1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// \brief This file implements a token annotator, i.e. creates
12 /// \c AnnotatedTokens out of \c FormatTokens with required extra information.
14 //===----------------------------------------------------------------------===//
16 #include "TokenAnnotator.h"
17 #include "clang/Basic/SourceManager.h"
18 #include "llvm/ADT/SmallPtrSet.h"
19 #include "llvm/Support/Debug.h"
21 #define DEBUG_TYPE "format-token-annotator"
28 /// \brief A parser that gathers additional information about tokens.
30 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
31 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
32 /// into template parameter lists.
33 class AnnotatingParser {
35 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line,
36 const AdditionalKeywords &Keywords)
37 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false),
39 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
40 resetTokenMetadata(CurrentToken);
45 if (!CurrentToken || !CurrentToken->Previous)
47 if (NonTemplateLess.count(CurrentToken->Previous))
50 const FormatToken& Previous = *CurrentToken->Previous;
51 if (Previous.Previous) {
52 if (Previous.Previous->Tok.isLiteral())
54 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 &&
55 (!Previous.Previous->MatchingParen ||
56 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen)))
60 FormatToken *Left = CurrentToken->Previous;
61 Left->ParentBracket = Contexts.back().ContextKind;
62 ScopedContextCreator ContextCreator(*this, tok::less, 12);
64 // If this angle is in the context of an expression, we need to be more
65 // hesitant to detect it as opening template parameters.
66 bool InExprContext = Contexts.back().IsExpression;
68 Contexts.back().IsExpression = false;
69 // If there's a template keyword before the opening angle bracket, this is a
70 // template parameter, not an argument.
71 Contexts.back().InTemplateArgument =
72 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
74 if (Style.Language == FormatStyle::LK_Java &&
75 CurrentToken->is(tok::question))
78 while (CurrentToken) {
79 if (CurrentToken->is(tok::greater)) {
80 Left->MatchingParen = CurrentToken;
81 CurrentToken->MatchingParen = Left;
82 CurrentToken->Type = TT_TemplateCloser;
86 if (CurrentToken->is(tok::question) &&
87 Style.Language == FormatStyle::LK_Java) {
91 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
92 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext))
94 // If a && or || is found and interpreted as a binary operator, this set
95 // of angles is likely part of something like "a < b && c > d". If the
96 // angles are inside an expression, the ||/&& might also be a binary
97 // operator that was misinterpreted because we are parsing template
99 // FIXME: This is getting out of hand, write a decent parser.
100 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
101 CurrentToken->Previous->is(TT_BinaryOperator) &&
102 Contexts[Contexts.size() - 2].IsExpression &&
103 !Line.startsWith(tok::kw_template))
105 updateParameterCount(Left, CurrentToken);
112 bool parseParens(bool LookForDecls = false) {
115 FormatToken *Left = CurrentToken->Previous;
116 Left->ParentBracket = Contexts.back().ContextKind;
117 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
119 // FIXME: This is a bit of a hack. Do better.
120 Contexts.back().ColonIsForRangeExpr =
121 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
123 bool StartsObjCMethodExpr = false;
124 if (CurrentToken->is(tok::caret)) {
125 // (^ can start a block type.
126 Left->Type = TT_ObjCBlockLParen;
127 } else if (FormatToken *MaybeSel = Left->Previous) {
128 // @selector( starts a selector.
129 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
130 MaybeSel->Previous->is(tok::at)) {
131 StartsObjCMethodExpr = true;
135 if (Left->is(TT_OverloadedOperatorLParen)) {
136 Contexts.back().IsExpression = false;
137 } else if (Style.Language == FormatStyle::LK_JavaScript &&
138 Line.startsWith(Keywords.kw_type, tok::identifier)) {
140 Contexts.back().IsExpression = false;
141 } else if (Left->Previous &&
142 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype,
143 tok::kw_if, tok::kw_while, tok::l_paren,
145 Left->Previous->is(TT_BinaryOperator))) {
146 // static_assert, if and while usually contain expressions.
147 Contexts.back().IsExpression = true;
148 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
149 (Left->Previous->is(Keywords.kw_function) ||
150 (Left->Previous->endsSequence(tok::identifier,
151 Keywords.kw_function)))) {
152 // function(...) or function f(...)
153 Contexts.back().IsExpression = false;
154 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
155 Left->Previous->is(TT_JsTypeColon)) {
156 // let x: (SomeType);
157 Contexts.back().IsExpression = false;
158 } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
159 Left->Previous->MatchingParen &&
160 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
161 // This is a parameter list of a lambda expression.
162 Contexts.back().IsExpression = false;
163 } else if (Line.InPPDirective &&
164 (!Left->Previous || !Left->Previous->is(tok::identifier))) {
165 Contexts.back().IsExpression = true;
166 } else if (Contexts[Contexts.size() - 2].CaretFound) {
167 // This is the parameter list of an ObjC block.
168 Contexts.back().IsExpression = false;
169 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
170 Left->Type = TT_AttributeParen;
171 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
172 // The first argument to a foreach macro is a declaration.
173 Contexts.back().IsForEachMacro = true;
174 Contexts.back().IsExpression = false;
175 } else if (Left->Previous && Left->Previous->MatchingParen &&
176 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
177 Contexts.back().IsExpression = false;
178 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
180 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
181 Contexts.back().IsExpression = !IsForOrCatch;
184 if (StartsObjCMethodExpr) {
185 Contexts.back().ColonIsObjCMethodExpr = true;
186 Left->Type = TT_ObjCMethodExpr;
189 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression;
190 bool ProbablyFunctionType = CurrentToken->isOneOf(tok::star, tok::amp);
191 bool HasMultipleLines = false;
192 bool HasMultipleParametersOnALine = false;
193 bool MightBeObjCForRangeLoop =
194 Left->Previous && Left->Previous->is(tok::kw_for);
195 while (CurrentToken) {
196 // LookForDecls is set when "if (" has been seen. Check for
197 // 'identifier' '*' 'identifier' followed by not '=' -- this
198 // '*' has to be a binary operator but determineStarAmpUsage() will
199 // categorize it as an unary operator, so set the right type here.
200 if (LookForDecls && CurrentToken->Next) {
201 FormatToken *Prev = CurrentToken->getPreviousNonComment();
203 FormatToken *PrevPrev = Prev->getPreviousNonComment();
204 FormatToken *Next = CurrentToken->Next;
205 if (PrevPrev && PrevPrev->is(tok::identifier) &&
206 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
207 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
208 Prev->Type = TT_BinaryOperator;
209 LookForDecls = false;
214 if (CurrentToken->Previous->is(TT_PointerOrReference) &&
215 CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
217 ProbablyFunctionType = true;
218 if (CurrentToken->is(tok::comma))
219 MightBeFunctionType = false;
220 if (CurrentToken->Previous->is(TT_BinaryOperator))
221 Contexts.back().IsExpression = true;
222 if (CurrentToken->is(tok::r_paren)) {
223 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next &&
224 (CurrentToken->Next->is(tok::l_paren) ||
225 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration)))
226 Left->Type = TT_FunctionTypeLParen;
227 Left->MatchingParen = CurrentToken;
228 CurrentToken->MatchingParen = Left;
230 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) &&
231 Left->Previous && Left->Previous->is(tok::l_paren)) {
232 // Detect the case where macros are used to generate lambdas or
233 // function bodies, e.g.:
234 // auto my_lambda = MARCO((Type *type, int i) { .. body .. });
235 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) {
236 if (Tok->is(TT_BinaryOperator) &&
237 Tok->isOneOf(tok::star, tok::amp, tok::ampamp))
238 Tok->Type = TT_PointerOrReference;
242 if (StartsObjCMethodExpr) {
243 CurrentToken->Type = TT_ObjCMethodExpr;
244 if (Contexts.back().FirstObjCSelectorName) {
245 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
246 Contexts.back().LongestObjCSelectorName;
250 if (Left->is(TT_AttributeParen))
251 CurrentToken->Type = TT_AttributeParen;
252 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
253 CurrentToken->Type = TT_JavaAnnotation;
254 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
255 CurrentToken->Type = TT_LeadingJavaAnnotation;
257 if (!HasMultipleLines)
258 Left->PackingKind = PPK_Inconclusive;
259 else if (HasMultipleParametersOnALine)
260 Left->PackingKind = PPK_BinPacked;
262 Left->PackingKind = PPK_OnePerLine;
267 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
270 if (CurrentToken->is(tok::l_brace))
271 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
272 if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
273 !CurrentToken->Next->HasUnescapedNewline &&
274 !CurrentToken->Next->isTrailingComment())
275 HasMultipleParametersOnALine = true;
276 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) ||
277 CurrentToken->Previous->isSimpleTypeSpecifier()) &&
278 !CurrentToken->is(tok::l_brace))
279 Contexts.back().IsExpression = false;
280 if (CurrentToken->isOneOf(tok::semi, tok::colon))
281 MightBeObjCForRangeLoop = false;
282 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
283 CurrentToken->Type = TT_ObjCForIn;
284 // When we discover a 'new', we set CanBeExpression to 'false' in order to
285 // parse the type correctly. Reset that after a comma.
