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.Language == FormatStyle::LK_Cpp && Parent &&
315 Parent->is(TT_TemplateCloser) &&
316 (Contexts.back().CanBeExpression || Contexts.back().IsExpression ||
317 Contexts.back().InTemplateArgument);
319 bool StartsObjCMethodExpr =
320 !CppArrayTemplates && (Style.Language == FormatStyle::LK_Cpp ||
321 Style.Language == FormatStyle::LK_ObjC) &&
322 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
323 CurrentToken->isNot(tok::l_brace) &&
325 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
326 tok::kw_return, tok::kw_throw) ||
327 Parent->isUnaryOperator() ||
328 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
329 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
330 bool ColonFound = false;
332 unsigned BindingIncrease = 1;
333 if (Left->is(TT_Unknown)) {
334 if (StartsObjCMethodExpr) {
335 Left->Type = TT_ObjCMethodExpr;
336 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
337 Contexts.back().ContextKind == tok::l_brace &&
338 Parent->isOneOf(tok::l_brace, tok::comma)) {
339 Left->Type = TT_JsComputedPropertyName;
340 } else if (Style.Language == FormatStyle::LK_Proto ||
341 (!CppArrayTemplates && Parent &&
342 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at,
343 tok::comma, tok::l_paren, tok::l_square,
344 tok::question, tok::colon, tok::kw_return,
345 // Should only be relevant to JavaScript:
347 Left->Type = TT_ArrayInitializerLSquare;
349 BindingIncrease = 10;
350 Left->Type = TT_ArraySubscriptLSquare;
354 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
355 Contexts.back().IsExpression = true;
356 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
358 while (CurrentToken) {
359 if (CurrentToken->is(tok::r_square)) {
360 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
361 Left->is(TT_ObjCMethodExpr)) {
362 // An ObjC method call is rarely followed by an open parenthesis.
363 // FIXME: Do we incorrectly label ":" with this?
364 StartsObjCMethodExpr = false;
365 Left->Type = TT_Unknown;
367 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
368 CurrentToken->Type = TT_ObjCMethodExpr;
369 // determineStarAmpUsage() thinks that '*' '[' is allocating an
370 // array of pointers, but if '[' starts a selector then '*' is a
372 if (Parent && Parent->is(TT_PointerOrReference))
373 Parent->Type = TT_BinaryOperator;
375 Left->MatchingParen = CurrentToken;
376 CurrentToken->MatchingParen = Left;
377 if (Contexts.back().FirstObjCSelectorName) {
378 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
379 Contexts.back().LongestObjCSelectorName;
380 if (Left->BlockParameterCount > 1)
381 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
386 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
388 if (CurrentToken->is(tok::colon)) {
389 if (Left->is(TT_ArraySubscriptLSquare)) {
390 Left->Type = TT_ObjCMethodExpr;
391 StartsObjCMethodExpr = true;
392 Contexts.back().ColonIsObjCMethodExpr = true;
393 if (Parent && Parent->is(tok::r_paren))
394 Parent->Type = TT_CastRParen;
398 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
400 Left->Type = TT_ArrayInitializerLSquare;
401 FormatToken *Tok = CurrentToken;
404 updateParameterCount(Left, Tok);
411 FormatToken *Left = CurrentToken->Previous;
412 Left->ParentBracket = Contexts.back().ContextKind;
414 if (Contexts.back().CaretFound)
415 Left->Type = TT_ObjCBlockLBrace;
416 Contexts.back().CaretFound = false;
418 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
419 Contexts.back().ColonIsDictLiteral = true;
420 if (Left->BlockKind == BK_BracedInit)
421 Contexts.back().IsExpression = true;
423 while (CurrentToken) {
424 if (CurrentToken->is(tok::r_brace)) {
425 Left->MatchingParen = CurrentToken;
426 CurrentToken->MatchingParen = Left;
430 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
432 updateParameterCount(Left, CurrentToken);
433 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) {
434 FormatToken *Previous = CurrentToken->getPreviousNonComment();
435 if (((CurrentToken->is(tok::colon) &&
436 (!Contexts.back().ColonIsDictLiteral ||
437 (Style.Language != FormatStyle::LK_Cpp &&
438 Style.Language != FormatStyle::LK_ObjC))) ||
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().FirstStartOfName &&
680 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
681 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
682 Line.IsMultiVariableDeclStmt = true;
684 if (Contexts.back().IsForEachMacro)
685 Contexts.back().IsExpression = true;
693 void parseIncludeDirective() {
694 if (CurrentToken && CurrentToken->is(tok::less)) {
696 while (CurrentToken) {
697 if (CurrentToken->isNot(tok::comment) || CurrentToken->Next)
698 CurrentToken->Type = TT_ImplicitStringLiteral;
704 void parseWarningOrError() {
706 // We still want to format the whitespace left of the first token of the
709 while (CurrentToken) {
710 CurrentToken->Type = TT_ImplicitStringLiteral;
716 next(); // Consume "pragma".
718 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
719 bool IsMark = CurrentToken->is(Keywords.kw_mark);
720 next(); // Consume "mark".
721 next(); // Consume first token (so we fix leading whitespace).
722 while (CurrentToken) {
723 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
724 CurrentToken->Type = TT_ImplicitStringLiteral;
730 LineType parsePreprocessorDirective() {
731 bool IsFirstToken = CurrentToken->IsFirst;
732 LineType Type = LT_PreprocessorDirective;
737 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) {
738 // JavaScript files can contain shebang lines of the form:
739 // #!/usr/bin/env node
740 // Treat these like C++ #include directives.
741 while (CurrentToken) {
742 // Tokens cannot be comments here.
743 CurrentToken->Type = TT_ImplicitStringLiteral;
746 return LT_ImportStatement;
749 if (CurrentToken->Tok.is(tok::numeric_constant)) {
750 CurrentToken->SpacesRequiredBefore = 1;
753 // Hashes in the middle of a line can lead to any strange token
755 if (!CurrentToken->Tok.getIdentifierInfo())
757 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
758 case tok::pp_include:
759 case tok::pp_include_next:
762 parseIncludeDirective();
763 Type = LT_ImportStatement;
766 case tok::pp_warning:
767 parseWarningOrError();
774 Contexts.back().IsExpression = true;
786 LineType parseLine() {
787 NonTemplateLess.clear();
788 if (CurrentToken->is(tok::hash))
789 return parsePreprocessorDirective();
791 // Directly allow to 'import <string-literal>' to support protocol buffer
792 // definitions (code.google.com/p/protobuf) or missing "#" (either way we
793 // should not break the line).
794 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
795 if ((Style.Language == FormatStyle::LK_Java &&
796 CurrentToken->is(Keywords.kw_package)) ||
797 (Info && Info->getPPKeywordID() == tok::pp_import &&
798 CurrentToken->Next &&
799 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
802 parseIncludeDirective();
803 return LT_ImportStatement;
806 // If this line starts and ends in '<' and '>', respectively, it is likely
807 // part of "#define <a/b.h>".
808 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
809 parseIncludeDirective();
810 return LT_ImportStatement;
813 // In .proto files, top-level options are very similar to import statements
814 // and should not be line-wrapped.
815 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
816 CurrentToken->is(Keywords.kw_option)) {
818 if (CurrentToken && CurrentToken->is(tok::identifier))
819 return LT_ImportStatement;
822 bool KeywordVirtualFound = false;
823 bool ImportStatement = false;
825 // import {...} from '...';
826 if (Style.Language == FormatStyle::LK_JavaScript &&
827 CurrentToken->is(Keywords.kw_import))
828 ImportStatement = true;
830 while (CurrentToken) {
831 if (CurrentToken->is(tok::kw_virtual))
832 KeywordVirtualFound = true;
833 if (Style.Language == FormatStyle::LK_JavaScript) {
834 // export {...} from '...';
835 // An export followed by "from 'some string';" is a re-export from
836 // another module identified by a URI and is treated as a
837 // LT_ImportStatement (i.e. prevent wraps on it for long URIs).
