1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// \brief This file implements a token annotator, i.e. creates
12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information.
14 //===----------------------------------------------------------------------===//
16 #include "TokenAnnotator.h"
17 #include "clang/Basic/SourceManager.h"
18 #include "llvm/ADT/SmallPtrSet.h"
19 #include "llvm/Support/Debug.h"
21 #define DEBUG_TYPE "format-token-annotator"
28 /// \brief A parser that gathers additional information about tokens.
30 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
31 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
32 /// into template parameter lists.
33 class AnnotatingParser {
35 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line,
36 const AdditionalKeywords &Keywords)
37 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false),
39 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
40 resetTokenMetadata(CurrentToken);
45 if (!CurrentToken || !CurrentToken->Previous)
47 if (NonTemplateLess.count(CurrentToken->Previous))
50 const FormatToken& Previous = *CurrentToken->Previous;
51 if (Previous.Previous) {
52 if (Previous.Previous->Tok.isLiteral())
54 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 &&
55 (!Previous.Previous->MatchingParen ||
56 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen)))
60 FormatToken *Left = CurrentToken->Previous;
61 Left->ParentBracket = Contexts.back().ContextKind;
62 ScopedContextCreator ContextCreator(*this, tok::less, 12);
64 // If this angle is in the context of an expression, we need to be more
65 // hesitant to detect it as opening template parameters.
66 bool InExprContext = Contexts.back().IsExpression;
68 Contexts.back().IsExpression = false;
69 // If there's a template keyword before the opening angle bracket, this is a
70 // template parameter, not an argument.
71 Contexts.back().InTemplateArgument =
72 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
74 if (Style.Language == FormatStyle::LK_Java &&
75 CurrentToken->is(tok::question))
78 while (CurrentToken) {
79 if (CurrentToken->is(tok::greater)) {
80 Left->MatchingParen = CurrentToken;
81 CurrentToken->MatchingParen = Left;
82 CurrentToken->Type = TT_TemplateCloser;
86 if (CurrentToken->is(tok::question) &&
87 Style.Language == FormatStyle::LK_Java) {
91 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
92 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext))
94 // If a && or || is found and interpreted as a binary operator, this set
95 // of angles is likely part of something like "a < b && c > d". If the
96 // angles are inside an expression, the ||/&& might also be a binary
97 // operator that was misinterpreted because we are parsing template
99 // FIXME: This is getting out of hand, write a decent parser.
100 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
101 CurrentToken->Previous->is(TT_BinaryOperator) &&
102 Contexts[Contexts.size() - 2].IsExpression &&
103 !Line.startsWith(tok::kw_template))
105 updateParameterCount(Left, CurrentToken);
112 bool parseParens(bool LookForDecls = false) {
115 FormatToken *Left = CurrentToken->Previous;
116 Left->ParentBracket = Contexts.back().ContextKind;
117 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
119 // FIXME: This is a bit of a hack. Do better.
120 Contexts.back().ColonIsForRangeExpr =
121 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
123 bool StartsObjCMethodExpr = false;
124 if (CurrentToken->is(tok::caret)) {
125 // (^ can start a block type.
126 Left->Type = TT_ObjCBlockLParen;
127 } else if (FormatToken *MaybeSel = Left->Previous) {
128 // @selector( starts a selector.
129 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
130 MaybeSel->Previous->is(tok::at)) {
131 StartsObjCMethodExpr = true;
135 if (Left->is(TT_OverloadedOperatorLParen)) {
136 Contexts.back().IsExpression = false;
137 } else if (Style.Language == FormatStyle::LK_JavaScript &&
138 Line.startsWith(Keywords.kw_type, tok::identifier)) {
140 Contexts.back().IsExpression = false;
141 } else if (Left->Previous &&
142 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype,
143 tok::kw_if, tok::kw_while, tok::l_paren,
145 Left->Previous->is(TT_BinaryOperator))) {
146 // static_assert, if and while usually contain expressions.
147 Contexts.back().IsExpression = true;
148 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
149 (Left->Previous->is(Keywords.kw_function) ||
150 (Left->Previous->endsSequence(tok::identifier,
151 Keywords.kw_function)))) {
152 // function(...) or function f(...)
153 Contexts.back().IsExpression = false;
154 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
155 Left->Previous->is(TT_JsTypeColon)) {
156 // let x: (SomeType);
157 Contexts.back().IsExpression = false;
158 } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
159 Left->Previous->MatchingParen &&
160 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
161 // This is a parameter list of a lambda expression.
162 Contexts.back().IsExpression = false;
163 } else if (Line.InPPDirective &&
164 (!Left->Previous || !Left->Previous->is(tok::identifier))) {
165 Contexts.back().IsExpression = true;
166 } else if (Contexts[Contexts.size() - 2].CaretFound) {
167 // This is the parameter list of an ObjC block.
168 Contexts.back().IsExpression = false;
169 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
170 Left->Type = TT_AttributeParen;
171 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
172 // The first argument to a foreach macro is a declaration.
173 Contexts.back().IsForEachMacro = true;
174 Contexts.back().IsExpression = false;
175 } else if (Left->Previous && Left->Previous->MatchingParen &&
176 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
177 Contexts.back().IsExpression = false;
178 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
180 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
181 Contexts.back().IsExpression = !IsForOrCatch;
184 if (StartsObjCMethodExpr) {
185 Contexts.back().ColonIsObjCMethodExpr = true;
186 Left->Type = TT_ObjCMethodExpr;
189 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression;
190 bool ProbablyFunctionType = CurrentToken->isOneOf(tok::star, tok::amp);
191 bool HasMultipleLines = false;
192 bool HasMultipleParametersOnALine = false;
193 bool MightBeObjCForRangeLoop =
194 Left->Previous && Left->Previous->is(tok::kw_for);
195 while (CurrentToken) {
196 // LookForDecls is set when "if (" has been seen. Check for
197 // 'identifier' '*' 'identifier' followed by not '=' -- this
198 // '*' has to be a binary operator but determineStarAmpUsage() will
199 // categorize it as an unary operator, so set the right type here.
200 if (LookForDecls && CurrentToken->Next) {
201 FormatToken *Prev = CurrentToken->getPreviousNonComment();
203 FormatToken *PrevPrev = Prev->getPreviousNonComment();
204 FormatToken *Next = CurrentToken->Next;
205 if (PrevPrev && PrevPrev->is(tok::identifier) &&
206 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
207 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
208 Prev->Type = TT_BinaryOperator;
209 LookForDecls = false;
214 if (CurrentToken->Previous->is(TT_PointerOrReference) &&
215 CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
217 ProbablyFunctionType = true;
218 if (CurrentToken->is(tok::comma))
219 MightBeFunctionType = false;
220 if (CurrentToken->Previous->is(TT_BinaryOperator))
221 Contexts.back().IsExpression = true;
222 if (CurrentToken->is(tok::r_paren)) {
223 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next &&
224 (CurrentToken->Next->is(tok::l_paren) ||
225 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration)))
226 Left->Type = TT_FunctionTypeLParen;
227 Left->MatchingParen = CurrentToken;
228 CurrentToken->MatchingParen = Left;
230 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) &&
231 Left->Previous && Left->Previous->is(tok::l_paren)) {
232 // Detect the case where macros are used to generate lambdas or
233 // function bodies, e.g.:
234 // auto my_lambda = MARCO((Type *type, int i) { .. body .. });
235 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) {
236 if (Tok->is(TT_BinaryOperator) &&
237 Tok->isOneOf(tok::star, tok::amp, tok::ampamp))
238 Tok->Type = TT_PointerOrReference;
242 if (StartsObjCMethodExpr) {
243 CurrentToken->Type = TT_ObjCMethodExpr;
244 if (Contexts.back().FirstObjCSelectorName) {
245 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
246 Contexts.back().LongestObjCSelectorName;
250 if (Left->is(TT_AttributeParen))
251 CurrentToken->Type = TT_AttributeParen;
252 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
253 CurrentToken->Type = TT_JavaAnnotation;
254 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
255 CurrentToken->Type = TT_LeadingJavaAnnotation;
257 if (!HasMultipleLines)
258 Left->PackingKind = PPK_Inconclusive;
259 else if (HasMultipleParametersOnALine)
260 Left->PackingKind = PPK_BinPacked;
262 Left->PackingKind = PPK_OnePerLine;
267 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
270 if (CurrentToken->is(tok::l_brace))
271 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
272 if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
273 !CurrentToken->Next->HasUnescapedNewline &&
274 !CurrentToken->Next->isTrailingComment())
275 HasMultipleParametersOnALine = true;
276 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) ||
277 CurrentToken->Previous->isSimpleTypeSpecifier()) &&
278 !CurrentToken->is(tok::l_brace))
279 Contexts.back().IsExpression = false;
280 if (CurrentToken->isOneOf(tok::semi, tok::colon))
281 MightBeObjCForRangeLoop = false;
282 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
283 CurrentToken->Type = TT_ObjCForIn;
284 // When we discover a 'new', we set CanBeExpression to 'false' in order to
285 // parse the type correctly. Reset that after a comma.
286 if (CurrentToken->is(tok::comma))
287 Contexts.back().CanBeExpression = true;
289 FormatToken *Tok = CurrentToken;
292 updateParameterCount(Left, Tok);
293 if (CurrentToken && CurrentToken->HasUnescapedNewline)
294 HasMultipleLines = true;
303 // A '[' could be an index subscript (after an identifier or after
304 // ')' or ']'), it could be the start of an Objective-C method
305 // expression, or it could the start of an Objective-C array literal.
306 FormatToken *Left = CurrentToken->Previous;
307 Left->ParentBracket = Contexts.back().ContextKind;
308 FormatToken *Parent = Left->getPreviousNonComment();
310 // Cases where '>' is followed by '['.
311 // In C++, this can happen either in array of templates (foo<int>[10])
312 // or when array is a nested template type (unique_ptr<type1<type2>[]>).
