1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// \brief This file contains the implementation of the UnwrappedLineParser,
12 /// which turns a stream of tokens into UnwrappedLines.
13 ///
14 //===----------------------------------------------------------------------===//
15
16 #include "UnwrappedLineParser.h"
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/Support/Debug.h"
19 #include "llvm/Support/raw_ostream.h"
20
21 #define DEBUG_TYPE "format-parser"
22
23 namespace clang {
24 namespace format {
25
26 class FormatTokenSource {
27 public:
~FormatTokenSource()28 virtual ~FormatTokenSource() {}
29 virtual FormatToken *getNextToken() = 0;
30
31 virtual unsigned getPosition() = 0;
32 virtual FormatToken *setPosition(unsigned Position) = 0;
33 };
34
35 namespace {
36
37 class ScopedDeclarationState {
38 public:
ScopedDeclarationState(UnwrappedLine & Line,std::vector<bool> & Stack,bool MustBeDeclaration)39 ScopedDeclarationState(UnwrappedLine &Line, std::vector<bool> &Stack,
40 bool MustBeDeclaration)
41 : Line(Line), Stack(Stack) {
42 Line.MustBeDeclaration = MustBeDeclaration;
43 Stack.push_back(MustBeDeclaration);
44 }
~ScopedDeclarationState()45 ~ScopedDeclarationState() {
46 Stack.pop_back();
47 if (!Stack.empty())
48 Line.MustBeDeclaration = Stack.back();
49 else
50 Line.MustBeDeclaration = true;
51 }
52
53 private:
54 UnwrappedLine &Line;
55 std::vector<bool> &Stack;
56 };
57
58 class ScopedMacroState : public FormatTokenSource {
59 public:
ScopedMacroState(UnwrappedLine & Line,FormatTokenSource * & TokenSource,FormatToken * & ResetToken)60 ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource,
61 FormatToken *&ResetToken)
62 : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken),
63 PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource),
64 Token(nullptr) {
65 TokenSource = this;
66 Line.Level = 0;
67 Line.InPPDirective = true;
68 }
69
~ScopedMacroState()70 ~ScopedMacroState() override {
71 TokenSource = PreviousTokenSource;
72 ResetToken = Token;
73 Line.InPPDirective = false;
74 Line.Level = PreviousLineLevel;
75 }
76
getNextToken()77 FormatToken *getNextToken() override {
78 // The \c UnwrappedLineParser guards against this by never calling
79 // \c getNextToken() after it has encountered the first eof token.
80 assert(!eof());
81 Token = PreviousTokenSource->getNextToken();
82 if (eof())
83 return getFakeEOF();
84 return Token;
85 }
86
getPosition()87 unsigned getPosition() override { return PreviousTokenSource->getPosition(); }
88
setPosition(unsigned Position)89 FormatToken *setPosition(unsigned Position) override {
90 Token = PreviousTokenSource->setPosition(Position);
91 return Token;
92 }
93
94 private:
eof()95 bool eof() { return Token && Token->HasUnescapedNewline; }
96
getFakeEOF()97 FormatToken *getFakeEOF() {
98 static bool EOFInitialized = false;
99 static FormatToken FormatTok;
100 if (!EOFInitialized) {
101 FormatTok.Tok.startToken();
102 FormatTok.Tok.setKind(tok::eof);
103 EOFInitialized = true;
104 }
105 return &FormatTok;
106 }
107
108 UnwrappedLine &Line;
109 FormatTokenSource *&TokenSource;
110 FormatToken *&ResetToken;
111 unsigned PreviousLineLevel;
112 FormatTokenSource *PreviousTokenSource;
113
114 FormatToken *Token;
115 };
116
117 } // end anonymous namespace
118
119 class ScopedLineState {
120 public:
ScopedLineState(UnwrappedLineParser & Parser,bool SwitchToPreprocessorLines=false)121 ScopedLineState(UnwrappedLineParser &Parser,
122 bool SwitchToPreprocessorLines = false)
123 : Parser(Parser), OriginalLines(Parser.CurrentLines) {
124 if (SwitchToPreprocessorLines)
125 Parser.CurrentLines = &Parser.PreprocessorDirectives;
126 else if (!Parser.Line->Tokens.empty())
127 Parser.CurrentLines = &Parser.Line->Tokens.back().Children;
128 PreBlockLine = std::move(Parser.Line);
129 Parser.Line = llvm::make_unique<UnwrappedLine>();
130 Parser.Line->Level = PreBlockLine->Level;
131 Parser.Line->InPPDirective = PreBlockLine->InPPDirective;
132 }
133
~ScopedLineState()134 ~ScopedLineState() {
135 if (!Parser.Line->Tokens.empty()) {
136 Parser.addUnwrappedLine();
137 }
138 assert(Parser.Line->Tokens.empty());
139 Parser.Line = std::move(PreBlockLine);
140 if (Parser.CurrentLines == &Parser.PreprocessorDirectives)
141 Parser.MustBreakBeforeNextToken = true;
142 Parser.CurrentLines = OriginalLines;
143 }
144
145 private:
146 UnwrappedLineParser &Parser;
147
148 std::unique_ptr<UnwrappedLine> PreBlockLine;
149 SmallVectorImpl<UnwrappedLine> *OriginalLines;
150 };
151
152 class CompoundStatementIndenter {
153 public:
CompoundStatementIndenter(UnwrappedLineParser * Parser,const FormatStyle & Style,unsigned & LineLevel)154 CompoundStatementIndenter(UnwrappedLineParser *Parser,
155 const FormatStyle &Style, unsigned &LineLevel)
156 : LineLevel(LineLevel), OldLineLevel(LineLevel) {
157 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman) {
158 Parser->addUnwrappedLine();
159 } else if (Style.BreakBeforeBraces == FormatStyle::BS_GNU) {
160 Parser->addUnwrappedLine();
161 ++LineLevel;
162 }
163 }
~CompoundStatementIndenter()164 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; }
165
166 private:
167 unsigned &LineLevel;
168 unsigned OldLineLevel;
169 };
170
171 namespace {
172
173 class IndexedTokenSource : public FormatTokenSource {
174 public:
IndexedTokenSource(ArrayRef<FormatToken * > Tokens)175 IndexedTokenSource(ArrayRef<FormatToken *> Tokens)
176 : Tokens(Tokens), Position(-1) {}
177
getNextToken()178 FormatToken *getNextToken() override {
179 ++Position;
180 return Tokens[Position];
181 }
182
getPosition()183 unsigned getPosition() override {
184 assert(Position >= 0);
185 return Position;
186 }
187
setPosition(unsigned P)188 FormatToken *setPosition(unsigned P) override {
189 Position = P;
190 return Tokens[Position];
191 }
192
reset()193 void reset() { Position = -1; }
194
195 private:
196 ArrayRef<FormatToken *> Tokens;
197 int Position;
198 };
199
200 } // end anonymous namespace
201
UnwrappedLineParser(const FormatStyle & Style,const AdditionalKeywords & Keywords,ArrayRef<FormatToken * > Tokens,UnwrappedLineConsumer & Callback)202 UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style,
203 const AdditionalKeywords &Keywords,
204 ArrayRef<FormatToken *> Tokens,
205 UnwrappedLineConsumer &Callback)
206 : Line(new UnwrappedLine), MustBreakBeforeNextToken(false),
207 CurrentLines(&Lines), Style(Style), Keywords(Keywords), Tokens(nullptr),
208 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {}
209
reset()210 void UnwrappedLineParser::reset() {
211 PPBranchLevel = -1;
212 Line.reset(new UnwrappedLine);
213 CommentsBeforeNextToken.clear();
214 FormatTok = nullptr;
215 MustBreakBeforeNextToken = false;
216 PreprocessorDirectives.clear();
217 CurrentLines = &Lines;
218 DeclarationScopeStack.clear();
219 PPStack.clear();
220 }
221
parse()222 void UnwrappedLineParser::parse() {
223 IndexedTokenSource TokenSource(AllTokens);
224 do {
225 DEBUG(llvm::dbgs() << "----\n");
226 reset();
227 Tokens = &TokenSource;
228 TokenSource.reset();
229
230 readToken();
231 parseFile();
232 // Create line with eof token.