286 if (CurrentToken->is(tok::comma))
287 Contexts.back().CanBeExpression = true;
289 FormatToken *Tok = CurrentToken;
292 updateParameterCount(Left, Tok);
293 if (CurrentToken && CurrentToken->HasUnescapedNewline)
294 HasMultipleLines = true;
303 // A '[' could be an index subscript (after an identifier or after
304 // ')' or ']'), it could be the start of an Objective-C method
305 // expression, or it could the start of an Objective-C array literal.
306 FormatToken *Left = CurrentToken->Previous;
307 Left->ParentBracket = Contexts.back().ContextKind;
308 FormatToken *Parent = Left->getPreviousNonComment();
310 // Cases where '>' is followed by '['.
311 // In C++, this can happen either in array of templates (foo<int>[10])
312 // or when array is a nested template type (unique_ptr<type1<type2>[]>).
313 bool CppArrayTemplates =
314 Style.isCpp() && Parent &&
315 Parent->is(TT_TemplateCloser) &&
316 (Contexts.back().CanBeExpression || Contexts.back().IsExpression ||
317 Contexts.back().InTemplateArgument);
319 bool StartsObjCMethodExpr =
320 !CppArrayTemplates && Style.isCpp() &&
321 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
322 CurrentToken->isNot(tok::l_brace) &&
324 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
325 tok::kw_return, tok::kw_throw) ||
326 Parent->isUnaryOperator() ||
327 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
328 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
329 bool ColonFound = false;
331 unsigned BindingIncrease = 1;
332 if (Left->is(TT_Unknown)) {
333 if (StartsObjCMethodExpr) {
334 Left->Type = TT_ObjCMethodExpr;
335 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
336 Contexts.back().ContextKind == tok::l_brace &&
337 Parent->isOneOf(tok::l_brace, tok::comma)) {
338 Left->Type = TT_JsComputedPropertyName;
339 } else if (CurrentToken->is(tok::r_square) && Parent &&
340 Parent->is(TT_TemplateCloser)) {
341 Left->Type = TT_ArraySubscriptLSquare;
342 } else if (Style.Language == FormatStyle::LK_Proto ||
343 (!CppArrayTemplates && Parent &&
344 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at,
345 tok::comma, tok::l_paren, tok::l_square,
346 tok::question, tok::colon, tok::kw_return,
347 // Should only be relevant to JavaScript:
349 Left->Type = TT_ArrayInitializerLSquare;
351 BindingIncrease = 10;
352 Left->Type = TT_ArraySubscriptLSquare;
356 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
357 Contexts.back().IsExpression = true;
358 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
360 while (CurrentToken) {
361 if (CurrentToken->is(tok::r_square)) {
362 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
363 Left->is(TT_ObjCMethodExpr)) {
364 // An ObjC method call is rarely followed by an open parenthesis.
365 // FIXME: Do we incorrectly label ":" with this?
366 StartsObjCMethodExpr = false;
367 Left->Type = TT_Unknown;
369 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
370 CurrentToken->Type = TT_ObjCMethodExpr;
371 // determineStarAmpUsage() thinks that '*' '[' is allocating an
372 // array of pointers, but if '[' starts a selector then '*' is a
374 if (Parent && Parent->is(TT_PointerOrReference))
375 Parent->Type = TT_BinaryOperator;
377 Left->MatchingParen = CurrentToken;
378 CurrentToken->MatchingParen = Left;
379 if (Contexts.back().FirstObjCSelectorName) {
380 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
381 Contexts.back().LongestObjCSelectorName;
382 if (Left->BlockParameterCount > 1)
383 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
388 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
390 if (CurrentToken->is(tok::colon)) {
391 if (Left->is(TT_ArraySubscriptLSquare)) {
392 Left->Type = TT_ObjCMethodExpr;
393 StartsObjCMethodExpr = true;
394 Contexts.back().ColonIsObjCMethodExpr = true;
395 if (Parent && Parent->is(tok::r_paren))
396 Parent->Type = TT_CastRParen;
400 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
402 Left->Type = TT_ArrayInitializerLSquare;
403 FormatToken *Tok = CurrentToken;
406 updateParameterCount(Left, Tok);
413 FormatToken *Left = CurrentToken->Previous;
414 Left->ParentBracket = Contexts.back().ContextKind;
416 if (Contexts.back().CaretFound)
417 Left->Type = TT_ObjCBlockLBrace;
418 Contexts.back().CaretFound = false;
420 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
421 Contexts.back().ColonIsDictLiteral = true;
422 if (Left->BlockKind == BK_BracedInit)
423 Contexts.back().IsExpression = true;
425 while (CurrentToken) {
426 if (CurrentToken->is(tok::r_brace)) {
427 Left->MatchingParen = CurrentToken;
428 CurrentToken->MatchingParen = Left;
432 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
434 updateParameterCount(Left, CurrentToken);
435 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) {
436 FormatToken *Previous = CurrentToken->getPreviousNonComment();
437 if (((CurrentToken->is(tok::colon) &&
438 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) ||
439 Style.Language == FormatStyle::LK_Proto) &&
440 (Previous->Tok.getIdentifierInfo() ||
441 Previous->is(tok::string_literal)))
442 Previous->Type = TT_SelectorName;
443 if (CurrentToken->is(tok::colon) ||
444 Style.Language == FormatStyle::LK_JavaScript)
445 Left->Type = TT_DictLiteral;
447 if (CurrentToken->is(tok::comma) &&
448 Style.Language == FormatStyle::LK_JavaScript)
449 Left->Type = TT_DictLiteral;
457 void updateParameterCount(FormatToken *Left, FormatToken *Current) {
458 if (Current->is(tok::l_brace) && Current->BlockKind == BK_Block)
459 ++Left->BlockParameterCount;
460 if (Current->is(tok::comma)) {
461 ++Left->ParameterCount;
463 Left->Role.reset(new CommaSeparatedList(Style));
464 Left->Role->CommaFound(Current);
465 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
466 Left->ParameterCount = 1;
470 bool parseConditional() {
471 while (CurrentToken) {
472 if (CurrentToken->is(tok::colon)) {
473 CurrentToken->Type = TT_ConditionalExpr;
483 bool parseTemplateDeclaration() {
484 if (CurrentToken && CurrentToken->is(tok::less)) {
485 CurrentToken->Type = TT_TemplateOpener;
490 CurrentToken->Previous->ClosesTemplateDeclaration = true;
496 bool consumeToken() {
497 FormatToken *Tok = CurrentToken;
499 switch (Tok->Tok.getKind()) {
502 if (!Tok->Previous && Line.MustBeDeclaration)
503 Tok->Type = TT_ObjCMethodSpecifier;
508 // Colons from ?: are handled in parseConditional().