313 bool CppArrayTemplates =
314 Style.isCpp() && Parent &&
315 Parent->is(TT_TemplateCloser) &&
316 (Contexts.back().CanBeExpression || Contexts.back().IsExpression ||
317 Contexts.back().InTemplateArgument);
319 bool StartsObjCMethodExpr =
320 !CppArrayTemplates && Style.isCpp() &&
321 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
322 CurrentToken->isNot(tok::l_brace) &&
324 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
325 tok::kw_return, tok::kw_throw) ||
326 Parent->isUnaryOperator() ||
327 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
328 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
329 bool ColonFound = false;
331 unsigned BindingIncrease = 1;
332 if (Left->is(TT_Unknown)) {
333 if (StartsObjCMethodExpr) {
334 Left->Type = TT_ObjCMethodExpr;
335 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
336 Contexts.back().ContextKind == tok::l_brace &&
337 Parent->isOneOf(tok::l_brace, tok::comma)) {
338 Left->Type = TT_JsComputedPropertyName;
339 } else if (CurrentToken->is(tok::r_square) && Parent &&
340 Parent->is(TT_TemplateCloser)) {
341 Left->Type = TT_ArraySubscriptLSquare;
342 } else if (Style.Language == FormatStyle::LK_Proto ||
343 (!CppArrayTemplates && Parent &&
344 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at,
345 tok::comma, tok::l_paren, tok::l_square,
346 tok::question, tok::colon, tok::kw_return,
347 // Should only be relevant to JavaScript:
349 Left->Type = TT_ArrayInitializerLSquare;
351 BindingIncrease = 10;
352 Left->Type = TT_ArraySubscriptLSquare;
356 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
357 Contexts.back().IsExpression = true;
358 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
360 while (CurrentToken) {
361 if (CurrentToken->is(tok::r_square)) {
362 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
363 Left->is(TT_ObjCMethodExpr)) {
364 // An ObjC method call is rarely followed by an open parenthesis.
365 // FIXME: Do we incorrectly label ":" with this?
366 StartsObjCMethodExpr = false;
367 Left->Type = TT_Unknown;
369 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
370 CurrentToken->Type = TT_ObjCMethodExpr;
371 // determineStarAmpUsage() thinks that '*' '[' is allocating an
372 // array of pointers, but if '[' starts a selector then '*' is a
374 if (Parent && Parent->is(TT_PointerOrReference))
375 Parent->Type = TT_BinaryOperator;
377 Left->MatchingParen = CurrentToken;
378 CurrentToken->MatchingParen = Left;
379 if (Contexts.back().FirstObjCSelectorName) {
380 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
381 Contexts.back().LongestObjCSelectorName;
382 if (Left->BlockParameterCount > 1)
383 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
388 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
390 if (CurrentToken->is(tok::colon)) {
391 if (Left->is(TT_ArraySubscriptLSquare)) {
392 Left->Type = TT_ObjCMethodExpr;
393 StartsObjCMethodExpr = true;
394 Contexts.back().ColonIsObjCMethodExpr = true;
395 if (Parent && Parent->is(tok::r_paren))
396 Parent->Type = TT_CastRParen;
400 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
402 Left->Type = TT_ArrayInitializerLSquare;
403 FormatToken *Tok = CurrentToken;
406 updateParameterCount(Left, Tok);
413 FormatToken *Left = CurrentToken->Previous;
414 Left->ParentBracket = Contexts.back().ContextKind;
416 if (Contexts.back().CaretFound)
417 Left->Type = TT_ObjCBlockLBrace;
418 Contexts.back().CaretFound = false;
420 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
421 Contexts.back().ColonIsDictLiteral = true;
422 if (Left->BlockKind == BK_BracedInit)
423 Contexts.back().IsExpression = true;
425 while (CurrentToken) {
426 if (CurrentToken->is(tok::r_brace)) {
427 Left->MatchingParen = CurrentToken;
428 CurrentToken->MatchingParen = Left;
432 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
434 updateParameterCount(Left, CurrentToken);
435 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) {
436 FormatToken *Previous = CurrentToken->getPreviousNonComment();
437 if (((CurrentToken->is(tok::colon) &&
438 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) ||
439 Style.Language == FormatStyle::LK_Proto) &&
440 (Previous->Tok.getIdentifierInfo() ||
441 Previous->is(tok::string_literal)))
442 Previous->Type = TT_SelectorName;
443 if (CurrentToken->is(tok::colon) ||
444 Style.Language == FormatStyle::LK_JavaScript)
445 Left->Type = TT_DictLiteral;
447 if (CurrentToken->is(tok::comma) &&
448 Style.Language == FormatStyle::LK_JavaScript)
449 Left->Type = TT_DictLiteral;
457 void updateParameterCount(FormatToken *Left, FormatToken *Current) {
458 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block)
459 ++Left->BlockParameterCount;
460 if (Current->is(tok::comma)) {
461 ++Left->ParameterCount;
463 Left->Role.reset(new CommaSeparatedList(Style));
464 Left->Role->CommaFound(Current);
465 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
466 Left->ParameterCount = 1;
470 bool parseConditional() {
471 while (CurrentToken) {
472 if (CurrentToken->is(tok::colon)) {
473 CurrentToken->Type = TT_ConditionalExpr;
483 bool parseTemplateDeclaration() {
484 if (CurrentToken && CurrentToken->is(tok::less)) {
485 CurrentToken->Type = TT_TemplateOpener;
490 CurrentToken->Previous->ClosesTemplateDeclaration = true;
496 bool consumeToken() {
497 FormatToken *Tok = CurrentToken;
499 switch (Tok->Tok.getKind()) {
502 if (!Tok->Previous && Line.MustBeDeclaration)
503 Tok->Type = TT_ObjCMethodSpecifier;
508 // Colons from ?: are handled in parseConditional().
509 if (Style.Language == FormatStyle::LK_JavaScript) {
510 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
511 (Contexts.size() == 1 && // switch/case labels
512 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
513 Contexts.back().ContextKind == tok::l_paren || // function params
514 Contexts.back().ContextKind == tok::l_square || // array type
515 (Contexts.size() == 1 &&
516 Line.MustBeDeclaration)) { // method/property declaration
517 Tok->Type = TT_JsTypeColon;
521 if (Contexts.back().ColonIsDictLiteral ||
522 Style.Language == FormatStyle::LK_Proto) {
523 Tok->Type = TT_DictLiteral;
524 } else if (Contexts.back().ColonIsObjCMethodExpr ||
525 Line.startsWith(TT_ObjCMethodSpecifier)) {
526 Tok->Type = TT_ObjCMethodExpr;
527 const FormatToken *BeforePrevious = Tok->Previous->Previous;
528 if (!BeforePrevious ||
529 !(BeforePrevious->is(TT_CastRParen) ||
530 (BeforePrevious->is(TT_ObjCMethodExpr) &&
531 BeforePrevious->is(tok::colon))) ||
532 BeforePrevious->is(tok::r_square) ||
533 Contexts.back().LongestObjCSelectorName == 0) {
534 Tok->Previous->Type = TT_SelectorName;
535 if (Tok->Previous->ColumnWidth >
536 Contexts.back().LongestObjCSelectorName)
537 Contexts.back().LongestObjCSelectorName =
538 Tok->Previous->ColumnWidth;
539 if (!Contexts.back().FirstObjCSelectorName)
540 Contexts.back().FirstObjCSelectorName = Tok->Previous;
542 } else if (Contexts.back().ColonIsForRangeExpr) {
543 Tok->Type = TT_RangeBasedForLoopColon;
544 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
545 Tok->Type = TT_BitFieldColon;
546 } else if (Contexts.size() == 1 &&
547 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
548 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren,
550 Tok->Type = TT_CtorInitializerColon;
552 Tok->Type = TT_InheritanceColon;
553 } else if (Tok->Previous->is(tok::identifier) && Tok->Next &&
554 Tok->Next->isOneOf(tok::r_paren, tok::comma)) {
555 // This handles a special macro in ObjC code where selectors including
556 // the colon are passed as macro arguments.
557 Tok->Type = TT_ObjCMethodExpr;
558 } else if (Contexts.back().ContextKind == tok::l_paren) {
559 Tok->Type = TT_InlineASMColon;
564 // | and & in declarations/type expressions represent union and
565 // intersection types, respectively.
566 if (Style.Language == FormatStyle::LK_JavaScript &&
567 !Contexts.back().IsExpression)
568 Tok->Type = TT_JsTypeOperator;
572 if (CurrentToken && CurrentToken->is(tok::l_paren)) {
574 if (!parseParens(/*LookForDecls=*/true))
579 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Previous &&
580 Tok->Previous->is(tok::period))
582 Contexts.back().ColonIsForRangeExpr = true;
588 // When faced with 'operator()()', the kw_operator handler incorrectly
589 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
590 // the first two parens OverloadedOperators and the second l_paren an
591 // OverloadedOperatorLParen.
593 Tok->Previous->is(tok::r_paren) &&
594 Tok->Previous->MatchingParen &&
595 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
596 Tok->Previous->Type = TT_OverloadedOperator;
597 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator;
598 Tok->Type = TT_OverloadedOperatorLParen;
603 if (Line.MustBeDeclaration && Contexts.size() == 1 &&
604 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
606 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute,
607 TT_LeadingJavaAnnotation)))
608 Line.MightBeFunctionDecl = true;
620 Tok->Type = TT_TemplateOpener;
622 Tok->Type = TT_BinaryOperator;
623 NonTemplateLess.insert(Tok);
632 // Lines can start with '}'.
637 Tok->Type = TT_BinaryOperator;
639 case tok::kw_operator:
640 while (CurrentToken &&
641 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
642 if (CurrentToken->isOneOf(tok::star, tok::amp))
643 CurrentToken->Type = TT_PointerOrReference;
646 CurrentToken->Previous->isOneOf(TT_BinaryOperator, tok::comma))
647 CurrentToken->Previous->Type = TT_OverloadedOperator;
650 CurrentToken->Type = TT_OverloadedOperatorLParen;
651 if (CurrentToken->Previous->is(TT_BinaryOperator))
652 CurrentToken->Previous->Type = TT_OverloadedOperator;
656 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
657 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
659 // Question marks before semicolons, colons, etc. indicate optional
660 // types (fields, parameters), e.g.