233 pushToken(FormatTok);
234 addUnwrappedLine();
235
236 for (SmallVectorImpl<UnwrappedLine>::iterator I = Lines.begin(),
237 E = Lines.end();
238 I != E; ++I) {
239 Callback.consumeUnwrappedLine(*I);
240 }
241 Callback.finishRun();
242 Lines.clear();
243 while (!PPLevelBranchIndex.empty() &&
244 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) {
245 PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1);
246 PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1);
247 }
248 if (!PPLevelBranchIndex.empty()) {
249 ++PPLevelBranchIndex.back();
250 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size());
251 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back());
252 }
253 } while (!PPLevelBranchIndex.empty());
254 }
255
parseFile()256 void UnwrappedLineParser::parseFile() {
257 // The top-level context in a file always has declarations, except for pre-
258 // processor directives and JavaScript files.
259 bool MustBeDeclaration =
260 !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript;
261 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
262 MustBeDeclaration);
263 parseLevel(/*HasOpeningBrace=*/false);
264 // Make sure to format the remaining tokens.
265 flushComments(true);
266 addUnwrappedLine();
267 }
268
parseLevel(bool HasOpeningBrace)269 void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) {
270 bool SwitchLabelEncountered = false;
271 do {
272 tok::TokenKind kind = FormatTok->Tok.getKind();
273 if (FormatTok->Type == TT_MacroBlockBegin) {
274 kind = tok::l_brace;
275 } else if (FormatTok->Type == TT_MacroBlockEnd) {
276 kind = tok::r_brace;
277 }
278
279 switch (kind) {
280 case tok::comment:
281 nextToken();
282 addUnwrappedLine();
283 break;
284 case tok::l_brace:
285 // FIXME: Add parameter whether this can happen - if this happens, we must
286 // be in a non-declaration context.
287 parseBlock(/*MustBeDeclaration=*/false);
288 addUnwrappedLine();
289 break;
290 case tok::r_brace:
291 if (HasOpeningBrace)
292 return;
293 nextToken();
294 addUnwrappedLine();
295 break;
296 case tok::kw_default:
297 case tok::kw_case:
298 if (!SwitchLabelEncountered &&
299 (Style.IndentCaseLabels || (Line->InPPDirective && Line->Level == 1)))
300 ++Line->Level;
301 SwitchLabelEncountered = true;
302 parseStructuralElement();
303 break;
304 default:
305 parseStructuralElement();
306 break;
307 }
308 } while (!eof());
309 }
310
calculateBraceTypes(bool ExpectClassBody)311 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
312 // We'll parse forward through the tokens until we hit
313 // a closing brace or eof - note that getNextToken() will
314 // parse macros, so this will magically work inside macro
315 // definitions, too.
316 unsigned StoredPosition = Tokens->getPosition();
317 FormatToken *Tok = FormatTok;
318 // Keep a stack of positions of lbrace tokens. We will
319 // update information about whether an lbrace starts a
320 // braced init list or a different block during the loop.
321 SmallVector<FormatToken *, 8> LBraceStack;
322 assert(Tok->Tok.is(tok::l_brace));
323 do {
324 // Get next none-comment token.
325 FormatToken *NextTok;
326 unsigned ReadTokens = 0;
327 do {
328 NextTok = Tokens->getNextToken();
329 ++ReadTokens;
330 } while (NextTok->is(tok::comment));
331
332 switch (Tok->Tok.getKind()) {
333 case tok::l_brace:
334 Tok->BlockKind = BK_Unknown;
335 LBraceStack.push_back(Tok);
336 break;
337 case tok::r_brace:
338 if (!LBraceStack.empty()) {
339 if (LBraceStack.back()->BlockKind == BK_Unknown) {
340 bool ProbablyBracedList = false;
341 if (Style.Language == FormatStyle::LK_Proto) {
342 ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square);
343 } else {
344 // Using OriginalColumn to distinguish between ObjC methods and
345 // binary operators is a bit hacky.
346 bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) &&
347 NextTok->OriginalColumn == 0;
348
349 // If there is a comma, semicolon or right paren after the closing
350 // brace, we assume this is a braced initializer list. Note that
351 // regardless how we mark inner braces here, we will overwrite the
352 // BlockKind later if we parse a braced list (where all blocks
353 // inside are by default braced lists), or when we explicitly detect
354 // blocks (for example while parsing lambdas).
355 //
356 // We exclude + and - as they can be ObjC visibility modifiers.
357 ProbablyBracedList =
358 NextTok->isOneOf(tok::comma, tok::period, tok::colon,
359 tok::r_paren, tok::r_square, tok::l_brace,
360 tok::l_paren, tok::ellipsis) ||
361 (NextTok->is(tok::semi) &&
362 (!ExpectClassBody || LBraceStack.size() != 1)) ||
363 (NextTok->isBinaryOperator() && !NextIsObjCMethod);
364 }
365 if (ProbablyBracedList) {
366 Tok->BlockKind = BK_BracedInit;
367 LBraceStack.back()->BlockKind = BK_BracedInit;
368 } else {
369 Tok->BlockKind = BK_Block;
370 LBraceStack.back()->BlockKind = BK_Block;
371 }
372 }
373 LBraceStack.pop_back();
374 }
375 break;
376 case tok::at:
377 case tok::semi:
378 case tok::kw_if:
379 case tok::kw_while:
380 case tok::kw_for:
381 case tok::kw_switch:
382 case tok::kw_try:
383 case tok::kw___try:
384 if (!LBraceStack.empty())
385 LBraceStack.back()->BlockKind = BK_Block;
386 break;
387 default:
388 break;
389 }
390 Tok = NextTok;
391 } while (Tok->Tok.isNot(tok::eof) && !LBraceStack.empty());
392 // Assume other blocks for all unclosed opening braces.
393 for (unsigned i = 0, e = LBraceStack.size(); i != e; ++i) {
394 if (LBraceStack[i]->BlockKind == BK_Unknown)
395 LBraceStack[i]->BlockKind = BK_Block;
396 }
397
398 FormatTok = Tokens->setPosition(StoredPosition);
399 }
400
parseBlock(bool MustBeDeclaration,bool AddLevel,bool MunchSemi)401 void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel,
402 bool MunchSemi) {
403 assert(FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) &&
404 "'{' or macro block token expected");
405 const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin);
406
407 unsigned InitialLevel = Line->Level;
408 nextToken();
409
410 if (MacroBlock && FormatTok->is(tok::l_paren))
411 parseParens();
412
413 addUnwrappedLine();
414
415 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
416 MustBeDeclaration);
417 if (AddLevel)
418 ++Line->Level;
419 parseLevel(/*HasOpeningBrace=*/true);
420
421 if (MacroBlock ? !FormatTok->is(TT_MacroBlockEnd)
422 : !FormatTok->is(tok::r_brace)) {
423 Line->Level = InitialLevel;
424 return;
425 }
426
427 nextToken(); // Munch the closing brace.
428
429 if (MacroBlock && FormatTok->is(tok::l_paren))
430 parseParens();
431
432 if (MunchSemi && FormatTok->Tok.is(tok::semi))
433 nextToken();
434 Line->Level = InitialLevel;
435 }
436
isGoogScope(const UnwrappedLine & Line)437 static bool isGoogScope(const UnwrappedLine &Line) {
438 // FIXME: Closure-library specific stuff should not be hard-coded but be
439 // configurable.