509 if (Style.Language == FormatStyle::LK_JavaScript) {
510 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
511 (Contexts.size() == 1 && // switch/case labels
512 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
513 Contexts.back().ContextKind == tok::l_paren || // function params
514 Contexts.back().ContextKind == tok::l_square || // array type
515 (Contexts.size() == 1 &&
516 Line.MustBeDeclaration)) { // method/property declaration
517 Tok->Type = TT_JsTypeColon;
521 if (Contexts.back().ColonIsDictLiteral ||
522 Style.Language == FormatStyle::LK_Proto) {
523 Tok->Type = TT_DictLiteral;
524 } else if (Contexts.back().ColonIsObjCMethodExpr ||
525 Line.startsWith(TT_ObjCMethodSpecifier)) {
526 Tok->Type = TT_ObjCMethodExpr;
527 const FormatToken *BeforePrevious = Tok->Previous->Previous;
528 if (!BeforePrevious ||
529 !(BeforePrevious->is(TT_CastRParen) ||
530 (BeforePrevious->is(TT_ObjCMethodExpr) &&
531 BeforePrevious->is(tok::colon))) ||
532 BeforePrevious->is(tok::r_square) ||
533 Contexts.back().LongestObjCSelectorName == 0) {
534 Tok->Previous->Type = TT_SelectorName;
535 if (Tok->Previous->ColumnWidth >
536 Contexts.back().LongestObjCSelectorName)
537 Contexts.back().LongestObjCSelectorName =
538 Tok->Previous->ColumnWidth;
539 if (!Contexts.back().FirstObjCSelectorName)
540 Contexts.back().FirstObjCSelectorName = Tok->Previous;
542 } else if (Contexts.back().ColonIsForRangeExpr) {
543 Tok->Type = TT_RangeBasedForLoopColon;
544 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
545 Tok->Type = TT_BitFieldColon;
546 } else if (Contexts.size() == 1 &&
547 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
548 if (Tok->getPreviousNonComment()->isOneOf(tok::r_paren,
550 Tok->Type = TT_CtorInitializerColon;
552 Tok->Type = TT_InheritanceColon;
553 } else if (Tok->Previous->is(tok::identifier) && Tok->Next &&
554 Tok->Next->isOneOf(tok::r_paren, tok::comma)) {
555 // This handles a special macro in ObjC code where selectors including
556 // the colon are passed as macro arguments.
557 Tok->Type = TT_ObjCMethodExpr;
558 } else if (Contexts.back().ContextKind == tok::l_paren) {
559 Tok->Type = TT_InlineASMColon;
564 // | and & in declarations/type expressions represent union and
565 // intersection types, respectively.
566 if (Style.Language == FormatStyle::LK_JavaScript &&
567 !Contexts.back().IsExpression)
568 Tok->Type = TT_JsTypeOperator;
572 if (CurrentToken && CurrentToken->is(tok::l_paren)) {
574 if (!parseParens(/*LookForDecls=*/true))
579 if (Style.Language == FormatStyle::LK_JavaScript) {
580 if (Tok->Previous && Tok->Previous->is(tok::period))
582 // JS' for await ( ...
583 if (CurrentToken && CurrentToken->is(Keywords.kw_await))
586 Contexts.back().ColonIsForRangeExpr = true;
592 // When faced with 'operator()()', the kw_operator handler incorrectly
593 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
594 // the first two parens OverloadedOperators and the second l_paren an
595 // OverloadedOperatorLParen.
597 Tok->Previous->is(tok::r_paren) &&
598 Tok->Previous->MatchingParen &&
599 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
600 Tok->Previous->Type = TT_OverloadedOperator;
601 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator;
602 Tok->Type = TT_OverloadedOperatorLParen;
607 if (Line.MustBeDeclaration && Contexts.size() == 1 &&
608 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
610 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute,
611 TT_LeadingJavaAnnotation)))
612 Line.MightBeFunctionDecl = true;
624 Tok->Type = TT_TemplateOpener;
626 Tok->Type = TT_BinaryOperator;
627 NonTemplateLess.insert(Tok);
636 // Lines can start with '}'.
641 Tok->Type = TT_BinaryOperator;
643 case tok::kw_operator:
644 while (CurrentToken &&
645 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
646 if (CurrentToken->isOneOf(tok::star, tok::amp))
647 CurrentToken->Type = TT_PointerOrReference;
650 CurrentToken->Previous->isOneOf(TT_BinaryOperator, tok::comma))
651 CurrentToken->Previous->Type = TT_OverloadedOperator;
654 CurrentToken->Type = TT_OverloadedOperatorLParen;
655 if (CurrentToken->Previous->is(TT_BinaryOperator))
656 CurrentToken->Previous->Type = TT_OverloadedOperator;
660 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
661 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
663 // Question marks before semicolons, colons, etc. indicate optional
664 // types (fields, parameters), e.g.
665 // function(x?: string, y?) {...}
667 Tok->Type = TT_JsTypeOptionalQuestion;
670 // Declarations cannot be conditional expressions, this can only be part
671 // of a type declaration.
672 if (Line.MustBeDeclaration && !Contexts.back().IsExpression &&
673 Style.Language == FormatStyle::LK_JavaScript)
677 case tok::kw_template:
678 parseTemplateDeclaration();
681 if (Contexts.back().InCtorInitializer)
682 Tok->Type = TT_CtorInitializerComma;
683 else if (Contexts.back().InInheritanceList)
684 Tok->Type = TT_InheritanceComma;
685 else if (Contexts.back().FirstStartOfName &&
686 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
687 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
688 Line.IsMultiVariableDeclStmt = true;
690 if (Contexts.back().IsForEachMacro)
691 Contexts.back().IsExpression = true;
693 case tok::identifier:
694 if (Tok->isOneOf(Keywords.kw___has_include,
695 Keywords.kw___has_include_next)) {
705 void parseIncludeDirective() {
706 if (CurrentToken && CurrentToken->is(tok::less)) {
708 while (CurrentToken) {
709 // Mark tokens up to the trailing line comments as implicit string
711 if (CurrentToken->isNot(tok::comment) &&
712 !CurrentToken->TokenText.startswith("//"))
713 CurrentToken->Type = TT_ImplicitStringLiteral;
719 void parseWarningOrError() {
721 // We still want to format the whitespace left of the first token of the
724 while (CurrentToken) {
725 CurrentToken->Type = TT_ImplicitStringLiteral;
731 next(); // Consume "pragma".
733 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
734 bool IsMark = CurrentToken->is(Keywords.kw_mark);
735 next(); // Consume "mark".
736 next(); // Consume first token (so we fix leading whitespace).
737 while (CurrentToken) {
738 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
739 CurrentToken->Type = TT_ImplicitStringLiteral;
745 void parseHasInclude() {
746 if (!CurrentToken || !CurrentToken->is(tok::l_paren))
749 parseIncludeDirective();
753 LineType parsePreprocessorDirective() {
754 bool IsFirstToken = CurrentToken->IsFirst;
755 LineType Type = LT_PreprocessorDirective;
760 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) {
761 // JavaScript files can contain shebang lines of the form:
762 // #!/usr/bin/env node
763 // Treat these like C++ #include directives.
