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 bool isRecordLBrace(const FormatToken &Tok) { 30 return Tok.isOneOf(TT_ClassLBrace, TT_EnumLBrace, TT_RecordLBrace, 31 TT_StructLBrace, TT_UnionLBrace); 32 } 33 34 /// Tracks the indent level of \c AnnotatedLines across levels. 35 /// 36 /// \c nextLine must be called for each \c AnnotatedLine, after which \c 37 /// getIndent() will return the indent for the last line \c nextLine was called 38 /// with. 39 /// If the line is not formatted (and thus the indent does not change), calling 40 /// \c adjustToUnmodifiedLine after the call to \c nextLine will cause 41 /// subsequent lines on the same level to be indented at the same level as the 42 /// given line. 43 class LevelIndentTracker { 44 public: 45 LevelIndentTracker(const FormatStyle &Style, 46 const AdditionalKeywords &Keywords, unsigned StartLevel, 47 int AdditionalIndent) 48 : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) { 49 for (unsigned i = 0; i != StartLevel; ++i) 50 IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent); 51 } 52 53 /// Returns the indent for the current line. 54 unsigned getIndent() const { return Indent; } 55 56 /// Update the indent state given that \p Line is going to be formatted 57 /// next. 58 void nextLine(const AnnotatedLine &Line) { 59 Offset = getIndentOffset(*Line.First); 60 // Update the indent level cache size so that we can rely on it 61 // having the right size in adjustToUnmodifiedline. 62 while (IndentForLevel.size() <= Line.Level) 63 IndentForLevel.push_back(-1); 64 if (Line.InPPDirective) { 65 unsigned IndentWidth = 66 (Style.PPIndentWidth >= 0) ? Style.PPIndentWidth : Style.IndentWidth; 67 Indent = Line.Level * IndentWidth + AdditionalIndent; 68 } else { 69 IndentForLevel.resize(Line.Level + 1); 70 Indent = getIndent(Line.Level); 71 } 72 if (static_cast<int>(Indent) + Offset >= 0) 73 Indent += Offset; 74 if (Line.First->is(TT_CSharpGenericTypeConstraint)) 75 Indent = Line.Level * Style.IndentWidth + Style.ContinuationIndentWidth; 76 } 77 78 /// Update the indent state given that \p Line indent should be 79 /// skipped. 80 void skipLine(const AnnotatedLine &Line) { 81 while (IndentForLevel.size() <= Line.Level) 82 IndentForLevel.push_back(Indent); 83 } 84 85 /// Update the level indent to adapt to the given \p Line. 86 /// 87 /// When a line is not formatted, we move the subsequent lines on the same 88 /// level to the same indent. 89 /// Note that \c nextLine must have been called before this method. 90 void adjustToUnmodifiedLine(const AnnotatedLine &Line) { 91 unsigned LevelIndent = Line.First->OriginalColumn; 92 if (static_cast<int>(LevelIndent) - Offset >= 0) 93 LevelIndent -= Offset; 94 if ((!Line.First->is(tok::comment) || IndentForLevel[Line.Level] == -1) && 95 !Line.InPPDirective) 96 IndentForLevel[Line.Level] = LevelIndent; 97 } 98 99 private: 100 /// Get the offset of the line relatively to the level. 101 /// 102 /// For example, 'public:' labels in classes are offset by 1 or 2 103 /// characters to the left from their level. 104 int getIndentOffset(const FormatToken &RootToken) { 105 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() || 106 Style.isCSharp()) 107 return 0; 108 109 auto IsAccessModifier = [this, &RootToken]() { 110 if (RootToken.isAccessSpecifier(Style.isCpp())) 111 return true; 112 else if (RootToken.isObjCAccessSpecifier()) 113 return true; 114 // Handle Qt signals. 115 else if ((RootToken.isOneOf(Keywords.kw_signals, Keywords.kw_qsignals) && 116 RootToken.Next && RootToken.Next->is(tok::colon))) 117 return true; 118 else if (RootToken.Next && 119 RootToken.Next->isOneOf(Keywords.kw_slots, Keywords.kw_qslots) && 120 RootToken.Next->Next && RootToken.Next->Next->is(tok::colon)) 121 return true; 122 // Handle malformed access specifier e.g. 'private' without trailing ':'. 123 else if (!RootToken.Next && RootToken.isAccessSpecifier(false)) 124 return true; 125 return false; 126 }; 127 128 if (IsAccessModifier()) { 129 // The AccessModifierOffset may be overridden by IndentAccessModifiers, 130 // in which case we take a negative value of the IndentWidth to simulate 131 // the upper indent level. 132 return Style.IndentAccessModifiers ? -Style.IndentWidth 133 : Style.AccessModifierOffset; 134 } 135 return 0; 136 } 137 138 /// Get the indent of \p Level from \p IndentForLevel. 139 /// 140 /// \p IndentForLevel must contain the indent for the level \c l 141 /// at \p IndentForLevel[l], or a value < 0 if the indent for 142 /// that level is unknown. 143 unsigned getIndent(unsigned Level) const { 144 if (IndentForLevel[Level] != -1) 145 return IndentForLevel[Level]; 146 if (Level == 0) 147 return 0; 148 return getIndent(Level - 1) + Style.IndentWidth; 149 } 150 151 const FormatStyle &Style; 152 const AdditionalKeywords &Keywords; 153 const unsigned AdditionalIndent; 154 155 /// The indent in characters for each level. 156 std::vector<int> IndentForLevel; 157 158 /// Offset of the current line relative to the indent level. 159 /// 160 /// For example, the 'public' keywords is often indented with a negative 161 /// offset. 162 int Offset = 0; 163 164 /// The current line's indent. 165 unsigned Indent = 0; 166 }; 167 168 const FormatToken *getMatchingNamespaceToken( 169 const AnnotatedLine *Line, 170 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) { 171 if (!Line->startsWith(tok::r_brace)) 172 return nullptr; 173 size_t StartLineIndex = Line->MatchingOpeningBlockLineIndex; 174 if (StartLineIndex == UnwrappedLine::kInvalidIndex) 175 return nullptr; 176 assert(StartLineIndex < AnnotatedLines.size()); 177 return AnnotatedLines[StartLineIndex]->First->getNamespaceToken(); 178 } 179 180 StringRef getNamespaceTokenText(const AnnotatedLine *Line) { 181 const FormatToken *NamespaceToken = Line->First->getNamespaceToken(); 182 return NamespaceToken ? NamespaceToken->TokenText : StringRef(); 183 } 184 185 StringRef getMatchingNamespaceTokenText( 186 const AnnotatedLine *Line, 187 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) { 188 const FormatToken *NamespaceToken = 189 getMatchingNamespaceToken(Line, AnnotatedLines); 190 return NamespaceToken ? NamespaceToken->TokenText : StringRef(); 191 } 192 193 class LineJoiner { 194 public: 195 LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords, 196 const SmallVectorImpl<AnnotatedLine *> &Lines) 197 : Style(Style), Keywords(Keywords), End(Lines.end()), Next(Lines.begin()), 198 AnnotatedLines(Lines) {} 199 200 /// Returns the next line, merging multiple lines into one if possible. 