440 if (Line.Tokens.size() < 4)
441 return false;
442 auto I = Line.Tokens.begin();
443 if (I->Tok->TokenText != "goog")
444 return false;
445 ++I;
446 if (I->Tok->isNot(tok::period))
447 return false;
448 ++I;
449 if (I->Tok->TokenText != "scope")
450 return false;
451 ++I;
452 return I->Tok->is(tok::l_paren);
453 }
454
ShouldBreakBeforeBrace(const FormatStyle & Style,const FormatToken & InitialToken)455 static bool ShouldBreakBeforeBrace(const FormatStyle &Style,
456 const FormatToken &InitialToken) {
457 switch (Style.BreakBeforeBraces) {
458 case FormatStyle::BS_Linux:
459 return InitialToken.isOneOf(tok::kw_namespace, tok::kw_class);
460 case FormatStyle::BS_Mozilla:
461 return InitialToken.isOneOf(tok::kw_class, tok::kw_struct, tok::kw_union);
462 case FormatStyle::BS_Allman:
463 case FormatStyle::BS_GNU:
464 return true;
465 default:
466 return false;
467 }
468 }
469
parseChildBlock()470 void UnwrappedLineParser::parseChildBlock() {
471 FormatTok->BlockKind = BK_Block;
472 nextToken();
473 {
474 bool GoogScope =
475 Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line);
476 ScopedLineState LineState(*this);
477 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
478 /*MustBeDeclaration=*/false);
479 Line->Level += GoogScope ? 0 : 1;
480 parseLevel(/*HasOpeningBrace=*/true);
481 flushComments(isOnNewLine(*FormatTok));
482 Line->Level -= GoogScope ? 0 : 1;
483 }
484 nextToken();
485 }
486
parsePPDirective()487 void UnwrappedLineParser::parsePPDirective() {
488 assert(FormatTok->Tok.is(tok::hash) && "'#' expected");
489 ScopedMacroState MacroState(*Line, Tokens, FormatTok);
490 nextToken();
491
492 if (!FormatTok->Tok.getIdentifierInfo()) {
493 parsePPUnknown();
494 return;
495 }
496
497 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) {
498 case tok::pp_define:
499 parsePPDefine();
500 return;
501 case tok::pp_if:
502 parsePPIf(/*IfDef=*/false);
503 break;
504 case tok::pp_ifdef:
505 case tok::pp_ifndef:
506 parsePPIf(/*IfDef=*/true);
507 break;
508 case tok::pp_else:
509 parsePPElse();
510 break;
511 case tok::pp_elif:
512 parsePPElIf();
513 break;
514 case tok::pp_endif:
515 parsePPEndIf();
516 break;
517 default:
518 parsePPUnknown();
519 break;
520 }
521 }
522
conditionalCompilationCondition(bool Unreachable)523 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) {
524 if (Unreachable || (!PPStack.empty() && PPStack.back() == PP_Unreachable))
525 PPStack.push_back(PP_Unreachable);
526 else
527 PPStack.push_back(PP_Conditional);
528 }
529
conditionalCompilationStart(bool Unreachable)530 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) {
531 ++PPBranchLevel;
532 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size());
533 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) {
534 PPLevelBranchIndex.push_back(0);
535 PPLevelBranchCount.push_back(0);
536 }
537 PPChainBranchIndex.push(0);
538 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0;
539 conditionalCompilationCondition(Unreachable || Skip);
540 }
541
conditionalCompilationAlternative()542 void UnwrappedLineParser::conditionalCompilationAlternative() {
543 if (!PPStack.empty())
544 PPStack.pop_back();
545 assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
546 if (!PPChainBranchIndex.empty())
547 ++PPChainBranchIndex.top();
548 conditionalCompilationCondition(
549 PPBranchLevel >= 0 && !PPChainBranchIndex.empty() &&
550 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top());
551 }
552
conditionalCompilationEnd()553 void UnwrappedLineParser::conditionalCompilationEnd() {
554 assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
555 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) {
556 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel]) {
557 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1;
558 }
559 }
560 // Guard against #endif's without #if.
561 if (PPBranchLevel > 0)
562 --PPBranchLevel;
563 if (!PPChainBranchIndex.empty())
564 PPChainBranchIndex.pop();
565 if (!PPStack.empty())
566 PPStack.pop_back();
567 }
568
parsePPIf(bool IfDef)569 void UnwrappedLineParser::parsePPIf(bool IfDef) {
570 nextToken();
571 bool IsLiteralFalse = (FormatTok->Tok.isLiteral() &&
572 FormatTok->Tok.getLiteralData() != nullptr &&
573 StringRef(FormatTok->Tok.getLiteralData(),
574 FormatTok->Tok.getLength()) == "0") ||
575 FormatTok->Tok.is(tok::kw_false);
576 conditionalCompilationStart(!IfDef && IsLiteralFalse);
577 parsePPUnknown();
578 }
579
parsePPElse()580 void UnwrappedLineParser::parsePPElse() {
581 conditionalCompilationAlternative();
582 parsePPUnknown();
583 }
584
parsePPElIf()585 void UnwrappedLineParser::parsePPElIf() { parsePPElse(); }
586
parsePPEndIf()587 void UnwrappedLineParser::parsePPEndIf() {
588 conditionalCompilationEnd();
589 parsePPUnknown();
590 }
591
parsePPDefine()592 void UnwrappedLineParser::parsePPDefine() {
593 nextToken();
594
595 if (FormatTok->Tok.getKind() != tok::identifier) {
596 parsePPUnknown();
597 return;
598 }
599 nextToken();
600 if (FormatTok->Tok.getKind() == tok::l_paren &&
601 FormatTok->WhitespaceRange.getBegin() ==
602 FormatTok->WhitespaceRange.getEnd()) {
603 parseParens();
604 }
605 addUnwrappedLine();
606 Line->Level = 1;
607
608 // Errors during a preprocessor directive can only affect the layout of the
609 // preprocessor directive, and thus we ignore them. An alternative approach
610 // would be to use the same approach we use on the file level (no
611 // re-indentation if there was a structural error) within the macro
612 // definition.
613 parseFile();
614 }
615
parsePPUnknown()616 void UnwrappedLineParser::parsePPUnknown() {
617 do {
618 nextToken();
619 } while (!eof());
620 addUnwrappedLine();
621 }
622
623 // Here we blacklist certain tokens that are not usually the first token in an
624 // unwrapped line. This is used in attempt to distinguish macro calls without
625 // trailing semicolons from other constructs split to several lines.
tokenCanStartNewLine(const clang::Token & Tok)626 static bool tokenCanStartNewLine(const clang::Token &Tok) {
627 // Semicolon can be a null-statement, l_square can be a start of a macro or
628 // a C++11 attribute, but this doesn't seem to be common.
629 return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) &&
630 Tok.isNot(tok::l_square) &&
631 // Tokens that can only be used as binary operators and a part of
632 // overloaded operator names.
633 Tok.isNot(tok::period) && Tok.isNot(tok::periodstar) &&
634 Tok.isNot(tok::arrow) && Tok.isNot(tok::arrowstar) &&
635 Tok.isNot(tok::less) && Tok.isNot(tok::greater) &&
636 Tok.isNot(tok::slash) && Tok.isNot(tok::percent) &&
637 Tok.isNot(tok::lessless) && Tok.isNot(tok::greatergreater) &&
638 Tok.isNot(tok::equal) && Tok.isNot(tok::plusequal) &&
639 Tok.isNot(tok::minusequal) && Tok.isNot(tok::starequal) &&
640 Tok.isNot(tok::slashequal) && Tok.isNot(tok::percentequal) &&
641 Tok.isNot(tok::ampequal) && Tok.isNot(tok::pipeequal) &&
642 Tok.isNot(tok::caretequal) && Tok.isNot(tok::greatergreaterequal) &&
643 Tok.isNot(tok::lesslessequal) &&
644 // Colon is used in labels, base class lists, initializer lists,
645 // range-based for loops, ternary operator, but should never be the
646 // first token in an unwrapped line.
647 Tok.isNot(tok::colon) &&
648 // 'noexcept' is a trailing annotation.
649 Tok.isNot(tok::kw_noexcept);
650 }
651
parseStructuralElement()652 void UnwrappedLineParser::parseStructuralElement() {
653 assert(!FormatTok->Tok.is(tok::l_brace));
654 switch (FormatTok->Tok.getKind()) {
655 case tok::at:
656 nextToken();
657 if (FormatTok->Tok.is(tok::l_brace)) {
658 parseBracedList();
659 break;
660 }
661 switch (FormatTok->Tok.getObjCKeywordID()) {
662 case tok::objc_public:
663 case tok::objc_protected:
664 case tok::objc_package:
665 case tok::objc_private:
666 return parseAccessSpecifier();
667 case tok::objc_interface:
668 case tok::objc_implementation:
669 return parseObjCInterfaceOrImplementation();
670 case tok::objc_protocol:
671 return parseObjCProtocol();
672 case tok::objc_end:
673 return; // Handled by the caller.
674 case tok::objc_optional:
675 case tok::objc_required:
676 nextToken();
677 addUnwrappedLine();
678 return;
679 case tok::objc_autoreleasepool:
680 nextToken();
681 if (FormatTok->Tok.is(tok::l_brace)) {
682 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
683 Style.BreakBeforeBraces == FormatStyle::BS_GNU)
684 addUnwrappedLine();
685 parseBlock(/*MustBeDeclaration=*/false);
686 }
687 addUnwrappedLine();
688 return;
689 case tok::objc_try:
690 // This branch isn't strictly necessary (the kw_try case below would
691 // do this too after the tok::at is parsed above). But be explicit.