764 while (CurrentToken) {
765 // Tokens cannot be comments here.
766 CurrentToken->Type = TT_ImplicitStringLiteral;
769 return LT_ImportStatement;
772 if (CurrentToken->Tok.is(tok::numeric_constant)) {
773 CurrentToken->SpacesRequiredBefore = 1;
776 // Hashes in the middle of a line can lead to any strange token
778 if (!CurrentToken->Tok.getIdentifierInfo())
780 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
781 case tok::pp_include:
782 case tok::pp_include_next:
785 parseIncludeDirective();
786 Type = LT_ImportStatement;
789 case tok::pp_warning:
790 parseWarningOrError();
797 Contexts.back().IsExpression = true;
803 while (CurrentToken) {
804 FormatToken *Tok = CurrentToken;
806 if (Tok->is(tok::l_paren))
808 else if (Tok->isOneOf(Keywords.kw___has_include,
809 Keywords.kw___has_include_next))
816 LineType parseLine() {
817 NonTemplateLess.clear();
818 if (CurrentToken->is(tok::hash))
819 return parsePreprocessorDirective();
821 // Directly allow to 'import <string-literal>' to support protocol buffer
822 // definitions (code.google.com/p/protobuf) or missing "#" (either way we
823 // should not break the line).
824 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
825 if ((Style.Language == FormatStyle::LK_Java &&
826 CurrentToken->is(Keywords.kw_package)) ||
827 (Info && Info->getPPKeywordID() == tok::pp_import &&
828 CurrentToken->Next &&
829 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
832 parseIncludeDirective();
833 return LT_ImportStatement;
836 // If this line starts and ends in '<' and '>', respectively, it is likely
837 // part of "#define <a/b.h>".
838 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
839 parseIncludeDirective();
840 return LT_ImportStatement;
843 // In .proto files, top-level options are very similar to import statements
844 // and should not be line-wrapped.
845 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
846 CurrentToken->is(Keywords.kw_option)) {
848 if (CurrentToken && CurrentToken->is(tok::identifier))
849 return LT_ImportStatement;
852 bool KeywordVirtualFound = false;
853 bool ImportStatement = false;
855 // import {...} from '...';
856 if (Style.Language == FormatStyle::LK_JavaScript &&
857 CurrentToken->is(Keywords.kw_import))
858 ImportStatement = true;
860 while (CurrentToken) {
861 if (CurrentToken->is(tok::kw_virtual))
862 KeywordVirtualFound = true;
863 if (Style.Language == FormatStyle::LK_JavaScript) {
864 // export {...} from '...';
865 // An export followed by "from 'some string';" is a re-export from
866 // another module identified by a URI and is treated as a
867 // LT_ImportStatement (i.e. prevent wraps on it for long URIs).
868 // Just "export {...};" or "export class ..." should not be treated as
869 // an import in this sense.
870 if (Line.First->is(tok::kw_export) &&
871 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next &&
872 CurrentToken->Next->isStringLiteral())
873 ImportStatement = true;
874 if (isClosureImportStatement(*CurrentToken))
875 ImportStatement = true;
880 if (KeywordVirtualFound)
881 return LT_VirtualFunctionDecl;
883 return LT_ImportStatement;
885 if (Line.startsWith(TT_ObjCMethodSpecifier)) {
886 if (Contexts.back().FirstObjCSelectorName)
887 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
888 Contexts.back().LongestObjCSelectorName;
889 return LT_ObjCMethodDecl;
896 bool isClosureImportStatement(const FormatToken &Tok) {
897 // FIXME: Closure-library specific stuff should not be hard-coded but be
899 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
900 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" ||
901 Tok.Next->Next->TokenText == "provide" ||
902 Tok.Next->Next->TokenText == "require" ||
903 Tok.Next->Next->TokenText == "setTestOnly" ||
904 Tok.Next->Next->TokenText == "forwardDeclare") &&
905 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
908 void resetTokenMetadata(FormatToken *Token) {
912 // Reset token type in case we have already looked at it and then
913 // recovered from an error (e.g. failure to find the matching >).
914 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
915 TT_FunctionLBrace, TT_ImplicitStringLiteral,
916 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
917 TT_OverloadedOperator, TT_RegexLiteral,
918 TT_TemplateString, TT_ObjCStringLiteral))
919 CurrentToken->Type = TT_Unknown;
920 CurrentToken->Role.reset();
921 CurrentToken->MatchingParen = nullptr;
922 CurrentToken->FakeLParens.clear();
923 CurrentToken->FakeRParens = 0;
928 CurrentToken->NestingLevel = Contexts.size() - 1;
929 CurrentToken->BindingStrength = Contexts.back().BindingStrength;
930 modifyContext(*CurrentToken);
931 determineTokenType(*CurrentToken);
932 CurrentToken = CurrentToken->Next;
935 resetTokenMetadata(CurrentToken);
938 /// \brief A struct to hold information valid in a specific context, e.g.
939 /// a pair of parenthesis.
941 Context(tok::TokenKind ContextKind, unsigned BindingStrength,
943 : ContextKind(ContextKind), BindingStrength(BindingStrength),
944 IsExpression(IsExpression) {}
946 tok::TokenKind ContextKind;
947 unsigned BindingStrength;
949 unsigned LongestObjCSelectorName = 0;
950 bool ColonIsForRangeExpr = false;
951 bool ColonIsDictLiteral = false;
952 bool ColonIsObjCMethodExpr = false;
953 FormatToken *FirstObjCSelectorName = nullptr;
954 FormatToken *FirstStartOfName = nullptr;
955 bool CanBeExpression = true;
956 bool InTemplateArgument = false;
957 bool InCtorInitializer = false;
958 bool InInheritanceList = false;
959 bool CaretFound = false;
960 bool IsForEachMacro = false;
963 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
964 /// of each instance.
965 struct ScopedContextCreator {
968 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
971 P.Contexts.push_back(Context(ContextKind,
972 P.Contexts.back().BindingStrength + Increase,
973 P.Contexts.back().IsExpression));
976 ~ScopedContextCreator() { P.Contexts.pop_back(); }
979 void modifyContext(const FormatToken &Current) {
980 if (Current.getPrecedence() == prec::Assignment &&
981 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
982 // Type aliases use `type X = ...;` in TypeScript.
983 !(Style.Language == FormatStyle::LK_JavaScript &&
984 Line.startsWith(Keywords.kw_type, tok::identifier)) &&
985 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
986 Contexts.back().IsExpression = true;
987 if (!Line.startsWith(TT_UnaryOperator)) {
988 for (FormatToken *Previous = Current.Previous;
989 Previous && Previous->Previous &&
990 !Previous->Previous->isOneOf(tok::comma, tok::semi);
991 Previous = Previous->Previous) {
992 if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
993 Previous = Previous->MatchingParen;
997 if (Previous->opensScope())
999 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
1000 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
1001 Previous->Previous && Previous->Previous->isNot(tok::equal))
1002 Previous->Type = TT_PointerOrReference;
1005 } else if (Current.is(tok::lessless) &&
1006 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
1007 Contexts.back().IsExpression = true;
1008 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
1009 Contexts.back().IsExpression = true;
1010 } else if (Current.is(TT_TrailingReturnArrow)) {
1011 Contexts.back().IsExpression = false;
1012 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
1013 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
1014 } else if (Current.Previous &&
1015 Current.Previous->is(TT_CtorInitializerColon)) {
1016 Contexts.back().IsExpression = true;
1017 Contexts.back().InCtorInitializer = true;
1018 } else if (Current.Previous &&
1019 Current.Previous->is(TT_InheritanceColon)) {
1020 Contexts.back().InInheritanceList = true;
1021 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
1022 for (FormatToken *Previous = Current.Previous;
1023 Previous && Previous->isOneOf(tok::star, tok::amp);
1024 Previous = Previous->Previous)
1025 Previous->Type = TT_PointerOrReference;
1026 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
1027 Contexts.back().IsExpression = false;
1028 } else if (Current.is(tok::kw_new)) {
1029 Contexts.back().CanBeExpression = false;
1030 } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
1031 // This should be the condition or increment in a for-loop.
