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();
1286 NextToken->isOneOf(tok::arrow, Keywords.kw_final, tok::equal,
1287 Keywords.kw_override) ||
1288 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1289 return TT_PointerOrReference;
1291 if (PrevToken->is(tok::coloncolon))
1292 return TT_PointerOrReference;
1294 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1295 tok::comma, tok::semi, tok::kw_return, tok::colon,
1296 tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1297 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1298 TT_UnaryOperator, TT_CastRParen))
1299 return TT_UnaryOperator;
1301 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1302 return TT_PointerOrReference;
1303 if (NextToken->is(tok::kw_operator) && !IsExpression)
1304 return TT_PointerOrReference;
1305 if (NextToken->isOneOf(tok::comma, tok::semi))
1306 return TT_PointerOrReference;
1308 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1309 PrevToken->MatchingParen->Previous &&
1310 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1312 return TT_PointerOrReference;
1314 if (PrevToken->Tok.isLiteral() ||
1315 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1316 tok::kw_false, tok::r_brace) ||
1317 NextToken->Tok.isLiteral() ||
1318 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1319 NextToken->isUnaryOperator() ||
1320 // If we know we're in a template argument, there are no named
1321 // declarations. Thus, having an identifier on the right-hand side
1322 // indicates a binary operator.
1323 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1324 return TT_BinaryOperator;
1326 // "&&(" is quite unlikely to be two successive unary "&".
1327 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1328 return TT_BinaryOperator;
1330 // This catches some cases where evaluation order is used as control flow:
1332 const FormatToken *NextNextToken = NextToken->getNextNonComment();
1333 if (NextNextToken && NextNextToken->is(tok::arrow))
1334 return TT_BinaryOperator;
1336 // It is very unlikely that we are going to find a pointer or reference type
1337 // definition on the RHS of an assignment.
1338 if (IsExpression && !Contexts.back().CaretFound)
1339 return TT_BinaryOperator;
1341 return TT_PointerOrReference;
1344 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1345 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1347 return TT_UnaryOperator;
1349 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator) &&
1350 !PrevToken->is(tok::exclaim))
1351 // There aren't any trailing unary operators except for TypeScript's
1352 // non-null operator (!). Thus, this must be squence of leading operators.
1353 return TT_UnaryOperator;
1355 // Use heuristics to recognize unary operators.
1356 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1357 tok::question, tok::colon, tok::kw_return,
1358 tok::kw_case, tok::at, tok::l_brace))
1359 return TT_UnaryOperator;
1361 // There can't be two consecutive binary operators.
1362 if (PrevToken->is(TT_BinaryOperator))
1363 return TT_UnaryOperator;
1365 // Fall back to marking the token as binary operator.
1366 return TT_BinaryOperator;
1369 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1370 TokenType determineIncrementUsage(const FormatToken &Tok) {
1371 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1372 if (!PrevToken || PrevToken->is(TT_CastRParen))
1373 return TT_UnaryOperator;
1374 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1375 return TT_TrailingUnaryOperator;
1377 return TT_UnaryOperator;
1380 SmallVector<Context, 8> Contexts;
1382 const FormatStyle &Style;
1383 AnnotatedLine &Line;
1384 FormatToken *CurrentToken;
1386 const AdditionalKeywords &Keywords;
1388 // Set of "<" tokens that do not open a template parameter list. If parseAngle
1389 // determines that a specific token can't be a template opener, it will make
1390 // same decision irrespective of the decisions for tokens leading up to it.
1391 // Store this information to prevent this from causing exponential runtime.
1392 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1395 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1396 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1398 /// \brief Parses binary expressions by inserting fake parenthesis based on
1399 /// operator precedence.
1400 class ExpressionParser {
1402 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1403 AnnotatedLine &Line)
1404 : Style(Style), Keywords(Keywords), Current(Line.First) {}
1406 /// \brief Parse expressions with the given operatore precedence.
1407 void parse(int Precedence = 0) {
1408 // Skip 'return' and ObjC selector colons as they are not part of a binary
1410 while (Current && (Current->is(tok::kw_return) ||
1411 (Current->is(tok::colon) &&
1412 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1415 if (!Current || Precedence > PrecedenceArrowAndPeriod)
1418 // Conditional expressions need to be parsed separately for proper nesting.
1419 if (Precedence == prec::Conditional) {
1420 parseConditionalExpr();
1424 // Parse unary operators, which all have a higher precedence than binary
1426 if (Precedence == PrecedenceUnaryOperator) {
1427 parseUnaryOperator();
1431 FormatToken *Start = Current;
1432 FormatToken *LatestOperator = nullptr;
1433 unsigned OperatorIndex = 0;
1436 // Consume operators with higher precedence.
1437 parse(Precedence + 1);
1439 int CurrentPrecedence = getCurrentPrecedence();
1441 if (Current && Current->is(TT_SelectorName) &&
1442 Precedence == CurrentPrecedence) {
1444 addFakeParenthesis(Start, prec::Level(Precedence));
1448 // At the end of the line or when an operator with higher precedence is
1449 // found, insert fake parenthesis and return.