692 parseTryCatch();
693 return;
694 default:
695 break;
696 }
697 break;
698 case tok::kw_asm:
699 nextToken();
700 if (FormatTok->is(tok::l_brace)) {
701 FormatTok->Type = TT_InlineASMBrace;
702 nextToken();
703 while (FormatTok && FormatTok->isNot(tok::eof)) {
704 if (FormatTok->is(tok::r_brace)) {
705 FormatTok->Type = TT_InlineASMBrace;
706 nextToken();
707 addUnwrappedLine();
708 break;
709 }
710 FormatTok->Finalized = true;
711 nextToken();
712 }
713 }
714 break;
715 case tok::kw_namespace:
716 parseNamespace();
717 return;
718 case tok::kw_inline:
719 nextToken();
720 if (FormatTok->Tok.is(tok::kw_namespace)) {
721 parseNamespace();
722 return;
723 }
724 break;
725 case tok::kw_public:
726 case tok::kw_protected:
727 case tok::kw_private:
728 if (Style.Language == FormatStyle::LK_Java ||
729 Style.Language == FormatStyle::LK_JavaScript)
730 nextToken();
731 else
732 parseAccessSpecifier();
733 return;
734 case tok::kw_if:
735 parseIfThenElse();
736 return;
737 case tok::kw_for:
738 case tok::kw_while:
739 parseForOrWhileLoop();
740 return;
741 case tok::kw_do:
742 parseDoWhile();
743 return;
744 case tok::kw_switch:
745 parseSwitch();
746 return;
747 case tok::kw_default:
748 nextToken();
749 parseLabel();
750 return;
751 case tok::kw_case:
752 parseCaseLabel();
753 return;
754 case tok::kw_try:
755 case tok::kw___try:
756 parseTryCatch();
757 return;
758 case tok::kw_extern:
759 nextToken();
760 if (FormatTok->Tok.is(tok::string_literal)) {
761 nextToken();
762 if (FormatTok->Tok.is(tok::l_brace)) {
763 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/false);
764 addUnwrappedLine();
765 return;
766 }
767 }
768 break;
769 case tok::kw_export:
770 if (Style.Language == FormatStyle::LK_JavaScript) {
771 parseJavaScriptEs6ImportExport();
772 return;
773 }
774 break;
775 case tok::identifier:
776 if (FormatTok->is(TT_ForEachMacro)) {
777 parseForOrWhileLoop();
778 return;
779 }
780 if (FormatTok->is(TT_MacroBlockBegin)) {
781 parseBlock(/*MustBeDeclaration=*/false, /*AddLevel=*/true,
782 /*MunchSemi=*/false);
783 return;
784 }
785 if (Style.Language == FormatStyle::LK_JavaScript &&
786 FormatTok->is(Keywords.kw_import)) {
787 parseJavaScriptEs6ImportExport();
788 return;
789 }
790 if (FormatTok->is(Keywords.kw_signals)) {
791 nextToken();
792 if (FormatTok->is(tok::colon)) {
793 nextToken();
794 addUnwrappedLine();
795 }
796 return;
797 }
798 // In all other cases, parse the declaration.
799 break;
800 default:
801 break;
802 }
803 do {
804 switch (FormatTok->Tok.getKind()) {
805 case tok::at:
806 nextToken();
807 if (FormatTok->Tok.is(tok::l_brace))
808 parseBracedList();
809 break;
810 case tok::kw_enum:
811 // parseEnum falls through and does not yet add an unwrapped line as an
812 // enum definition can start a structural element.
813 parseEnum();
814 // This does not apply for Java and JavaScript.
815 if (Style.Language == FormatStyle::LK_Java ||
816 Style.Language == FormatStyle::LK_JavaScript) {
817 addUnwrappedLine();
818 return;
819 }
820 break;
821 case tok::kw_typedef:
822 nextToken();
823 if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS,
824 Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS))
825 parseEnum();
826 break;
827 case tok::kw_struct:
828 case tok::kw_union:
829 case tok::kw_class:
830 // parseRecord falls through and does not yet add an unwrapped line as a
831 // record declaration or definition can start a structural element.
832 parseRecord();
833 // This does not apply for Java and JavaScript.
834 if (Style.Language == FormatStyle::LK_Java ||
835 Style.Language == FormatStyle::LK_JavaScript) {
836 addUnwrappedLine();
837 return;
838 }
839 break;
840 case tok::period:
841 nextToken();
842 // In Java, classes have an implicit static member "class".
843 if (Style.Language == FormatStyle::LK_Java && FormatTok &&
844 FormatTok->is(tok::kw_class))
845 nextToken();
846 break;
847 case tok::semi:
848 nextToken();
849 addUnwrappedLine();
850 return;
851 case tok::r_brace:
852 addUnwrappedLine();
853 return;
854 case tok::l_paren:
855 parseParens();
856 break;
857 case tok::caret:
858 nextToken();
859 if (FormatTok->Tok.isAnyIdentifier() ||
860 FormatTok->isSimpleTypeSpecifier())
861 nextToken();
862 if (FormatTok->is(tok::l_paren))
863 parseParens();
864 if (FormatTok->is(tok::l_brace))
865 parseChildBlock();
866 break;
867 case tok::l_brace:
868 if (!tryToParseBracedList()) {
869 // A block outside of parentheses must be the last part of a
870 // structural element.
871 // FIXME: Figure out cases where this is not true, and add projections
872 // for them (the one we know is missing are lambdas).
873 if (Style.BreakBeforeBraces != FormatStyle::BS_Attach)
874 addUnwrappedLine();
875 FormatTok->Type = TT_FunctionLBrace;
876 parseBlock(/*MustBeDeclaration=*/false);
877 addUnwrappedLine();
878 return;
879 }
880 // Otherwise this was a braced init list, and the structural
881 // element continues.
882 break;
883 case tok::kw_try:
884 // We arrive here when parsing function-try blocks.
885 parseTryCatch();
886 return;
887 case tok::identifier: {
888 if (FormatTok->is(TT_MacroBlockEnd)) {
889 addUnwrappedLine();
890 return;
891 }
892
893 // Parse function literal unless 'function' is the first token in a line
894 // in which case this should be treated as a free-standing function.
895 if (Style.Language == FormatStyle::LK_JavaScript &&
896 FormatTok->is(Keywords.kw_function) && Line->Tokens.size() > 0) {
897 tryToParseJSFunction();
898 break;
899 }
900 if ((Style.Language == FormatStyle::LK_JavaScript ||
901 Style.Language == FormatStyle::LK_Java) &&
902 FormatTok->is(Keywords.kw_interface)) {
903 parseRecord();
904 addUnwrappedLine();
905 return;
906 }
907
908 StringRef Text = FormatTok->TokenText;
909 nextToken();
910 if (Line->Tokens.size() == 1 &&
911 // JS doesn't have macros, and within classes colons indicate fields,
912 // not labels.
913 Style.Language != FormatStyle::LK_JavaScript) {
914 if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) {
915 parseLabel();
916 return;
917 }
918 // Recognize function-like macro usages without trailing semicolon as
919 // well as free-standing macros like Q_OBJECT.
920 bool FunctionLike = FormatTok->is(tok::l_paren);
921 if (FunctionLike)
922 parseParens();
923
924 bool FollowedByNewline =
925 CommentsBeforeNextToken.empty()
926 ? FormatTok->NewlinesBefore > 0
927 : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
928
929 if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) &&
930 tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) {
931 addUnwrappedLine();
932 return;
933 }
934 }
935 break;
936 }
937 case tok::equal:
938 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType
939 // TT_JsFatArrow. The always start an expression or a child block if
940 // followed by a curly.
941 if (FormatTok->is(TT_JsFatArrow)) {
942 nextToken();
943 if (FormatTok->is(tok::l_brace))
944 parseChildBlock();
945 break;
946 }
947
948 nextToken();
949 if (FormatTok->Tok.is(tok::l_brace)) {
950 parseBracedList();
951 }
952 break;
953 case tok::l_square:
954 parseSquare();
955 break;
956 case tok::kw_new:
957 parseNew();
958 break;
959 default:
960 nextToken();
961 break;
962 }
963 } while (!eof());
964 }
965
tryToParseLambda()966 bool UnwrappedLineParser::tryToParseLambda() {
967 if (Style.Language != FormatStyle::LK_Cpp) {
968 nextToken();
969 return false;
970 }
971 // FIXME: This is a dirty way to access the previous token. Find a better
972 // solution.