1032 Contexts.back().IsExpression = true;
1036 void determineTokenType(FormatToken &Current) {
1037 if (!Current.is(TT_Unknown))
1038 // The token type is already known.
1041 if (Style.Language == FormatStyle::LK_JavaScript) {
1042 if (Current.is(tok::exclaim)) {
1043 if (Current.Previous &&
1044 (Current.Previous->isOneOf(tok::identifier, tok::kw_namespace,
1045 tok::r_paren, tok::r_square,
1047 Current.Previous->Tok.isLiteral())) {
1048 Current.Type = TT_JsNonNullAssertion;
1052 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) {
1053 Current.Type = TT_JsNonNullAssertion;
1059 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1060 // function declaration have been found. In this case, 'Current' is a
1061 // trailing token of this declaration and thus cannot be a name.
1062 if (Current.is(Keywords.kw_instanceof)) {
1063 Current.Type = TT_BinaryOperator;
1064 } else if (isStartOfName(Current) &&
1065 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
1066 Contexts.back().FirstStartOfName = &Current;
1067 Current.Type = TT_StartOfName;
1068 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
1070 } else if (Current.is(tok::arrow) &&
1071 Style.Language == FormatStyle::LK_Java) {
1072 Current.Type = TT_LambdaArrow;
1073 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
1074 Current.NestingLevel == 0) {
1075 Current.Type = TT_TrailingReturnArrow;
1076 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
1078 determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
1079 Contexts.back().IsExpression,
1080 Contexts.back().InTemplateArgument);
1081 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
1082 Current.Type = determinePlusMinusCaretUsage(Current);
1083 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
1084 Contexts.back().CaretFound = true;
1085 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
1086 Current.Type = determineIncrementUsage(Current);
1087 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
1088 Current.Type = TT_UnaryOperator;
1089 } else if (Current.is(tok::question)) {
1090 if (Style.Language == FormatStyle::LK_JavaScript &&
1091 Line.MustBeDeclaration && !Contexts.back().IsExpression) {
1092 // In JavaScript, `interface X { foo?(): bar; }` is an optional method
1093 // on the interface, not a ternary expression.
1094 Current.Type = TT_JsTypeOptionalQuestion;
1096 Current.Type = TT_ConditionalExpr;
1098 } else if (Current.isBinaryOperator() &&
1099 (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
1100 Current.Type = TT_BinaryOperator;
1101 } else if (Current.is(tok::comment)) {
1102 if (Current.TokenText.startswith("/*")) {
1103 if (Current.TokenText.endswith("*/"))
1104 Current.Type = TT_BlockComment;
1106 // The lexer has for some reason determined a comment here. But we
1107 // cannot really handle it, if it isn't properly terminated.
1108 Current.Tok.setKind(tok::unknown);
1110 Current.Type = TT_LineComment;
1112 } else if (Current.is(tok::r_paren)) {
1113 if (rParenEndsCast(Current))
1114 Current.Type = TT_CastRParen;
1115 if (Current.MatchingParen && Current.Next &&
1116 !Current.Next->isBinaryOperator() &&
1117 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace,
1118 tok::comma, tok::period, tok::arrow,
1120 if (FormatToken *AfterParen = Current.MatchingParen->Next) {
1121 // Make sure this isn't the return type of an Obj-C block declaration
1122 if (AfterParen->Tok.isNot(tok::caret)) {
1123 if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
1124 if (BeforeParen->is(tok::identifier) &&
1125 BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
1126 (!BeforeParen->Previous ||
1127 BeforeParen->Previous->ClosesTemplateDeclaration))
1128 Current.Type = TT_FunctionAnnotationRParen;
1131 } else if (Current.is(tok::at) && Current.Next &&
1132 Style.Language != FormatStyle::LK_JavaScript &&
1133 Style.Language != FormatStyle::LK_Java) {
1134 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it
1135 // marks declarations and properties that need special formatting.
1136 switch (Current.Next->Tok.getObjCKeywordID()) {
1137 case tok::objc_interface:
1138 case tok::objc_implementation:
1139 case tok::objc_protocol:
1140 Current.Type = TT_ObjCDecl;
1142 case tok::objc_property:
1143 Current.Type = TT_ObjCProperty;
1148 } else if (Current.is(tok::period)) {
1149 FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1150 if (PreviousNoComment &&
1151 PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1152 Current.Type = TT_DesignatedInitializerPeriod;
1153 else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1154 Current.Previous->isOneOf(TT_JavaAnnotation,
1155 TT_LeadingJavaAnnotation)) {
1156 Current.Type = Current.Previous->Type;
1158 } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
1160 !Current.Previous->isOneOf(tok::equal, tok::at) &&
1161 Line.MightBeFunctionDecl && Contexts.size() == 1) {
1162 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1163 // function declaration have been found.
1164 Current.Type = TT_TrailingAnnotation;
1165 } else if ((Style.Language == FormatStyle::LK_Java ||
1166 Style.Language == FormatStyle::LK_JavaScript) &&
1168 if (Current.Previous->is(tok::at) &&
1169 Current.isNot(Keywords.kw_interface)) {
1170 const FormatToken &AtToken = *Current.Previous;
1171 const FormatToken *Previous = AtToken.getPreviousNonComment();
1172 if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1173 Current.Type = TT_LeadingJavaAnnotation;
1175 Current.Type = TT_JavaAnnotation;
1176 } else if (Current.Previous->is(tok::period) &&
1177 Current.Previous->isOneOf(TT_JavaAnnotation,
1178 TT_LeadingJavaAnnotation)) {
1179 Current.Type = Current.Previous->Type;
1184 /// \brief Take a guess at whether \p Tok starts a name of a function or
1185 /// variable declaration.
1187 /// This is a heuristic based on whether \p Tok is an identifier following
1188 /// something that is likely a type.
1189 bool isStartOfName(const FormatToken &Tok) {
1190 if (Tok.isNot(tok::identifier) || !Tok.Previous)
1193 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof,
1196 if (Style.Language == FormatStyle::LK_JavaScript &&
1197 Tok.Previous->is(Keywords.kw_in))
1200 // Skip "const" as it does not have an influence on whether this is a name.
1201 FormatToken *PreviousNotConst = Tok.getPreviousNonComment();
1202 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1203 PreviousNotConst = PreviousNotConst->getPreviousNonComment();
1205 if (!PreviousNotConst)
1208 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1209 PreviousNotConst->Previous &&
1210 PreviousNotConst->Previous->is(tok::hash);
1212 if (PreviousNotConst->is(TT_TemplateCloser))
1213 return PreviousNotConst && PreviousNotConst->MatchingParen &&
1214 PreviousNotConst->MatchingParen->Previous &&
1215 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1216 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1218 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1219 PreviousNotConst->MatchingParen->Previous &&
1220 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1223 return (!IsPPKeyword &&
1224 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1225 PreviousNotConst->is(TT_PointerOrReference) ||
1226 PreviousNotConst->isSimpleTypeSpecifier();
1229 /// \brief Determine whether ')' is ending a cast.
1230 bool rParenEndsCast(const FormatToken &Tok) {
1231 // C-style casts are only used in C++ and Java.