201 const AnnotatedLine *getNextMergedLine(bool DryRun, 202 LevelIndentTracker &IndentTracker) { 203 if (Next == End) 204 return nullptr; 205 const AnnotatedLine *Current = *Next; 206 IndentTracker.nextLine(*Current); 207 unsigned MergedLines = tryFitMultipleLinesInOne(IndentTracker, Next, End); 208 if (MergedLines > 0 && Style.ColumnLimit == 0) 209 // Disallow line merging if there is a break at the start of one of the 210 // input lines. 211 for (unsigned i = 0; i < MergedLines; ++i) 212 if (Next[i + 1]->First->NewlinesBefore > 0) 213 MergedLines = 0; 214 if (!DryRun) 215 for (unsigned i = 0; i < MergedLines; ++i) 216 join(*Next[0], *Next[i + 1]); 217 Next = Next + MergedLines + 1; 218 return Current; 219 } 220 221 private: 222 /// Calculates how many lines can be merged into 1 starting at \p I. 223 unsigned 224 tryFitMultipleLinesInOne(LevelIndentTracker &IndentTracker, 225 SmallVectorImpl<AnnotatedLine *>::const_iterator I, 226 SmallVectorImpl<AnnotatedLine *>::const_iterator E) { 227 const unsigned Indent = IndentTracker.getIndent(); 228 229 // Can't join the last line with anything. 230 if (I + 1 == E) 231 return 0; 232 // We can never merge stuff if there are trailing line comments. 233 const AnnotatedLine *TheLine = *I; 234 if (TheLine->Last->is(TT_LineComment)) 235 return 0; 236 const auto &NextLine = *I[1]; 237 if (NextLine.Type == LT_Invalid || NextLine.First->MustBreakBefore) 238 return 0; 239 if (TheLine->InPPDirective && 240 (!NextLine.InPPDirective || NextLine.First->HasUnescapedNewline)) 241 return 0; 242 243 if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit) 244 return 0; 245 246 unsigned Limit = 247 Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent; 248 // If we already exceed the column limit, we set 'Limit' to 0. The different 249 // tryMerge..() functions can then decide whether to still do merging. 250 Limit = TheLine->Last->TotalLength > Limit 251 ? 0 252 : Limit - TheLine->Last->TotalLength; 253 254 if (TheLine->Last->is(TT_FunctionLBrace) && 255 TheLine->First == TheLine->Last && 256 !Style.BraceWrapping.SplitEmptyFunction && 257 NextLine.First->is(tok::r_brace)) 258 return tryMergeSimpleBlock(I, E, Limit); 259 260 const auto *PreviousLine = I != AnnotatedLines.begin() ? I[-1] : nullptr; 261 // Handle empty record blocks where the brace has already been wrapped. 262 if (PreviousLine && TheLine->Last->is(tok::l_brace) && 263 TheLine->First == TheLine->Last) { 264 bool EmptyBlock = NextLine.First->is(tok::r_brace); 265 266 const FormatToken *Tok = PreviousLine->First; 267 if (Tok && Tok->is(tok::comment)) 268 Tok = Tok->getNextNonComment(); 269 270 if (Tok && Tok->getNamespaceToken()) 271 return !Style.BraceWrapping.SplitEmptyNamespace && EmptyBlock 272 ? tryMergeSimpleBlock(I, E, Limit) 273 : 0; 274 275 if (Tok && Tok->is(tok::kw_typedef)) 276 Tok = Tok->getNextNonComment(); 277 if (Tok && Tok->isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union, 278 tok::kw_extern, Keywords.kw_interface)) 279 return !Style.BraceWrapping.SplitEmptyRecord && EmptyBlock 280 ? tryMergeSimpleBlock(I, E, Limit) 281 : 0; 282 283 if (Tok && Tok->is(tok::kw_template) && 284 Style.BraceWrapping.SplitEmptyRecord && EmptyBlock) 285 return 0; 286 } 287 288 auto ShouldMergeShortFunctions = [this, &I, &NextLine, PreviousLine, 289 TheLine]() { 290 if (Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_All) 291 return true; 292 if (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty && 293 NextLine.First->is(tok::r_brace)) 294 return true; 295 296 if (Style.AllowShortFunctionsOnASingleLine & 297 FormatStyle::SFS_InlineOnly) { 298 // Just checking TheLine->Level != 0 is not enough, because it 299 // provokes treating functions inside indented namespaces as short. 300 if (Style.isJavaScript() && TheLine->Last->is(TT_FunctionLBrace)) 301 return true; 302 303 if (TheLine->Level != 0) { 304 if (!PreviousLine) 305 return false; 306 307 // TODO: Use IndentTracker to avoid loop? 308 // Find the last line with lower level. 309 auto J = I - 1; 310 if (!TheLine->InPPDirective) { 311 for (; J != AnnotatedLines.begin(); --J) { 312 if (!(*J)->InPPDirective && (*J)->Level < TheLine->Level) 313 break; 314 } 315 } 316 317 if ((*J)->Level >= TheLine->Level) 318 return false; 319 320 // Check if the found line starts a record. 321 const FormatToken *LastNonComment = (*J)->Last; 322 assert(LastNonComment); 323 if (LastNonComment->is(tok::comment)) { 324 LastNonComment = LastNonComment->getPreviousNonComment(); 325 // There must be another token (usually `{`), because we chose a 326 // line that has a smaller level. 327 assert(LastNonComment); 328 } 329 return isRecordLBrace(*LastNonComment); 330 } 331 } 332 333 return false; 334 }; 335 336 bool MergeShortFunctions = ShouldMergeShortFunctions(); 337 338 const FormatToken *FirstNonComment = TheLine->First; 339 if (FirstNonComment->is(tok::comment)) { 340 FirstNonComment = FirstNonComment->getNextNonComment(); 341 if (!FirstNonComment) 342 return 0; 343 } 344 // FIXME: There are probably cases where we should use FirstNonComment 345 // instead of TheLine->First. 346 347 if (Style.CompactNamespaces) { 348 if (auto nsToken = TheLine->First->getNamespaceToken()) { 349 int i = 0; 350 unsigned closingLine = TheLine->MatchingClosingBlockLineIndex - 1; 351 for (; I + 1 + i != E && 352 nsToken->TokenText == getNamespaceTokenText(I[i + 1]) && 353 closingLine == I[i + 1]->MatchingClosingBlockLineIndex && 354 I[i + 1]->Last->TotalLength < Limit; 355 i++, --closingLine) { 356 // No extra indent for compacted namespaces. 357 IndentTracker.skipLine(*I[i + 1]); 358 359 Limit -= I[i + 1]->Last->TotalLength; 360 } 361 return i; 362 } 363 364 if (auto nsToken = getMatchingNamespaceToken(TheLine, AnnotatedLines)) { 365 int i = 0; 366 unsigned openingLine = TheLine->MatchingOpeningBlockLineIndex - 1; 367 for (; I + 1 + i != E && 368 nsToken->TokenText == 369 getMatchingNamespaceTokenText(I[i + 1], AnnotatedLines) && 370 openingLine == I[i + 1]->MatchingOpeningBlockLineIndex; 371 i++, --openingLine) { 372 // No space between consecutive braces. 373 I[i + 1]->First->SpacesRequiredBefore = !I[i]->Last->is(tok::r_brace); 374 375 // Indent like the outer-most namespace. 376 IndentTracker.nextLine(*I[i + 1]); 377 } 378 return i; 379 } 380 } 381 382 // Try to merge a function block with left brace unwrapped. 383 if (TheLine->Last->is(TT_FunctionLBrace) && TheLine->First != TheLine->Last) 384 return MergeShortFunctions ? tryMergeSimpleBlock(I, E, Limit) : 0; 385 // Try to merge a control statement block with left brace unwrapped. 386 if (TheLine->Last->is(tok::l_brace) && FirstNonComment != TheLine->Last && 387 FirstNonComment->isOneOf(tok::kw_if, tok::kw_while, tok::kw_for, 388 TT_ForEachMacro)) { 389 return Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never 390 ? tryMergeSimpleBlock(I, E, Limit) 391 : 0; 392 } 393 // Try to merge a control statement block with left brace wrapped. 