838 // Just "export {...};" or "export class ..." should not be treated as
839 // an import in this sense.
840 if (Line.First->is(tok::kw_export) &&
841 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next &&
842 CurrentToken->Next->isStringLiteral())
843 ImportStatement = true;
844 if (isClosureImportStatement(*CurrentToken))
845 ImportStatement = true;
850 if (KeywordVirtualFound)
851 return LT_VirtualFunctionDecl;
853 return LT_ImportStatement;
855 if (Line.startsWith(TT_ObjCMethodSpecifier)) {
856 if (Contexts.back().FirstObjCSelectorName)
857 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
858 Contexts.back().LongestObjCSelectorName;
859 return LT_ObjCMethodDecl;
866 bool isClosureImportStatement(const FormatToken &Tok) {
867 // FIXME: Closure-library specific stuff should not be hard-coded but be
869 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
870 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" ||
871 Tok.Next->Next->TokenText == "provide" ||
872 Tok.Next->Next->TokenText == "require" ||
873 Tok.Next->Next->TokenText == "setTestOnly" ||
874 Tok.Next->Next->TokenText == "forwardDeclare") &&
875 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
878 void resetTokenMetadata(FormatToken *Token) {
882 // Reset token type in case we have already looked at it and then
883 // recovered from an error (e.g. failure to find the matching >).
884 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
885 TT_FunctionLBrace, TT_ImplicitStringLiteral,
886 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
887 TT_OverloadedOperator, TT_RegexLiteral,
889 CurrentToken->Type = TT_Unknown;
890 CurrentToken->Role.reset();
891 CurrentToken->MatchingParen = nullptr;
892 CurrentToken->FakeLParens.clear();
893 CurrentToken->FakeRParens = 0;
898 CurrentToken->NestingLevel = Contexts.size() - 1;
899 CurrentToken->BindingStrength = Contexts.back().BindingStrength;
900 modifyContext(*CurrentToken);
901 determineTokenType(*CurrentToken);
902 CurrentToken = CurrentToken->Next;
905 resetTokenMetadata(CurrentToken);
908 /// \brief A struct to hold information valid in a specific context, e.g.
909 /// a pair of parenthesis.
911 Context(tok::TokenKind ContextKind, unsigned BindingStrength,
913 : ContextKind(ContextKind), BindingStrength(BindingStrength),
914 IsExpression(IsExpression) {}
916 tok::TokenKind ContextKind;
917 unsigned BindingStrength;
919 unsigned LongestObjCSelectorName = 0;
920 bool ColonIsForRangeExpr = false;
921 bool ColonIsDictLiteral = false;
922 bool ColonIsObjCMethodExpr = false;
923 FormatToken *FirstObjCSelectorName = nullptr;
924 FormatToken *FirstStartOfName = nullptr;
925 bool CanBeExpression = true;
926 bool InTemplateArgument = false;
927 bool InCtorInitializer = false;
928 bool CaretFound = false;
929 bool IsForEachMacro = false;
932 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
933 /// of each instance.
934 struct ScopedContextCreator {
937 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
940 P.Contexts.push_back(Context(ContextKind,
941 P.Contexts.back().BindingStrength + Increase,
942 P.Contexts.back().IsExpression));
945 ~ScopedContextCreator() { P.Contexts.pop_back(); }
948 void modifyContext(const FormatToken &Current) {
949 if (Current.getPrecedence() == prec::Assignment &&
950 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
951 // Type aliases use `type X = ...;` in TypeScript.
952 !(Style.Language == FormatStyle::LK_JavaScript &&
953 Line.startsWith(Keywords.kw_type, tok::identifier)) &&
954 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
955 Contexts.back().IsExpression = true;
956 if (!Line.startsWith(TT_UnaryOperator)) {
957 for (FormatToken *Previous = Current.Previous;
958 Previous && Previous->Previous &&
959 !Previous->Previous->isOneOf(tok::comma, tok::semi);
960 Previous = Previous->Previous) {
961 if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
962 Previous = Previous->MatchingParen;
966 if (Previous->opensScope())
968 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
969 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
970 Previous->Previous && Previous->Previous->isNot(tok::equal))
971 Previous->Type = TT_PointerOrReference;
974 } else if (Current.is(tok::lessless) &&
975 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
976 Contexts.back().IsExpression = true;
977 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
978 Contexts.back().IsExpression = true;
979 } else if (Current.is(TT_TrailingReturnArrow)) {
980 Contexts.back().IsExpression = false;
981 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
982 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
983 } else if (Current.Previous &&
984 Current.Previous->is(TT_CtorInitializerColon)) {
985 Contexts.back().IsExpression = true;
986 Contexts.back().InCtorInitializer = true;
987 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
988 for (FormatToken *Previous = Current.Previous;
989 Previous && Previous->isOneOf(tok::star, tok::amp);
990 Previous = Previous->Previous)
991 Previous->Type = TT_PointerOrReference;
992 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
993 Contexts.back().IsExpression = false;
994 } else if (Current.is(tok::kw_new)) {
995 Contexts.back().CanBeExpression = false;
996 } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
997 // This should be the condition or increment in a for-loop.
998 Contexts.back().IsExpression = true;
1002 void determineTokenType(FormatToken &Current) {
1003 if (!Current.is(TT_Unknown))
1004 // The token type is already known.
1007 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1008 // function declaration have been found. In this case, 'Current' is a
1009 // trailing token of this declaration and thus cannot be a name.
1010 if (Current.is(Keywords.kw_instanceof)) {
1011 Current.Type = TT_BinaryOperator;
1012 } else if (isStartOfName(Current) &&
1013 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
1014 Contexts.back().FirstStartOfName = &Current;
1015 Current.Type = TT_StartOfName;
1016 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
1018 } else if (Current.is(tok::arrow) &&
1019 Style.Language == FormatStyle::LK_Java) {
1020 Current.Type = TT_LambdaArrow;
1021 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
1022 Current.NestingLevel == 0) {
1023 Current.Type = TT_TrailingReturnArrow;
1024 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
1026 determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
1027 Contexts.back().IsExpression,
1028 Contexts.back().InTemplateArgument);
1029 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
1030 Current.Type = determinePlusMinusCaretUsage(Current);
1031 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
1032 Contexts.back().CaretFound = true;
1033 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
1034 Current.Type = determineIncrementUsage(Current);
1035 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
1036 Current.Type = TT_UnaryOperator;
1037 } else if (Current.is(tok::question)) {
1038 if (Style.Language == FormatStyle::LK_JavaScript &&
1039 Line.MustBeDeclaration && !Contexts.back().IsExpression) {
1040 // In JavaScript, `interface X { foo?(): bar; }` is an optional method
1041 // on the interface, not a ternary expression.
1042 Current.Type = TT_JsTypeOptionalQuestion;
1044 Current.Type = TT_ConditionalExpr;
1046 } else if (Current.isBinaryOperator() &&
1047 (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
1048 Current.Type = TT_BinaryOperator;
1049 } else if (Current.is(tok::comment)) {
1050 if (Current.TokenText.startswith("/*")) {
1051 if (Current.TokenText.endswith("*/"))
1052 Current.Type = TT_BlockComment;
1054 // The lexer has for some reason determined a comment here. But we
1055 // cannot really handle it, if it isn't properly terminated.
