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