1450 if (!Current || (Current->closesScope() && Current->MatchingParen) ||
1451 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1452 (CurrentPrecedence == prec::Conditional &&
1453 Precedence == prec::Assignment && Current->is(tok::colon))) {
1457 // Consume scopes: (), [], <> and {}
1458 if (Current->opensScope()) {
1459 while (Current && !Current->closesScope()) {
1466 if (CurrentPrecedence == Precedence) {
1468 LatestOperator->NextOperator = Current;
1469 LatestOperator = Current;
1470 Current->OperatorIndex = OperatorIndex;
1473 next(/*SkipPastLeadingComments=*/Precedence > 0);
1477 if (LatestOperator && (Current || Precedence > 0)) {
1478 // LatestOperator->LastOperator = true;
1479 if (Precedence == PrecedenceArrowAndPeriod) {
1480 // Call expressions don't have a binary operator precedence.
1481 addFakeParenthesis(Start, prec::Unknown);
1483 addFakeParenthesis(Start, prec::Level(Precedence));
1489 /// \brief Gets the precedence (+1) of the given token for binary operators
1490 /// and other tokens that we treat like binary operators.
1491 int getCurrentPrecedence() {
1493 const FormatToken *NextNonComment = Current->getNextNonComment();
1494 if (Current->is(TT_ConditionalExpr))
1495 return prec::Conditional;
1496 if (NextNonComment && NextNonComment->is(tok::colon) &&
1497 NextNonComment->is(TT_DictLiteral))
1499 if (Current->is(TT_LambdaArrow))
1501 if (Current->is(TT_JsFatArrow))
1502 return prec::Assignment;
1503 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName,
1504 TT_JsComputedPropertyName) ||
1505 (Current->is(tok::comment) && NextNonComment &&
1506 NextNonComment->is(TT_SelectorName)))
1508 if (Current->is(TT_RangeBasedForLoopColon))
1510 if ((Style.Language == FormatStyle::LK_Java ||
1511 Style.Language == FormatStyle::LK_JavaScript) &&
1512 Current->is(Keywords.kw_instanceof))
1513 return prec::Relational;
1514 if (Style.Language == FormatStyle::LK_JavaScript &&
1515 Current->is(Keywords.kw_in))
1516 return prec::Relational;
1517 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1518 return Current->getPrecedence();
1519 if (Current->isOneOf(tok::period, tok::arrow))
1520 return PrecedenceArrowAndPeriod;
1521 if ((Style.Language == FormatStyle::LK_Java ||
1522 Style.Language == FormatStyle::LK_JavaScript) &&
1523 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1524 Keywords.kw_throws))
1530 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1531 Start->FakeLParens.push_back(Precedence);
1532 if (Precedence > prec::Unknown)
1533 Start->StartsBinaryExpression = true;
1535 FormatToken *Previous = Current->Previous;
1536 while (Previous->is(tok::comment) && Previous->Previous)
1537 Previous = Previous->Previous;
1538 ++Previous->FakeRParens;
1539 if (Precedence > prec::Unknown)
1540 Previous->EndsBinaryExpression = true;
1544 /// \brief Parse unary operator expressions and surround them with fake
1545 /// parentheses if appropriate.
1546 void parseUnaryOperator() {
1547 if (!Current || Current->isNot(TT_UnaryOperator)) {
1548 parse(PrecedenceArrowAndPeriod);
1552 FormatToken *Start = Current;
1554 parseUnaryOperator();
1556 // The actual precedence doesn't matter.
1557 addFakeParenthesis(Start, prec::Unknown);
1560 void parseConditionalExpr() {
1561 while (Current && Current->isTrailingComment()) {
1564 FormatToken *Start = Current;
1565 parse(prec::LogicalOr);
1566 if (!Current || !Current->is(tok::question))
1569 parse(prec::Assignment);
1570 if (!Current || Current->isNot(TT_ConditionalExpr))
1573 parse(prec::Assignment);
1574 addFakeParenthesis(Start, prec::Conditional);
1577 void next(bool SkipPastLeadingComments = true) {
1579 Current = Current->Next;
1581 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1582 Current->isTrailingComment())
1583 Current = Current->Next;
1586 const FormatStyle &Style;
1587 const AdditionalKeywords &Keywords;
1588 FormatToken *Current;
1591 } // end anonymous namespace
1593 void TokenAnnotator::setCommentLineLevels(
1594 SmallVectorImpl<AnnotatedLine *> &Lines) {
1595 const AnnotatedLine *NextNonCommentLine = nullptr;
1596 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1599 if (NextNonCommentLine && (*I)->First->is(tok::comment) &&
1600 (*I)->First->Next == nullptr)
1601 (*I)->Level = NextNonCommentLine->Level;
1603 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1605 setCommentLineLevels((*I)->Children);
1609 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
1610 unsigned Result = 0;
1611 for (const auto* Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
1612 Result = std::max(Result, Tok->NestingLevel);
1616 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1617 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1618 E = Line.Children.end();
1622 AnnotatingParser Parser(Style, Line, Keywords);
1623 Line.Type = Parser.parseLine();
1625 // With very deep nesting, ExpressionParser uses lots of stack and the
1626 // formatting algorithm is very slow. We're not going to do a good job here
1627 // anyway - it's probably generated code being formatted by mistake.