1056 Current.Tok.setKind(tok::unknown);
1058 Current.Type = TT_LineComment;
1060 } else if (Current.is(tok::r_paren)) {
1061 if (rParenEndsCast(Current))
1062 Current.Type = TT_CastRParen;
1063 if (Current.MatchingParen && Current.Next &&
1064 !Current.Next->isBinaryOperator() &&
1065 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace,
1066 tok::period, tok::arrow, tok::coloncolon))
1067 if (FormatToken *AfterParen = Current.MatchingParen->Next) {
1068 // Make sure this isn't the return type of an Obj-C block declaration
1069 if (AfterParen->Tok.isNot(tok::caret)) {
1070 if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
1071 if (BeforeParen->is(tok::identifier) &&
1072 BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
1073 (!BeforeParen->Previous ||
1074 BeforeParen->Previous->ClosesTemplateDeclaration))
1075 Current.Type = TT_FunctionAnnotationRParen;
1078 } else if (Current.is(tok::at) && Current.Next) {
1079 if (Current.Next->isStringLiteral()) {
1080 Current.Type = TT_ObjCStringLiteral;
1082 switch (Current.Next->Tok.getObjCKeywordID()) {
1083 case tok::objc_interface:
1084 case tok::objc_implementation:
1085 case tok::objc_protocol:
1086 Current.Type = TT_ObjCDecl;
1088 case tok::objc_property:
1089 Current.Type = TT_ObjCProperty;
1095 } else if (Current.is(tok::period)) {
1096 FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1097 if (PreviousNoComment &&
1098 PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1099 Current.Type = TT_DesignatedInitializerPeriod;
1100 else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1101 Current.Previous->isOneOf(TT_JavaAnnotation,
1102 TT_LeadingJavaAnnotation)) {
1103 Current.Type = Current.Previous->Type;
1105 } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
1107 !Current.Previous->isOneOf(tok::equal, tok::at) &&
1108 Line.MightBeFunctionDecl && Contexts.size() == 1) {
1109 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1110 // function declaration have been found.
1111 Current.Type = TT_TrailingAnnotation;
1112 } else if ((Style.Language == FormatStyle::LK_Java ||
1113 Style.Language == FormatStyle::LK_JavaScript) &&
1115 if (Current.Previous->is(tok::at) &&
1116 Current.isNot(Keywords.kw_interface)) {
1117 const FormatToken &AtToken = *Current.Previous;
1118 const FormatToken *Previous = AtToken.getPreviousNonComment();
1119 if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1120 Current.Type = TT_LeadingJavaAnnotation;
1122 Current.Type = TT_JavaAnnotation;
1123 } else if (Current.Previous->is(tok::period) &&
1124 Current.Previous->isOneOf(TT_JavaAnnotation,
1125 TT_LeadingJavaAnnotation)) {
1126 Current.Type = Current.Previous->Type;
1131 /// \brief Take a guess at whether \p Tok starts a name of a function or
1132 /// variable declaration.
1134 /// This is a heuristic based on whether \p Tok is an identifier following
1135 /// something that is likely a type.
1136 bool isStartOfName(const FormatToken &Tok) {
1137 if (Tok.isNot(tok::identifier) || !Tok.Previous)
1140 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof))
1142 if (Style.Language == FormatStyle::LK_JavaScript &&
1143 Tok.Previous->is(Keywords.kw_in))
1146 // Skip "const" as it does not have an influence on whether this is a name.
1147 FormatToken *PreviousNotConst = Tok.Previous;
1148 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1149 PreviousNotConst = PreviousNotConst->Previous;
1151 if (!PreviousNotConst)
1154 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1155 PreviousNotConst->Previous &&
1156 PreviousNotConst->Previous->is(tok::hash);
1158 if (PreviousNotConst->is(TT_TemplateCloser))
1159 return PreviousNotConst && PreviousNotConst->MatchingParen &&
1160 PreviousNotConst->MatchingParen->Previous &&
1161 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1162 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1164 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1165 PreviousNotConst->MatchingParen->Previous &&
1166 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1169 return (!IsPPKeyword &&
1170 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1171 PreviousNotConst->is(TT_PointerOrReference) ||
1172 PreviousNotConst->isSimpleTypeSpecifier();
1175 /// \brief Determine whether ')' is ending a cast.
1176 bool rParenEndsCast(const FormatToken &Tok) {
1177 // C-style casts are only used in C++ and Java.
1178 if (Style.Language != FormatStyle::LK_Cpp &&
1179 Style.Language != FormatStyle::LK_ObjC &&
1180 Style.Language != FormatStyle::LK_Java)
1183 // Empty parens aren't casts and there are no casts at the end of the line.
1184 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1187 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1189 // If there is a closing parenthesis left of the current parentheses,
1190 // look past it as these might be chained casts.
1191 if (LeftOfParens->is(tok::r_paren)) {
1192 if (!LeftOfParens->MatchingParen ||
1193 !LeftOfParens->MatchingParen->Previous)
1195 LeftOfParens = LeftOfParens->MatchingParen->Previous;
1198 // If there is an identifier (or with a few exceptions a keyword) right
1199 // before the parentheses, this is unlikely to be a cast.
1200 if (LeftOfParens->Tok.getIdentifierInfo() &&
1201 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1205 // Certain other tokens right before the parentheses are also signals that
1206 // this cannot be a cast.
1207 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1208 TT_TemplateCloser, tok::ellipsis))
1212 if (Tok.Next->is(tok::question))
1215 // As Java has no function types, a "(" after the ")" likely means that this
1217 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1220 // If a (non-string) literal follows, this is likely a cast.
1221 if (Tok.Next->isNot(tok::string_literal) &&
1222 (Tok.Next->Tok.isLiteral() ||
1223 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1226 // Heuristically try to determine whether the parentheses contain a type.
1227 bool ParensAreType =
1229 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1230 Tok.Previous->isSimpleTypeSpecifier();
1231 bool ParensCouldEndDecl =
1232 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1233 if (ParensAreType && !ParensCouldEndDecl)
1236 // At this point, we heuristically assume that there are no casts at the
1237 // start of the line. We assume that we have found most cases where there
1238 // are by the logic above, e.g. "(void)x;".
1242 // Certain token types inside the parentheses mean that this can't be a
1244 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok;
1245 Token = Token->Next)
1246 if (Token->is(TT_BinaryOperator))
1249 // If the following token is an identifier or 'this', this is a cast. All
1250 // cases where this can be something else are handled above.
1251 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this))
1254 if (!Tok.Next->Next)
1257 // If the next token after the parenthesis is a unary operator, assume
1258 // that this is cast, unless there are unexpected tokens inside the
1261 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1262 if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1263 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1265 // Search for unexpected tokens.
1266 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1267 Prev = Prev->Previous) {
1268 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1274 /// \brief Return the type of the given token assuming it is * or &.
1275 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1276 bool InTemplateArgument) {
1277 if (Style.Language == FormatStyle::LK_JavaScript)
1278 return TT_BinaryOperator;
1280 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1282 return TT_UnaryOperator;
1284 const FormatToken *NextToken = Tok.getNextNonComment();
1285 if (!NextToken || NextToken->isOneOf(tok::arrow, tok::equal) ||
1286 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1287 return TT_PointerOrReference;
1289 if (PrevToken->is(tok::coloncolon))
1290 return TT_PointerOrReference;
1292 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1293 tok::comma, tok::semi, tok::kw_return, tok::colon,
1294 tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1295 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1296 TT_UnaryOperator, TT_CastRParen))
1297 return TT_UnaryOperator;
1299 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1300 return TT_PointerOrReference;
1301 if (NextToken->is(tok::kw_operator) && !IsExpression)
1302 return TT_PointerOrReference;
1303 if (NextToken->isOneOf(tok::comma, tok::semi))
1304 return TT_PointerOrReference;
1306 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1307 PrevToken->MatchingParen->Previous &&
1308 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1310 return TT_PointerOrReference;
1312 if (PrevToken->Tok.isLiteral() ||
1313 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1314 tok::kw_false, tok::r_brace) ||
1315 NextToken->Tok.isLiteral() ||
1316 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1317 NextToken->isUnaryOperator() ||
1318 // If we know we're in a template argument, there are no named
1319 // declarations. Thus, having an identifier on the right-hand side
1320 // indicates a binary operator.