661 // function(x?: string, y?) {...}
663 Tok->Type = TT_JsTypeOptionalQuestion;
666 // Declarations cannot be conditional expressions, this can only be part
667 // of a type declaration.
668 if (Line.MustBeDeclaration && !Contexts.back().IsExpression &&
669 Style.Language == FormatStyle::LK_JavaScript)
673 case tok::kw_template:
674 parseTemplateDeclaration();
677 if (Contexts.back().InCtorInitializer)
678 Tok->Type = TT_CtorInitializerComma;
679 else if (Contexts.back().InInheritanceList)
680 Tok->Type = TT_InheritanceComma;
681 else if (Contexts.back().FirstStartOfName &&
682 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
683 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
684 Line.IsMultiVariableDeclStmt = true;
686 if (Contexts.back().IsForEachMacro)
687 Contexts.back().IsExpression = true;
689 case tok::identifier:
690 if (Tok->isOneOf(Keywords.kw___has_include,
691 Keywords.kw___has_include_next)) {
701 void parseIncludeDirective() {
702 if (CurrentToken && CurrentToken->is(tok::less)) {
704 while (CurrentToken) {
705 if (CurrentToken->isNot(tok::comment) || CurrentToken->Next)
706 CurrentToken->Type = TT_ImplicitStringLiteral;
712 void parseWarningOrError() {
714 // We still want to format the whitespace left of the first token of the
717 while (CurrentToken) {
718 CurrentToken->Type = TT_ImplicitStringLiteral;
724 next(); // Consume "pragma".
726 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
727 bool IsMark = CurrentToken->is(Keywords.kw_mark);
728 next(); // Consume "mark".
729 next(); // Consume first token (so we fix leading whitespace).
730 while (CurrentToken) {
731 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
732 CurrentToken->Type = TT_ImplicitStringLiteral;
738 void parseHasInclude() {
739 if (!CurrentToken || !CurrentToken->is(tok::l_paren))
742 parseIncludeDirective();
746 LineType parsePreprocessorDirective() {
747 bool IsFirstToken = CurrentToken->IsFirst;
748 LineType Type = LT_PreprocessorDirective;
753 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) {
754 // JavaScript files can contain shebang lines of the form:
755 // #!/usr/bin/env node
756 // Treat these like C++ #include directives.
757 while (CurrentToken) {
758 // Tokens cannot be comments here.
759 CurrentToken->Type = TT_ImplicitStringLiteral;
762 return LT_ImportStatement;
765 if (CurrentToken->Tok.is(tok::numeric_constant)) {
766 CurrentToken->SpacesRequiredBefore = 1;
769 // Hashes in the middle of a line can lead to any strange token
771 if (!CurrentToken->Tok.getIdentifierInfo())
773 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
774 case tok::pp_include:
775 case tok::pp_include_next:
778 parseIncludeDirective();
779 Type = LT_ImportStatement;
782 case tok::pp_warning:
783 parseWarningOrError();
790 Contexts.back().IsExpression = true;
796 while (CurrentToken) {
797 FormatToken *Tok = CurrentToken;
799 if (Tok->isOneOf(Keywords.kw___has_include,
800 Keywords.kw___has_include_next)) {
808 LineType parseLine() {
809 NonTemplateLess.clear();
810 if (CurrentToken->is(tok::hash))
811 return parsePreprocessorDirective();
813 // Directly allow to 'import <string-literal>' to support protocol buffer
814 // definitions (code.google.com/p/protobuf) or missing "#" (either way we
815 // should not break the line).
816 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
817 if ((Style.Language == FormatStyle::LK_Java &&
818 CurrentToken->is(Keywords.kw_package)) ||
819 (Info && Info->getPPKeywordID() == tok::pp_import &&
820 CurrentToken->Next &&
821 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
824 parseIncludeDirective();
825 return LT_ImportStatement;
828 // If this line starts and ends in '<' and '>', respectively, it is likely
829 // part of "#define <a/b.h>".
830 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
831 parseIncludeDirective();
832 return LT_ImportStatement;
835 // In .proto files, top-level options are very similar to import statements
836 // and should not be line-wrapped.
837 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
838 CurrentToken->is(Keywords.kw_option)) {
840 if (CurrentToken && CurrentToken->is(tok::identifier))
841 return LT_ImportStatement;
844 bool KeywordVirtualFound = false;
845 bool ImportStatement = false;
847 // import {...} from '...';
848 if (Style.Language == FormatStyle::LK_JavaScript &&
849 CurrentToken->is(Keywords.kw_import))
850 ImportStatement = true;
852 while (CurrentToken) {
853 if (CurrentToken->is(tok::kw_virtual))
854 KeywordVirtualFound = true;
855 if (Style.Language == FormatStyle::LK_JavaScript) {
856 // export {...} from '...';
857 // An export followed by "from 'some string';" is a re-export from
858 // another module identified by a URI and is treated as a
859 // LT_ImportStatement (i.e. prevent wraps on it for long URIs).
860 // Just "export {...};" or "export class ..." should not be treated as
861 // an import in this sense.
862 if (Line.First->is(tok::kw_export) &&
863 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next &&
864 CurrentToken->Next->isStringLiteral())
865 ImportStatement = true;
866 if (isClosureImportStatement(*CurrentToken))
867 ImportStatement = true;
872 if (KeywordVirtualFound)
873 return LT_VirtualFunctionDecl;
875 return LT_ImportStatement;
877 if (Line.startsWith(TT_ObjCMethodSpecifier)) {
878 if (Contexts.back().FirstObjCSelectorName)
879 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
880 Contexts.back().LongestObjCSelectorName;
881 return LT_ObjCMethodDecl;
888 bool isClosureImportStatement(const FormatToken &Tok) {
889 // FIXME: Closure-library specific stuff should not be hard-coded but be
891 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
892 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" ||
893 Tok.Next->Next->TokenText == "provide" ||
894 Tok.Next->Next->TokenText == "require" ||
895 Tok.Next->Next->TokenText == "setTestOnly" ||
896 Tok.Next->Next->TokenText == "forwardDeclare") &&
897 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
900 void resetTokenMetadata(FormatToken *Token) {
904 // Reset token type in case we have already looked at it and then
905 // recovered from an error (e.g. failure to find the matching >).
906 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
907 TT_FunctionLBrace, TT_ImplicitStringLiteral,
908 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
909 TT_OverloadedOperator, TT_RegexLiteral,
910 TT_TemplateString, TT_ObjCStringLiteral))
911 CurrentToken->Type = TT_Unknown;
912 CurrentToken->Role.reset();
913 CurrentToken->MatchingParen = nullptr;
914 CurrentToken->FakeLParens.clear();
915 CurrentToken->FakeRParens = 0;
920 CurrentToken->NestingLevel = Contexts.size() - 1;
921 CurrentToken->BindingStrength = Contexts.back().BindingStrength;
922 modifyContext(*CurrentToken);
923 determineTokenType(*CurrentToken);
924 CurrentToken = CurrentToken->Next;
927 resetTokenMetadata(CurrentToken);
930 /// \brief A struct to hold information valid in a specific context, e.g.
931 /// a pair of parenthesis.
933 Context(tok::TokenKind ContextKind, unsigned BindingStrength,
935 : ContextKind(ContextKind), BindingStrength(BindingStrength),
936 IsExpression(IsExpression) {}
938 tok::TokenKind ContextKind;
939 unsigned BindingStrength;
941 unsigned LongestObjCSelectorName = 0;
942 bool ColonIsForRangeExpr = false;
943 bool ColonIsDictLiteral = false;
944 bool ColonIsObjCMethodExpr = false;
945 FormatToken *FirstObjCSelectorName = nullptr;
946 FormatToken *FirstStartOfName = nullptr;
947 bool CanBeExpression = true;
948 bool InTemplateArgument = false;
949 bool InCtorInitializer = false;
950 bool InInheritanceList = false;
951 bool CaretFound = false;
952 bool IsForEachMacro = false;
955 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
956 /// of each instance.
957 struct ScopedContextCreator {
960 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
963 P.Contexts.push_back(Context(ContextKind,
964 P.Contexts.back().BindingStrength + Increase,
965 P.Contexts.back().IsExpression));
968 ~ScopedContextCreator() { P.Contexts.pop_back(); }
971 void modifyContext(const FormatToken &Current) {
972 if (Current.getPrecedence() == prec::Assignment &&
973 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
974 // Type aliases use `type X = ...;` in TypeScript.
975 !(Style.Language == FormatStyle::LK_JavaScript &&
976 Line.startsWith(Keywords.kw_type, tok::identifier)) &&
977 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
978 Contexts.back().IsExpression = true;
979 if (!Line.startsWith(TT_UnaryOperator)) {
980 for (FormatToken *Previous = Current.Previous;
981 Previous && Previous->Previous &&
982 !Previous->Previous->isOneOf(tok::comma, tok::semi);
983 Previous = Previous->Previous) {
984 if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
985 Previous = Previous->MatchingParen;
989 if (Previous->opensScope())
991 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
992 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
993 Previous->Previous && Previous->Previous->isNot(tok::equal))
994 Previous->Type = TT_PointerOrReference;
997 } else if (Current.is(tok::lessless) &&
998 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
999 Contexts.back().IsExpression = true;
1000 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
1001 Contexts.back().IsExpression = true;
1002 } else if (Current.is(TT_TrailingReturnArrow)) {
1003 Contexts.back().IsExpression = false;
1004 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
1005 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
1006 } else if (Current.Previous &&
1007 Current.Previous->is(TT_CtorInitializerColon)) {
1008 Contexts.back().IsExpression = true;
1009 Contexts.back().InCtorInitializer = true;
1010 } else if (Current.Previous &&
1011 Current.Previous->is(TT_InheritanceColon)) {
1012 Contexts.back().InInheritanceList = true;
1013 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
1014 for (FormatToken *Previous = Current.Previous;
1015 Previous && Previous->isOneOf(tok::star, tok::amp);
1016 Previous = Previous->Previous)
1017 Previous->Type = TT_PointerOrReference;
1018 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
1019 Contexts.back().IsExpression = false;
1020 } else if (Current.is(tok::kw_new)) {
1021 Contexts.back().CanBeExpression = false;
1022 } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
1023 // This should be the condition or increment in a for-loop.