1628 // Just skip the whole line.
1629 if (maxNestingDepth(Line) > 50)
1630 Line.Type = LT_Invalid;
1632 if (Line.Type == LT_Invalid)
1635 ExpressionParser ExprParser(Style, Keywords, Line);
1638 if (Line.startsWith(TT_ObjCMethodSpecifier))
1639 Line.Type = LT_ObjCMethodDecl;
1640 else if (Line.startsWith(TT_ObjCDecl))
1641 Line.Type = LT_ObjCDecl;
1642 else if (Line.startsWith(TT_ObjCProperty))
1643 Line.Type = LT_ObjCProperty;
1645 Line.First->SpacesRequiredBefore = 1;
1646 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1649 // This function heuristically determines whether 'Current' starts the name of a
1650 // function declaration.
1651 static bool isFunctionDeclarationName(const FormatToken &Current,
1652 const AnnotatedLine &Line) {
1653 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1654 for (; Next; Next = Next->Next) {
1655 if (Next->is(TT_OverloadedOperatorLParen))
1657 if (Next->is(TT_OverloadedOperator))
1659 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1660 // For 'new[]' and 'delete[]'.
1661 if (Next->Next && Next->Next->is(tok::l_square) &&
1662 Next->Next->Next && Next->Next->Next->is(tok::r_square))
1663 Next = Next->Next->Next;
1672 // Find parentheses of parameter list.
1673 const FormatToken *Next = Current.Next;
1674 if (Current.is(tok::kw_operator)) {
1675 if (Current.Previous && Current.Previous->is(tok::coloncolon))
1677 Next = skipOperatorName(Next);
1679 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1681 for (; Next; Next = Next->Next) {
1682 if (Next->is(TT_TemplateOpener)) {
1683 Next = Next->MatchingParen;
1684 } else if (Next->is(tok::coloncolon)) {
1688 if (Next->is(tok::kw_operator)) {
1689 Next = skipOperatorName(Next->Next);
1692 if (!Next->is(tok::identifier))
1694 } else if (Next->is(tok::l_paren)) {
1702 // Check whether parameter list can be long to a function declaration.
1703 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
1705 // If the lines ends with "{", this is likely an function definition.
1706 if (Line.Last->is(tok::l_brace))
1708 if (Next->Next == Next->MatchingParen)
1709 return true; // Empty parentheses.
1710 // If there is an &/&& after the r_paren, this is likely a function.
1711 if (Next->MatchingParen->Next &&
1712 Next->MatchingParen->Next->is(TT_PointerOrReference))
1714 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1716 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1717 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
1719 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1720 Tok->Tok.isLiteral())
1726 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1727 assert(Line.MightBeFunctionDecl);
1729 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1730 Style.AlwaysBreakAfterReturnType ==
1731 FormatStyle::RTBS_TopLevelDefinitions) &&
1735 switch (Style.AlwaysBreakAfterReturnType) {
1736 case FormatStyle::RTBS_None:
1738 case FormatStyle::RTBS_All:
1739 case FormatStyle::RTBS_TopLevel:
1741 case FormatStyle::RTBS_AllDefinitions:
1742 case FormatStyle::RTBS_TopLevelDefinitions:
1743 return Line.mightBeFunctionDefinition();
1749 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1750 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1751 E = Line.Children.end();
1753 calculateFormattingInformation(**I);
1756 Line.First->TotalLength =
1757 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1758 if (!Line.First->Next)
1760 FormatToken *Current = Line.First->Next;
1761 bool InFunctionDecl = Line.MightBeFunctionDecl;
1763 if (isFunctionDeclarationName(*Current, Line))
1764 Current->Type = TT_FunctionDeclarationName;
1765 if (Current->is(TT_LineComment)) {
1766 if (Current->Previous->BlockKind == BK_BracedInit &&
1767 Current->Previous->opensScope())
1768 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1770 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1772 // If we find a trailing comment, iterate backwards to determine whether
1773 // it seems to relate to a specific parameter. If so, break before that
1774 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1775 // to the previous line in:
1779 if (!Current->HasUnescapedNewline) {
1780 for (FormatToken *Parameter = Current->Previous; Parameter;
1781 Parameter = Parameter->Previous) {
1782 if (Parameter->isOneOf(tok::comment, tok::r_brace))
1784 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1785 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1786 Parameter->HasUnescapedNewline)
1787 Parameter->MustBreakBefore = true;
1792 } else if (Current->SpacesRequiredBefore == 0 &&
1793 spaceRequiredBefore(Line, *Current)) {
1794 Current->SpacesRequiredBefore = 1;
1797 Current->MustBreakBefore =
1798 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1800 if (!Current->MustBreakBefore && InFunctionDecl &&
1801 Current->is(TT_FunctionDeclarationName))
1802 Current->MustBreakBefore = mustBreakForReturnType(Line);
1804 Current->CanBreakBefore =
1805 Current->MustBreakBefore || canBreakBefore(Line, *Current);
1806 unsigned ChildSize = 0;
1807 if (Current->Previous->Children.size() == 1) {
1808 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1809 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1810 : LastOfChild.TotalLength + 1;
1812 const FormatToken *Prev = Current->Previous;
1813 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1814 (Prev->Children.size() == 1 &&
1815 Prev->Children[0]->First->MustBreakBefore) ||
1816 Current->IsMultiline)
1817 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1819 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1820 ChildSize + Current->SpacesRequiredBefore;
1822 if (Current->is(TT_CtorInitializerColon))
1823 InFunctionDecl = false;
1825 // FIXME: Only calculate this if CanBreakBefore is true once static
1826 // initializers etc. are sorted out.