1321 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1322 return TT_BinaryOperator;
1324 // "&&(" is quite unlikely to be two successive unary "&".
1325 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1326 return TT_BinaryOperator;
1328 // This catches some cases where evaluation order is used as control flow:
1330 const FormatToken *NextNextToken = NextToken->getNextNonComment();
1331 if (NextNextToken && NextNextToken->is(tok::arrow))
1332 return TT_BinaryOperator;
1334 // It is very unlikely that we are going to find a pointer or reference type
1335 // definition on the RHS of an assignment.
1336 if (IsExpression && !Contexts.back().CaretFound)
1337 return TT_BinaryOperator;
1339 return TT_PointerOrReference;
1342 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1343 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1345 return TT_UnaryOperator;
1347 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator) &&
1348 !PrevToken->is(tok::exclaim))
1349 // There aren't any trailing unary operators except for TypeScript's
1350 // non-null operator (!). Thus, this must be squence of leading operators.
1351 return TT_UnaryOperator;
1353 // Use heuristics to recognize unary operators.
1354 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1355 tok::question, tok::colon, tok::kw_return,
1356 tok::kw_case, tok::at, tok::l_brace))
1357 return TT_UnaryOperator;
1359 // There can't be two consecutive binary operators.
1360 if (PrevToken->is(TT_BinaryOperator))
1361 return TT_UnaryOperator;
1363 // Fall back to marking the token as binary operator.
1364 return TT_BinaryOperator;
1367 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1368 TokenType determineIncrementUsage(const FormatToken &Tok) {
1369 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1370 if (!PrevToken || PrevToken->is(TT_CastRParen))
1371 return TT_UnaryOperator;
1372 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1373 return TT_TrailingUnaryOperator;
1375 return TT_UnaryOperator;
1378 SmallVector<Context, 8> Contexts;
1380 const FormatStyle &Style;
1381 AnnotatedLine &Line;
1382 FormatToken *CurrentToken;
1384 const AdditionalKeywords &Keywords;
1386 // Set of "<" tokens that do not open a template parameter list. If parseAngle
1387 // determines that a specific token can't be a template opener, it will make
1388 // same decision irrespective of the decisions for tokens leading up to it.
1389 // Store this information to prevent this from causing exponential runtime.
1390 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1393 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1394 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1396 /// \brief Parses binary expressions by inserting fake parenthesis based on
1397 /// operator precedence.
1398 class ExpressionParser {
1400 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1401 AnnotatedLine &Line)
1402 : Style(Style), Keywords(Keywords), Current(Line.First) {}
1404 /// \brief Parse expressions with the given operatore precedence.
1405 void parse(int Precedence = 0) {
1406 // Skip 'return' and ObjC selector colons as they are not part of a binary
1408 while (Current && (Current->is(tok::kw_return) ||
1409 (Current->is(tok::colon) &&
1410 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1413 if (!Current || Precedence > PrecedenceArrowAndPeriod)
1416 // Conditional expressions need to be parsed separately for proper nesting.
1417 if (Precedence == prec::Conditional) {
1418 parseConditionalExpr();
1422 // Parse unary operators, which all have a higher precedence than binary
1424 if (Precedence == PrecedenceUnaryOperator) {
1425 parseUnaryOperator();
1429 FormatToken *Start = Current;
1430 FormatToken *LatestOperator = nullptr;
1431 unsigned OperatorIndex = 0;
1434 // Consume operators with higher precedence.
1435 parse(Precedence + 1);
1437 int CurrentPrecedence = getCurrentPrecedence();
1439 if (Current && Current->is(TT_SelectorName) &&
1440 Precedence == CurrentPrecedence) {
1442 addFakeParenthesis(Start, prec::Level(Precedence));
1446 // At the end of the line or when an operator with higher precedence is
1447 // found, insert fake parenthesis and return.
1448 if (!Current || (Current->closesScope() && Current->MatchingParen) ||
1449 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1450 (CurrentPrecedence == prec::Conditional &&
1451 Precedence == prec::Assignment && Current->is(tok::colon))) {
1455 // Consume scopes: (), [], <> and {}
1456 if (Current->opensScope()) {
1457 while (Current && !Current->closesScope()) {
1464 if (CurrentPrecedence == Precedence) {
1466 LatestOperator->NextOperator = Current;
1467 LatestOperator = Current;
1468 Current->OperatorIndex = OperatorIndex;
1471 next(/*SkipPastLeadingComments=*/Precedence > 0);
1475 if (LatestOperator && (Current || Precedence > 0)) {
1476 // LatestOperator->LastOperator = true;
1477 if (Precedence == PrecedenceArrowAndPeriod) {
1478 // Call expressions don't have a binary operator precedence.
1479 addFakeParenthesis(Start, prec::Unknown);
1481 addFakeParenthesis(Start, prec::Level(Precedence));
1487 /// \brief Gets the precedence (+1) of the given token for binary operators
1488 /// and other tokens that we treat like binary operators.
1489 int getCurrentPrecedence() {
1491 const FormatToken *NextNonComment = Current->getNextNonComment();
1492 if (Current->is(TT_ConditionalExpr))
1493 return prec::Conditional;
1494 if (NextNonComment && NextNonComment->is(tok::colon) &&
1495 NextNonComment->is(TT_DictLiteral))
1497 if (Current->is(TT_LambdaArrow))
1499 if (Current->is(TT_JsFatArrow))
1500 return prec::Assignment;
1501 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName,
1502 TT_JsComputedPropertyName) ||
1503 (Current->is(tok::comment) && NextNonComment &&
1504 NextNonComment->is(TT_SelectorName)))
1506 if (Current->is(TT_RangeBasedForLoopColon))
1508 if ((Style.Language == FormatStyle::LK_Java ||
1509 Style.Language == FormatStyle::LK_JavaScript) &&
1510 Current->is(Keywords.kw_instanceof))
1511 return prec::Relational;
1512 if (Style.Language == FormatStyle::LK_JavaScript &&
1513 Current->is(Keywords.kw_in))
1514 return prec::Relational;
1515 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1516 return Current->getPrecedence();
1517 if (Current->isOneOf(tok::period, tok::arrow))
1518 return PrecedenceArrowAndPeriod;
1519 if ((Style.Language == FormatStyle::LK_Java ||
1520 Style.Language == FormatStyle::LK_JavaScript) &&
1521 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1522 Keywords.kw_throws))
1528 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1529 Start->FakeLParens.push_back(Precedence);
1530 if (Precedence > prec::Unknown)
1531 Start->StartsBinaryExpression = true;
1533 FormatToken *Previous = Current->Previous;
1534 while (Previous->is(tok::comment) && Previous->Previous)
1535 Previous = Previous->Previous;
1536 ++Previous->FakeRParens;
1537 if (Precedence > prec::Unknown)
1538 Previous->EndsBinaryExpression = true;
1542 /// \brief Parse unary operator expressions and surround them with fake
1543 /// parentheses if appropriate.
1544 void parseUnaryOperator() {
1545 if (!Current || Current->isNot(TT_UnaryOperator)) {
1546 parse(PrecedenceArrowAndPeriod);
1550 FormatToken *Start = Current;
1552 parseUnaryOperator();
1554 // The actual precedence doesn't matter.