1024 Contexts.back().IsExpression = true;
1028 void determineTokenType(FormatToken &Current) {
1029 if (!Current.is(TT_Unknown))
1030 // The token type is already known.
1033 if (Style.Language == FormatStyle::LK_JavaScript) {
1034 if (Current.is(tok::exclaim)) {
1035 if (Current.Previous &&
1036 (Current.Previous->isOneOf(tok::identifier, tok::r_paren,
1037 tok::r_square, tok::r_brace) ||
1038 Current.Previous->Tok.isLiteral())) {
1039 Current.Type = TT_JsNonNullAssertion;
1043 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) {
1044 Current.Type = TT_JsNonNullAssertion;
1050 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1051 // function declaration have been found. In this case, 'Current' is a
1052 // trailing token of this declaration and thus cannot be a name.
1053 if (Current.is(Keywords.kw_instanceof)) {
1054 Current.Type = TT_BinaryOperator;
1055 } else if (isStartOfName(Current) &&
1056 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
1057 Contexts.back().FirstStartOfName = &Current;
1058 Current.Type = TT_StartOfName;
1059 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
1061 } else if (Current.is(tok::arrow) &&
1062 Style.Language == FormatStyle::LK_Java) {
1063 Current.Type = TT_LambdaArrow;
1064 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
1065 Current.NestingLevel == 0) {
1066 Current.Type = TT_TrailingReturnArrow;
1067 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
1069 determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
1070 Contexts.back().IsExpression,
1071 Contexts.back().InTemplateArgument);
1072 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
1073 Current.Type = determinePlusMinusCaretUsage(Current);
1074 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
1075 Contexts.back().CaretFound = true;
1076 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
1077 Current.Type = determineIncrementUsage(Current);
1078 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
1079 Current.Type = TT_UnaryOperator;
1080 } else if (Current.is(tok::question)) {
1081 if (Style.Language == FormatStyle::LK_JavaScript &&
1082 Line.MustBeDeclaration && !Contexts.back().IsExpression) {
1083 // In JavaScript, `interface X { foo?(): bar; }` is an optional method
1084 // on the interface, not a ternary expression.
1085 Current.Type = TT_JsTypeOptionalQuestion;
1087 Current.Type = TT_ConditionalExpr;
1089 } else if (Current.isBinaryOperator() &&
1090 (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
1091 Current.Type = TT_BinaryOperator;
1092 } else if (Current.is(tok::comment)) {
1093 if (Current.TokenText.startswith("/*")) {
1094 if (Current.TokenText.endswith("*/"))
1095 Current.Type = TT_BlockComment;
1097 // The lexer has for some reason determined a comment here. But we
1098 // cannot really handle it, if it isn't properly terminated.
1099 Current.Tok.setKind(tok::unknown);
1101 Current.Type = TT_LineComment;
1103 } else if (Current.is(tok::r_paren)) {
1104 if (rParenEndsCast(Current))
1105 Current.Type = TT_CastRParen;
1106 if (Current.MatchingParen && Current.Next &&
1107 !Current.Next->isBinaryOperator() &&
1108 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace,
1109 tok::comma, tok::period, tok::arrow,
1111 if (FormatToken *AfterParen = Current.MatchingParen->Next) {
1112 // Make sure this isn't the return type of an Obj-C block declaration
1113 if (AfterParen->Tok.isNot(tok::caret)) {
1114 if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
1115 if (BeforeParen->is(tok::identifier) &&
1116 BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
1117 (!BeforeParen->Previous ||
1118 BeforeParen->Previous->ClosesTemplateDeclaration))
1119 Current.Type = TT_FunctionAnnotationRParen;
1122 } else if (Current.is(tok::at) && Current.Next) {
1123 switch (Current.Next->Tok.getObjCKeywordID()) {
1124 case tok::objc_interface:
1125 case tok::objc_implementation:
1126 case tok::objc_protocol:
1127 Current.Type = TT_ObjCDecl;
1129 case tok::objc_property:
1130 Current.Type = TT_ObjCProperty;
1135 } else if (Current.is(tok::period)) {
1136 FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1137 if (PreviousNoComment &&
1138 PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1139 Current.Type = TT_DesignatedInitializerPeriod;
1140 else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1141 Current.Previous->isOneOf(TT_JavaAnnotation,
1142 TT_LeadingJavaAnnotation)) {
1143 Current.Type = Current.Previous->Type;
1145 } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
1147 !Current.Previous->isOneOf(tok::equal, tok::at) &&
1148 Line.MightBeFunctionDecl && Contexts.size() == 1) {
1149 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1150 // function declaration have been found.
1151 Current.Type = TT_TrailingAnnotation;
1152 } else if ((Style.Language == FormatStyle::LK_Java ||
1153 Style.Language == FormatStyle::LK_JavaScript) &&
1155 if (Current.Previous->is(tok::at) &&
1156 Current.isNot(Keywords.kw_interface)) {
1157 const FormatToken &AtToken = *Current.Previous;
1158 const FormatToken *Previous = AtToken.getPreviousNonComment();
1159 if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1160 Current.Type = TT_LeadingJavaAnnotation;
1162 Current.Type = TT_JavaAnnotation;
1163 } else if (Current.Previous->is(tok::period) &&
1164 Current.Previous->isOneOf(TT_JavaAnnotation,
1165 TT_LeadingJavaAnnotation)) {
1166 Current.Type = Current.Previous->Type;
1171 /// \brief Take a guess at whether \p Tok starts a name of a function or
1172 /// variable declaration.
1174 /// This is a heuristic based on whether \p Tok is an identifier following
1175 /// something that is likely a type.
1176 bool isStartOfName(const FormatToken &Tok) {
1177 if (Tok.isNot(tok::identifier) || !Tok.Previous)
1180 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof,
1183 if (Style.Language == FormatStyle::LK_JavaScript &&
1184 Tok.Previous->is(Keywords.kw_in))
1187 // Skip "const" as it does not have an influence on whether this is a name.
1188 FormatToken *PreviousNotConst = Tok.getPreviousNonComment();
1189 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1190 PreviousNotConst = PreviousNotConst->getPreviousNonComment();
1192 if (!PreviousNotConst)
1195 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1196 PreviousNotConst->Previous &&
1197 PreviousNotConst->Previous->is(tok::hash);
1199 if (PreviousNotConst->is(TT_TemplateCloser))
1200 return PreviousNotConst && PreviousNotConst->MatchingParen &&
1201 PreviousNotConst->MatchingParen->Previous &&
1202 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1203 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1205 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1206 PreviousNotConst->MatchingParen->Previous &&
1207 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1210 return (!IsPPKeyword &&
1211 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1212 PreviousNotConst->is(TT_PointerOrReference) ||
1213 PreviousNotConst->isSimpleTypeSpecifier();
1216 /// \brief Determine whether ')' is ending a cast.
1217 bool rParenEndsCast(const FormatToken &Tok) {
1218 // C-style casts are only used in C++ and Java.
1219 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java)
1222 // Empty parens aren't casts and there are no casts at the end of the line.
1223 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1226 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1228 // If there is a closing parenthesis left of the current parentheses,
1229 // look past it as these might be chained casts.
1230 if (LeftOfParens->is(tok::r_paren)) {
1231 if (!LeftOfParens->MatchingParen ||
1232 !LeftOfParens->MatchingParen->Previous)
1234 LeftOfParens = LeftOfParens->MatchingParen->Previous;
1237 // If there is an identifier (or with a few exceptions a keyword) right
1238 // before the parentheses, this is unlikely to be a cast.
1239 if (LeftOfParens->Tok.getIdentifierInfo() &&
1240 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1244 // Certain other tokens right before the parentheses are also signals that
1245 // this cannot be a cast.
1246 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1247 TT_TemplateCloser, tok::ellipsis))
1251 if (Tok.Next->is(tok::question))
1254 // As Java has no function types, a "(" after the ")" likely means that this
1256 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1259 // If a (non-string) literal follows, this is likely a cast.
1260 if (Tok.Next->isNot(tok::string_literal) &&
1261 (Tok.Next->Tok.isLiteral() ||
1262 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1265 // Heuristically try to determine whether the parentheses contain a type.
1266 bool ParensAreType =
1268 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1269 Tok.Previous->isSimpleTypeSpecifier();
1270 bool ParensCouldEndDecl =
1271 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1272 if (ParensAreType && !ParensCouldEndDecl)
1275 // At this point, we heuristically assume that there are no casts at the
1276 // start of the line. We assume that we have found most cases where there
1277 // are by the logic above, e.g. "(void)x;".
1281 // Certain token types inside the parentheses mean that this can't be a
1283 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok;
1284 Token = Token->Next)
1285 if (Token->is(TT_BinaryOperator))
1288 // If the following token is an identifier or 'this', this is a cast. All
1289 // cases where this can be something else are handled above.
1290 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this))
1293 if (!Tok.Next->Next)
1296 // If the next token after the parenthesis is a unary operator, assume
1297 // that this is cast, unless there are unexpected tokens inside the
1300 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1301 if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1302 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1304 // Search for unexpected tokens.
1305 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1306 Prev = Prev->Previous) {
1307 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1313 /// \brief Return the type of the given token assuming it is * or &.
1314 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1315 bool InTemplateArgument) {
1316 if (Style.Language == FormatStyle::LK_JavaScript)
1317 return TT_BinaryOperator;
1319 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1321 return TT_UnaryOperator;
1323 const FormatToken *NextToken = Tok.getNextNonComment();
1325 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) ||
1326 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1327 return TT_PointerOrReference;
1329 if (PrevToken->is(tok::coloncolon))
1330 return TT_PointerOrReference;
1332 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1333 tok::comma, tok::semi, tok::kw_return, tok::colon,
1334 tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1335 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1336 TT_UnaryOperator, TT_CastRParen))
1337 return TT_UnaryOperator;
1339 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1340 return TT_PointerOrReference;
1341 if (NextToken->is(tok::kw_operator) && !IsExpression)
1342 return TT_PointerOrReference;
1343 if (NextToken->isOneOf(tok::comma, tok::semi))
1344 return TT_PointerOrReference;
1346 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1347 PrevToken->MatchingParen->Previous &&
1348 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1350 return TT_PointerOrReference;
1352 if (PrevToken->Tok.isLiteral() ||
1353 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1354 tok::kw_false, tok::r_brace) ||
1355 NextToken->Tok.isLiteral() ||
1356 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1357 NextToken->isUnaryOperator() ||
1358 // If we know we're in a template argument, there are no named
1359 // declarations. Thus, having an identifier on the right-hand side
1360 // indicates a binary operator.