1232 if (!Style.isCpp() && Style.Language != FormatStyle::LK_Java)
1235 // Empty parens aren't casts and there are no casts at the end of the line.
1236 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1239 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1241 // If there is a closing parenthesis left of the current parentheses,
1242 // look past it as these might be chained casts.
1243 if (LeftOfParens->is(tok::r_paren)) {
1244 if (!LeftOfParens->MatchingParen ||
1245 !LeftOfParens->MatchingParen->Previous)
1247 LeftOfParens = LeftOfParens->MatchingParen->Previous;
1250 // If there is an identifier (or with a few exceptions a keyword) right
1251 // before the parentheses, this is unlikely to be a cast.
1252 if (LeftOfParens->Tok.getIdentifierInfo() &&
1253 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1257 // Certain other tokens right before the parentheses are also signals that
1258 // this cannot be a cast.
1259 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1260 TT_TemplateCloser, tok::ellipsis))
1264 if (Tok.Next->is(tok::question))
1267 // As Java has no function types, a "(" after the ")" likely means that this
1269 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1272 // If a (non-string) literal follows, this is likely a cast.
1273 if (Tok.Next->isNot(tok::string_literal) &&
1274 (Tok.Next->Tok.isLiteral() ||
1275 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1278 // Heuristically try to determine whether the parentheses contain a type.
1279 bool ParensAreType =
1281 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1282 Tok.Previous->isSimpleTypeSpecifier();
1283 bool ParensCouldEndDecl =
1284 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1285 if (ParensAreType && !ParensCouldEndDecl)
1288 // At this point, we heuristically assume that there are no casts at the
1289 // start of the line. We assume that we have found most cases where there
1290 // are by the logic above, e.g. "(void)x;".
1294 // Certain token types inside the parentheses mean that this can't be a
1296 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok;
1297 Token = Token->Next)
1298 if (Token->is(TT_BinaryOperator))
1301 // If the following token is an identifier or 'this', this is a cast. All
1302 // cases where this can be something else are handled above.
1303 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this))
1306 if (!Tok.Next->Next)
1309 // If the next token after the parenthesis is a unary operator, assume
1310 // that this is cast, unless there are unexpected tokens inside the
1313 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1314 if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1315 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1317 // Search for unexpected tokens.
1318 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1319 Prev = Prev->Previous) {
1320 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1326 /// \brief Return the type of the given token assuming it is * or &.
1327 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1328 bool InTemplateArgument) {
1329 if (Style.Language == FormatStyle::LK_JavaScript)
1330 return TT_BinaryOperator;
1332 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1334 return TT_UnaryOperator;
1336 const FormatToken *NextToken = Tok.getNextNonComment();
1338 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_const) ||
1339 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1340 return TT_PointerOrReference;
1342 if (PrevToken->is(tok::coloncolon))
1343 return TT_PointerOrReference;
1345 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1346 tok::comma, tok::semi, tok::kw_return, tok::colon,
1347 tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1348 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1349 TT_UnaryOperator, TT_CastRParen))
1350 return TT_UnaryOperator;
1352 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1353 return TT_PointerOrReference;
1354 if (NextToken->is(tok::kw_operator) && !IsExpression)
1355 return TT_PointerOrReference;
1356 if (NextToken->isOneOf(tok::comma, tok::semi))
1357 return TT_PointerOrReference;
1359 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1360 PrevToken->MatchingParen->Previous &&
1361 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1363 return TT_PointerOrReference;
1365 if (PrevToken->Tok.isLiteral() ||
1366 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1367 tok::kw_false, tok::r_brace) ||
1368 NextToken->Tok.isLiteral() ||
1369 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1370 NextToken->isUnaryOperator() ||
1371 // If we know we're in a template argument, there are no named
1372 // declarations. Thus, having an identifier on the right-hand side
1373 // indicates a binary operator.
1374 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1375 return TT_BinaryOperator;
1377 // "&&(" is quite unlikely to be two successive unary "&".
1378 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1379 return TT_BinaryOperator;
1381 // This catches some cases where evaluation order is used as control flow:
1383 const FormatToken *NextNextToken = NextToken->getNextNonComment();
1384 if (NextNextToken && NextNextToken->is(tok::arrow))
1385 return TT_BinaryOperator;
1387 // It is very unlikely that we are going to find a pointer or reference type
1388 // definition on the RHS of an assignment.
1389 if (IsExpression && !Contexts.back().CaretFound)
1390 return TT_BinaryOperator;
1392 return TT_PointerOrReference;
1395 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1396 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1398 return TT_UnaryOperator;
1400 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator) &&
1401 !PrevToken->is(tok::exclaim))
1402 // There aren't any trailing unary operators except for TypeScript's
1403 // non-null operator (!). Thus, this must be squence of leading operators.
1404 return TT_UnaryOperator;
1406 // Use heuristics to recognize unary operators.
1407 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1408 tok::question, tok::colon, tok::kw_return,
1409 tok::kw_case, tok::at, tok::l_brace))
1410 return TT_UnaryOperator;
1412 // There can't be two consecutive binary operators.
1413 if (PrevToken->is(TT_BinaryOperator))
1414 return TT_UnaryOperator;
1416 // Fall back to marking the token as binary operator.
1417 return TT_BinaryOperator;
1420 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1421 TokenType determineIncrementUsage(const FormatToken &Tok) {
1422 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1423 if (!PrevToken || PrevToken->is(TT_CastRParen))
1424 return TT_UnaryOperator;
1425 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1426 return TT_TrailingUnaryOperator;
1428 return TT_UnaryOperator;
1431 SmallVector<Context, 8> Contexts;
1433 const FormatStyle &Style;
1434 AnnotatedLine &Line;
1435 FormatToken *CurrentToken;
1437 const AdditionalKeywords &Keywords;
1439 // Set of "<" tokens that do not open a template parameter list. If parseAngle
1440 // determines that a specific token can't be a template opener, it will make
1441 // same decision irrespective of the decisions for tokens leading up to it.
1442 // Store this information to prevent this from causing exponential runtime.
1443 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1446 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1447 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1449 /// \brief Parses binary expressions by inserting fake parenthesis based on
1450 /// operator precedence.
1451 class ExpressionParser {
1453 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1454 AnnotatedLine &Line)
1455 : Style(Style), Keywords(Keywords), Current(Line.First) {}
1457 /// \brief Parse expressions with the given operatore precedence.
1458 void parse(int Precedence = 0) {
1459 // Skip 'return' and ObjC selector colons as they are not part of a binary
1461 while (Current && (Current->is(tok::kw_return) ||
1462 (Current->is(tok::colon) &&
1463 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1466 if (!Current || Precedence > PrecedenceArrowAndPeriod)
1469 // Conditional expressions need to be parsed separately for proper nesting.
1470 if (Precedence == prec::Conditional) {
1471 parseConditionalExpr();
1475 // Parse unary operators, which all have a higher precedence than binary
1477 if (Precedence == PrecedenceUnaryOperator) {
1478 parseUnaryOperator();
1482 FormatToken *Start = Current;
1483 FormatToken *LatestOperator = nullptr;
1484 unsigned OperatorIndex = 0;
1487 // Consume operators with higher precedence.
1488 parse(Precedence + 1);
1490 int CurrentPrecedence = getCurrentPrecedence();
1492 if (Current && Current->is(TT_SelectorName) &&
1493 Precedence == CurrentPrecedence) {
1495 addFakeParenthesis(Start, prec::Level(Precedence));
1499 // At the end of the line or when an operator with higher precedence is
1500 // found, insert fake parenthesis and return.