1555 addFakeParenthesis(Start, prec::Unknown);
1558 void parseConditionalExpr() {
1559 while (Current && Current->isTrailingComment()) {
1562 FormatToken *Start = Current;
1563 parse(prec::LogicalOr);
1564 if (!Current || !Current->is(tok::question))
1567 parse(prec::Assignment);
1568 if (!Current || Current->isNot(TT_ConditionalExpr))
1571 parse(prec::Assignment);
1572 addFakeParenthesis(Start, prec::Conditional);
1575 void next(bool SkipPastLeadingComments = true) {
1577 Current = Current->Next;
1579 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1580 Current->isTrailingComment())
1581 Current = Current->Next;
1584 const FormatStyle &Style;
1585 const AdditionalKeywords &Keywords;
1586 FormatToken *Current;
1589 } // end anonymous namespace
1591 void TokenAnnotator::setCommentLineLevels(
1592 SmallVectorImpl<AnnotatedLine *> &Lines) {
1593 const AnnotatedLine *NextNonCommentLine = nullptr;
1594 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1597 if (NextNonCommentLine && (*I)->First->is(tok::comment) &&
1598 (*I)->First->Next == nullptr)
1599 (*I)->Level = NextNonCommentLine->Level;
1601 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1603 setCommentLineLevels((*I)->Children);
1607 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
1608 unsigned Result = 0;
1609 for (const auto* Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
1610 Result = std::max(Result, Tok->NestingLevel);
1614 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1615 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1616 E = Line.Children.end();
1620 AnnotatingParser Parser(Style, Line, Keywords);
1621 Line.Type = Parser.parseLine();
1623 // With very deep nesting, ExpressionParser uses lots of stack and the
1624 // formatting algorithm is very slow. We're not going to do a good job here
1625 // anyway - it's probably generated code being formatted by mistake.
1626 // Just skip the whole line.
1627 if (maxNestingDepth(Line) > 50)
1628 Line.Type = LT_Invalid;
1630 if (Line.Type == LT_Invalid)
1633 ExpressionParser ExprParser(Style, Keywords, Line);
1636 if (Line.startsWith(TT_ObjCMethodSpecifier))
1637 Line.Type = LT_ObjCMethodDecl;
1638 else if (Line.startsWith(TT_ObjCDecl))
1639 Line.Type = LT_ObjCDecl;
1640 else if (Line.startsWith(TT_ObjCProperty))
1641 Line.Type = LT_ObjCProperty;
1643 Line.First->SpacesRequiredBefore = 1;
1644 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1647 // This function heuristically determines whether 'Current' starts the name of a
1648 // function declaration.
1649 static bool isFunctionDeclarationName(const FormatToken &Current,
1650 const AnnotatedLine &Line) {
1651 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1652 for (; Next; Next = Next->Next) {
1653 if (Next->is(TT_OverloadedOperatorLParen))
1655 if (Next->is(TT_OverloadedOperator))
1657 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1658 // For 'new[]' and 'delete[]'.
1659 if (Next->Next && Next->Next->is(tok::l_square) &&
1660 Next->Next->Next && Next->Next->Next->is(tok::r_square))
1661 Next = Next->Next->Next;
1670 // Find parentheses of parameter list.
1671 const FormatToken *Next = Current.Next;
1672 if (Current.is(tok::kw_operator)) {
1673 if (Current.Previous && Current.Previous->is(tok::coloncolon))
1675 Next = skipOperatorName(Next);
1677 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1679 for (; Next; Next = Next->Next) {
1680 if (Next->is(TT_TemplateOpener)) {
1681 Next = Next->MatchingParen;
1682 } else if (Next->is(tok::coloncolon)) {
1686 if (Next->is(tok::kw_operator)) {
1687 Next = skipOperatorName(Next->Next);
1690 if (!Next->is(tok::identifier))
1692 } else if (Next->is(tok::l_paren)) {
1700 // Check whether parameter list can be long to a function declaration.
1701 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
1703 // If the lines ends with "{", this is likely an function definition.
1704 if (Line.Last->is(tok::l_brace))
1706 if (Next->Next == Next->MatchingParen)
1707 return true; // Empty parentheses.
1708 // If there is an &/&& after the r_paren, this is likely a function.
1709 if (Next->MatchingParen->Next &&
1710 Next->MatchingParen->Next->is(TT_PointerOrReference))
1712 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1714 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1715 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
1717 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1718 Tok->Tok.isLiteral())
1724 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1725 assert(Line.MightBeFunctionDecl);
1727 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1728 Style.AlwaysBreakAfterReturnType ==
1729 FormatStyle::RTBS_TopLevelDefinitions) &&
1733 switch (Style.AlwaysBreakAfterReturnType) {
1734 case FormatStyle::RTBS_None:
1736 case FormatStyle::RTBS_All:
1737 case FormatStyle::RTBS_TopLevel:
1739 case FormatStyle::RTBS_AllDefinitions:
1740 case FormatStyle::RTBS_TopLevelDefinitions:
1741 return Line.mightBeFunctionDefinition();
1747 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1748 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1749 E = Line.Children.end();
1751 calculateFormattingInformation(**I);
1754 Line.First->TotalLength =
1755 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1756 if (!Line.First->Next)
1758 FormatToken *Current = Line.First->Next;
1759 bool InFunctionDecl = Line.MightBeFunctionDecl;
1761 if (isFunctionDeclarationName(*Current, Line))
1762 Current->Type = TT_FunctionDeclarationName;
1763 if (Current->is(TT_LineComment)) {
1764 if (Current->Previous->BlockKind == BK_BracedInit &&
1765 Current->Previous->opensScope())
1766 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1768 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1770 // If we find a trailing comment, iterate backwards to determine whether
1771 // it seems to relate to a specific parameter. If so, break before that
1772 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1773 // to the previous line in:
1777 if (!Current->HasUnescapedNewline) {
1778 for (FormatToken *Parameter = Current->Previous; Parameter;
1779 Parameter = Parameter->Previous) {
1780 if (Parameter->isOneOf(tok::comment, tok::r_brace))
1782 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1783 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1784 Parameter->HasUnescapedNewline)
1785 Parameter->MustBreakBefore = true;
1790 } else if (Current->SpacesRequiredBefore == 0 &&
1791 spaceRequiredBefore(Line, *Current)) {
1792 Current->SpacesRequiredBefore = 1;
1795 Current->MustBreakBefore =
1796 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1798 if (!Current->MustBreakBefore && InFunctionDecl &&
1799 Current->is(TT_FunctionDeclarationName))
1800 Current->MustBreakBefore = mustBreakForReturnType(Line);
1802 Current->CanBreakBefore =
1803 Current->MustBreakBefore || canBreakBefore(Line, *Current);
1804 unsigned ChildSize = 0;
1805 if (Current->Previous->Children.size() == 1) {
1806 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1807 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1808 : LastOfChild.TotalLength + 1;
1810 const FormatToken *Prev = Current->Previous;
1811 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1812 (Prev->Children.size() == 1 &&
1813 Prev->Children[0]->First->MustBreakBefore) ||
1814 Current->IsMultiline)
1815 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1817 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1818 ChildSize + Current->SpacesRequiredBefore;
1820 if (Current->is(TT_CtorInitializerColon))
1821 InFunctionDecl = false;
1823 // FIXME: Only calculate this if CanBreakBefore is true once static
1824 // initializers etc. are sorted out.
1825 // FIXME: Move magic numbers to a better place.