1827 // FIXME: Move magic numbers to a better place.
1828 Current->SplitPenalty = 20 * Current->BindingStrength +
1829 splitPenalty(Line, *Current, InFunctionDecl);
1831 Current = Current->Next;
1834 calculateUnbreakableTailLengths(Line);
1835 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1837 Current->Role->precomputeFormattingInfos(Current);
1840 DEBUG({ printDebugInfo(Line); });
1843 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1844 unsigned UnbreakableTailLength = 0;
1845 FormatToken *Current = Line.Last;
1847 Current->UnbreakableTailLength = UnbreakableTailLength;
1848 if (Current->CanBreakBefore ||
1849 Current->isOneOf(tok::comment, tok::string_literal)) {
1850 UnbreakableTailLength = 0;
1852 UnbreakableTailLength +=
1853 Current->ColumnWidth + Current->SpacesRequiredBefore;
1855 Current = Current->Previous;
1859 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1860 const FormatToken &Tok,
1861 bool InFunctionDecl) {
1862 const FormatToken &Left = *Tok.Previous;
1863 const FormatToken &Right = Tok;
1865 if (Left.is(tok::semi))
1868 if (Style.Language == FormatStyle::LK_Java) {
1869 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1871 if (Right.is(Keywords.kw_implements))
1873 if (Left.is(tok::comma) && Left.NestingLevel == 0)
1875 } else if (Style.Language == FormatStyle::LK_JavaScript) {
1876 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1878 if (Left.is(TT_JsTypeColon))
1880 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
1881 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
1885 if (Left.is(tok::comma) || (Right.is(tok::identifier) && Right.Next &&
1886 Right.Next->is(TT_DictLiteral)))
1888 if (Right.is(tok::l_square)) {
1889 if (Style.Language == FormatStyle::LK_Proto)
1891 if (Left.is(tok::r_square))
1893 // Slightly prefer formatting local lambda definitions like functions.
1894 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1896 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1897 TT_ArrayInitializerLSquare))
1901 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1902 Right.is(tok::kw_operator)) {
1903 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1905 if (Left.is(TT_StartOfName))
1907 if (InFunctionDecl && Right.NestingLevel == 0)
1908 return Style.PenaltyReturnTypeOnItsOwnLine;
1911 if (Right.is(TT_PointerOrReference))
1913 if (Right.is(TT_LambdaArrow))
1915 if (Left.is(tok::equal) && Right.is(tok::l_brace))
1917 if (Left.is(TT_CastRParen))
1919 if (Left.is(tok::coloncolon) ||
1920 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1922 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1924 if (Left.is(tok::comment))
1927 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
1930 if (Right.isMemberAccess()) {
1931 // Breaking before the "./->" of a chained call/member access is reasonably
1932 // cheap, as formatting those with one call per line is generally
1933 // desirable. In particular, it should be cheaper to break before the call
1934 // than it is to break inside a call's parameters, which could lead to weird
1935 // "hanging" indents. The exception is the very last "./->" to support this
1936 // frequent pattern:
1938 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
1941 // which might otherwise be blown up onto many lines. Here, clang-format
1942 // won't produce "hanging" indents anyway as there is no other trailing
1945 // Also apply higher penalty is not a call as that might lead to a wrapping
1949 // .aaaaaaaaa.bbbbbbbb(cccccccc);
1950 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
1955 if (Right.is(TT_TrailingAnnotation) &&
1956 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
1957 // Moving trailing annotations to the next line is fine for ObjC method
1959 if (Line.startsWith(TT_ObjCMethodSpecifier))
1961 // Generally, breaking before a trailing annotation is bad unless it is
1962 // function-like. It seems to be especially preferable to keep standard
1963 // annotations (i.e. "const", "final" and "override") on the same line.
1964 // Use a slightly higher penalty after ")" so that annotations like
1965 // "const override" are kept together.