973 if (!Line->Tokens.empty() &&
974 (Line->Tokens.back().Tok->isOneOf(tok::identifier, tok::kw_operator,
975 tok::kw_new, tok::kw_delete) ||
976 Line->Tokens.back().Tok->closesScope() ||
977 Line->Tokens.back().Tok->isSimpleTypeSpecifier())) {
978 nextToken();
979 return false;
980 }
981 assert(FormatTok->is(tok::l_square));
982 FormatToken &LSquare = *FormatTok;
983 if (!tryToParseLambdaIntroducer())
984 return false;
985
986 while (FormatTok->isNot(tok::l_brace)) {
987 if (FormatTok->isSimpleTypeSpecifier()) {
988 nextToken();
989 continue;
990 }
991 switch (FormatTok->Tok.getKind()) {
992 case tok::l_brace:
993 break;
994 case tok::l_paren:
995 parseParens();
996 break;
997 case tok::amp:
998 case tok::star:
999 case tok::kw_const:
1000 case tok::comma:
1001 case tok::less:
1002 case tok::greater:
1003 case tok::identifier:
1004 case tok::coloncolon:
1005 case tok::kw_mutable:
1006 nextToken();
1007 break;
1008 case tok::arrow:
1009 FormatTok->Type = TT_LambdaArrow;
1010 nextToken();
1011 break;
1012 default:
1013 return true;
1014 }
1015 }
1016 LSquare.Type = TT_LambdaLSquare;
1017 parseChildBlock();
1018 return true;
1019 }
1020
tryToParseLambdaIntroducer()1021 bool UnwrappedLineParser::tryToParseLambdaIntroducer() {
1022 nextToken();
1023 if (FormatTok->is(tok::equal)) {
1024 nextToken();
1025 if (FormatTok->is(tok::r_square)) {
1026 nextToken();
1027 return true;
1028 }
1029 if (FormatTok->isNot(tok::comma))
1030 return false;
1031 nextToken();
1032 } else if (FormatTok->is(tok::amp)) {
1033 nextToken();
1034 if (FormatTok->is(tok::r_square)) {
1035 nextToken();
1036 return true;
1037 }
1038 if (!FormatTok->isOneOf(tok::comma, tok::identifier)) {
1039 return false;
1040 }
1041 if (FormatTok->is(tok::comma))
1042 nextToken();
1043 } else if (FormatTok->is(tok::r_square)) {
1044 nextToken();
1045 return true;
1046 }
1047 do {
1048 if (FormatTok->is(tok::amp))
1049 nextToken();
1050 if (!FormatTok->isOneOf(tok::identifier, tok::kw_this))
1051 return false;
1052 nextToken();
1053 if (FormatTok->is(tok::ellipsis))
1054 nextToken();
1055 if (FormatTok->is(tok::comma)) {
1056 nextToken();
1057 } else if (FormatTok->is(tok::r_square)) {
1058 nextToken();
1059 return true;
1060 } else {
1061 return false;
1062 }
1063 } while (!eof());
1064 return false;
1065 }
1066
tryToParseJSFunction()1067 void UnwrappedLineParser::tryToParseJSFunction() {
1068 nextToken();
1069
1070 // Consume function name.
1071 if (FormatTok->is(tok::identifier))
1072 nextToken();
1073
1074 if (FormatTok->isNot(tok::l_paren))
1075 return;
1076
1077 // Parse formal parameter list.
1078 parseParens();
1079
1080 if (FormatTok->is(tok::colon)) {
1081 // Parse a type definition.
1082 nextToken();
1083
1084 // Eat the type declaration. For braced inline object types, balance braces,
1085 // otherwise just parse until finding an l_brace for the function body.
1086 if (FormatTok->is(tok::l_brace))
1087 tryToParseBracedList();
1088 else
1089 while (FormatTok->isNot(tok::l_brace) && !eof())
1090 nextToken();
1091 }
1092
1093 parseChildBlock();
1094 }
1095
tryToParseBracedList()1096 bool UnwrappedLineParser::tryToParseBracedList() {
1097 if (FormatTok->BlockKind == BK_Unknown)
1098 calculateBraceTypes();
1099 assert(FormatTok->BlockKind != BK_Unknown);
1100 if (FormatTok->BlockKind == BK_Block)
1101 return false;
1102 parseBracedList();
1103 return true;
1104 }
1105
parseBracedList(bool ContinueOnSemicolons)1106 bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) {
1107 bool HasError = false;
1108 nextToken();
1109
1110 // FIXME: Once we have an expression parser in the UnwrappedLineParser,
1111 // replace this by using parseAssigmentExpression() inside.
1112 do {
1113 if (Style.Language == FormatStyle::LK_JavaScript) {
1114 if (FormatTok->is(Keywords.kw_function)) {
1115 tryToParseJSFunction();
1116 continue;
1117 }
1118 if (FormatTok->is(TT_JsFatArrow)) {
1119 nextToken();
1120 // Fat arrows can be followed by simple expressions or by child blocks
1121 // in curly braces.
1122 if (FormatTok->is(tok::l_brace)) {
1123 parseChildBlock();
1124 continue;
1125 }
1126 }
1127 }
1128 switch (FormatTok->Tok.getKind()) {
1129 case tok::caret:
1130 nextToken();
1131 if (FormatTok->is(tok::l_brace)) {
1132 parseChildBlock();
1133 }
1134 break;
1135 case tok::l_square:
1136 tryToParseLambda();
1137 break;
1138 case tok::l_brace:
1139 // Assume there are no blocks inside a braced init list apart
1140 // from the ones we explicitly parse out (like lambdas).
1141 FormatTok->BlockKind = BK_BracedInit;
1142 parseBracedList();
1143 break;
1144 case tok::l_paren:
1145 parseParens();
1146 // JavaScript can just have free standing methods and getters/setters in
1147 // object literals. Detect them by a "{" following ")".
1148 if (Style.Language == FormatStyle::LK_JavaScript) {
1149 if (FormatTok->is(tok::l_brace))
1150 parseChildBlock();
1151 break;
1152 }
1153 break;
1154 case tok::r_brace:
1155 nextToken();
1156 return !HasError;
1157 case tok::semi:
1158 HasError = true;
1159 if (!ContinueOnSemicolons)
1160 return !HasError;
1161 nextToken();
1162 break;
1163 case tok::comma:
1164 nextToken();
1165 break;
1166 default:
1167 nextToken();
1168 break;
1169 }
1170 } while (!eof());
1171 return false;
1172 }
1173
parseParens()1174 void UnwrappedLineParser::parseParens() {
1175 assert(FormatTok->Tok.is(tok::l_paren) && "'(' expected.");
1176 nextToken();
1177 do {
1178 switch (FormatTok->Tok.getKind()) {
1179 case tok::l_paren:
1180 parseParens();
1181 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace))
1182 parseChildBlock();
1183 break;
1184 case tok::r_paren:
1185 nextToken();
1186 return;
1187 case tok::r_brace:
1188 // A "}" inside parenthesis is an error if there wasn't a matching "{".
1189 return;
1190 case tok::l_square:
1191 tryToParseLambda();
1192 break;
1193 case tok::l_brace:
1194 if (!tryToParseBracedList())
1195 parseChildBlock();
1196 break;
1197 case tok::at:
1198 nextToken();
1199 if (FormatTok->Tok.is(tok::l_brace))
1200 parseBracedList();
1201 break;
1202 case tok::identifier:
1203 if (Style.Language == FormatStyle::LK_JavaScript &&
1204 FormatTok->is(Keywords.kw_function))
1205 tryToParseJSFunction();
1206 else
1207 nextToken();
1208 break;
1209 default:
1210 nextToken();
1211 break;
1212 }
1213 } while (!eof());
1214 }
1215
parseSquare()1216 void UnwrappedLineParser::parseSquare() {
1217 assert(FormatTok->Tok.is(tok::l_square) && "'[' expected.");
1218 if (tryToParseLambda())
1219 return;
1220 do {
1221 switch (FormatTok->Tok.getKind()) {
1222 case tok::l_paren:
1223 parseParens();
1224 break;
1225 case tok::r_square:
1226 nextToken();
1227 return;
1228 case tok::r_brace:
1229 // A "}" inside parenthesis is an error if there wasn't a matching "{".
