259 } 1260 unsigned Newlines = 1261 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1); 1262 // Remove empty lines before "}" where applicable. 1263 if (RootToken.is(tok::r_brace) && 1264 (!RootToken.Next || 1265 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) && 1266 // Do not remove empty lines before namespace closing "}". 1267 !getNamespaceToken(&Line, Lines)) 1268 Newlines = std::min(Newlines, 1u); 1269 // Remove empty lines at the start of nested blocks (lambdas/arrow functions) 1270 if (PreviousLine == nullptr && Line.Level > 0) 1271 Newlines = std::min(Newlines, 1u); 1272 if (Newlines == 0 && !RootToken.IsFirst) 1273 Newlines = 1; 1274 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline) 1275 Newlines = 0; 1276 1277 // Remove empty lines after "{". 1278 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine && 1279 PreviousLine->Last->is(tok::l_brace) && 1280 !PreviousLine->startsWithNamespace() && 1281 !(PrevPrevLine && PrevPrevLine->startsWithNamespace() && 1282 PreviousLine->startsWith(tok::l_brace)) && 1283 !startsExternCBlock(*PreviousLine)) 1284 Newlines = 1; 1285 1286 // Insert or remove empty line before access specifiers. 1287 if (PreviousLine && RootToken.isAccessSpecifier()) { 1288 switch (Style.EmptyLineBeforeAccessModifier) { 1289 case FormatStyle::ELBAMS_Never: 1290 if (Newlines > 1) 1291 Newlines = 1; 1292 break; 1293 case FormatStyle::ELBAMS_Leave: 1294 Newlines = std::max(RootToken.NewlinesBefore, 1u); 1295 break; 1296 case FormatStyle::ELBAMS_LogicalBlock: 1297 if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1) 1298 Newlines = 2; 1299 if (PreviousLine->First->isAccessSpecifier()) 1300 Newlines = 1; // Previous is an access modifier remove all new lines. 1301 break; 1302 case FormatStyle::ELBAMS_Always: { 1303 const FormatToken *previousToken; 1304 if (PreviousLine->Last->is(tok::comment)) 1305 previousToken = PreviousLine->Last->getPreviousNonComment(); 1306 else 1307 previousToken = PreviousLine->Last; 1308 if ((!previousToken || !previousToken->is(tok::l_brace)) && Newlines <= 1) 1309 Newlines = 2; 1310 } break; 1311 } 1312 } 1313 1314 // Insert or remove empty line after access specifiers. 1315 if (PreviousLine && PreviousLine->First->isAccessSpecifier() && 1316 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) { 1317 // EmptyLineBeforeAccessModifier is handling the case when two access 1318 // modifiers follow each other. 1319 if (!RootToken.isAccessSpecifier()) { 1320 switch (Style.EmptyLineAfterAccessModifier) { 1321 case FormatStyle::ELAAMS_Never: 1322 Newlines = 1; 1323 break; 1324 case FormatStyle::ELAAMS_Leave: 1325 Newlines = std::max(Newlines, 1u); 1326 break; 1327 case FormatStyle::ELAAMS_Always: 1328 if (RootToken.is(tok::r_brace)) // Do not add at end of class. 1329 Newlines = 1u; 1330 else 1331 Newlines = std::max(Newlines, 2u); 1332 break; 1333 } 1334 } 1335 } 1336 1337 if (Newlines) 1338 Indent = NewlineIndent; 1339 1340 // Preprocessor directives get indented before the hash only if specified 1341 if (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 1342 (Line.Type == LT_PreprocessorDirective || 1343 Line.Type == LT_ImportStatement)) 1344 Indent = 0; 1345 1346 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent, 1347 /*IsAligned=*/false, 1348 Line.InPPDirective && 1349 !RootToken.HasUnescapedNewline); 1350 } 1351 1352 unsigned 1353 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective, 1354 const AnnotatedLine *NextLine) const { 1355 // In preprocessor directives reserve two chars for trailing " \" if the 1356 // next line continues the preprocessor directive. 1357 bool ContinuesPPDirective = 1358 InPPDirective && 1359 // If there is no next line, this is likely a child line and the parent 1360 // continues the preprocessor directive. 1361 (!NextLine || 1362 (NextLine->InPPDirective && 1363 // If there is an unescaped newline between this line and the next, the 1364 // next line starts a new preprocessor directive. 1365 !NextLine->First->HasUnescapedNewline)); 1366 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0); 1367 } 1368 1369 } // namespace format 1370 } // namespace clang 1371