1361 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1362 return TT_BinaryOperator;
1364 // "&&(" is quite unlikely to be two successive unary "&".
1365 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1366 return TT_BinaryOperator;
1368 // This catches some cases where evaluation order is used as control flow:
1370 const FormatToken *NextNextToken = NextToken->getNextNonComment();
1371 if (NextNextToken && NextNextToken->is(tok::arrow))
1372 return TT_BinaryOperator;
1374 // It is very unlikely that we are going to find a pointer or reference type
1375 // definition on the RHS of an assignment.
1376 if (IsExpression && !Contexts.back().CaretFound)
1377 return TT_BinaryOperator;
1379 return TT_PointerOrReference;
1382 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1383 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1385 return TT_UnaryOperator;
1387 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator) &&
1388 !PrevToken->is(tok::exclaim))
1389 // There aren't any trailing unary operators except for TypeScript's
1390 // non-null operator (!). Thus, this must be squence of leading operators.
1391 return TT_UnaryOperator;
1393 // Use heuristics to recognize unary operators.
1394 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1395 tok::question, tok::colon, tok::kw_return,
1396 tok::kw_case, tok::at, tok::l_brace))
1397 return TT_UnaryOperator;
1399 // There can't be two consecutive binary operators.
1400 if (PrevToken->is(TT_BinaryOperator))
1401 return TT_UnaryOperator;
1403 // Fall back to marking the token as binary operator.
1404 return TT_BinaryOperator;
1407 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1408 TokenType determineIncrementUsage(const FormatToken &Tok) {
1409 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1410 if (!PrevToken || PrevToken->is(TT_CastRParen))
1411 return TT_UnaryOperator;
1412 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1413 return TT_TrailingUnaryOperator;
1415 return TT_UnaryOperator;
1418 SmallVector<Context, 8> Contexts;
1420 const FormatStyle &Style;
1421 AnnotatedLine &Line;
1422 FormatToken *CurrentToken;
1424 const AdditionalKeywords &Keywords;
1426 // Set of "<" tokens that do not open a template parameter list. If parseAngle
1427 // determines that a specific token can't be a template opener, it will make
1428 // same decision irrespective of the decisions for tokens leading up to it.
1429 // Store this information to prevent this from causing exponential runtime.
1430 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1433 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1434 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1436 /// \brief Parses binary expressions by inserting fake parenthesis based on
1437 /// operator precedence.
1438 class ExpressionParser {
1440 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1441 AnnotatedLine &Line)
1442 : Style(Style), Keywords(Keywords), Current(Line.First) {}
1444 /// \brief Parse expressions with the given operatore precedence.
1445 void parse(int Precedence = 0) {
1446 // Skip 'return' and ObjC selector colons as they are not part of a binary
1448 while (Current && (Current->is(tok::kw_return) ||
1449 (Current->is(tok::colon) &&
1450 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1453 if (!Current || Precedence > PrecedenceArrowAndPeriod)
1456 // Conditional expressions need to be parsed separately for proper nesting.
1457 if (Precedence == prec::Conditional) {
1458 parseConditionalExpr();
1462 // Parse unary operators, which all have a higher precedence than binary
1464 if (Precedence == PrecedenceUnaryOperator) {
1465 parseUnaryOperator();
1469 FormatToken *Start = Current;
1470 FormatToken *LatestOperator = nullptr;
1471 unsigned OperatorIndex = 0;
1474 // Consume operators with higher precedence.
1475 parse(Precedence + 1);
1477 int CurrentPrecedence = getCurrentPrecedence();
1479 if (Current && Current->is(TT_SelectorName) &&
1480 Precedence == CurrentPrecedence) {
1482 addFakeParenthesis(Start, prec::Level(Precedence));
1486 // At the end of the line or when an operator with higher precedence is
1487 // found, insert fake parenthesis and return.
1489 (Current->closesScope() &&
1490 (Current->MatchingParen || Current->is(TT_TemplateString))) ||
1491 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1492 (CurrentPrecedence == prec::Conditional &&
1493 Precedence == prec::Assignment && Current->is(tok::colon))) {
1497 // Consume scopes: (), [], <> and {}
1498 if (Current->opensScope()) {
1499 // In fragment of a JavaScript template string can look like '}..${' and
1500 // thus close a scope and open a new one at the same time.
1501 while (Current && (!Current->closesScope() || Current->opensScope())) {
1508 if (CurrentPrecedence == Precedence) {
1510 LatestOperator->NextOperator = Current;
1511 LatestOperator = Current;
1512 Current->OperatorIndex = OperatorIndex;
1515 next(/*SkipPastLeadingComments=*/Precedence > 0);
1519 if (LatestOperator && (Current || Precedence > 0)) {
1520 // LatestOperator->LastOperator = true;
1521 if (Precedence == PrecedenceArrowAndPeriod) {
1522 // Call expressions don't have a binary operator precedence.
1523 addFakeParenthesis(Start, prec::Unknown);
1525 addFakeParenthesis(Start, prec::Level(Precedence));
1531 /// \brief Gets the precedence (+1) of the given token for binary operators
1532 /// and other tokens that we treat like binary operators.
1533 int getCurrentPrecedence() {
1535 const FormatToken *NextNonComment = Current->getNextNonComment();
1536 if (Current->is(TT_ConditionalExpr))
1537 return prec::Conditional;
1538 if (NextNonComment && NextNonComment->is(tok::colon) &&
1539 NextNonComment->is(TT_DictLiteral))
1540 return prec::Assignment;
1541 if (Current->is(TT_JsComputedPropertyName))
1542 return prec::Assignment;
1543 if (Current->is(TT_LambdaArrow))
1545 if (Current->is(TT_JsFatArrow))
1546 return prec::Assignment;
1547 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) ||
1548 (Current->is(tok::comment) && NextNonComment &&
1549 NextNonComment->is(TT_SelectorName)))
1551 if (Current->is(TT_RangeBasedForLoopColon))
1553 if ((Style.Language == FormatStyle::LK_Java ||
1554 Style.Language == FormatStyle::LK_JavaScript) &&
1555 Current->is(Keywords.kw_instanceof))
1556 return prec::Relational;
1557 if (Style.Language == FormatStyle::LK_JavaScript &&
1558 Current->isOneOf(Keywords.kw_in, Keywords.kw_as))
1559 return prec::Relational;
1560 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1561 return Current->getPrecedence();
1562 if (Current->isOneOf(tok::period, tok::arrow))
1563 return PrecedenceArrowAndPeriod;
1564 if ((Style.Language == FormatStyle::LK_Java ||
1565 Style.Language == FormatStyle::LK_JavaScript) &&
1566 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1567 Keywords.kw_throws))
1573 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1574 Start->FakeLParens.push_back(Precedence);
1575 if (Precedence > prec::Unknown)
1576 Start->StartsBinaryExpression = true;
1578 FormatToken *Previous = Current->Previous;
1579 while (Previous->is(tok::comment) && Previous->Previous)
1580 Previous = Previous->Previous;
1581 ++Previous->FakeRParens;
1582 if (Precedence > prec::Unknown)
1583 Previous->EndsBinaryExpression = true;
1587 /// \brief Parse unary operator expressions and surround them with fake
1588 /// parentheses if appropriate.
1589 void parseUnaryOperator() {
1590 if (!Current || Current->isNot(TT_UnaryOperator)) {
1591 parse(PrecedenceArrowAndPeriod);
1595 FormatToken *Start = Current;
1597 parseUnaryOperator();
1599 // The actual precedence doesn't matter.
1600 addFakeParenthesis(Start, prec::Unknown);
1603 void parseConditionalExpr() {
1604 while (Current && Current->isTrailingComment()) {
1607 FormatToken *Start = Current;
1608 parse(prec::LogicalOr);
1609 if (!Current || !Current->is(tok::question))
1612 parse(prec::Assignment);
1613 if (!Current || Current->isNot(TT_ConditionalExpr))
1616 parse(prec::Assignment);
1617 addFakeParenthesis(Start, prec::Conditional);
1620 void next(bool SkipPastLeadingComments = true) {
1622 Current = Current->Next;
1624 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1625 Current->isTrailingComment())
1626 Current = Current->Next;
1629 const FormatStyle &Style;
1630 const AdditionalKeywords &Keywords;
1631 FormatToken *Current;
1634 } // end anonymous namespace
1636 void TokenAnnotator::setCommentLineLevels(
1637 SmallVectorImpl<AnnotatedLine *> &Lines) {
1638 const AnnotatedLine *NextNonCommentLine = nullptr;
1639 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1642 bool CommentLine = (*I)->First;
1643 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) {
1644 if (!Tok->is(tok::comment)) {
1645 CommentLine = false;
1649 if (NextNonCommentLine && CommentLine)
1650 (*I)->Level = NextNonCommentLine->Level;
1652 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1654 setCommentLineLevels((*I)->Children);
1658 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
1659 unsigned Result = 0;
1660 for (const auto* Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
1661 Result = std::max(Result, Tok->NestingLevel);
1665 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1666 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1667 E = Line.Children.end();
1671 AnnotatingParser Parser(Style, Line, Keywords);
1672 Line.Type = Parser.parseLine();
1674 // With very deep nesting, ExpressionParser uses lots of stack and the
1675 // formatting algorithm is very slow. We're not going to do a good job here
1676 // anyway - it's probably generated code being formatted by mistake.
1677 // Just skip the whole line.