1230 return;
1231 case tok::l_square:
1232 parseSquare();
1233 break;
1234 case tok::l_brace: {
1235 if (!tryToParseBracedList())
1236 parseChildBlock();
1237 break;
1238 }
1239 case tok::at:
1240 nextToken();
1241 if (FormatTok->Tok.is(tok::l_brace))
1242 parseBracedList();
1243 break;
1244 default:
1245 nextToken();
1246 break;
1247 }
1248 } while (!eof());
1249 }
1250
parseIfThenElse()1251 void UnwrappedLineParser::parseIfThenElse() {
1252 assert(FormatTok->Tok.is(tok::kw_if) && "'if' expected");
1253 nextToken();
1254 if (FormatTok->Tok.is(tok::l_paren))
1255 parseParens();
1256 bool NeedsUnwrappedLine = false;
1257 if (FormatTok->Tok.is(tok::l_brace)) {
1258 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1259 parseBlock(/*MustBeDeclaration=*/false);
1260 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
1261 Style.BreakBeforeBraces == FormatStyle::BS_GNU) {
1262 addUnwrappedLine();
1263 } else {
1264 NeedsUnwrappedLine = true;
1265 }
1266 } else {
1267 addUnwrappedLine();
1268 ++Line->Level;
1269 parseStructuralElement();
1270 --Line->Level;
1271 }
1272 if (FormatTok->Tok.is(tok::kw_else)) {
1273 if (Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup)
1274 addUnwrappedLine();
1275 nextToken();
1276 if (FormatTok->Tok.is(tok::l_brace)) {
1277 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1278 parseBlock(/*MustBeDeclaration=*/false);
1279 addUnwrappedLine();
1280 } else if (FormatTok->Tok.is(tok::kw_if)) {
1281 parseIfThenElse();
1282 } else {
1283 addUnwrappedLine();
1284 ++Line->Level;
1285 parseStructuralElement();
1286 --Line->Level;
1287 }
1288 } else if (NeedsUnwrappedLine) {
1289 addUnwrappedLine();
1290 }
1291 }
1292
parseTryCatch()1293 void UnwrappedLineParser::parseTryCatch() {
1294 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
1295 nextToken();
1296 bool NeedsUnwrappedLine = false;
1297 if (FormatTok->is(tok::colon)) {
1298 // We are in a function try block, what comes is an initializer list.
1299 nextToken();
1300 while (FormatTok->is(tok::identifier)) {
1301 nextToken();
1302 if (FormatTok->is(tok::l_paren))
1303 parseParens();
1304 if (FormatTok->is(tok::comma))
1305 nextToken();
1306 }
1307 }
1308 // Parse try with resource.
1309 if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren)) {
1310 parseParens();
1311 }
1312 if (FormatTok->is(tok::l_brace)) {
1313 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1314 parseBlock(/*MustBeDeclaration=*/false);
1315 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
1316 Style.BreakBeforeBraces == FormatStyle::BS_GNU ||
1317 Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup) {
1318 addUnwrappedLine();
1319 } else {
1320 NeedsUnwrappedLine = true;
1321 }
1322 } else if (!FormatTok->is(tok::kw_catch)) {
1323 // The C++ standard requires a compound-statement after a try.
1324 // If there's none, we try to assume there's a structuralElement
1325 // and try to continue.
1326 addUnwrappedLine();
1327 ++Line->Level;
1328 parseStructuralElement();
1329 --Line->Level;
1330 }
1331 while (1) {
1332 if (FormatTok->is(tok::at))
1333 nextToken();
1334 if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except,
1335 tok::kw___finally) ||
1336 ((Style.Language == FormatStyle::LK_Java ||
1337 Style.Language == FormatStyle::LK_JavaScript) &&
1338 FormatTok->is(Keywords.kw_finally)) ||
1339 (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) ||
1340 FormatTok->Tok.isObjCAtKeyword(tok::objc_finally))))
1341 break;
1342 nextToken();
1343 while (FormatTok->isNot(tok::l_brace)) {
1344 if (FormatTok->is(tok::l_paren)) {
1345 parseParens();
1346 continue;
1347 }
1348 if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof))
1349 return;
1350 nextToken();
1351 }
1352 NeedsUnwrappedLine = false;
1353 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1354 parseBlock(/*MustBeDeclaration=*/false);
1355 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
1356 Style.BreakBeforeBraces == FormatStyle::BS_GNU ||
1357 Style.BreakBeforeBraces == FormatStyle::BS_Stroustrup) {
1358 addUnwrappedLine();
1359 } else {
1360 NeedsUnwrappedLine = true;
1361 }
1362 }
1363 if (NeedsUnwrappedLine) {
1364 addUnwrappedLine();
1365 }
1366 }
1367
parseNamespace()1368 void UnwrappedLineParser::parseNamespace() {
1369 assert(FormatTok->Tok.is(tok::kw_namespace) && "'namespace' expected");
1370
1371 const FormatToken &InitialToken = *FormatTok;
1372 nextToken();
1373 if (FormatTok->Tok.is(tok::identifier))
1374 nextToken();
1375 if (FormatTok->Tok.is(tok::l_brace)) {
1376 if (ShouldBreakBeforeBrace(Style, InitialToken))
1377 addUnwrappedLine();
1378
1379 bool AddLevel = Style.NamespaceIndentation == FormatStyle::NI_All ||
1380 (Style.NamespaceIndentation == FormatStyle::NI_Inner &&
1381 DeclarationScopeStack.size() > 1);
1382 parseBlock(/*MustBeDeclaration=*/true, AddLevel);
1383 // Munch the semicolon after a namespace. This is more common than one would
1384 // think. Puttin the semicolon into its own line is very ugly.
1385 if (FormatTok->Tok.is(tok::semi))
1386 nextToken();
1387 addUnwrappedLine();
1388 }
1389 // FIXME: Add error handling.
1390 }
1391
parseNew()1392 void UnwrappedLineParser::parseNew() {
1393 assert(FormatTok->is(tok::kw_new) && "'new' expected");
1394 nextToken();
1395 if (Style.Language != FormatStyle::LK_Java)
1396 return;
1397
1398 // In Java, we can parse everything up to the parens, which aren't optional.
1399 do {
1400 // There should not be a ;, { or } before the new's open paren.
1401 if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace))
1402 return;
1403
1404 // Consume the parens.
1405 if (FormatTok->is(tok::l_paren)) {
1406 parseParens();
1407
1408 // If there is a class body of an anonymous class, consume that as child.
1409 if (FormatTok->is(tok::l_brace))
1410 parseChildBlock();
1411 return;
1412 }
1413 nextToken();
1414 } while (!eof());
1415 }
1416
parseForOrWhileLoop()1417 void UnwrappedLineParser::parseForOrWhileLoop() {
1418 assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) &&
1419 "'for', 'while' or foreach macro expected");
1420 nextToken();
1421 if (FormatTok->Tok.is(tok::l_paren))
1422 parseParens();
1423 if (FormatTok->Tok.is(tok::l_brace)) {
1424 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1425 parseBlock(/*MustBeDeclaration=*/false);
1426 addUnwrappedLine();
1427 } else {
1428 addUnwrappedLine();
1429 ++Line->Level;
1430 parseStructuralElement();
1431 --Line->Level;
1432 }
1433 }
1434
parseDoWhile()1435 void UnwrappedLineParser::parseDoWhile() {
1436 assert(FormatTok->Tok.is(tok::kw_do) && "'do' expected");
1437 nextToken();
1438 if (FormatTok->Tok.is(tok::l_brace)) {
1439 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1440 parseBlock(/*MustBeDeclaration=*/false);
1441 if (Style.BreakBeforeBraces == FormatStyle::BS_GNU)
1442 addUnwrappedLine();
1443 } else {
1444 addUnwrappedLine();
1445 ++Line->Level;
1446 parseStructuralElement();
1447 --Line->Level;
1448 }
1449
1450 // FIXME: Add error handling.
1451 if (!FormatTok->Tok.is(tok::kw_while)) {
1452 addUnwrappedLine();
1453 return;
1454 }
1455
1456 nextToken();
1457 parseStructuralElement();
1458 }
1459
parseLabel()1460 void UnwrappedLineParser::parseLabel() {
1461 nextToken();
1462 unsigned OldLineLevel = Line->Level;
1463 if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0))
1464 --Line->Level;
1465 if (CommentsBeforeNextToken.empty() && FormatTok->Tok.is(tok::l_brace)) {
1466 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1467 parseBlock(/*MustBeDeclaration=*/false);
1468 if (FormatTok->Tok.is(tok::kw_break)) {
1469 // "break;" after "}" on its own line only for BS_Allman and BS_GNU
1470 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
1471 Style.BreakBeforeBraces == FormatStyle::BS_GNU) {
1472 addUnwrappedLine();
1473 }
1474 parseStructuralElement();
1475 }
1476 addUnwrappedLine();
1477 } else {
1478 if (FormatTok->is(tok::semi))
1479 nextToken();
1480 addUnwrappedLine();
1481 }
1482 Line->Level = OldLineLevel;
1483 }
1484
parseCaseLabel()1485 void UnwrappedLineParser::parseCaseLabel() {
1486 assert(FormatTok->Tok.is(tok::kw_case) && "'case' expected");
1487 // FIXME: fix handling of complex expressions here.