394 if (NextLine.First->is(tok::l_brace)) { 395 if ((TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, 396 tok::kw_for, tok::kw_switch, tok::kw_try, 397 tok::kw_do, TT_ForEachMacro) || 398 (TheLine->First->is(tok::r_brace) && TheLine->First->Next && 399 TheLine->First->Next->isOneOf(tok::kw_else, tok::kw_catch))) && 400 Style.BraceWrapping.AfterControlStatement == 401 FormatStyle::BWACS_MultiLine) { 402 // If possible, merge the next line's wrapped left brace with the 403 // current line. Otherwise, leave it on the next line, as this is a 404 // multi-line control statement. 405 return (Style.ColumnLimit == 0 || 406 TheLine->Last->TotalLength <= Style.ColumnLimit) 407 ? 1 408 : 0; 409 } 410 if (TheLine->First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, 411 tok::kw_for, TT_ForEachMacro)) { 412 return (Style.BraceWrapping.AfterControlStatement == 413 FormatStyle::BWACS_Always) 414 ? tryMergeSimpleBlock(I, E, Limit) 415 : 0; 416 } 417 if (TheLine->First->isOneOf(tok::kw_else, tok::kw_catch) && 418 Style.BraceWrapping.AfterControlStatement == 419 FormatStyle::BWACS_MultiLine) { 420 // This case if different from the upper BWACS_MultiLine processing 421 // in that a preceding r_brace is not on the same line as else/catch 422 // most likely because of BeforeElse/BeforeCatch set to true. 423 // If the line length doesn't fit ColumnLimit, leave l_brace on the 424 // next line to respect the BWACS_MultiLine. 425 return (Style.ColumnLimit == 0 || 426 TheLine->Last->TotalLength <= Style.ColumnLimit) 427 ? 1 428 : 0; 429 } 430 } 431 if (PreviousLine && TheLine->First->is(tok::l_brace)) { 432 switch (PreviousLine->First->Tok.getKind()) { 433 case tok::at: 434 // Don't merge block with left brace wrapped after ObjC special blocks. 435 if (PreviousLine->First->Next) { 436 tok::ObjCKeywordKind kwId = 437 PreviousLine->First->Next->Tok.getObjCKeywordID(); 438 if (kwId == tok::objc_autoreleasepool || 439 kwId == tok::objc_synchronized) 440 return 0; 441 } 442 break; 443 444 case tok::kw_case: 445 case tok::kw_default: 446 // Don't merge block with left brace wrapped after case labels. 447 return 0; 448 449 default: 450 break; 451 } 452 } 453 454 // Don't merge an empty template class or struct if SplitEmptyRecords 455 // is defined. 456 if (PreviousLine && Style.BraceWrapping.SplitEmptyRecord && 457 TheLine->Last->is(tok::l_brace) && PreviousLine->Last) { 458 const FormatToken *Previous = PreviousLine->Last; 459 if (Previous) { 460 if (Previous->is(tok::comment)) 461 Previous = Previous->getPreviousNonComment(); 462 if (Previous) { 463 if (Previous->is(tok::greater) && !PreviousLine->InPPDirective) 464 return 0; 465 if (Previous->is(tok::identifier)) { 466 const FormatToken *PreviousPrevious = 467 Previous->getPreviousNonComment(); 468 if (PreviousPrevious && 469 PreviousPrevious->isOneOf(tok::kw_class, tok::kw_struct)) 470 return 0; 471 } 472 } 473 } 474 } 475 476 if (TheLine->Last->is(tok::l_brace)) { 477 bool ShouldMerge = false; 478 // Try to merge records. 479 if (TheLine->Last->is(TT_EnumLBrace)) { 480 ShouldMerge = Style.AllowShortEnumsOnASingleLine; 481 } else if (TheLine->Last->isOneOf(TT_ClassLBrace, TT_StructLBrace)) { 482 // NOTE: We use AfterClass (whereas AfterStruct exists) for both classes 483 // and structs, but it seems that wrapping is still handled correctly 484 // elsewhere. 485 ShouldMerge = !Style.BraceWrapping.AfterClass || 486 (NextLine.First->is(tok::r_brace) && 487 !Style.BraceWrapping.SplitEmptyRecord); 488 } else { 489 // Try to merge a block with left brace unwrapped that wasn't yet 490 // covered. 491 assert(TheLine->InPPDirective || 492 !TheLine->First->isOneOf(tok::kw_class, tok::kw_enum, 493 tok::kw_struct)); 494 ShouldMerge = !Style.BraceWrapping.AfterFunction || 495 (NextLine.First->is(tok::r_brace) && 496 !Style.BraceWrapping.SplitEmptyFunction); 497 } 498 return ShouldMerge ? tryMergeSimpleBlock(I, E, Limit) : 0; 499 } 500 501 // Try to merge a function block with left brace wrapped. 502 if (NextLine.First->is(TT_FunctionLBrace) && 503 Style.BraceWrapping.AfterFunction) { 504 if (NextLine.Last->is(TT_LineComment)) 505 return 0; 506 507 // Check for Limit <= 2 to account for the " {". 508 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(TheLine))) 509 return 0; 510 Limit -= 2; 511 512 unsigned MergedLines = 0; 513 if (MergeShortFunctions || 514 (Style.AllowShortFunctionsOnASingleLine >= FormatStyle::SFS_Empty && 515 NextLine.First == NextLine.Last && I + 2 != E && 516 I[2]->First->is(tok::r_brace))) { 517 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); 518 // If we managed to merge the block, count the function header, which is 519 // on a separate line. 520 if (MergedLines > 0) 521 ++MergedLines; 522 } 523 return MergedLines; 524 } 525 auto IsElseLine = [&TheLine]() -> bool { 526 const FormatToken *First = TheLine->First; 527 if (First->is(tok::kw_else)) 528 return true; 529 530 return First->is(tok::r_brace) && First->Next && 531 First->Next->is(tok::kw_else); 532 }; 533 if (TheLine->First->is(tok::kw_if) || 534 (IsElseLine() && (Style.AllowShortIfStatementsOnASingleLine == 535 FormatStyle::SIS_AllIfsAndElse))) { 536 return Style.AllowShortIfStatementsOnASingleLine 537 ? tryMergeSimpleControlStatement(I, E, Limit) 538 : 0; 539 } 540 if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while, tok::kw_do, 541 TT_ForEachMacro)) { 542 return Style.AllowShortLoopsOnASingleLine 543 ? tryMergeSimpleControlStatement(I, E, Limit) 544 : 0; 545 } 546 if (TheLine->First->isOneOf(tok::kw_case, tok::kw_default)) { 547 return Style.AllowShortCaseLabelsOnASingleLine 548 ? tryMergeShortCaseLabels(I, E, Limit) 549 : 0; 550 } 551 if (TheLine->InPPDirective && 552 (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) 553 return tryMergeSimplePPDirective(I, E, Limit); 554 return 0; 555 } 556 557 unsigned 558 tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 559 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 560 unsigned Limit) { 561 if (Limit == 0) 562 return 0; 563 if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline) 564 return 0; 565 if (1 + I[1]->Last->TotalLength > Limit) 566 return 0; 567 return 1; 568 } 569 570 unsigned tryMergeSimpleControlStatement( 571 SmallVectorImpl<AnnotatedLine *>::const_iterator I, 572 SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) { 573 if (Limit == 0) 574 return 0; 575 if (Style.BraceWrapping.AfterControlStatement == 576 FormatStyle::BWACS_Always && 577 I[1]->First->is(tok::l_brace) && 578 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 579 return 0; 580 if (I[1]->InPPDirective != (*I)->InPPDirective || 581 (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline)) 582 return 0; 583 Limit = limitConsideringMacros(I + 1, E, Limit); 584 AnnotatedLine &Line = **I; 585 if (!Line.First->is(tok::kw_do) && !Line.First->is(tok::kw_else) && 586 !Line.Last->is(tok::kw_else) && Line.Last->isNot(tok::r_paren)) 587 return 0; 588 // Only merge `do while` if `do` is the only statement on the line. 589 if (Line.First->is(tok::kw_do) && !Line.Last->is(tok::kw_do)) 590 return 0; 591 if (1 + I[1]->Last->TotalLength > Limit) 592 return 0; 593 // Don't merge with loops, ifs, a single semicolon or a line comment. 