1678 if (maxNestingDepth(Line) > 50)
1679 Line.Type = LT_Invalid;
1681 if (Line.Type == LT_Invalid)
1684 ExpressionParser ExprParser(Style, Keywords, Line);
1687 if (Line.startsWith(TT_ObjCMethodSpecifier))
1688 Line.Type = LT_ObjCMethodDecl;
1689 else if (Line.startsWith(TT_ObjCDecl))
1690 Line.Type = LT_ObjCDecl;
1691 else if (Line.startsWith(TT_ObjCProperty))
1692 Line.Type = LT_ObjCProperty;
1694 Line.First->SpacesRequiredBefore = 1;
1695 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1698 // This function heuristically determines whether 'Current' starts the name of a
1699 // function declaration.
1700 static bool isFunctionDeclarationName(const FormatToken &Current,
1701 const AnnotatedLine &Line) {
1702 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1703 for (; Next; Next = Next->Next) {
1704 if (Next->is(TT_OverloadedOperatorLParen))
1706 if (Next->is(TT_OverloadedOperator))
1708 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1709 // For 'new[]' and 'delete[]'.
1710 if (Next->Next && Next->Next->is(tok::l_square) &&
1711 Next->Next->Next && Next->Next->Next->is(tok::r_square))
1712 Next = Next->Next->Next;
1721 // Find parentheses of parameter list.
1722 const FormatToken *Next = Current.Next;
1723 if (Current.is(tok::kw_operator)) {
1724 if (Current.Previous && Current.Previous->is(tok::coloncolon))
1726 Next = skipOperatorName(Next);
1728 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1730 for (; Next; Next = Next->Next) {
1731 if (Next->is(TT_TemplateOpener)) {
1732 Next = Next->MatchingParen;
1733 } else if (Next->is(tok::coloncolon)) {
1737 if (Next->is(tok::kw_operator)) {
1738 Next = skipOperatorName(Next->Next);
1741 if (!Next->is(tok::identifier))
1743 } else if (Next->is(tok::l_paren)) {
1751 // Check whether parameter list can belong to a function declaration.
1752 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
1754 // If the lines ends with "{", this is likely an function definition.
1755 if (Line.Last->is(tok::l_brace))
1757 if (Next->Next == Next->MatchingParen)
1758 return true; // Empty parentheses.
1759 // If there is an &/&& after the r_paren, this is likely a function.
1760 if (Next->MatchingParen->Next &&
1761 Next->MatchingParen->Next->is(TT_PointerOrReference))
1763 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1765 if (Tok->is(tok::l_paren) && Tok->MatchingParen) {
1766 Tok = Tok->MatchingParen;
1769 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1770 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
1772 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1773 Tok->Tok.isLiteral())
1779 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1780 assert(Line.MightBeFunctionDecl);
1782 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1783 Style.AlwaysBreakAfterReturnType ==
1784 FormatStyle::RTBS_TopLevelDefinitions) &&
1788 switch (Style.AlwaysBreakAfterReturnType) {
1789 case FormatStyle::RTBS_None:
1791 case FormatStyle::RTBS_All:
1792 case FormatStyle::RTBS_TopLevel:
1794 case FormatStyle::RTBS_AllDefinitions:
1795 case FormatStyle::RTBS_TopLevelDefinitions:
1796 return Line.mightBeFunctionDefinition();
1802 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1803 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1804 E = Line.Children.end();
1806 calculateFormattingInformation(**I);
1809 Line.First->TotalLength =
1810 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1811 FormatToken *Current = Line.First->Next;
1812 bool InFunctionDecl = Line.MightBeFunctionDecl;
1814 if (isFunctionDeclarationName(*Current, Line))
1815 Current->Type = TT_FunctionDeclarationName;
1816 if (Current->is(TT_LineComment)) {
1817 if (Current->Previous->BlockKind == BK_BracedInit &&
1818 Current->Previous->opensScope())
1819 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1821 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1823 // If we find a trailing comment, iterate backwards to determine whether
1824 // it seems to relate to a specific parameter. If so, break before that
1825 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1826 // to the previous line in:
1830 if (!Current->HasUnescapedNewline) {
1831 for (FormatToken *Parameter = Current->Previous; Parameter;
1832 Parameter = Parameter->Previous) {
1833 if (Parameter->isOneOf(tok::comment, tok::r_brace))
1835 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1836 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1837 Parameter->HasUnescapedNewline)
1838 Parameter->MustBreakBefore = true;
1843 } else if (Current->SpacesRequiredBefore == 0 &&
1844 spaceRequiredBefore(Line, *Current)) {
1845 Current->SpacesRequiredBefore = 1;
1848 Current->MustBreakBefore =
1849 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1851 if (!Current->MustBreakBefore && InFunctionDecl &&
1852 Current->is(TT_FunctionDeclarationName))
1853 Current->MustBreakBefore = mustBreakForReturnType(Line);
1855 Current->CanBreakBefore =
1856 Current->MustBreakBefore || canBreakBefore(Line, *Current);
1857 unsigned ChildSize = 0;
1858 if (Current->Previous->Children.size() == 1) {
1859 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1860 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1861 : LastOfChild.TotalLength + 1;
1863 const FormatToken *Prev = Current->Previous;
1864 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1865 (Prev->Children.size() == 1 &&
1866 Prev->Children[0]->First->MustBreakBefore) ||
1867 Current->IsMultiline)
1868 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1870 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1871 ChildSize + Current->SpacesRequiredBefore;
1873 if (Current->is(TT_CtorInitializerColon))
1874 InFunctionDecl = false;
1876 // FIXME: Only calculate this if CanBreakBefore is true once static
1877 // initializers etc. are sorted out.
1878 // FIXME: Move magic numbers to a better place.
1879 Current->SplitPenalty = 20 * Current->BindingStrength +
1880 splitPenalty(Line, *Current, InFunctionDecl);
1882 Current = Current->Next;
1885 calculateUnbreakableTailLengths(Line);
1886 unsigned IndentLevel = Line.Level;
1887 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1889 Current->Role->precomputeFormattingInfos(Current);
1890 if (Current->MatchingParen &&
1891 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) {
1892 assert(IndentLevel > 0);
1895 Current->IndentLevel = IndentLevel;
1896 if (Current->opensBlockOrBlockTypeList(Style))
1900 DEBUG({ printDebugInfo(Line); });
1903 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1904 unsigned UnbreakableTailLength = 0;
1905 FormatToken *Current = Line.Last;
1907 Current->UnbreakableTailLength = UnbreakableTailLength;
1908 if (Current->CanBreakBefore ||
1909 Current->isOneOf(tok::comment, tok::string_literal)) {
1910 UnbreakableTailLength = 0;
1912 UnbreakableTailLength +=
1913 Current->ColumnWidth + Current->SpacesRequiredBefore;
1915 Current = Current->Previous;
1919 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1920 const FormatToken &Tok,
1921 bool InFunctionDecl) {
1922 const FormatToken &Left = *Tok.Previous;
1923 const FormatToken &Right = Tok;
1925 if (Left.is(tok::semi))
1928 if (Style.Language == FormatStyle::LK_Java) {
1929 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1931 if (Right.is(Keywords.kw_implements))
1933 if (Left.is(tok::comma) && Left.NestingLevel == 0)
1935 } else if (Style.Language == FormatStyle::LK_JavaScript) {
1936 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1938 if (Left.is(TT_JsTypeColon))
1940 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
1941 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
1945 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
1947 if (Right.is(tok::l_square)) {
1948 if (Style.Language == FormatStyle::LK_Proto)
1950 if (Left.is(tok::r_square))
1952 // Slightly prefer formatting local lambda definitions like functions.
1953 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1955 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1956 TT_ArrayInitializerLSquare))
1960 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1961 Right.is(tok::kw_operator)) {
1962 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1964 if (Left.is(TT_StartOfName))
1966 if (InFunctionDecl && Right.NestingLevel == 0)
1967 return Style.PenaltyReturnTypeOnItsOwnLine;
1970 if (Right.is(TT_PointerOrReference))
1972 if (Right.is(TT_LambdaArrow))
1974 if (Left.is(tok::equal) && Right.is(tok::l_brace))
1976 if (Left.is(TT_CastRParen))
1978 if (Left.is(tok::coloncolon) ||
1979 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1981 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1983 if (Left.is(tok::comment))
1986 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
1989 if (Right.isMemberAccess()) {
1990 // Breaking before the "./->" of a chained call/member access is reasonably
1991 // cheap, as formatting those with one call per line is generally
1992 // desirable. In particular, it should be cheaper to break before the call
1993 // than it is to break inside a call's parameters, which could lead to weird
1994 // "hanging" indents. The exception is the very last "./->" to support this
1995 // frequent pattern:
1997 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
2000 // which might otherwise be blown up onto many lines. Here, clang-format
2001 // won't produce "hanging" indents anyway as there is no other trailing
2004 // Also apply higher penalty is not a call as that might lead to a wrapping
2008 // .aaaaaaaaa.bbbbbbbb(cccccccc);
2009 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
2014 if (Right.is(TT_TrailingAnnotation) &&
2015 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
2016 // Moving trailing annotations to the next line is fine for ObjC method
2018 if (Line.startsWith(TT_ObjCMethodSpecifier))
2020 // Generally, breaking before a trailing annotation is bad unless it is
2021 // function-like. It seems to be especially preferable to keep standard
2022 // annotations (i.e. "const", "final" and "override") on the same line.
2023 // Use a slightly higher penalty after ")" so that annotations like
2024 // "const override" are kept together.
2025 bool is_short_annotation = Right.TokenText.size() < 10;
2026 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
2029 // In for-loops, prefer breaking at ',' and ';'.
2030 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
2033 // In Objective-C method expressions, prefer breaking before "param:" over
2034 // breaking after it.