1966 bool is_short_annotation = Right.TokenText.size() < 10;
1967 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
1970 // In for-loops, prefer breaking at ',' and ';'.
1971 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
1974 // In Objective-C method expressions, prefer breaking before "param:" over
1975 // breaking after it.
1976 if (Right.is(TT_SelectorName))
1978 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
1979 return Line.MightBeFunctionDecl ? 50 : 500;
1981 if (Left.is(tok::l_paren) && InFunctionDecl &&
1982 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1984 if (Left.is(tok::l_paren) && Left.Previous &&
1985 Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
1987 if (Left.is(tok::equal) && InFunctionDecl)
1989 if (Right.is(tok::r_brace))
1991 if (Left.is(TT_TemplateOpener))
1993 if (Left.opensScope()) {
1994 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
1996 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
1999 if (Left.is(TT_JavaAnnotation))
2002 if (Right.is(tok::lessless)) {
2003 if (Left.is(tok::string_literal) &&
2004 (Right.NextOperator || Right.OperatorIndex != 1)) {
2005 StringRef Content = Left.TokenText;
2006 if (Content.startswith("\""))
2007 Content = Content.drop_front(1);
2008 if (Content.endswith("\""))
2009 Content = Content.drop_back(1);
2010 Content = Content.trim();
2011 if (Content.size() > 1 &&
2012 (Content.back() == ':' || Content.back() == '='))
2015 return 1; // Breaking at a << is really cheap.
2017 if (Left.is(TT_ConditionalExpr))
2018 return prec::Conditional;
2019 prec::Level Level = Left.getPrecedence();
2020 if (Level != prec::Unknown)
2022 Level = Right.getPrecedence();
2023 if (Level != prec::Unknown)
2029 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2030 const FormatToken &Left,
2031 const FormatToken &Right) {
2032 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2034 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
2035 Left.Tok.getObjCKeywordID() == tok::objc_property)
2037 if (Right.is(tok::hashhash))
2038 return Left.is(tok::hash);
2039 if (Left.isOneOf(tok::hashhash, tok::hash))
2040 return Right.is(tok::hash);
2041 if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
2042 return Style.SpaceInEmptyParentheses;
2043 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2044 return (Right.is(TT_CastRParen) ||
2045 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2046 ? Style.SpacesInCStyleCastParentheses
2047 : Style.SpacesInParentheses;
2048 if (Right.isOneOf(tok::semi, tok::comma))
2050 if (Right.is(tok::less) &&
2051 Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)
2053 if (Right.is(tok::less) && Left.is(tok::kw_template))
2054 return Style.SpaceAfterTemplateKeyword;
2055 if (Left.isOneOf(tok::exclaim, tok::tilde))
2057 if (Left.is(tok::at) &&
2058 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2059 tok::numeric_constant, tok::l_paren, tok::l_brace,
2060 tok::kw_true, tok::kw_false))
2062 if (Left.is(tok::colon))
2063 return !Left.is(TT_ObjCMethodExpr);
2064 if (Left.is(tok::coloncolon))
2066 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
2068 if (Right.is(tok::ellipsis))
2069 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2070 Left.Previous->is(tok::kw_case));
2071 if (Left.is(tok::l_square) && Right.is(tok::amp))
2073 if (Right.is(TT_PointerOrReference))
2074 return (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) ||
2075 (Left.Tok.isLiteral() || (Left.is(tok::kw_const) && Left.Previous &&
2076 Left.Previous->is(tok::r_paren)) ||
2077 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
2078 (Style.PointerAlignment != FormatStyle::PAS_Left ||
2079 (Line.IsMultiVariableDeclStmt &&
2080 (Left.NestingLevel == 0 ||
2081 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
2082 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
2083 (!Left.is(TT_PointerOrReference) ||
2084 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2085 !Line.IsMultiVariableDeclStmt)))
2087 if (Left.is(TT_PointerOrReference))
2088 return Right.Tok.isLiteral() ||
2089 Right.isOneOf(TT_BlockComment, Keywords.kw_final,
2090 Keywords.kw_override) ||
2091 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
2092 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
2094 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2095 !Line.IsMultiVariableDeclStmt) &&
2097 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
2098 if (Right.is(tok::star) && Left.is(tok::l_paren))
2100 if (Left.is(tok::l_square))
2101 return (Left.is(TT_ArrayInitializerLSquare) &&
2102 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
2103 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
2104 Right.isNot(tok::r_square));
2105 if (Right.is(tok::r_square))
2106 return Right.MatchingParen &&
2107 ((Style.SpacesInContainerLiterals &&
2108 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
2109 (Style.SpacesInSquareBrackets &&
2110 Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
2111 if (Right.is(tok::l_square) &&
2112 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
2113 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
2115 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
2116 return !Left.Children.empty(); // No spaces in "{}".