594 if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while, 595 TT_ForEachMacro, TT_LineComment)) 596 return 0; 597 // Only inline simple if's (no nested if or else), unless specified 598 if (Style.AllowShortIfStatementsOnASingleLine == 599 FormatStyle::SIS_WithoutElse) { 600 if (I + 2 != E && Line.startsWith(tok::kw_if) && 601 I[2]->First->is(tok::kw_else)) 602 return 0; 603 } 604 return 1; 605 } 606 607 unsigned 608 tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 609 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 610 unsigned Limit) { 611 if (Limit == 0 || I + 1 == E || 612 I[1]->First->isOneOf(tok::kw_case, tok::kw_default)) 613 return 0; 614 if (I[0]->Last->is(tok::l_brace) || I[1]->First->is(tok::l_brace)) 615 return 0; 616 unsigned NumStmts = 0; 617 unsigned Length = 0; 618 bool EndsWithComment = false; 619 bool InPPDirective = I[0]->InPPDirective; 620 const unsigned Level = I[0]->Level; 621 for (; NumStmts < 3; ++NumStmts) { 622 if (I + 1 + NumStmts == E) 623 break; 624 const AnnotatedLine *Line = I[1 + NumStmts]; 625 if (Line->InPPDirective != InPPDirective) 626 break; 627 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace)) 628 break; 629 if (Line->First->isOneOf(tok::kw_if, tok::kw_for, tok::kw_switch, 630 tok::kw_while) || 631 EndsWithComment) 632 return 0; 633 if (Line->First->is(tok::comment)) { 634 if (Level != Line->Level) 635 return 0; 636 SmallVectorImpl<AnnotatedLine *>::const_iterator J = I + 2 + NumStmts; 637 for (; J != E; ++J) { 638 Line = *J; 639 if (Line->InPPDirective != InPPDirective) 640 break; 641 if (Line->First->isOneOf(tok::kw_case, tok::kw_default, tok::r_brace)) 642 break; 643 if (Line->First->isNot(tok::comment) || Level != Line->Level) 644 return 0; 645 } 646 break; 647 } 648 if (Line->Last->is(tok::comment)) 649 EndsWithComment = true; 650 Length += I[1 + NumStmts]->Last->TotalLength + 1; // 1 for the space. 651 } 652 if (NumStmts == 0 || NumStmts == 3 || Length > Limit) 653 return 0; 654 return NumStmts; 655 } 656 657 unsigned 658 tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 659 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 660 unsigned Limit) { 661 // Don't merge with a preprocessor directive. 662 if (I[1]->Type == LT_PreprocessorDirective) 663 return 0; 664 665 AnnotatedLine &Line = **I; 666 667 // Don't merge ObjC @ keywords and methods. 668 // FIXME: If an option to allow short exception handling clauses on a single 669 // line is added, change this to not return for @try and friends. 670 if (Style.Language != FormatStyle::LK_Java && 671 Line.First->isOneOf(tok::at, tok::minus, tok::plus)) 672 return 0; 673 674 // Check that the current line allows merging. This depends on whether we 675 // are in a control flow statements as well as several style flags. 676 if (Line.First->is(tok::kw_case) || 677 (Line.First->Next && Line.First->Next->is(tok::kw_else))) 678 return 0; 679 // default: in switch statement 680 if (Line.First->is(tok::kw_default)) { 681 const FormatToken *Tok = Line.First->getNextNonComment(); 682 if (Tok && Tok->is(tok::colon)) 683 return 0; 684 } 685 if (Line.First->isOneOf(tok::kw_if, tok::kw_else, tok::kw_while, tok::kw_do, 686 tok::kw_try, tok::kw___try, tok::kw_catch, 687 tok::kw___finally, tok::kw_for, TT_ForEachMacro, 688 tok::r_brace, Keywords.kw___except)) { 689 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never) 690 return 0; 691 if (Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Empty && 692 !I[1]->First->is(tok::r_brace)) 693 return 0; 694 // Don't merge when we can't except the case when 695 // the control statement block is empty 696 if (!Style.AllowShortIfStatementsOnASingleLine && 697 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 698 !Style.BraceWrapping.AfterControlStatement && 699 !I[1]->First->is(tok::r_brace)) 700 return 0; 701 if (!Style.AllowShortIfStatementsOnASingleLine && 702 Line.First->isOneOf(tok::kw_if, tok::kw_else) && 703 Style.BraceWrapping.AfterControlStatement == 704 FormatStyle::BWACS_Always && 705 I + 2 != E && !I[2]->First->is(tok::r_brace)) 706 return 0; 707 if (!Style.AllowShortLoopsOnASingleLine && 708 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for, 709 TT_ForEachMacro) && 710 !Style.BraceWrapping.AfterControlStatement && 711 !I[1]->First->is(tok::r_brace)) 712 return 0; 713 if (!Style.AllowShortLoopsOnASingleLine && 714 Line.First->isOneOf(tok::kw_while, tok::kw_do, tok::kw_for, 715 TT_ForEachMacro) && 716 Style.BraceWrapping.AfterControlStatement == 717 FormatStyle::BWACS_Always && 718 I + 2 != E && !I[2]->First->is(tok::r_brace)) 719 return 0; 720 // FIXME: Consider an option to allow short exception handling clauses on 721 // a single line. 722 // FIXME: This isn't covered by tests. 723 // FIXME: For catch, __except, __finally the first token on the line 724 // is '}', so this isn't correct here. 725 if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch, 726 Keywords.kw___except, tok::kw___finally)) 727 return 0; 728 } 729 730 if (Line.Last->is(tok::l_brace)) { 731 FormatToken *Tok = I[1]->First; 732 auto ShouldMerge = [Tok]() { 733 if (Tok->isNot(tok::r_brace) || Tok->MustBreakBefore) 734 return false; 735 const FormatToken *Next = Tok->getNextNonComment(); 736 return !Next || Next->is(tok::semi); 737 }; 738 739 if (ShouldMerge()) { 740 // We merge empty blocks even if the line exceeds the column limit. 741 Tok->SpacesRequiredBefore = Style.SpaceInEmptyBlock ? 1 : 0; 742 Tok->CanBreakBefore = true; 743 return 1; 744 } else if (Limit != 0 && !Line.startsWithNamespace() && 745 !startsExternCBlock(Line)) { 746 // We don't merge short records. 747 if (isRecordLBrace(*Line.Last)) 748 return 0; 749 750 // Check that we still have three lines and they fit into the limit. 