1488 do {
1489 nextToken();
1490 } while (!eof() && !FormatTok->Tok.is(tok::colon));
1491 parseLabel();
1492 }
1493
parseSwitch()1494 void UnwrappedLineParser::parseSwitch() {
1495 assert(FormatTok->Tok.is(tok::kw_switch) && "'switch' expected");
1496 nextToken();
1497 if (FormatTok->Tok.is(tok::l_paren))
1498 parseParens();
1499 if (FormatTok->Tok.is(tok::l_brace)) {
1500 CompoundStatementIndenter Indenter(this, Style, Line->Level);
1501 parseBlock(/*MustBeDeclaration=*/false);
1502 addUnwrappedLine();
1503 } else {
1504 addUnwrappedLine();
1505 ++Line->Level;
1506 parseStructuralElement();
1507 --Line->Level;
1508 }
1509 }
1510
parseAccessSpecifier()1511 void UnwrappedLineParser::parseAccessSpecifier() {
1512 nextToken();
1513 // Understand Qt's slots.
1514 if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots))
1515 nextToken();
1516 // Otherwise, we don't know what it is, and we'd better keep the next token.
1517 if (FormatTok->Tok.is(tok::colon))
1518 nextToken();
1519 addUnwrappedLine();
1520 }
1521
parseEnum()1522 void UnwrappedLineParser::parseEnum() {
1523 // Won't be 'enum' for NS_ENUMs.
1524 if (FormatTok->Tok.is(tok::kw_enum))
1525 nextToken();
1526
1527 // Eat up enum class ...
1528 if (FormatTok->Tok.is(tok::kw_class) || FormatTok->Tok.is(tok::kw_struct))
1529 nextToken();
1530
1531 while (FormatTok->Tok.getIdentifierInfo() ||
1532 FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less,
1533 tok::greater, tok::comma, tok::question)) {
1534 nextToken();
1535 // We can have macros or attributes in between 'enum' and the enum name.
1536 if (FormatTok->is(tok::l_paren))
1537 parseParens();
1538 if (FormatTok->is(tok::identifier)) {
1539 nextToken();
1540 // If there are two identifiers in a row, this is likely an elaborate
1541 // return type. In Java, this can be "implements", etc.
1542 if (Style.Language == FormatStyle::LK_Cpp &&
1543 FormatTok->is(tok::identifier))
1544 return;
1545 }
1546 }
1547
1548 // Just a declaration or something is wrong.
1549 if (FormatTok->isNot(tok::l_brace))
1550 return;
1551 FormatTok->BlockKind = BK_Block;
1552
1553 if (Style.Language == FormatStyle::LK_Java) {
1554 // Java enums are different.
1555 parseJavaEnumBody();
1556 return;
1557 }
1558
1559 // Parse enum body.
1560 bool HasError = !parseBracedList(/*ContinueOnSemicolons=*/true);
1561 if (HasError) {
1562 if (FormatTok->is(tok::semi))
1563 nextToken();
1564 addUnwrappedLine();
1565 }
1566
1567 // There is no addUnwrappedLine() here so that we fall through to parsing a
1568 // structural element afterwards. Thus, in "enum A {} n, m;",
1569 // "} n, m;" will end up in one unwrapped line.
1570 }
1571
parseJavaEnumBody()1572 void UnwrappedLineParser::parseJavaEnumBody() {
1573 // Determine whether the enum is simple, i.e. does not have a semicolon or
1574 // constants with class bodies. Simple enums can be formatted like braced
1575 // lists, contracted to a single line, etc.
1576 unsigned StoredPosition = Tokens->getPosition();
1577 bool IsSimple = true;
1578 FormatToken *Tok = Tokens->getNextToken();
1579 while (Tok) {
1580 if (Tok->is(tok::r_brace))
1581 break;
1582 if (Tok->isOneOf(tok::l_brace, tok::semi)) {
1583 IsSimple = false;
1584 break;
1585 }
1586 // FIXME: This will also mark enums with braces in the arguments to enum
1587 // constants as "not simple". This is probably fine in practice, though.
1588 Tok = Tokens->getNextToken();
1589 }
1590 FormatTok = Tokens->setPosition(StoredPosition);
1591
1592 if (IsSimple) {
1593 parseBracedList();
1594 addUnwrappedLine();
1595 return;
1596 }
1597
1598 // Parse the body of a more complex enum.
1599 // First add a line for everything up to the "{".
1600 nextToken();
1601 addUnwrappedLine();
1602 ++Line->Level;
1603
1604 // Parse the enum constants.
1605 while (FormatTok) {
1606 if (FormatTok->is(tok::l_brace)) {
1607 // Parse the constant's class body.
1608 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true,
1609 /*MunchSemi=*/false);
1610 } else if (FormatTok->is(tok::l_paren)) {
1611 parseParens();
1612 } else if (FormatTok->is(tok::comma)) {
1613 nextToken();
1614 addUnwrappedLine();
1615 } else if (FormatTok->is(tok::semi)) {
1616 nextToken();
1617 addUnwrappedLine();
1618 break;
1619 } else if (FormatTok->is(tok::r_brace)) {
1620 addUnwrappedLine();
1621 break;
1622 } else {
1623 nextToken();
1624 }
1625 }
1626
1627 // Parse the class body after the enum's ";" if any.
1628 parseLevel(/*HasOpeningBrace=*/true);
1629 nextToken();
1630 --Line->Level;
1631 addUnwrappedLine();
1632 }
1633
parseRecord()1634 void UnwrappedLineParser::parseRecord() {
1635 const FormatToken &InitialToken = *FormatTok;
1636 nextToken();
1637
1638 // The actual identifier can be a nested name specifier, and in macros
1639 // it is often token-pasted.
1640 while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash,
1641 tok::kw___attribute, tok::kw___declspec,
1642 tok::kw_alignas) ||
1643 ((Style.Language == FormatStyle::LK_Java ||
1644 Style.Language == FormatStyle::LK_JavaScript) &&
1645 FormatTok->isOneOf(tok::period, tok::comma))) {
1646 bool IsNonMacroIdentifier =
1647 FormatTok->is(tok::identifier) &&
1648 FormatTok->TokenText != FormatTok->TokenText.upper();
1649 nextToken();
1650 // We can have macros or attributes in between 'class' and the class name.
1651 if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren))
1652 parseParens();
1653 }
1654
1655 // Note that parsing away template declarations here leads to incorrectly
1656 // accepting function declarations as record declarations.
1657 // In general, we cannot solve this problem. Consider:
1658 // class A<int> B() {}
1659 // which can be a function definition or a class definition when B() is a
1660 // macro. If we find enough real-world cases where this is a problem, we
1661 // can parse for the 'template' keyword in the beginning of the statement,
1662 // and thus rule out the record production in case there is no template
1663 // (this would still leave us with an ambiguity between template function
1664 // and class declarations).
1665 if (FormatTok->isOneOf(tok::colon, tok::less)) {
1666 while (!eof()) {
1667 if (FormatTok->is(tok::l_brace)) {
1668 calculateBraceTypes(/*ExpectClassBody=*/true);
1669 if (!tryToParseBracedList())
1670 break;
1671 }
1672 if (FormatTok->Tok.is(tok::semi))
1673 return;
1674 nextToken();
1675 }
1676 }
1677 if (FormatTok->Tok.is(tok::l_brace)) {
1678 if (ShouldBreakBeforeBrace(Style, InitialToken))
1679 addUnwrappedLine();
1680
1681 parseBlock(/*MustBeDeclaration=*/true, /*AddLevel=*/true,
1682 /*MunchSemi=*/false);
1683 }
1684 // There is no addUnwrappedLine() here so that we fall through to parsing a
1685 // structural element afterwards. Thus, in "class A {} n, m;",
1686 // "} n, m;" will end up in one unwrapped line.
1687 }
1688
parseObjCProtocolList()1689 void UnwrappedLineParser::parseObjCProtocolList() {
1690 assert(FormatTok->Tok.is(tok::less) && "'<' expected.");
1691 do
1692 nextToken();
1693 while (!eof() && FormatTok->Tok.isNot(tok::greater));
1694 nextToken(); // Skip '>'.
1695 }
1696
parseObjCUntilAtEnd()1697 void UnwrappedLineParser::parseObjCUntilAtEnd() {
1698 do {
1699 if (FormatTok->Tok.isObjCAtKeyword(tok::objc_end)) {
1700 nextToken();
1701 addUnwrappedLine();
1702 break;
1703 }
1704 if (FormatTok->is(tok::l_brace)) {
1705 parseBlock(/*MustBeDeclaration=*/false);
1706 // In ObjC interfaces, nothing should be following the "}".