1826 Current->SplitPenalty = 20 * Current->BindingStrength +
1827 splitPenalty(Line, *Current, InFunctionDecl);
1829 Current = Current->Next;
1832 calculateUnbreakableTailLengths(Line);
1833 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1835 Current->Role->precomputeFormattingInfos(Current);
1838 DEBUG({ printDebugInfo(Line); });
1841 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1842 unsigned UnbreakableTailLength = 0;
1843 FormatToken *Current = Line.Last;
1845 Current->UnbreakableTailLength = UnbreakableTailLength;
1846 if (Current->CanBreakBefore ||
1847 Current->isOneOf(tok::comment, tok::string_literal)) {
1848 UnbreakableTailLength = 0;
1850 UnbreakableTailLength +=
1851 Current->ColumnWidth + Current->SpacesRequiredBefore;
1853 Current = Current->Previous;
1857 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1858 const FormatToken &Tok,
1859 bool InFunctionDecl) {
1860 const FormatToken &Left = *Tok.Previous;
1861 const FormatToken &Right = Tok;
1863 if (Left.is(tok::semi))
1866 if (Style.Language == FormatStyle::LK_Java) {
1867 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1869 if (Right.is(Keywords.kw_implements))
1871 if (Left.is(tok::comma) && Left.NestingLevel == 0)
1873 } else if (Style.Language == FormatStyle::LK_JavaScript) {
1874 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1876 if (Left.is(TT_JsTypeColon))
1878 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
1879 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
1883 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
1885 if (Right.is(tok::l_square)) {
1886 if (Style.Language == FormatStyle::LK_Proto)
1888 if (Left.is(tok::r_square))
1890 // Slightly prefer formatting local lambda definitions like functions.
1891 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1893 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1894 TT_ArrayInitializerLSquare))
1898 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1899 Right.is(tok::kw_operator)) {
1900 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1902 if (Left.is(TT_StartOfName))
1904 if (InFunctionDecl && Right.NestingLevel == 0)
1905 return Style.PenaltyReturnTypeOnItsOwnLine;
1908 if (Right.is(TT_PointerOrReference))
1910 if (Right.is(TT_LambdaArrow))
1912 if (Left.is(tok::equal) && Right.is(tok::l_brace))
1914 if (Left.is(TT_CastRParen))
1916 if (Left.is(tok::coloncolon) ||
1917 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1919 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1921 if (Left.is(tok::comment))
1924 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
1927 if (Right.isMemberAccess()) {
1928 // Breaking before the "./->" of a chained call/member access is reasonably
1929 // cheap, as formatting those with one call per line is generally
1930 // desirable. In particular, it should be cheaper to break before the call
1931 // than it is to break inside a call's parameters, which could lead to weird
1932 // "hanging" indents. The exception is the very last "./->" to support this
1933 // frequent pattern:
1935 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
1938 // which might otherwise be blown up onto many lines. Here, clang-format
1939 // won't produce "hanging" indents anyway as there is no other trailing
1942 // Also apply higher penalty is not a call as that might lead to a wrapping
1946 // .aaaaaaaaa.bbbbbbbb(cccccccc);
1947 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
1952 if (Right.is(TT_TrailingAnnotation) &&
1953 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
1954 // Moving trailing annotations to the next line is fine for ObjC method
1956 if (Line.startsWith(TT_ObjCMethodSpecifier))
1958 // Generally, breaking before a trailing annotation is bad unless it is
1959 // function-like. It seems to be especially preferable to keep standard
1960 // annotations (i.e. "const", "final" and "override") on the same line.
1961 // Use a slightly higher penalty after ")" so that annotations like
1962 // "const override" are kept together.
1963 bool is_short_annotation = Right.TokenText.size() < 10;
1964 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
1967 // In for-loops, prefer breaking at ',' and ';'.
1968 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
1971 // In Objective-C method expressions, prefer breaking before "param:" over
1972 // breaking after it.
1973 if (Right.is(TT_SelectorName))
1975 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
1976 return Line.MightBeFunctionDecl ? 50 : 500;
1978 if (Left.is(tok::l_paren) && InFunctionDecl &&
1979 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1981 if (Left.is(tok::l_paren) && Left.Previous &&
1982 Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
1984 if (Left.is(tok::equal) && InFunctionDecl)
1986 if (Right.is(tok::r_brace))
1988 if (Left.is(TT_TemplateOpener))
1990 if (Left.opensScope()) {
1991 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
1993 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
1996 if (Left.is(TT_JavaAnnotation))
1999 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous &&
2000 Left.Previous->isLabelString() &&
2001 (Left.NextOperator || Left.OperatorIndex != 0))
2003 if (Right.is(tok::plus) && Left.isLabelString() &&
2004 (Right.NextOperator || Right.OperatorIndex != 0))
2006 if (Left.is(tok::comma))
2008 if (Right.is(tok::lessless) && Left.isLabelString() &&
2009 (Right.NextOperator || Right.OperatorIndex != 1))
2011 if (Right.is(tok::lessless)) {
2012 // Breaking at a << is really cheap.
2013 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0)
2014 // Slightly prefer to break before the first one in log-like statements.
2018 if (Left.is(TT_ConditionalExpr))
2019 return prec::Conditional;
2020 prec::Level Level = Left.getPrecedence();
2021 if (Level != prec::Unknown)
2023 Level = Right.getPrecedence();
2024 if (Level != prec::Unknown)
2030 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2031 const FormatToken &Left,
2032 const FormatToken &Right) {
2033 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2035 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
2036 Left.Tok.getObjCKeywordID() == tok::objc_property)
2038 if (Right.is(tok::hashhash))
2039 return Left.is(tok::hash);
2040 if (Left.isOneOf(tok::hashhash, tok::hash))
2041 return Right.is(tok::hash);
2042 if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
2043 return Style.SpaceInEmptyParentheses;
2044 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2045 return (Right.is(TT_CastRParen) ||
2046 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2047 ? Style.SpacesInCStyleCastParentheses
2048 : Style.SpacesInParentheses;
2049 if (Right.isOneOf(tok::semi, tok::comma))
2051 if (Right.is(tok::less) &&
2052 Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)
2054 if (Right.is(tok::less) && Left.is(tok::kw_template))
2055 return Style.SpaceAfterTemplateKeyword;
2056 if (Left.isOneOf(tok::exclaim, tok::tilde))
2058 if (Left.is(tok::at) &&
2059 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2060 tok::numeric_constant, tok::l_paren, tok::l_brace,
2061 tok::kw_true, tok::kw_false))
2063 if (Left.is(tok::colon))
2064 return !Left.is(TT_ObjCMethodExpr);
2065 if (Left.is(tok::coloncolon))
2067 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
2069 if (Right.is(tok::ellipsis))
2070 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2071 Left.Previous->is(tok::kw_case));
2072 if (Left.is(tok::l_square) && Right.is(tok::amp))
2074 if (Right.is(TT_PointerOrReference))
2075 return (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) ||
2076 (Left.Tok.isLiteral() || (Left.is(tok::kw_const) && Left.Previous &&
2077 Left.Previous->is(tok::r_paren)) ||
2078 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
2079 (Style.PointerAlignment != FormatStyle::PAS_Left ||
2080 (Line.IsMultiVariableDeclStmt &&
2081 (Left.NestingLevel == 0 ||
2082 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
2083 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
2084 (!Left.is(TT_PointerOrReference) ||
2085 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2086 !Line.IsMultiVariableDeclStmt)))
2088 if (Left.is(TT_PointerOrReference))
2089 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
2090 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) &&
2091 !Right.is(TT_StartOfName)) ||
2092 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
2093 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
2095 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2096 !Line.IsMultiVariableDeclStmt) &&
2098 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
2099 if (Right.is(tok::star) && Left.is(tok::l_paren))
2101 if (Left.is(tok::l_square))
2102 return (Left.is(TT_ArrayInitializerLSquare) &&
2103 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
2104 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
2105 Right.isNot(tok::r_square));
2106 if (Right.is(tok::r_square))
2107 return Right.MatchingParen &&
2108 ((Style.SpacesInContainerLiterals &&
2109 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
2110 (Style.SpacesInSquareBrackets &&
2111 Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
2112 if (Right.is(tok::l_square) &&
2113 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
2114 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
2116 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
2117 return !Left.Children.empty(); // No spaces in "{}".