2117 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
2118 (Right.is(tok::r_brace) && Right.MatchingParen &&
2119 Right.MatchingParen->BlockKind != BK_Block))
2120 return !Style.Cpp11BracedListStyle;
2121 if (Left.is(TT_BlockComment))
2122 return !Left.TokenText.endswith("=*/");
2123 if (Right.is(tok::l_paren)) {
2124 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
2126 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
2127 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
2128 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
2129 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
2131 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
2132 tok::kw_new, tok::kw_delete) &&
2133 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
2134 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
2135 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
2136 Left.is(tok::r_paren)) &&
2137 Line.Type != LT_PreprocessorDirective);
2139 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
2141 if (Right.is(TT_UnaryOperator))
2142 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2143 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2144 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2146 Left.isSimpleTypeSpecifier()) &&
2147 Right.is(tok::l_brace) && Right.getNextNonComment() &&
2148 Right.BlockKind != BK_Block)
2150 if (Left.is(tok::period) || Right.is(tok::period))
2152 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2154 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2155 Left.MatchingParen->Previous &&
2156 Left.MatchingParen->Previous->is(tok::period))
2157 // A.<B>DoSomething();
2159 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2164 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2165 const FormatToken &Right) {
2166 const FormatToken &Left = *Right.Previous;
2167 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2168 return true; // Never ever merge two identifiers.
2169 if (Style.Language == FormatStyle::LK_Cpp) {
2170 if (Left.is(tok::kw_operator))
2171 return Right.is(tok::coloncolon);
2172 } else if (Style.Language == FormatStyle::LK_Proto) {
2173 if (Right.is(tok::period) &&
2174 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2175 Keywords.kw_repeated, Keywords.kw_extend))
2177 if (Right.is(tok::l_paren) &&
2178 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2180 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2181 if (Left.is(TT_JsFatArrow))
2183 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2184 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2186 if (Left.is(tok::identifier) && Right.is(TT_TemplateString))
2188 if (Right.is(tok::star) &&
2189 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2191 if (Right.isOneOf(tok::l_brace, tok::l_square) &&
2192 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2194 // JS methods can use some keywords as names (e.g. `delete()`).
2195 if (Right.is(tok::l_paren) && Line.MustBeDeclaration &&
2196 Left.Tok.getIdentifierInfo())
2198 if (Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
2199 Keywords.kw_of, tok::kw_const) &&
2200 (!Left.Previous || !Left.Previous->is(tok::period)))
2202 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
2203 Left.Previous->is(tok::period) && Right.is(tok::l_paren))
2205 if (Left.is(Keywords.kw_as) &&
2206 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
2208 if (Left.is(tok::kw_default) && Left.Previous &&
2209 Left.Previous->is(tok::kw_export))
2211 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2213 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2215 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
2217 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2218 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2220 if (Left.is(tok::ellipsis))
2222 if (Left.is(TT_TemplateCloser) &&
2223 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2224 Keywords.kw_implements, Keywords.kw_extends))
2225 // Type assertions ('<type>expr') are not followed by whitespace. Other
2226 // locations that should have whitespace following are identified by the
2227 // above set of follower tokens.
2229 // Postfix non-null assertion operator, as in `foo!.bar()`.
2230 if (Right.is(tok::exclaim) && (Left.isOneOf(tok::identifier, tok::r_paren,
2231 tok::r_square, tok::r_brace) ||
2232 Left.Tok.isLiteral()))
2234 if (Left.is(tok::exclaim) && Right.is(Keywords.kw_as))
2235 return true; // "x! as string"
2236 } else if (Style.Language == FormatStyle::LK_Java) {
2237 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2239 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2240 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2241 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2242 tok::kw_protected) ||
2243 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2244 Keywords.kw_native)) &&
2245 Right.is(TT_TemplateOpener))
2248 if (Left.is(TT_ImplicitStringLiteral))
2249 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2250 if (Line.Type == LT_ObjCMethodDecl) {
2251 if (Left.is(TT_ObjCMethodSpecifier))
2253 if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2254 // Don't space between ')' and <id>
2257 if (Line.Type == LT_ObjCProperty &&
2258 (Right.is(tok::equal) || Left.is(tok::equal)))
2261 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2262 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2264 if (Right.is(TT_OverloadedOperatorLParen))
2265 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2266 if (Left.is(tok::comma))
2268 if (Right.is(tok::comma))
2270 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2272 if (Right.is(tok::colon)) {
2273 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2274 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2276 if (Right.is(TT_ObjCMethodExpr))
2278 if (Left.is(tok::question))
2280 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2282 if (Right.is(TT_DictLiteral))
2283 return Style.SpacesInContainerLiterals;
2286 if (Left.is(TT_UnaryOperator))
2287 return Right.is(TT_BinaryOperator);
2289 // If the next token is a binary operator or a selector name, we have
2290 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2291 if (Left.is(TT_CastRParen))
2292 return Style.SpaceAfterCStyleCast ||
2293 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2295 if (Left.is(tok::greater) && Right.is(tok::greater))
2296 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2297 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2298 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2299 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2301 if (!Style.SpaceBeforeAssignmentOperators &&
2302 Right.getPrecedence() == prec::Assignment)
2304 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment))
2305 return (Left.is(TT_TemplateOpener) &&
2306 Style.Standard == FormatStyle::LS_Cpp03) ||
2307 !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren,
2309 Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener));
2310 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2311 return Style.SpacesInAngles;
2312 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2313 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2314 !Right.is(tok::r_paren)))
2316 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2317 Right.isNot(TT_FunctionTypeLParen))
2318 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2319 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2320 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2322 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2323 Line.startsWith(tok::hash))
2325 if (Right.is(TT_TrailingUnaryOperator))
2327 if (Left.is(TT_RegexLiteral))
2329 return spaceRequiredBetween(Line, Left, Right);
2332 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2333 static bool isAllmanBrace(const FormatToken &Tok) {
2334 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2335 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2338 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2339 const FormatToken &Right) {
2340 const FormatToken &Left = *Right.Previous;
2341 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2344 if (Style.Language == FormatStyle::LK_JavaScript) {
2345 // FIXME: This might apply to other languages and token kinds.