1707 addUnwrappedLine();
1708 } else if (FormatTok->is(tok::r_brace)) {
1709 // Ignore stray "}". parseStructuralElement doesn't consume them.
1710 nextToken();
1711 addUnwrappedLine();
1712 } else {
1713 parseStructuralElement();
1714 }
1715 } while (!eof());
1716 }
1717
parseObjCInterfaceOrImplementation()1718 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() {
1719 nextToken();
1720 nextToken(); // interface name
1721
1722 // @interface can be followed by either a base class, or a category.
1723 if (FormatTok->Tok.is(tok::colon)) {
1724 nextToken();
1725 nextToken(); // base class name
1726 } else if (FormatTok->Tok.is(tok::l_paren))
1727 // Skip category, if present.
1728 parseParens();
1729
1730 if (FormatTok->Tok.is(tok::less))
1731 parseObjCProtocolList();
1732
1733 if (FormatTok->Tok.is(tok::l_brace)) {
1734 if (Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
1735 Style.BreakBeforeBraces == FormatStyle::BS_GNU)
1736 addUnwrappedLine();
1737 parseBlock(/*MustBeDeclaration=*/true);
1738 }
1739
1740 // With instance variables, this puts '}' on its own line. Without instance
1741 // variables, this ends the @interface line.
1742 addUnwrappedLine();
1743
1744 parseObjCUntilAtEnd();
1745 }
1746
parseObjCProtocol()1747 void UnwrappedLineParser::parseObjCProtocol() {
1748 nextToken();
1749 nextToken(); // protocol name
1750
1751 if (FormatTok->Tok.is(tok::less))
1752 parseObjCProtocolList();
1753
1754 // Check for protocol declaration.
1755 if (FormatTok->Tok.is(tok::semi)) {
1756 nextToken();
1757 return addUnwrappedLine();
1758 }
1759
1760 addUnwrappedLine();
1761 parseObjCUntilAtEnd();
1762 }
1763
parseJavaScriptEs6ImportExport()1764 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
1765 assert(FormatTok->isOneOf(Keywords.kw_import, tok::kw_export));
1766 nextToken();
1767
1768 // Consume the "default" in "export default class/function".
1769 if (FormatTok->is(tok::kw_default))
1770 nextToken();
1771
1772 // Consume "function" and "default function", so that these get parsed as
1773 // free-standing JS functions, i.e. do not require a trailing semicolon.
1774 if (FormatTok->is(Keywords.kw_function)) {
1775 nextToken();
1776 return;
1777 }
1778
1779 if (FormatTok->isOneOf(tok::kw_const, tok::kw_class, tok::kw_enum,
1780 Keywords.kw_var))
1781 return; // Fall through to parsing the corresponding structure.
1782
1783 if (FormatTok->is(tok::l_brace)) {
1784 FormatTok->BlockKind = BK_Block;
1785 parseBracedList();
1786 }
1787
1788 while (!eof() && FormatTok->isNot(tok::semi) &&
1789 FormatTok->isNot(tok::l_brace)) {
1790 nextToken();
1791 }
1792 }
1793
printDebugInfo(const UnwrappedLine & Line,StringRef Prefix="")1794 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line,
1795 StringRef Prefix = "") {
1796 llvm::dbgs() << Prefix << "Line(" << Line.Level << ")"
1797 << (Line.InPPDirective ? " MACRO" : "") << ": ";
1798 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
1799 E = Line.Tokens.end();
1800 I != E; ++I) {
1801 llvm::dbgs() << I->Tok->Tok.getName() << "[" << I->Tok->Type << "] ";
1802 }
1803 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
1804 E = Line.Tokens.end();
1805 I != E; ++I) {
1806 const UnwrappedLineNode &Node = *I;
1807 for (SmallVectorImpl<UnwrappedLine>::const_iterator
1808 I = Node.Children.begin(),
1809 E = Node.Children.end();
1810 I != E; ++I) {
1811 printDebugInfo(*I, "\nChild: ");
1812 }
1813 }
1814 llvm::dbgs() << "\n";
1815 }
1816
addUnwrappedLine()1817 void UnwrappedLineParser::addUnwrappedLine() {
1818 if (Line->Tokens.empty())
1819 return;
1820 DEBUG({
1821 if (CurrentLines == &Lines)
1822 printDebugInfo(*Line);
1823 });
1824 CurrentLines->push_back(std::move(*Line));
1825 Line->Tokens.clear();
1826 if (CurrentLines == &Lines && !PreprocessorDirectives.empty()) {
1827 CurrentLines->append(
1828 std::make_move_iterator(PreprocessorDirectives.begin()),
1829 std::make_move_iterator(PreprocessorDirectives.end()));
1830 PreprocessorDirectives.clear();
1831 }
1832 }
1833
eof() const1834 bool UnwrappedLineParser::eof() const { return FormatTok->Tok.is(tok::eof); }
1835
isOnNewLine(const FormatToken & FormatTok)1836 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) {
1837 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) &&
1838 FormatTok.NewlinesBefore > 0;
1839 }
1840
flushComments(bool NewlineBeforeNext)1841 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
1842 bool JustComments = Line->Tokens.empty();
1843 for (SmallVectorImpl<FormatToken *>::const_iterator
1844 I = CommentsBeforeNextToken.begin(),
1845 E = CommentsBeforeNextToken.end();
1846 I != E; ++I) {
1847 if (isOnNewLine(**I) && JustComments)
1848 addUnwrappedLine();
1849 pushToken(*I);
1850 }
1851 if (NewlineBeforeNext && JustComments)
1852 addUnwrappedLine();
1853 CommentsBeforeNextToken.clear();
1854 }
1855
nextToken()1856 void UnwrappedLineParser::nextToken() {
1857 if (eof())
1858 return;
1859 flushComments(isOnNewLine(*FormatTok));
1860 pushToken(FormatTok);
1861 readToken();
1862 }
1863
readToken()1864 void UnwrappedLineParser::readToken() {
1865 bool CommentsInCurrentLine = true;
1866 do {
1867 FormatTok = Tokens->getNextToken();
1868 assert(FormatTok);
1869 while (!Line->InPPDirective && FormatTok->Tok.is(tok::hash) &&
1870 (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) {
1871 // If there is an unfinished unwrapped line, we flush the preprocessor
1872 // directives only after that unwrapped line was finished later.
1873 bool SwitchToPreprocessorLines = !Line->Tokens.empty();
1874 ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
1875 // Comments stored before the preprocessor directive need to be output
1876 // before the preprocessor directive, at the same level as the
1877 // preprocessor directive, as we consider them to apply to the directive.
1878 flushComments(isOnNewLine(*FormatTok));
1879 parsePPDirective();
1880 }
1881 while (FormatTok->Type == TT_ConflictStart ||
1882 FormatTok->Type == TT_ConflictEnd ||
1883 FormatTok->Type == TT_ConflictAlternative) {
1884 if (FormatTok->Type == TT_ConflictStart) {
1885 conditionalCompilationStart(/*Unreachable=*/false);
1886 } else if (FormatTok->Type == TT_ConflictAlternative) {
1887 conditionalCompilationAlternative();
1888 } else if (FormatTok->Type == TT_ConflictEnd) {
1889 conditionalCompilationEnd();
1890 }
1891 FormatTok = Tokens->getNextToken();
1892 FormatTok->MustBreakBefore = true;
1893 }
1894
1895 if (!PPStack.empty() && (PPStack.back() == PP_Unreachable) &&
1896 !Line->InPPDirective) {
1897 continue;
1898 }
1899
1900 if (!FormatTok->Tok.is(tok::comment))
1901 return;
1902 if (isOnNewLine(*FormatTok) || FormatTok->IsFirst) {
1903 CommentsInCurrentLine = false;
1904 }
1905 if (CommentsInCurrentLine) {
1906 pushToken(FormatTok);
1907 } else {
1908 CommentsBeforeNextToken.push_back(FormatTok);
1909 }
1910 } while (!eof());
1911 }
1912
pushToken(FormatToken * Tok)1913 void UnwrappedLineParser::pushToken(FormatToken *Tok) {
1914 Line->Tokens.push_back(UnwrappedLineNode(Tok));
1915 if (MustBreakBeforeNextToken) {
1916 Line->Tokens.back().Tok->MustBreakBefore = true;
1917 MustBreakBeforeNextToken = false;
1918 }
1919 }
1920
1921 } // end namespace format
1922 } // end namespace clang
1923