1502 (Current->closesScope() &&
1503 (Current->MatchingParen || Current->is(TT_TemplateString))) ||
1504 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1505 (CurrentPrecedence == prec::Conditional &&
1506 Precedence == prec::Assignment && Current->is(tok::colon))) {
1510 // Consume scopes: (), [], <> and {}
1511 if (Current->opensScope()) {
1512 // In fragment of a JavaScript template string can look like '}..${' and
1513 // thus close a scope and open a new one at the same time.
1514 while (Current && (!Current->closesScope() || Current->opensScope())) {
1521 if (CurrentPrecedence == Precedence) {
1523 LatestOperator->NextOperator = Current;
1524 LatestOperator = Current;
1525 Current->OperatorIndex = OperatorIndex;
1528 next(/*SkipPastLeadingComments=*/Precedence > 0);
1532 if (LatestOperator && (Current || Precedence > 0)) {
1533 // LatestOperator->LastOperator = true;
1534 if (Precedence == PrecedenceArrowAndPeriod) {
1535 // Call expressions don't have a binary operator precedence.
1536 addFakeParenthesis(Start, prec::Unknown);
1538 addFakeParenthesis(Start, prec::Level(Precedence));
1544 /// \brief Gets the precedence (+1) of the given token for binary operators
1545 /// and other tokens that we treat like binary operators.
1546 int getCurrentPrecedence() {
1548 const FormatToken *NextNonComment = Current->getNextNonComment();
1549 if (Current->is(TT_ConditionalExpr))
1550 return prec::Conditional;
1551 if (NextNonComment && NextNonComment->is(tok::colon) &&
1552 NextNonComment->is(TT_DictLiteral))
1553 return prec::Assignment;
1554 if (Current->is(TT_JsComputedPropertyName))
1555 return prec::Assignment;
1556 if (Current->is(TT_LambdaArrow))
1558 if (Current->is(TT_JsFatArrow))
1559 return prec::Assignment;
1560 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) ||
1561 (Current->is(tok::comment) && NextNonComment &&
1562 NextNonComment->is(TT_SelectorName)))
1564 if (Current->is(TT_RangeBasedForLoopColon))
1566 if ((Style.Language == FormatStyle::LK_Java ||
1567 Style.Language == FormatStyle::LK_JavaScript) &&
1568 Current->is(Keywords.kw_instanceof))
1569 return prec::Relational;
1570 if (Style.Language == FormatStyle::LK_JavaScript &&
1571 Current->isOneOf(Keywords.kw_in, Keywords.kw_as))
1572 return prec::Relational;
1573 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1574 return Current->getPrecedence();
1575 if (Current->isOneOf(tok::period, tok::arrow))
1576 return PrecedenceArrowAndPeriod;
1577 if ((Style.Language == FormatStyle::LK_Java ||
1578 Style.Language == FormatStyle::LK_JavaScript) &&
1579 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1580 Keywords.kw_throws))
1586 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1587 Start->FakeLParens.push_back(Precedence);
1588 if (Precedence > prec::Unknown)
1589 Start->StartsBinaryExpression = true;
1591 FormatToken *Previous = Current->Previous;
1592 while (Previous->is(tok::comment) && Previous->Previous)
1593 Previous = Previous->Previous;
1594 ++Previous->FakeRParens;
1595 if (Precedence > prec::Unknown)
1596 Previous->EndsBinaryExpression = true;
1600 /// \brief Parse unary operator expressions and surround them with fake
1601 /// parentheses if appropriate.
1602 void parseUnaryOperator() {
1603 if (!Current || Current->isNot(TT_UnaryOperator)) {
1604 parse(PrecedenceArrowAndPeriod);
1608 FormatToken *Start = Current;
1610 parseUnaryOperator();
1612 // The actual precedence doesn't matter.
1613 addFakeParenthesis(Start, prec::Unknown);
1616 void parseConditionalExpr() {
1617 while (Current && Current->isTrailingComment()) {
1620 FormatToken *Start = Current;
1621 parse(prec::LogicalOr);
1622 if (!Current || !Current->is(tok::question))
1625 parse(prec::Assignment);
1626 if (!Current || Current->isNot(TT_ConditionalExpr))
1629 parse(prec::Assignment);
1630 addFakeParenthesis(Start, prec::Conditional);
1633 void next(bool SkipPastLeadingComments = true) {
1635 Current = Current->Next;
1637 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1638 Current->isTrailingComment())
1639 Current = Current->Next;
1642 const FormatStyle &Style;
1643 const AdditionalKeywords &Keywords;
1644 FormatToken *Current;
1647 } // end anonymous namespace
1649 void TokenAnnotator::setCommentLineLevels(
1650 SmallVectorImpl<AnnotatedLine *> &Lines) {
1651 const AnnotatedLine *NextNonCommentLine = nullptr;
1652 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1655 bool CommentLine = (*I)->First;
1656 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) {
1657 if (!Tok->is(tok::comment)) {
1658 CommentLine = false;
1662 if (NextNonCommentLine && CommentLine)
1663 (*I)->Level = NextNonCommentLine->Level;
1665 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1667 setCommentLineLevels((*I)->Children);
1671 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
1672 unsigned Result = 0;
1673 for (const auto* Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
1674 Result = std::max(Result, Tok->NestingLevel);
1678 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1679 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1680 E = Line.Children.end();
1684 AnnotatingParser Parser(Style, Line, Keywords);
1685 Line.Type = Parser.parseLine();
1687 // With very deep nesting, ExpressionParser uses lots of stack and the
1688 // formatting algorithm is very slow. We're not going to do a good job here
1689 // anyway - it's probably generated code being formatted by mistake.
1690 // Just skip the whole line.
1691 if (maxNestingDepth(Line) > 50)
1692 Line.Type = LT_Invalid;
1694 if (Line.Type == LT_Invalid)
1697 ExpressionParser ExprParser(Style, Keywords, Line);
1700 if (Line.startsWith(TT_ObjCMethodSpecifier))
1701 Line.Type = LT_ObjCMethodDecl;
1702 else if (Line.startsWith(TT_ObjCDecl))
1703 Line.Type = LT_ObjCDecl;
1704 else if (Line.startsWith(TT_ObjCProperty))
1705 Line.Type = LT_ObjCProperty;
1707 Line.First->SpacesRequiredBefore = 1;
1708 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1711 // This function heuristically determines whether 'Current' starts the name of a
1712 // function declaration.
1713 static bool isFunctionDeclarationName(const FormatToken &Current,
1714 const AnnotatedLine &Line) {
1715 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1716 for (; Next; Next = Next->Next) {
1717 if (Next->is(TT_OverloadedOperatorLParen))
1719 if (Next->is(TT_OverloadedOperator))
1721 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1722 // For 'new[]' and 'delete[]'.
1723 if (Next->Next && Next->Next->is(tok::l_square) &&
1724 Next->Next->Next && Next->Next->Next->is(tok::r_square))
1725 Next = Next->Next->Next;
1734 // Find parentheses of parameter list.
1735 const FormatToken *Next = Current.Next;
1736 if (Current.is(tok::kw_operator)) {
1737 if (Current.Previous && Current.Previous->is(tok::coloncolon))
1739 Next = skipOperatorName(Next);
1741 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1743 for (; Next; Next = Next->Next) {
1744 if (Next->is(TT_TemplateOpener)) {
1745 Next = Next->MatchingParen;
1746 } else if (Next->is(tok::coloncolon)) {
1750 if (Next->is(tok::kw_operator)) {
1751 Next = skipOperatorName(Next->Next);
1754 if (!Next->is(tok::identifier))
1756 } else if (Next->is(tok::l_paren)) {
1764 // Check whether parameter list can belong to a function declaration.
1765 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
1767 // If the lines ends with "{", this is likely an function definition.