2118 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
2119 (Right.is(tok::r_brace) && Right.MatchingParen &&
2120 Right.MatchingParen->BlockKind != BK_Block))
2121 return !Style.Cpp11BracedListStyle;
2122 if (Left.is(TT_BlockComment))
2123 return !Left.TokenText.endswith("=*/");
2124 if (Right.is(tok::l_paren)) {
2125 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
2127 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
2128 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
2129 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
2130 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
2132 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
2133 tok::kw_new, tok::kw_delete) &&
2134 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
2135 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
2136 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
2137 Left.is(tok::r_paren)) &&
2138 Line.Type != LT_PreprocessorDirective);
2140 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
2142 if (Right.is(TT_UnaryOperator))
2143 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2144 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2145 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2147 Left.isSimpleTypeSpecifier()) &&
2148 Right.is(tok::l_brace) && Right.getNextNonComment() &&
2149 Right.BlockKind != BK_Block)
2151 if (Left.is(tok::period) || Right.is(tok::period))
2153 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2155 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2156 Left.MatchingParen->Previous &&
2157 Left.MatchingParen->Previous->is(tok::period))
2158 // A.<B>DoSomething();
2160 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2165 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2166 const FormatToken &Right) {
2167 const FormatToken &Left = *Right.Previous;
2168 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2169 return true; // Never ever merge two identifiers.
2170 if (Style.Language == FormatStyle::LK_Cpp) {
2171 if (Left.is(tok::kw_operator))
2172 return Right.is(tok::coloncolon);
2173 } else if (Style.Language == FormatStyle::LK_Proto) {
2174 if (Right.is(tok::period) &&
2175 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2176 Keywords.kw_repeated, Keywords.kw_extend))
2178 if (Right.is(tok::l_paren) &&
2179 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2181 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2182 if (Left.is(TT_JsFatArrow))
2184 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2185 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2187 if (Left.is(tok::identifier) && Right.is(TT_TemplateString))
2189 if (Right.is(tok::star) &&
2190 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2192 if (Right.isOneOf(tok::l_brace, tok::l_square) &&
2193 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2195 // JS methods can use some keywords as names (e.g. `delete()`).
2196 if (Right.is(tok::l_paren) && Line.MustBeDeclaration &&
2197 Left.Tok.getIdentifierInfo())
2199 if (Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
2200 Keywords.kw_of, tok::kw_const) &&
2201 (!Left.Previous || !Left.Previous->is(tok::period)))
2203 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
2204 Left.Previous->is(tok::period) && Right.is(tok::l_paren))
2206 if (Left.is(Keywords.kw_as) &&
2207 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
2209 if (Left.is(tok::kw_default) && Left.Previous &&
2210 Left.Previous->is(tok::kw_export))
2212 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2214 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2216 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
2218 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2219 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2221 if (Left.is(tok::ellipsis))
2223 if (Left.is(TT_TemplateCloser) &&
2224 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2225 Keywords.kw_implements, Keywords.kw_extends))
2226 // Type assertions ('<type>expr') are not followed by whitespace. Other
2227 // locations that should have whitespace following are identified by the
2228 // above set of follower tokens.
2230 // Postfix non-null assertion operator, as in `foo!.bar()`.
2231 if (Right.is(tok::exclaim) && (Left.isOneOf(tok::identifier, tok::r_paren,
2232 tok::r_square, tok::r_brace) ||
2233 Left.Tok.isLiteral()))
2235 if (Left.is(tok::exclaim) && Right.is(Keywords.kw_as))
2236 return true; // "x! as string"
2237 } else if (Style.Language == FormatStyle::LK_Java) {
2238 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2240 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2241 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2242 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2243 tok::kw_protected) ||
2244 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2245 Keywords.kw_native)) &&
2246 Right.is(TT_TemplateOpener))
2249 if (Left.is(TT_ImplicitStringLiteral))
2250 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2251 if (Line.Type == LT_ObjCMethodDecl) {
2252 if (Left.is(TT_ObjCMethodSpecifier))
2254 if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2255 // Don't space between ')' and <id>
2258 if (Line.Type == LT_ObjCProperty &&
2259 (Right.is(tok::equal) || Left.is(tok::equal)))
2262 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2263 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2265 if (Right.is(TT_OverloadedOperatorLParen))
2266 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2267 if (Left.is(tok::comma))
2269 if (Right.is(tok::comma))
2271 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2273 if (Right.is(tok::colon)) {
2274 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2275 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2277 if (Right.is(TT_ObjCMethodExpr))
2279 if (Left.is(tok::question))
2281 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2283 if (Right.is(TT_DictLiteral))
2284 return Style.SpacesInContainerLiterals;
2287 if (Left.is(TT_UnaryOperator))
2288 return Right.is(TT_BinaryOperator);
2290 // If the next token is a binary operator or a selector name, we have
2291 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2292 if (Left.is(TT_CastRParen))
2293 return Style.SpaceAfterCStyleCast ||
2294 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2296 if (Left.is(tok::greater) && Right.is(tok::greater))
2297 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2298 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2299 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2300 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2302 if (!Style.SpaceBeforeAssignmentOperators &&
2303 Right.getPrecedence() == prec::Assignment)
2305 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment))
2306 return (Left.is(TT_TemplateOpener) &&
2307 Style.Standard == FormatStyle::LS_Cpp03) ||
2308 !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren,
2310 Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener));
2311 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2312 return Style.SpacesInAngles;
2313 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2314 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2315 !Right.is(tok::r_paren)))
2317 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2318 Right.isNot(TT_FunctionTypeLParen))
2319 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2320 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2321 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2323 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2324 Line.startsWith(tok::hash))
2326 if (Right.is(TT_TrailingUnaryOperator))
2328 if (Left.is(TT_RegexLiteral))
2330 return spaceRequiredBetween(Line, Left, Right);
2333 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2334 static bool isAllmanBrace(const FormatToken &Tok) {
2335 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2336 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2339 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2340 const FormatToken &Right) {
2341 const FormatToken &Left = *Right.Previous;
2342 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2345 if (Style.Language == FormatStyle::LK_JavaScript) {
2346 // FIXME: This might apply to other languages and token kinds.
2347 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
2348 Left.Previous->is(tok::string_literal))
2350 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2351 Left.Previous && Left.Previous->is(tok::equal) &&
2352 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2354 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2356 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2357 // Object literals on the top level of a file are treated as "enum-style".
2358 // Each key/value pair is put on a separate line, instead of bin-packing.
2360 if (Left.is(tok::l_brace) && Line.Level == 0 &&
2361 (Line.startsWith(tok::kw_enum) ||
2362 Line.startsWith(tok::kw_export, tok::kw_enum)))
2363 // JavaScript top-level enum key/value pairs are put on separate lines
2364 // instead of bin-packing.
2366 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2367 !Left.Children.empty())
2368 // Support AllowShortFunctionsOnASingleLine for JavaScript.
2369 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2370 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2371 (Left.NestingLevel == 0 && Line.Level == 0 &&
2372 Style.AllowShortFunctionsOnASingleLine ==
2373 FormatStyle::SFS_Inline);
2374 } else if (Style.Language == FormatStyle::LK_Java) {
2375 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2376 Right.Next->is(tok::string_literal))
2380 // If the last token before a '}' is a comma or a trailing comment, the
2381 // intention is to insert a line break after it in order to make shuffling
2382 // around entries easier.