2346 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
2347 Left.Previous->is(tok::string_literal))
2349 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2350 Left.Previous && Left.Previous->is(tok::equal) &&
2351 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2353 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2355 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2356 // Object literals on the top level of a file are treated as "enum-style".
2357 // Each key/value pair is put on a separate line, instead of bin-packing.
2359 if (Left.is(tok::l_brace) && Line.Level == 0 &&
2360 (Line.startsWith(tok::kw_enum) ||
2361 Line.startsWith(tok::kw_export, tok::kw_enum)))
2362 // JavaScript top-level enum key/value pairs are put on separate lines
2363 // instead of bin-packing.
2365 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2366 !Left.Children.empty())
2367 // Support AllowShortFunctionsOnASingleLine for JavaScript.
2368 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2369 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2370 (Left.NestingLevel == 0 && Line.Level == 0 &&
2371 Style.AllowShortFunctionsOnASingleLine ==
2372 FormatStyle::SFS_Inline);
2373 } else if (Style.Language == FormatStyle::LK_Java) {
2374 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2375 Right.Next->is(tok::string_literal))
2379 // If the last token before a '}' is a comma or a trailing comment, the
2380 // intention is to insert a line break after it in order to make shuffling
2381 // around entries easier.
2382 const FormatToken *BeforeClosingBrace = nullptr;
2383 if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
2384 Left.BlockKind != BK_Block && Left.MatchingParen)
2385 BeforeClosingBrace = Left.MatchingParen->Previous;
2386 else if (Right.MatchingParen &&
2387 Right.MatchingParen->isOneOf(tok::l_brace,
2388 TT_ArrayInitializerLSquare))
2389 BeforeClosingBrace = &Left;
2390 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2391 BeforeClosingBrace->isTrailingComment()))
2394 if (Right.is(tok::comment))
2395 return Left.BlockKind != BK_BracedInit &&
2396 Left.isNot(TT_CtorInitializerColon) &&
2397 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2398 if (Left.isTrailingComment())
2400 if (Left.isStringLiteral() &&
2401 (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral)))
2403 if (Right.Previous->IsUnterminatedLiteral)
2405 if (Right.is(tok::lessless) && Right.Next &&
2406 Right.Previous->is(tok::string_literal) &&
2407 Right.Next->is(tok::string_literal))
2409 if (Right.Previous->ClosesTemplateDeclaration &&
2410 Right.Previous->MatchingParen &&
2411 Right.Previous->MatchingParen->NestingLevel == 0 &&
2412 Style.AlwaysBreakTemplateDeclarations)
2414 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2415 Style.BreakConstructorInitializersBeforeComma &&
2416 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2418 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2419 // Raw string literals are special wrt. line breaks. The author has made a
2420 // deliberate choice and might have aligned the contents of the string
2421 // literal accordingly. Thus, we try keep existing line breaks.
2422 return Right.NewlinesBefore > 0;
2423 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2424 Style.Language == FormatStyle::LK_Proto)
2425 // Don't put enums onto single lines in protocol buffers.
2427 if (Right.is(TT_InlineASMBrace))
2428 return Right.HasUnescapedNewline;
2429 if (isAllmanBrace(Left) || isAllmanBrace(Right))
2430 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2431 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2432 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2433 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2436 if ((Style.Language == FormatStyle::LK_Java ||
2437 Style.Language == FormatStyle::LK_JavaScript) &&
2438 Left.is(TT_LeadingJavaAnnotation) &&
2439 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2440 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2446 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2447 const FormatToken &Right) {
2448 const FormatToken &Left = *Right.Previous;
2450 // Language-specific stuff.
2451 if (Style.Language == FormatStyle::LK_Java) {
2452 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2453 Keywords.kw_implements))
2455 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2456 Keywords.kw_implements))
2458 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2459 const FormatToken *NonComment = Right.getPreviousNonComment();
2460 if (Left.isOneOf(tok::kw_return, tok::kw_continue, tok::kw_break,
2463 NonComment->isOneOf(tok::kw_return, tok::kw_continue, tok::kw_break,
2465 return false; // Otherwise a semicolon is inserted.