2035 if (Right.is(TT_SelectorName))
2037 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
2038 return Line.MightBeFunctionDecl ? 50 : 500;
2040 if (Left.is(tok::l_paren) && InFunctionDecl &&
2041 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
2043 if (Left.is(tok::l_paren) && Left.Previous &&
2044 Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
2046 if (Left.is(tok::equal) && InFunctionDecl)
2048 if (Right.is(tok::r_brace))
2050 if (Left.is(TT_TemplateOpener))
2052 if (Left.opensScope()) {
2053 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
2055 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
2058 if (Left.is(TT_JavaAnnotation))
2061 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous &&
2062 Left.Previous->isLabelString() &&
2063 (Left.NextOperator || Left.OperatorIndex != 0))
2065 if (Right.is(tok::plus) && Left.isLabelString() &&
2066 (Right.NextOperator || Right.OperatorIndex != 0))
2068 if (Left.is(tok::comma))
2070 if (Right.is(tok::lessless) && Left.isLabelString() &&
2071 (Right.NextOperator || Right.OperatorIndex != 1))
2073 if (Right.is(tok::lessless)) {
2074 // Breaking at a << is really cheap.
2075 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0)
2076 // Slightly prefer to break before the first one in log-like statements.
2080 if (Left.is(TT_ConditionalExpr))
2081 return prec::Conditional;
2082 prec::Level Level = Left.getPrecedence();
2083 if (Level != prec::Unknown)
2085 Level = Right.getPrecedence();
2086 if (Level != prec::Unknown)
2092 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2093 const FormatToken &Left,
2094 const FormatToken &Right) {
2095 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2097 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
2098 Left.Tok.getObjCKeywordID() == tok::objc_property)
2100 if (Right.is(tok::hashhash))
2101 return Left.is(tok::hash);
2102 if (Left.isOneOf(tok::hashhash, tok::hash))
2103 return Right.is(tok::hash);
2104 if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
2105 return Style.SpaceInEmptyParentheses;
2106 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2107 return (Right.is(TT_CastRParen) ||
2108 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2109 ? Style.SpacesInCStyleCastParentheses
2110 : Style.SpacesInParentheses;
2111 if (Right.isOneOf(tok::semi, tok::comma))
2113 if (Right.is(tok::less) &&
2114 Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)
2116 if (Right.is(tok::less) && Left.is(tok::kw_template))
2117 return Style.SpaceAfterTemplateKeyword;
2118 if (Left.isOneOf(tok::exclaim, tok::tilde))
2120 if (Left.is(tok::at) &&
2121 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2122 tok::numeric_constant, tok::l_paren, tok::l_brace,
2123 tok::kw_true, tok::kw_false))
2125 if (Left.is(tok::colon))
2126 return !Left.is(TT_ObjCMethodExpr);
2127 if (Left.is(tok::coloncolon))
2129 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
2131 if (Right.is(tok::ellipsis))
2132 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2133 Left.Previous->is(tok::kw_case));
2134 if (Left.is(tok::l_square) && Right.is(tok::amp))
2136 if (Right.is(TT_PointerOrReference))
2137 return (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) ||
2138 (Left.Tok.isLiteral() || (Left.is(tok::kw_const) && Left.Previous &&
2139 Left.Previous->is(tok::r_paren)) ||
2140 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
2141 (Style.PointerAlignment != FormatStyle::PAS_Left ||
2142 (Line.IsMultiVariableDeclStmt &&
2143 (Left.NestingLevel == 0 ||
2144 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
2145 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
2146 (!Left.is(TT_PointerOrReference) ||
2147 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2148 !Line.IsMultiVariableDeclStmt)))
2150 if (Left.is(TT_PointerOrReference))
2151 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
2152 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) &&
2153 !Right.is(TT_StartOfName)) ||
2154 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
2155 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
2157 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2158 !Line.IsMultiVariableDeclStmt) &&
2160 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
2161 if (Right.is(tok::star) && Left.is(tok::l_paren))
2163 if (Left.is(tok::l_square))
2164 return (Left.is(TT_ArrayInitializerLSquare) &&
2165 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
2166 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
2167 Right.isNot(tok::r_square));
2168 if (Right.is(tok::r_square))
2169 return Right.MatchingParen &&
2170 ((Style.SpacesInContainerLiterals &&
2171 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
2172 (Style.SpacesInSquareBrackets &&
2173 Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
2174 if (Right.is(tok::l_square) &&
2175 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
2176 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
2178 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
2179 return !Left.Children.empty(); // No spaces in "{}".
2180 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
2181 (Right.is(tok::r_brace) && Right.MatchingParen &&
2182 Right.MatchingParen->BlockKind != BK_Block))
2183 return !Style.Cpp11BracedListStyle;
2184 if (Left.is(TT_BlockComment))
2185 return !Left.TokenText.endswith("=*/");
2186 if (Right.is(tok::l_paren)) {
2187 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
2189 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
2190 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
2191 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
2192 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
2194 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
2195 tok::kw_new, tok::kw_delete) &&
2196 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
2197 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
2198 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
2199 Left.is(tok::r_paren)) &&
2200 Line.Type != LT_PreprocessorDirective);
2202 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
2204 if (Right.is(TT_UnaryOperator))
2205 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2206 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2207 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2209 Left.isSimpleTypeSpecifier()) &&
2210 Right.is(tok::l_brace) && Right.getNextNonComment() &&
2211 Right.BlockKind != BK_Block)
2213 if (Left.is(tok::period) || Right.is(tok::period))
2215 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2217 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2218 Left.MatchingParen->Previous &&
2219 Left.MatchingParen->Previous->is(tok::period))
2220 // A.<B>DoSomething();
2222 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2227 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2228 const FormatToken &Right) {
2229 const FormatToken &Left = *Right.Previous;
2230 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2231 return true; // Never ever merge two identifiers.
2232 if (Style.isCpp()) {
2233 if (Left.is(tok::kw_operator))
2234 return Right.is(tok::coloncolon);
2235 } else if (Style.Language == FormatStyle::LK_Proto) {
2236 if (Right.is(tok::period) &&
2237 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2238 Keywords.kw_repeated, Keywords.kw_extend))
2240 if (Right.is(tok::l_paren) &&
2241 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2243 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2244 if (Left.is(TT_JsFatArrow))
2246 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) &&
2247 Right.MatchingParen) {
2248 const FormatToken *Next = Right.MatchingParen->getNextNonComment();
2249 // An async arrow function, for example: `x = async () => foo();`,
2250 // as opposed to calling a function called async: `x = async();`
2251 if (Next && Next->is(TT_JsFatArrow))
2254 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2255 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2257 if (Left.is(tok::identifier) && Right.is(TT_TemplateString))
2259 if (Right.is(tok::star) &&
2260 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2262 if (Right.isOneOf(tok::l_brace, tok::l_square) &&
2263 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2265 // JS methods can use some keywords as names (e.g. `delete()`).
2266 if (Right.is(tok::l_paren) && Line.MustBeDeclaration &&
2267 Left.Tok.getIdentifierInfo())
2269 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
2271 // "of" is only a keyword if it appears after another identifier
2272 // (e.g. as "const x of y" in a for loop).
2273 (Left.is(Keywords.kw_of) && Left.Previous &&
2274 Left.Previous->Tok.getIdentifierInfo())) &&
2275 (!Left.Previous || !Left.Previous->is(tok::period)))
2277 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
2278 Left.Previous->is(tok::period) && Right.is(tok::l_paren))
2280 if (Left.is(Keywords.kw_as) &&
2281 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
2283 if (Left.is(tok::kw_default) && Left.Previous &&
2284 Left.Previous->is(tok::kw_export))
2286 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2288 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2290 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
2292 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2293 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2295 if (Left.is(tok::ellipsis))
2297 if (Left.is(TT_TemplateCloser) &&
2298 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2299 Keywords.kw_implements, Keywords.kw_extends))
2300 // Type assertions ('<type>expr') are not followed by whitespace. Other
2301 // locations that should have whitespace following are identified by the
2302 // above set of follower tokens.
2304 if (Right.is(TT_JsNonNullAssertion))
2306 if (Left.is(TT_JsNonNullAssertion) && Right.is(Keywords.kw_as))
2307 return true; // "x! as string"
2308 } else if (Style.Language == FormatStyle::LK_Java) {
2309 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2311 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2312 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2313 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2314 tok::kw_protected) ||
2315 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2316 Keywords.kw_native)) &&
2317 Right.is(TT_TemplateOpener))
2320 if (Left.is(TT_ImplicitStringLiteral))
2321 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2322 if (Line.Type == LT_ObjCMethodDecl) {
2323 if (Left.is(TT_ObjCMethodSpecifier))
2325 if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2326 // Don't space between ')' and <id>
2329 if (Line.Type == LT_ObjCProperty &&
2330 (Right.is(tok::equal) || Left.is(tok::equal)))
2333 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2334 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2336 if (Right.is(TT_OverloadedOperatorLParen))
2337 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2338 if (Left.is(tok::comma))
2340 if (Right.is(tok::comma))
2342 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2344 if (Right.is(tok::colon)) {
2345 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2346 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2348 if (Right.is(TT_ObjCMethodExpr))
2350 if (Left.is(tok::question))
2352 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2354 if (Right.is(TT_DictLiteral))
2355 return Style.SpacesInContainerLiterals;
2358 if (Left.is(TT_UnaryOperator))
2359 return Right.is(TT_BinaryOperator);
2361 // If the next token is a binary operator or a selector name, we have
2362 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2363 if (Left.is(TT_CastRParen))
2364 return Style.SpaceAfterCStyleCast ||
2365 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2367 if (Left.is(tok::greater) && Right.is(tok::greater))
2368 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2369 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2370 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2371 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2373 if (!Style.SpaceBeforeAssignmentOperators &&
2374 Right.getPrecedence() == prec::Assignment)
2376 if (Right.is(tok::coloncolon) && Left.is(tok::identifier))
2377 // Generally don't remove existing spaces between an identifier and "::".
2378 // The identifier might actually be a macro name such as ALWAYS_INLINE. If
2379 // this turns out to be too lenient, add analysis of the identifier itself.