751 if (I + 2 == E || I[2]->Type == LT_Invalid) 752 return 0; 753 Limit = limitConsideringMacros(I + 2, E, Limit); 754 755 if (!nextTwoLinesFitInto(I, Limit)) 756 return 0; 757 758 // Second, check that the next line does not contain any braces - if it 759 // does, readability declines when putting it into a single line. 760 if (I[1]->Last->is(TT_LineComment)) 761 return 0; 762 do { 763 if (Tok->is(tok::l_brace) && Tok->isNot(BK_BracedInit)) 764 return 0; 765 Tok = Tok->Next; 766 } while (Tok); 767 768 // Last, check that the third line starts with a closing brace. 769 Tok = I[2]->First; 770 if (Tok->isNot(tok::r_brace)) 771 return 0; 772 773 // Don't merge "if (a) { .. } else {". 774 if (Tok->Next && Tok->Next->is(tok::kw_else)) 775 return 0; 776 777 // Don't merge a trailing multi-line control statement block like: 778 // } else if (foo && 779 // bar) 780 // { <-- current Line 781 // baz(); 782 // } 783 if (Line.First == Line.Last && Line.First->isNot(TT_FunctionLBrace) && 784 Style.BraceWrapping.AfterControlStatement == 785 FormatStyle::BWACS_MultiLine) 786 return 0; 787 788 return 2; 789 } 790 } else if (I[1]->First->is(tok::l_brace)) { 791 if (I[1]->Last->is(TT_LineComment)) 792 return 0; 793 794 // Check for Limit <= 2 to account for the " {". 795 if (Limit <= 2 || (Style.ColumnLimit == 0 && containsMustBreak(*I))) 796 return 0; 797 Limit -= 2; 798 unsigned MergedLines = 0; 799 if (Style.AllowShortBlocksOnASingleLine != FormatStyle::SBS_Never || 800 (I[1]->First == I[1]->Last && I + 2 != E && 801 I[2]->First->is(tok::r_brace))) { 802 MergedLines = tryMergeSimpleBlock(I + 1, E, Limit); 803 // If we managed to merge the block, count the statement header, which 804 // is on a separate line. 805 if (MergedLines > 0) 806 ++MergedLines; 807 } 808 return MergedLines; 809 } 810 return 0; 811 } 812 813 /// Returns the modified column limit for \p I if it is inside a macro and 814 /// needs a trailing '\'. 815 unsigned 816 limitConsideringMacros(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 817 SmallVectorImpl<AnnotatedLine *>::const_iterator E, 818 unsigned Limit) { 819 if (I[0]->InPPDirective && I + 1 != E && 820 !I[1]->First->HasUnescapedNewline && !I[1]->First->is(tok::eof)) 821 return Limit < 2 ? 0 : Limit - 2; 822 return Limit; 823 } 824 825 bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I, 826 unsigned Limit) { 827 if (I[1]->First->MustBreakBefore || I[2]->First->MustBreakBefore) 828 return false; 829 return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit; 830 } 831 832 bool containsMustBreak(const AnnotatedLine *Line) { 833 for (const FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) 834 if (Tok->MustBreakBefore) 835 return true; 836 return false; 837 } 838 839 void join(AnnotatedLine &A, const AnnotatedLine &B) { 840 assert(!A.Last->Next); 841 assert(!B.First->Previous); 842 if (B.Affected) 843 A.Affected = true; 844 A.Last->Next = B.First; 845 B.First->Previous = A.Last; 846 B.First->CanBreakBefore = true; 847 unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore; 848 for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) { 849 Tok->TotalLength += LengthA; 850 A.Last = Tok; 851 } 852 } 853 854 const FormatStyle &Style; 855 const AdditionalKeywords &Keywords; 856 const SmallVectorImpl<AnnotatedLine *>::const_iterator End; 857 858 SmallVectorImpl<AnnotatedLine *>::const_iterator Next; 859 const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines; 860 }; 861 862 static void markFinalized(FormatToken *Tok) { 863 for (; Tok; Tok = Tok->Next) { 864 Tok->Finalized = true; 865 for (AnnotatedLine *Child : Tok->Children) 866 markFinalized(Child->First); 867 } 868 } 869 870 #ifndef NDEBUG 871 static void printLineState(const LineState &State) { 872 llvm::dbgs() << "State: "; 873 for (const ParenState &P : State.Stack) { 874 llvm::dbgs() << (P.Tok ? P.Tok->TokenText : "F") << "|" << P.Indent << "|" 875 << P.LastSpace << "|" << P.NestedBlockIndent << " "; 876 } 877 llvm::dbgs() << State.NextToken->TokenText << "\n"; 878 } 879 #endif 880 881 /// Base class for classes that format one \c AnnotatedLine. 882 class LineFormatter { 883 public: 884 LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces, 885 const FormatStyle &Style, 886 UnwrappedLineFormatter *BlockFormatter) 887 : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style), 888 BlockFormatter(BlockFormatter) {} 889 virtual ~LineFormatter() {} 890 891 /// Formats an \c AnnotatedLine and returns the penalty. 892 /// 893 /// If \p DryRun is \c false, directly applies the changes. 894 virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 895 unsigned FirstStartColumn, bool DryRun) = 0; 896 897 protected: 898 /// If the \p State's next token is an r_brace closing a nested block, 899 /// format the nested block before it. 900 /// 901 /// Returns \c true if all children could be placed successfully and adapts 902 /// \p Penalty as well as \p State. If \p DryRun is false, also directly 903 /// creates changes using \c Whitespaces. 904 /// 905 /// The crucial idea here is that children always get formatted upon 906 /// encountering the closing brace right after the nested block. Now, if we 907 /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is 908 /// \c false), the entire block has to be kept on the same line (which is only 909 /// possible if it fits on the line, only contains a single statement, etc. 910 /// 911 /// If \p NewLine is true, we format the nested block on separate lines, i.e. 912 /// break after the "{", format all lines with correct indentation and the put 913 /// the closing "}" on yet another new line. 914 /// 915 /// This enables us to keep the simple structure of the 916 /// \c UnwrappedLineFormatter, where we only have two options for each token: 917 /// break or don't break. 918 bool formatChildren(LineState &State, bool NewLine, bool DryRun, 919 unsigned &Penalty) { 920 const FormatToken *LBrace = State.NextToken->getPreviousNonComment(); 921 FormatToken &Previous = *State.NextToken->Previous; 922 if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->isNot(BK_Block) || 923 Previous.Children.size() == 0) 924 // The previous token does not open a block. Nothing to do. We don't 925 // assert so that we can simply call this function for all tokens. 926 return true; 927 928 if (NewLine) { 929 const ParenState &P = State.Stack.back(); 930 931 int AdditionalIndent = 932 P.Indent - Previous.Children[0]->Level * Style.IndentWidth; 933 934 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && 935 P.NestedBlockIndent == P.LastSpace) { 936 if (State.NextToken->MatchingParen && 937 State.NextToken->MatchingParen->is(TT_LambdaLBrace)) 938 State.Stack.pop_back(); 939 if (LBrace->is(TT_LambdaLBrace)) 940 AdditionalIndent = 0; 941 } 942 943 Penalty += 944 BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent, 945 /*FixBadIndentation=*/true); 946 return true; 947 } 948 949 if (Previous.Children[0]->First->MustBreakBefore) 950 return false; 951 952 // Cannot merge into one line if this line ends on a comment. 