2466 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2468 if (Left.is(TT_JsTypeColon))
2470 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2472 if (Left.is(Keywords.kw_in))
2473 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
2474 if (Right.is(Keywords.kw_in))
2475 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
2476 if (Right.is(Keywords.kw_as))
2477 return false; // must not break before as in 'x as type' casts
2478 if (Left.is(Keywords.kw_declare) &&
2479 Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
2480 Keywords.kw_function, tok::kw_class, tok::kw_enum,
2481 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
2482 Keywords.kw_let, tok::kw_const))
2483 // See grammar for 'declare' statements at:
2484 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
2486 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
2487 Right.isOneOf(tok::identifier, tok::string_literal)) {
2488 return false; // must not break in "module foo { ...}"
2492 if (Left.is(tok::at))
2494 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2496 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2497 return !Right.is(tok::l_paren);
2498 if (Right.is(TT_PointerOrReference))
2499 return Line.IsMultiVariableDeclStmt ||
2500 (Style.PointerAlignment == FormatStyle::PAS_Right &&
2501 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2502 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2503 Right.is(tok::kw_operator))
2505 if (Left.is(TT_PointerOrReference))
2507 if (Right.isTrailingComment())
2508 // We rely on MustBreakBefore being set correctly here as we should not
2509 // change the "binding" behavior of a comment.
2510 // The first comment in a braced lists is always interpreted as belonging to
2511 // the first list element. Otherwise, it should be placed outside of the
2513 return Left.BlockKind == BK_BracedInit;
2514 if (Left.is(tok::question) && Right.is(tok::colon))
2516 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2517 return Style.BreakBeforeTernaryOperators;
2518 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2519 return !Style.BreakBeforeTernaryOperators;
2520 if (Right.is(TT_InheritanceColon))
2522 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
2523 Left.isNot(TT_SelectorName))
2525 if (Right.is(tok::colon) &&
2526 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2528 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))
2530 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2531 Right.Next->is(TT_ObjCMethodExpr)))
2532 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2533 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2535 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2537 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2538 TT_OverloadedOperator))
2540 if (Left.is(TT_RangeBasedForLoopColon))
2542 if (Right.is(TT_RangeBasedForLoopColon))
2544 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
2546 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2547 Left.is(tok::kw_operator))
2549 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2550 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2552 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2554 if (Left.is(tok::l_paren) && Left.Previous &&
2555 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2557 if (Right.is(TT_ImplicitStringLiteral))
2560 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2562 if (Right.is(tok::r_square) && Right.MatchingParen &&
2563 Right.MatchingParen->is(TT_LambdaLSquare))
2566 // We only break before r_brace if there was a corresponding break before
2567 // the l_brace, which is tracked by BreakBeforeClosingBrace.
2568 if (Right.is(tok::r_brace))
2569 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2571 // Allow breaking after a trailing annotation, e.g. after a method
2573 if (Left.is(TT_TrailingAnnotation))
2574 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2575 tok::less, tok::coloncolon);
2577 if (Right.is(tok::kw___attribute))
2580 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2583 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2586 if (Left.is(TT_CtorInitializerComma) &&
2587 Style.BreakConstructorInitializersBeforeComma)
2589 if (Right.is(TT_CtorInitializerComma) &&
2590 Style.BreakConstructorInitializersBeforeComma)
2592 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2593 (Left.is(tok::less) && Right.is(tok::less)))
2595 if (Right.is(TT_BinaryOperator) &&
2596 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2597 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2598 Right.getPrecedence() != prec::Assignment))
2600 if (Left.is(TT_ArrayInitializerLSquare))
2602 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2604 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2605 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2606 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2607 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2608 Left.getPrecedence() == prec::Assignment))
2610 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2611 tok::kw_class, tok::kw_struct, tok::comment) ||
2612 Right.isMemberAccess() ||
2613 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2614 tok::colon, tok::l_square, tok::at) ||
2615 (Left.is(tok::r_paren) &&
2616 Right.isOneOf(tok::identifier, tok::kw_const)) ||
2617 (Left.is(tok::l_paren) && !Right.is(tok::r_paren));
2620 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2621 llvm::errs() << "AnnotatedTokens:\n";
2622 const FormatToken *Tok = Line.First;
2624 llvm::errs() << " M=" << Tok->MustBreakBefore
2625 << " C=" << Tok->CanBreakBefore
2626 << " T=" << getTokenTypeName(Tok->Type)
2627 << " S=" << Tok->SpacesRequiredBefore
2628 << " B=" << Tok->BlockParameterCount
2629 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2630 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2632 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2633 llvm::errs() << Tok->FakeLParens[i] << "/";
2634 llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n";
2636 assert(Tok == Line.Last);
2639 llvm::errs() << "----\n";
2642 } // namespace format
2643 } // namespace clang