2380 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2381 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment))
2382 return (Left.is(TT_TemplateOpener) &&
2383 Style.Standard == FormatStyle::LS_Cpp03) ||
2384 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square,
2385 tok::kw___super, TT_TemplateCloser, TT_TemplateOpener));
2386 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2387 return Style.SpacesInAngles;
2388 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2389 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2390 !Right.is(tok::r_paren)))
2392 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2393 Right.isNot(TT_FunctionTypeLParen))
2394 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2395 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2396 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2398 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2399 Line.startsWith(tok::hash))
2401 if (Right.is(TT_TrailingUnaryOperator))
2403 if (Left.is(TT_RegexLiteral))
2405 return spaceRequiredBetween(Line, Left, Right);
2408 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2409 static bool isAllmanBrace(const FormatToken &Tok) {
2410 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2411 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2414 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2415 const FormatToken &Right) {
2416 const FormatToken &Left = *Right.Previous;
2417 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2420 if (Style.Language == FormatStyle::LK_JavaScript) {
2421 // FIXME: This might apply to other languages and token kinds.
2422 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
2423 Left.Previous->is(tok::string_literal))
2425 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2426 Left.Previous && Left.Previous->is(tok::equal) &&
2427 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2429 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2431 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2432 // Object literals on the top level of a file are treated as "enum-style".
2433 // Each key/value pair is put on a separate line, instead of bin-packing.
2435 if (Left.is(tok::l_brace) && Line.Level == 0 &&
2436 (Line.startsWith(tok::kw_enum) ||
2437 Line.startsWith(tok::kw_export, tok::kw_enum)))
2438 // JavaScript top-level enum key/value pairs are put on separate lines
2439 // instead of bin-packing.
2441 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2442 !Left.Children.empty())
2443 // Support AllowShortFunctionsOnASingleLine for JavaScript.
2444 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2445 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2446 (Left.NestingLevel == 0 && Line.Level == 0 &&
2447 Style.AllowShortFunctionsOnASingleLine ==
2448 FormatStyle::SFS_Inline);
2449 } else if (Style.Language == FormatStyle::LK_Java) {
2450 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2451 Right.Next->is(tok::string_literal))
2453 } else if (Style.Language == FormatStyle::LK_Cpp ||
2454 Style.Language == FormatStyle::LK_ObjC ||
2455 Style.Language == FormatStyle::LK_Proto) {
2456 if (Left.isStringLiteral() && Right.isStringLiteral())
2460 // If the last token before a '}' is a comma or a trailing comment, the
2461 // intention is to insert a line break after it in order to make shuffling
2462 // around entries easier.
2463 const FormatToken *BeforeClosingBrace = nullptr;
2464 if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
2465 Left.BlockKind != BK_Block && Left.MatchingParen)
2466 BeforeClosingBrace = Left.MatchingParen->Previous;
2467 else if (Right.MatchingParen &&
2468 Right.MatchingParen->isOneOf(tok::l_brace,
2469 TT_ArrayInitializerLSquare))
2470 BeforeClosingBrace = &Left;
2471 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2472 BeforeClosingBrace->isTrailingComment()))
2475 if (Right.is(tok::comment))
2476 return Left.BlockKind != BK_BracedInit &&
2477 Left.isNot(TT_CtorInitializerColon) &&
2478 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2479 if (Left.isTrailingComment())
2481 if (Right.Previous->IsUnterminatedLiteral)
2483 if (Right.is(tok::lessless) && Right.Next &&
2484 Right.Previous->is(tok::string_literal) &&
2485 Right.Next->is(tok::string_literal))
2487 if (Right.Previous->ClosesTemplateDeclaration &&
2488 Right.Previous->MatchingParen &&
2489 Right.Previous->MatchingParen->NestingLevel == 0 &&
2490 Style.AlwaysBreakTemplateDeclarations)
2492 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2493 Style.BreakConstructorInitializersBeforeComma &&
2494 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2496 // Break only if we have multiple inheritance.
2497 if (Style.BreakBeforeInheritanceComma &&
2498 Right.is(TT_InheritanceComma))
2500 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2501 // Raw string literals are special wrt. line breaks. The author has made a
2502 // deliberate choice and might have aligned the contents of the string
2503 // literal accordingly. Thus, we try keep existing line breaks.
2504 return Right.NewlinesBefore > 0;
2505 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2506 Style.Language == FormatStyle::LK_Proto)
2507 // Don't put enums onto single lines in protocol buffers.
2509 if (Right.is(TT_InlineASMBrace))
2510 return Right.HasUnescapedNewline;
2511 if (isAllmanBrace(Left) || isAllmanBrace(Right))
2512 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2513 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2514 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2515 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2518 if ((Style.Language == FormatStyle::LK_Java ||
2519 Style.Language == FormatStyle::LK_JavaScript) &&
2520 Left.is(TT_LeadingJavaAnnotation) &&
2521 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2522 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2528 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2529 const FormatToken &Right) {
2530 const FormatToken &Left = *Right.Previous;
2532 // Language-specific stuff.
2533 if (Style.Language == FormatStyle::LK_Java) {
2534 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2535 Keywords.kw_implements))
2537 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2538 Keywords.kw_implements))
2540 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2541 const FormatToken *NonComment = Right.getPreviousNonComment();
2543 NonComment->isOneOf(tok::kw_return, tok::kw_continue, tok::kw_break,
2544 tok::kw_throw, Keywords.kw_interface,
2546 return false; // Otherwise a semicolon is inserted.
2547 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2549 if (Left.is(TT_JsTypeColon))
2551 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2553 if (Left.is(Keywords.kw_in))
2554 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
2555 if (Right.is(Keywords.kw_in))
2556 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
2557 if (Right.is(Keywords.kw_as))
2558 return false; // must not break before as in 'x as type' casts
2559 if (Left.is(Keywords.kw_as))
2561 if (Left.is(TT_JsNonNullAssertion))
2563 if (Left.is(Keywords.kw_declare) &&
2564 Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
2565 Keywords.kw_function, tok::kw_class, tok::kw_enum,
2566 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
2567 Keywords.kw_let, tok::kw_const))
2568 // See grammar for 'declare' statements at:
2569 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
2571 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
2572 Right.isOneOf(tok::identifier, tok::string_literal))
2573 return false; // must not break in "module foo { ...}"
2574 if (Right.is(TT_TemplateString) && Right.closesScope())
2576 if (Left.is(TT_TemplateString) && Left.opensScope())
2580 if (Left.is(tok::at))
2582 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2584 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2585 return !Right.is(tok::l_paren);
2586 if (Right.is(TT_PointerOrReference))
2587 return Line.IsMultiVariableDeclStmt ||
2588 (Style.PointerAlignment == FormatStyle::PAS_Right &&
2589 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2590 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2591 Right.is(tok::kw_operator))
2593 if (Left.is(TT_PointerOrReference))
2595 if (Right.isTrailingComment())
2596 // We rely on MustBreakBefore being set correctly here as we should not
2597 // change the "binding" behavior of a comment.
2598 // The first comment in a braced lists is always interpreted as belonging to
2599 // the first list element. Otherwise, it should be placed outside of the
2601 return Left.BlockKind == BK_BracedInit;
2602 if (Left.is(tok::question) && Right.is(tok::colon))
2604 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2605 return Style.BreakBeforeTernaryOperators;
2606 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2607 return !Style.BreakBeforeTernaryOperators;
2608 if (Right.is(TT_InheritanceColon))
2610 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
2611 Left.isNot(TT_SelectorName))
2613 if (Right.is(tok::colon) &&
2614 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2616 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))
2618 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2619 Right.Next->is(TT_ObjCMethodExpr)))
2620 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2621 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2623 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2625 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2626 TT_OverloadedOperator))
2628 if (Left.is(TT_RangeBasedForLoopColon))
2630 if (Right.is(TT_RangeBasedForLoopColon))
2632 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
2634 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2635 Left.is(tok::kw_operator))
2637 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2638 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2640 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2642 if (Left.is(tok::l_paren) && Left.Previous &&
2643 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2645 if (Right.is(TT_ImplicitStringLiteral))
2648 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2650 if (Right.is(tok::r_square) && Right.MatchingParen &&
2651 Right.MatchingParen->is(TT_LambdaLSquare))
2654 // We only break before r_brace if there was a corresponding break before
2655 // the l_brace, which is tracked by BreakBeforeClosingBrace.
2656 if (Right.is(tok::r_brace))
2657 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2659 // Allow breaking after a trailing annotation, e.g. after a method
2661 if (Left.is(TT_TrailingAnnotation))
2662 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2663 tok::less, tok::coloncolon);
2665 if (Right.is(tok::kw___attribute))
2668 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2671 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2674 if (Left.is(TT_CtorInitializerComma) &&
2675 Style.BreakConstructorInitializersBeforeComma)
2677 if (Right.is(TT_CtorInitializerComma) &&
2678 Style.BreakConstructorInitializersBeforeComma)
2680 if (Left.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma)
2682 if (Right.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma)
2684 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2685 (Left.is(tok::less) && Right.is(tok::less)))
2687 if (Right.is(TT_BinaryOperator) &&
2688 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2689 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2690 Right.getPrecedence() != prec::Assignment))
2692 if (Left.is(TT_ArrayInitializerLSquare))
2694 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2696 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2697 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2698 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2699 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2700 Left.getPrecedence() == prec::Assignment))
2702 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2703 tok::kw_class, tok::kw_struct, tok::comment) ||
2704 Right.isMemberAccess() ||
2705 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2706 tok::colon, tok::l_square, tok::at) ||
2707 (Left.is(tok::r_paren) &&
2708 Right.isOneOf(tok::identifier, tok::kw_const)) ||
2709 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) ||
2710 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser));
2713 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2714 llvm::errs() << "AnnotatedTokens:\n";
2715 const FormatToken *Tok = Line.First;
2717 llvm::errs() << " M=" << Tok->MustBreakBefore
2718 << " C=" << Tok->CanBreakBefore
2719 << " T=" << getTokenTypeName(Tok->Type)
2720 << " S=" << Tok->SpacesRequiredBefore
2721 << " B=" << Tok->BlockParameterCount
2722 << " BK=" << Tok->BlockKind
2723 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2724 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2726 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2727 llvm::errs() << Tok->FakeLParens[i] << "/";
2728 llvm::errs() << " FakeRParens=" << Tok->FakeRParens;
2729 llvm::errs() << " Text='" << Tok->TokenText << "'\n";
2731 assert(Tok == Line.Last);
2734 llvm::errs() << "----\n";
2737 } // namespace format
2738 } // namespace clang