953 if (Previous.is(tok::comment)) 954 return false; 955 956 // Cannot merge multiple statements into a single line. 957 if (Previous.Children.size() > 1) 958 return false; 959 960 const AnnotatedLine *Child = Previous.Children[0]; 961 // We can't put the closing "}" on a line with a trailing comment. 962 if (Child->Last->isTrailingComment()) 963 return false; 964 965 // If the child line exceeds the column limit, we wouldn't want to merge it. 966 // We add +2 for the trailing " }". 967 if (Style.ColumnLimit > 0 && 968 Child->Last->TotalLength + State.Column + 2 > Style.ColumnLimit) 969 return false; 970 971 if (!DryRun) { 972 Whitespaces->replaceWhitespace( 973 *Child->First, /*Newlines=*/0, /*Spaces=*/1, 974 /*StartOfTokenColumn=*/State.Column, /*IsAligned=*/false, 975 State.Line->InPPDirective); 976 } 977 Penalty += 978 formatLine(*Child, State.Column + 1, /*FirstStartColumn=*/0, DryRun); 979 980 State.Column += 1 + Child->Last->TotalLength; 981 return true; 982 } 983 984 ContinuationIndenter *Indenter; 985 986 private: 987 WhitespaceManager *Whitespaces; 988 const FormatStyle &Style; 989 UnwrappedLineFormatter *BlockFormatter; 990 }; 991 992 /// Formatter that keeps the existing line breaks. 993 class NoColumnLimitLineFormatter : public LineFormatter { 994 public: 995 NoColumnLimitLineFormatter(ContinuationIndenter *Indenter, 996 WhitespaceManager *Whitespaces, 997 const FormatStyle &Style, 998 UnwrappedLineFormatter *BlockFormatter) 999 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 1000 1001 /// Formats the line, simply keeping all of the input's line breaking 1002 /// decisions. 1003 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 1004 unsigned FirstStartColumn, bool DryRun) override { 1005 assert(!DryRun); 1006 LineState State = Indenter->getInitialState(FirstIndent, FirstStartColumn, 1007 &Line, /*DryRun=*/false); 1008 while (State.NextToken) { 1009 bool Newline = 1010 Indenter->mustBreak(State) || 1011 (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0); 1012 unsigned Penalty = 0; 1013 formatChildren(State, Newline, /*DryRun=*/false, Penalty); 1014 Indenter->addTokenToState(State, Newline, /*DryRun=*/false); 1015 } 1016 return 0; 1017 } 1018 }; 1019 1020 /// Formatter that puts all tokens into a single line without breaks. 1021 class NoLineBreakFormatter : public LineFormatter { 1022 public: 1023 NoLineBreakFormatter(ContinuationIndenter *Indenter, 1024 WhitespaceManager *Whitespaces, const FormatStyle &Style, 1025 UnwrappedLineFormatter *BlockFormatter) 1026 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 1027 1028 /// Puts all tokens into a single line. 1029 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 1030 unsigned FirstStartColumn, bool DryRun) override { 1031 unsigned Penalty = 0; 1032 LineState State = 1033 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 1034 while (State.NextToken) { 1035 formatChildren(State, /*NewLine=*/false, DryRun, Penalty); 1036 Indenter->addTokenToState( 1037 State, /*Newline=*/State.NextToken->MustBreakBefore, DryRun); 1038 } 1039 return Penalty; 1040 } 1041 }; 1042 1043 /// Finds the best way to break lines. 1044 class OptimizingLineFormatter : public LineFormatter { 1045 public: 1046 OptimizingLineFormatter(ContinuationIndenter *Indenter, 1047 WhitespaceManager *Whitespaces, 1048 const FormatStyle &Style, 1049 UnwrappedLineFormatter *BlockFormatter) 1050 : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {} 1051 1052 /// Formats the line by finding the best line breaks with line lengths 1053 /// below the column limit. 1054 unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent, 1055 unsigned FirstStartColumn, bool DryRun) override { 1056 LineState State = 1057 Indenter->getInitialState(FirstIndent, FirstStartColumn, &Line, DryRun); 1058 1059 // If the ObjC method declaration does not fit on a line, we should format 1060 // it with one arg per line. 1061 if (State.Line->Type == LT_ObjCMethodDecl) 1062 State.Stack.back().BreakBeforeParameter = true; 1063 1064 // Find best solution in solution space. 1065 return analyzeSolutionSpace(State, DryRun); 1066 } 1067 1068 private: 1069 struct CompareLineStatePointers { 1070 bool operator()(LineState *obj1, LineState *obj2) const { 1071 return *obj1 < *obj2; 1072 } 1073 }; 1074 1075 /// A pair of <penalty, count> that is used to prioritize the BFS on. 1076 /// 1077 /// In case of equal penalties, we want to prefer states that were inserted 1078 /// first. During state generation we make sure that we insert states first 1079 /// that break the line as late as possible. 1080 typedef std::pair<unsigned, unsigned> OrderedPenalty; 1081 1082 /// An edge in the solution space from \c Previous->State to \c State, 1083 /// inserting a newline dependent on the \c NewLine. 1084 struct StateNode { 1085 StateNode(const LineState &State, bool NewLine, StateNode *Previous) 1086 : State(State), NewLine(NewLine), Previous(Previous) {} 1087 LineState State; 1088 bool NewLine; 1089 StateNode *Previous; 1090 }; 1091 1092 /// An item in the prioritized BFS search queue. The \c StateNode's 1093 /// \c State has the given \c OrderedPenalty. 1094 typedef std::pair<OrderedPenalty, StateNode *> QueueItem; 1095 1096 /// The BFS queue type. 1097 typedef std::priority_queue<QueueItem, std::vector<QueueItem>, 1098 std::greater<QueueItem>> 1099 QueueType; 1100 1101 /// Analyze the entire solution space starting from \p InitialState. 1102 /// 1103 /// This implements a variant of Dijkstra's algorithm on the graph that spans 1104 /// the solution space (\c LineStates are the nodes). The algorithm tries to 1105 /// find the shortest path (the one with lowest penalty) from \p InitialState 1106 /// to a state where all tokens are placed. Returns the penalty. 1107 /// 1108 /// If \p DryRun is \c false, directly applies the changes. 1109 unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) { 1110 std::set<LineState *, CompareLineStatePointers> Seen; 1111 1112 // Increasing count of \c StateNode items we have created. This is used to 1113 // create a deterministic order independent of the container. 1114 unsigned Count = 0; 1115 QueueType Queue; 1116 1117 // Insert start element into queue. 1118 StateNode *RootNode = 1119 new (Allocator.Allocate()) StateNode(InitialState, false, nullptr); 1120 Queue.push(QueueItem(OrderedPenalty(0, Count), RootNode)); 1121 ++Count; 1122 1123 unsigned Penalty = 0; 1124 1125 // While not empty, take first element and follow edges. 