2383 const FormatToken *BeforeClosingBrace = nullptr;
2384 if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
2385 Left.BlockKind != BK_Block && Left.MatchingParen)
2386 BeforeClosingBrace = Left.MatchingParen->Previous;
2387 else if (Right.MatchingParen &&
2388 Right.MatchingParen->isOneOf(tok::l_brace,
2389 TT_ArrayInitializerLSquare))
2390 BeforeClosingBrace = &Left;
2391 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2392 BeforeClosingBrace->isTrailingComment()))
2395 if (Right.is(tok::comment))
2396 return Left.BlockKind != BK_BracedInit &&
2397 Left.isNot(TT_CtorInitializerColon) &&
2398 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2399 if (Left.isTrailingComment())
2401 if (Left.isStringLiteral() &&
2402 (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral)))
2404 if (Right.Previous->IsUnterminatedLiteral)
2406 if (Right.is(tok::lessless) && Right.Next &&
2407 Right.Previous->is(tok::string_literal) &&
2408 Right.Next->is(tok::string_literal))
2410 if (Right.Previous->ClosesTemplateDeclaration &&
2411 Right.Previous->MatchingParen &&
2412 Right.Previous->MatchingParen->NestingLevel == 0 &&
2413 Style.AlwaysBreakTemplateDeclarations)
2415 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2416 Style.BreakConstructorInitializersBeforeComma &&
2417 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2419 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2420 // Raw string literals are special wrt. line breaks. The author has made a
2421 // deliberate choice and might have aligned the contents of the string
2422 // literal accordingly. Thus, we try keep existing line breaks.
2423 return Right.NewlinesBefore > 0;
2424 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2425 Style.Language == FormatStyle::LK_Proto)
2426 // Don't put enums onto single lines in protocol buffers.
2428 if (Right.is(TT_InlineASMBrace))
2429 return Right.HasUnescapedNewline;
2430 if (isAllmanBrace(Left) || isAllmanBrace(Right))
2431 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2432 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2433 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2434 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2437 if ((Style.Language == FormatStyle::LK_Java ||
2438 Style.Language == FormatStyle::LK_JavaScript) &&
2439 Left.is(TT_LeadingJavaAnnotation) &&
2440 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2441 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2447 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2448 const FormatToken &Right) {
2449 const FormatToken &Left = *Right.Previous;
2451 // Language-specific stuff.
2452 if (Style.Language == FormatStyle::LK_Java) {
2453 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2454 Keywords.kw_implements))
2456 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2457 Keywords.kw_implements))
2459 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2460 const FormatToken *NonComment = Right.getPreviousNonComment();
2461 if (Left.isOneOf(tok::kw_return, tok::kw_continue, tok::kw_break,
2464 NonComment->isOneOf(tok::kw_return, tok::kw_continue, tok::kw_break,
2466 return false; // Otherwise a semicolon is inserted.
2467 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2469 if (Left.is(TT_JsTypeColon))
2471 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2473 if (Left.is(Keywords.kw_in))
2474 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
2475 if (Right.is(Keywords.kw_in))
2476 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
2477 if (Right.is(Keywords.kw_as))
2478 return false; // must not break before as in 'x as type' casts
2479 if (Left.is(Keywords.kw_declare) &&
2480 Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
2481 Keywords.kw_function, tok::kw_class, tok::kw_enum,
2482 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
2483 Keywords.kw_let, tok::kw_const))
2484 // See grammar for 'declare' statements at:
2485 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
2487 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
2488 Right.isOneOf(tok::identifier, tok::string_literal)) {
2489 return false; // must not break in "module foo { ...}"
2493 if (Left.is(tok::at))
2495 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2497 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2498 return !Right.is(tok::l_paren);
2499 if (Right.is(TT_PointerOrReference))
2500 return Line.IsMultiVariableDeclStmt ||
2501 (Style.PointerAlignment == FormatStyle::PAS_Right &&
2502 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2503 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2504 Right.is(tok::kw_operator))
2506 if (Left.is(TT_PointerOrReference))
2508 if (Right.isTrailingComment())
2509 // We rely on MustBreakBefore being set correctly here as we should not
2510 // change the "binding" behavior of a comment.
2511 // The first comment in a braced lists is always interpreted as belonging to
2512 // the first list element. Otherwise, it should be placed outside of the
2514 return Left.BlockKind == BK_BracedInit;
2515 if (Left.is(tok::question) && Right.is(tok::colon))
2517 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2518 return Style.BreakBeforeTernaryOperators;
2519 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2520 return !Style.BreakBeforeTernaryOperators;
2521 if (Right.is(TT_InheritanceColon))
2523 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
2524 Left.isNot(TT_SelectorName))
2526 if (Right.is(tok::colon) &&
2527 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2529 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))
2531 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2532 Right.Next->is(TT_ObjCMethodExpr)))
2533 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2534 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2536 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2538 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2539 TT_OverloadedOperator))
2541 if (Left.is(TT_RangeBasedForLoopColon))
2543 if (Right.is(TT_RangeBasedForLoopColon))
2545 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
2547 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2548 Left.is(tok::kw_operator))
2550 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2551 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2553 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2555 if (Left.is(tok::l_paren) && Left.Previous &&
2556 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2558 if (Right.is(TT_ImplicitStringLiteral))
2561 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2563 if (Right.is(tok::r_square) && Right.MatchingParen &&
2564 Right.MatchingParen->is(TT_LambdaLSquare))
2567 // We only break before r_brace if there was a corresponding break before
2568 // the l_brace, which is tracked by BreakBeforeClosingBrace.
2569 if (Right.is(tok::r_brace))
2570 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2572 // Allow breaking after a trailing annotation, e.g. after a method
2574 if (Left.is(TT_TrailingAnnotation))
2575 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2576 tok::less, tok::coloncolon);
2578 if (Right.is(tok::kw___attribute))
2581 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2584 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2587 if (Left.is(TT_CtorInitializerComma) &&
2588 Style.BreakConstructorInitializersBeforeComma)
2590 if (Right.is(TT_CtorInitializerComma) &&
2591 Style.BreakConstructorInitializersBeforeComma)
2593 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2594 (Left.is(tok::less) && Right.is(tok::less)))
2596 if (Right.is(TT_BinaryOperator) &&
2597 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2598 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2599 Right.getPrecedence() != prec::Assignment))
2601 if (Left.is(TT_ArrayInitializerLSquare))
2603 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2605 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2606 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2607 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2608 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2609 Left.getPrecedence() == prec::Assignment))
2611 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2612 tok::kw_class, tok::kw_struct, tok::comment) ||
2613 Right.isMemberAccess() ||
2614 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2615 tok::colon, tok::l_square, tok::at) ||
2616 (Left.is(tok::r_paren) &&
2617 Right.isOneOf(tok::identifier, tok::kw_const)) ||
2618 (Left.is(tok::l_paren) && !Right.is(tok::r_paren));
2621 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2622 llvm::errs() << "AnnotatedTokens:\n";
2623 const FormatToken *Tok = Line.First;
2625 llvm::errs() << " M=" << Tok->MustBreakBefore
2626 << " C=" << Tok->CanBreakBefore
2627 << " T=" << getTokenTypeName(Tok->Type)
2628 << " S=" << Tok->SpacesRequiredBefore
2629 << " B=" << Tok->BlockParameterCount
2630 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2631 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2633 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2634 llvm::errs() << Tok->FakeLParens[i] << "/";
2635 llvm::errs() << " FakeRParens=" << Tok->FakeRParens;
2636 llvm::errs() << " Text='" << Tok->TokenText << "'\n";
2638 assert(Tok == Line.Last);
2641 llvm::errs() << "----\n";
2644 } // namespace format
2645 } // namespace clang