1126 while (!Queue.empty()) { 1127 Penalty = Queue.top().first.first; 1128 StateNode *Node = Queue.top().second; 1129 if (!Node->State.NextToken) { 1130 LLVM_DEBUG(llvm::dbgs() 1131 << "\n---\nPenalty for line: " << Penalty << "\n"); 1132 break; 1133 } 1134 Queue.pop(); 1135 1136 // Cut off the analysis of certain solutions if the analysis gets too 1137 // complex. See description of IgnoreStackForComparison. 1138 if (Count > 50000) 1139 Node->State.IgnoreStackForComparison = true; 1140 1141 if (!Seen.insert(&Node->State).second) 1142 // State already examined with lower penalty. 1143 continue; 1144 1145 FormatDecision LastFormat = Node->State.NextToken->getDecision(); 1146 if (LastFormat == FD_Unformatted || LastFormat == FD_Continue) 1147 addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue); 1148 if (LastFormat == FD_Unformatted || LastFormat == FD_Break) 1149 addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue); 1150 } 1151 1152 if (Queue.empty()) { 1153 // We were unable to find a solution, do nothing. 1154 // FIXME: Add diagnostic? 1155 LLVM_DEBUG(llvm::dbgs() << "Could not find a solution.\n"); 1156 return 0; 1157 } 1158 1159 // Reconstruct the solution. 1160 if (!DryRun) 1161 reconstructPath(InitialState, Queue.top().second); 1162 1163 LLVM_DEBUG(llvm::dbgs() 1164 << "Total number of analyzed states: " << Count << "\n"); 1165 LLVM_DEBUG(llvm::dbgs() << "---\n"); 1166 1167 return Penalty; 1168 } 1169 1170 /// Add the following state to the analysis queue \c Queue. 1171 /// 1172 /// Assume the current state is \p PreviousNode and has been reached with a 1173 /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true. 1174 void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode, 1175 bool NewLine, unsigned *Count, QueueType *Queue) { 1176 if (NewLine && !Indenter->canBreak(PreviousNode->State)) 1177 return; 1178 if (!NewLine && Indenter->mustBreak(PreviousNode->State)) 1179 return; 1180 1181 StateNode *Node = new (Allocator.Allocate()) 1182 StateNode(PreviousNode->State, NewLine, PreviousNode); 1183 if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty)) 1184 return; 1185 1186 Penalty += Indenter->addTokenToState(Node->State, NewLine, true); 1187 1188 Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node)); 1189 ++(*Count); 1190 } 1191 1192 /// Applies the best formatting by reconstructing the path in the 1193 /// solution space that leads to \c Best. 1194 void reconstructPath(LineState &State, StateNode *Best) { 1195 llvm::SmallVector<StateNode *> Path; 1196 // We do not need a break before the initial token. 1197 while (Best->Previous) { 1198 Path.push_back(Best); 1199 Best = Best->Previous; 1200 } 1201 for (const auto &Node : llvm::reverse(Path)) { 1202 unsigned Penalty = 0; 1203 formatChildren(State, Node->NewLine, /*DryRun=*/false, Penalty); 1204 Penalty += Indenter->addTokenToState(State, Node->NewLine, false); 1205 1206 LLVM_DEBUG({ 1207 printLineState(Node->Previous->State); 1208 if (Node->NewLine) 1209 llvm::dbgs() << "Penalty for placing " 1210 << Node->Previous->State.NextToken->Tok.getName() 1211 << " on a new line: " << Penalty << "\n"; 1212 }); 1213 } 1214 } 1215 1216 llvm::SpecificBumpPtrAllocator<StateNode> Allocator; 1217 }; 1218 1219 } // anonymous namespace 1220 1221 unsigned UnwrappedLineFormatter::format( 1222 const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun, 1223 int AdditionalIndent, bool FixBadIndentation, unsigned FirstStartColumn, 1224 unsigned NextStartColumn, unsigned LastStartColumn) { 1225 LineJoiner Joiner(Style, Keywords, Lines); 1226 1227 // Try to look up already computed penalty in DryRun-mode. 1228 std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey( 1229 &Lines, AdditionalIndent); 1230 auto CacheIt = PenaltyCache.find(CacheKey); 1231 if (DryRun && CacheIt != PenaltyCache.end()) 1232 return CacheIt->second; 1233 1234 assert(!Lines.empty()); 1235 unsigned Penalty = 0; 1236 LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level, 1237 AdditionalIndent); 1238 const AnnotatedLine *PrevPrevLine = nullptr; 1239 const AnnotatedLine *PreviousLine = nullptr; 1240 const AnnotatedLine *NextLine = nullptr; 1241 1242 // The minimum level of consecutive lines that have been formatted. 1243 unsigned RangeMinLevel = UINT_MAX; 1244 1245 bool FirstLine = true; 1246 for (const AnnotatedLine *Line = 1247 Joiner.getNextMergedLine(DryRun, IndentTracker); 1248 Line; PrevPrevLine = PreviousLine, PreviousLine = Line, Line = NextLine, 1249 FirstLine = false) { 1250 assert(Line->First); 1251 const AnnotatedLine &TheLine = *Line; 1252 unsigned Indent = IndentTracker.getIndent(); 1253 1254 // We continue formatting unchanged lines to adjust their indent, e.g. if a 1255 // scope was added. However, we need to carefully stop doing this when we 1256 // exit the scope of affected lines to prevent indenting a the entire 1257 // remaining file if it currently missing a closing brace. 1258 bool PreviousRBrace = 1259 PreviousLine && PreviousLine->startsWith(tok::r_brace); 1260 bool ContinueFormatting = 1261 TheLine.Level > RangeMinLevel || 1262 (TheLine.Level == RangeMinLevel && !PreviousRBrace && 1263 !TheLine.startsWith(tok::r_brace)); 1264 1265 bool FixIndentation = (FixBadIndentation || ContinueFormatting) && 1266 Indent != TheLine.First->OriginalColumn; 1267 bool ShouldFormat = TheLine.Affected || FixIndentation; 1268 // We cannot format this line; if the reason is that the line had a 1269 // parsing error, remember that. 1270 if (ShouldFormat && TheLine.Type == LT_Invalid && Status) { 1271 Status->FormatComplete = false; 1272 Status->Line = 1273 SourceMgr.getSpellingLineNumber(TheLine.First->Tok.getLocation()); 1274 } 1275 1276 if (ShouldFormat && TheLine.Type != LT_Invalid) { 1277 if (!DryRun) { 1278 bool LastLine = TheLine.First->is(tok::eof); 1279 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, Indent, 1280 LastLine ? LastStartColumn : NextStartColumn + Indent); 1281 } 1282 1283 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1284 unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine); 1285 bool FitsIntoOneLine = 1286 TheLine.Last->TotalLength + Indent <= ColumnLimit || 1287 (TheLine.Type == LT_ImportStatement && 1288 (!Style.isJavaScript() || !Style.JavaScriptWrapImports)) || 1289 (Style.isCSharp() && 1290 TheLine.InPPDirective); // don't split #regions in C# 1291 if (Style.ColumnLimit == 0) 1292 NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this) 1293 .formatLine(TheLine, NextStartColumn + Indent, 1294 FirstLine ? FirstStartColumn : 0, DryRun); 1295 else if (FitsIntoOneLine) 1296 Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this) 1297 .formatLine(TheLine, NextStartColumn + Indent, 1298 FirstLine ? FirstStartColumn : 0, DryRun); 1299 else 1300 Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this) 1301 .formatLine(TheLine, NextStartColumn + Indent, 1302 FirstLine ? FirstStartColumn : 0, DryRun); 1303 RangeMinLevel = std::min(RangeMinLevel, TheLine.Level); 1304 } else { 1305 // If no token in the current line is affected, we still need to format 1306 // affected children. 1307 if (TheLine.ChildrenAffected) 1308 for (const FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) 1309 if (!Tok->Children.empty()) 1310 format(Tok->Children, DryRun); 1311 1312 // Adapt following lines on the current indent level to the same level 1313 // unless the current \c AnnotatedLine is not at the beginning of a line. 1314 bool StartsNewLine = 1315 TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst; 1316 if (StartsNewLine) 1317 IndentTracker.adjustToUnmodifiedLine(TheLine); 1318 if (!DryRun) { 1319 bool ReformatLeadingWhitespace = 1320 StartsNewLine && ((PreviousLine && PreviousLine->Affected) || 1321 TheLine.LeadingEmptyLinesAffected); 1322 // Format the first token. 1323 if (ReformatLeadingWhitespace) 1324 formatFirstToken(TheLine, PreviousLine, PrevPrevLine, Lines, 1325 TheLine.First->OriginalColumn, 1326 TheLine.First->OriginalColumn); 1327 else 1328 Whitespaces->addUntouchableToken(*TheLine.First, 1329 TheLine.InPPDirective); 1330 1331 // Notify the WhitespaceManager about the unchanged whitespace. 1332 for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next) 1333 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective); 1334 } 1335 NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker); 1336 RangeMinLevel = UINT_MAX; 1337 } 1338 if (!DryRun) 1339 markFinalized(TheLine.First); 1340 } 1341 PenaltyCache[CacheKey] = Penalty; 1342 return Penalty; 1343 } 1344 1345 void UnwrappedLineFormatter::formatFirstToken( 1346 const AnnotatedLine &Line, const AnnotatedLine *PreviousLine, 1347 const AnnotatedLine *PrevPrevLine, 1348 const SmallVectorImpl<AnnotatedLine *> &Lines, unsigned Indent, 1349 unsigned NewlineIndent) { 1350 FormatToken &RootToken = *Line.First; 1351 if (RootToken.is(tok::eof)) { 1352 unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u); 1353 unsigned TokenIndent = Newlines ? NewlineIndent : 0; 1354 Whitespaces->replaceWhitespace(RootToken, Newlines, TokenIndent, 1355 TokenIndent); 1356 return; 1357 } 1358 unsigned Newlines = 1359 std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1); 1360 // Remove empty lines before "}" where applicable. 1361 if (RootToken.is(tok::r_brace) && 1362 (!RootToken.Next || 1363 (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)) && 1364 // Do not remove empty lines before namespace closing "}". 1365 !getNamespaceToken(&Line, Lines)) 1366 Newlines = std::min(Newlines, 1u); 1367 // Remove empty lines at the start of nested blocks (lambdas/arrow functions) 1368 if (PreviousLine == nullptr && Line.Level > 0) 1369 Newlines = std::min(Newlines, 1u); 1370 if (Newlines == 0 && !RootToken.IsFirst) 1371 Newlines = 1; 1372 if (RootToken.IsFirst && !RootToken.HasUnescapedNewline) 1373 Newlines = 0; 1374 1375 // Remove empty lines after "{". 1376 if (!Style.KeepEmptyLinesAtTheStartOfBlocks && PreviousLine && 1377 PreviousLine->Last->is(tok::l_brace) && 1378 !PreviousLine->startsWithNamespace() && 1379 !(PrevPrevLine && PrevPrevLine->startsWithNamespace() && 1380 PreviousLine->startsWith(tok::l_brace)) && 1381 !startsExternCBlock(*PreviousLine)) 1382 Newlines = 1; 1383 1384 // Insert or remove empty line before access specifiers. 1385 if (PreviousLine && RootToken.isAccessSpecifier()) { 1386 switch (Style.EmptyLineBeforeAccessModifier) { 1387 case FormatStyle::ELBAMS_Never: 1388 if (Newlines > 1) 1389 Newlines = 1; 1390 break; 1391 case FormatStyle::ELBAMS_Leave: 1392 Newlines = std::max(RootToken.NewlinesBefore, 1u); 1393 break; 1394 case FormatStyle::ELBAMS_LogicalBlock: 1395 if (PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) && Newlines <= 1) 1396 Newlines = 2; 1397 if (PreviousLine->First->isAccessSpecifier()) 1398 Newlines = 1; // Previous is an access modifier remove all new lines. 1399 break; 1400 case FormatStyle::ELBAMS_Always: { 1401 const FormatToken *previousToken; 1402 if (PreviousLine->Last->is(tok::comment)) 1403 previousToken = PreviousLine->Last->getPreviousNonComment(); 1404 else 1405 previousToken = PreviousLine->Last; 1406 if ((!previousToken || !previousToken->is(tok::l_brace)) && Newlines <= 1) 1407 Newlines = 2; 1408 } break; 1409 } 1410 } 1411 1412 // Insert or remove empty line after access specifiers. 1413 if (PreviousLine && PreviousLine->First->isAccessSpecifier() && 1414 (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline)) { 1415 // EmptyLineBeforeAccessModifier is handling the case when two access 1416 // modifiers follow each other. 1417 if (!RootToken.isAccessSpecifier()) { 1418 switch (Style.EmptyLineAfterAccessModifier) { 1419 case FormatStyle::ELAAMS_Never: 1420 Newlines = 1; 1421 break; 1422 case FormatStyle::ELAAMS_Leave: 1423 Newlines = std::max(Newlines, 1u); 1424 break; 1425 case FormatStyle::ELAAMS_Always: 1426 if (RootToken.is(tok::r_brace)) // Do not add at end of class. 1427 Newlines = 1u; 1428 else 1429 Newlines = std::max(Newlines, 2u); 1430 break; 1431 } 1432 } 1433 } 1434 1435 if (Newlines) 1436 Indent = NewlineIndent; 1437 1438 // Preprocessor directives get indented before the hash only if specified. In 1439 // Javascript import statements are indented like normal statements. 1440 if (!Style.isJavaScript() && 1441 Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash && 1442 (Line.Type == LT_PreprocessorDirective || 1443 Line.Type == LT_ImportStatement)) 1444 Indent = 0; 1445 1446 Whitespaces->replaceWhitespace(RootToken, Newlines, Indent, Indent, 1447 /*IsAligned=*/false, 1448 Line.InPPDirective && 1449 !RootToken.HasUnescapedNewline); 1450 } 1451 1452 unsigned 1453 UnwrappedLineFormatter::getColumnLimit(bool InPPDirective, 1454 const AnnotatedLine *NextLine) const { 1455 // In preprocessor directives reserve two chars for trailing " \" if the 1456 // next line continues the preprocessor directive. 1457 bool ContinuesPPDirective = 1458 InPPDirective && 1459 // If there is no next line, this is likely a child line and the parent 1460 // continues the preprocessor directive. 1461 (!NextLine || 1462 (NextLine->InPPDirective && 1463 // If there is an unescaped newline between this line and the next, the 1464 // next line starts a new preprocessor directive. 1465 !NextLine->First->HasUnescapedNewline)); 1466 return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0); 1467 } 1468 1469 } // namespace format 1470 } // namespace clang 1471