1768 if (Line.Last->is(tok::l_brace))
1770 if (Next->Next == Next->MatchingParen)
1771 return true; // Empty parentheses.
1772 // If there is an &/&& after the r_paren, this is likely a function.
1773 if (Next->MatchingParen->Next &&
1774 Next->MatchingParen->Next->is(TT_PointerOrReference))
1776 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1778 if (Tok->is(tok::l_paren) && Tok->MatchingParen) {
1779 Tok = Tok->MatchingParen;
1782 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1783 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
1785 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1786 Tok->Tok.isLiteral())
1792 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1793 assert(Line.MightBeFunctionDecl);
1795 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1796 Style.AlwaysBreakAfterReturnType ==
1797 FormatStyle::RTBS_TopLevelDefinitions) &&
1801 switch (Style.AlwaysBreakAfterReturnType) {
1802 case FormatStyle::RTBS_None:
1804 case FormatStyle::RTBS_All:
1805 case FormatStyle::RTBS_TopLevel:
1807 case FormatStyle::RTBS_AllDefinitions:
1808 case FormatStyle::RTBS_TopLevelDefinitions:
1809 return Line.mightBeFunctionDefinition();
1815 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1816 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1817 E = Line.Children.end();
1819 calculateFormattingInformation(**I);
1822 Line.First->TotalLength =
1823 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1824 FormatToken *Current = Line.First->Next;
1825 bool InFunctionDecl = Line.MightBeFunctionDecl;
1827 if (isFunctionDeclarationName(*Current, Line))
1828 Current->Type = TT_FunctionDeclarationName;
1829 if (Current->is(TT_LineComment)) {
1830 if (Current->Previous->BlockKind == BK_BracedInit &&
1831 Current->Previous->opensScope())
1832 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1834 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1836 // If we find a trailing comment, iterate backwards to determine whether
1837 // it seems to relate to a specific parameter. If so, break before that
1838 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1839 // to the previous line in:
1843 if (!Current->HasUnescapedNewline) {
1844 for (FormatToken *Parameter = Current->Previous; Parameter;
1845 Parameter = Parameter->Previous) {
1846 if (Parameter->isOneOf(tok::comment, tok::r_brace))
1848 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1849 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1850 Parameter->HasUnescapedNewline)
1851 Parameter->MustBreakBefore = true;
1856 } else if (Current->SpacesRequiredBefore == 0 &&
1857 spaceRequiredBefore(Line, *Current)) {
1858 Current->SpacesRequiredBefore = 1;
1861 Current->MustBreakBefore =
1862 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1864 if (!Current->MustBreakBefore && InFunctionDecl &&
1865 Current->is(TT_FunctionDeclarationName))
1866 Current->MustBreakBefore = mustBreakForReturnType(Line);
1868 Current->CanBreakBefore =
1869 Current->MustBreakBefore || canBreakBefore(Line, *Current);
1870 unsigned ChildSize = 0;
1871 if (Current->Previous->Children.size() == 1) {
1872 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1873 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1874 : LastOfChild.TotalLength + 1;
1876 const FormatToken *Prev = Current->Previous;
1877 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1878 (Prev->Children.size() == 1 &&
1879 Prev->Children[0]->First->MustBreakBefore) ||
1880 Current->IsMultiline)
1881 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1883 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1884 ChildSize + Current->SpacesRequiredBefore;
1886 if (Current->is(TT_CtorInitializerColon))
1887 InFunctionDecl = false;
1889 // FIXME: Only calculate this if CanBreakBefore is true once static
1890 // initializers etc. are sorted out.
1891 // FIXME: Move magic numbers to a better place.
1892 Current->SplitPenalty = 20 * Current->BindingStrength +
1893 splitPenalty(Line, *Current, InFunctionDecl);
1895 Current = Current->Next;
1898 calculateUnbreakableTailLengths(Line);
1899 unsigned IndentLevel = Line.Level;
1900 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1902 Current->Role->precomputeFormattingInfos(Current);
1903 if (Current->MatchingParen &&
1904 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) {
1905 assert(IndentLevel > 0);
1908 Current->IndentLevel = IndentLevel;
1909 if (Current->opensBlockOrBlockTypeList(Style))
1913 DEBUG({ printDebugInfo(Line); });
1916 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1917 unsigned UnbreakableTailLength = 0;
1918 FormatToken *Current = Line.Last;
1920 Current->UnbreakableTailLength = UnbreakableTailLength;
1921 if (Current->CanBreakBefore ||
1922 Current->isOneOf(tok::comment, tok::string_literal)) {
1923 UnbreakableTailLength = 0;
1925 UnbreakableTailLength +=
1926 Current->ColumnWidth + Current->SpacesRequiredBefore;
1928 Current = Current->Previous;
1932 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1933 const FormatToken &Tok,
1934 bool InFunctionDecl) {
1935 const FormatToken &Left = *Tok.Previous;
1936 const FormatToken &Right = Tok;
1938 if (Left.is(tok::semi))
1941 if (Style.Language == FormatStyle::LK_Java) {
1942 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1944 if (Right.is(Keywords.kw_implements))
1946 if (Left.is(tok::comma) && Left.NestingLevel == 0)
1948 } else if (Style.Language == FormatStyle::LK_JavaScript) {
1949 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1951 if (Left.is(TT_JsTypeColon))
1953 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
1954 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
1958 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
1960 if (Right.is(tok::l_square)) {
1961 if (Style.Language == FormatStyle::LK_Proto)
1963 if (Left.is(tok::r_square))
1965 // Slightly prefer formatting local lambda definitions like functions.
1966 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1968 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1969 TT_ArrayInitializerLSquare))
1973 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1974 Right.is(tok::kw_operator)) {
1975 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1977 if (Left.is(TT_StartOfName))
1979 if (InFunctionDecl && Right.NestingLevel == 0)
1980 return Style.PenaltyReturnTypeOnItsOwnLine;
1983 if (Right.is(TT_PointerOrReference))
1985 if (Right.is(TT_LambdaArrow))
1987 if (Left.is(tok::equal) && Right.is(tok::l_brace))
1989 if (Left.is(TT_CastRParen))
1991 if (Left.is(tok::coloncolon) ||
1992 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1994 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1996 if (Left.is(tok::comment))
1999 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
2002 if (Right.isMemberAccess()) {
2003 // Breaking before the "./->" of a chained call/member access is reasonably
2004 // cheap, as formatting those with one call per line is generally
2005 // desirable. In particular, it should be cheaper to break before the call
2006 // than it is to break inside a call's parameters, which could lead to weird
2007 // "hanging" indents. The exception is the very last "./->" to support this
2008 // frequent pattern:
2010 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
2013 // which might otherwise be blown up onto many lines. Here, clang-format
2014 // won't produce "hanging" indents anyway as there is no other trailing
2017 // Also apply higher penalty is not a call as that might lead to a wrapping
2021 // .aaaaaaaaa.bbbbbbbb(cccccccc);
2022 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
2027 if (Right.is(TT_TrailingAnnotation) &&
2028 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
2029 // Moving trailing annotations to the next line is fine for ObjC method
2031 if (Line.startsWith(TT_ObjCMethodSpecifier))
2033 // Generally, breaking before a trailing annotation is bad unless it is
2034 // function-like. It seems to be especially preferable to keep standard
2035 // annotations (i.e. "const", "final" and "override") on the same line.
2036 // Use a slightly higher penalty after ")" so that annotations like
2037 // "const override" are kept together.
2038 bool is_short_annotation = Right.TokenText.size() < 10;
2039 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
2042 // In for-loops, prefer breaking at ',' and ';'.
2043 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
2046 // In Objective-C method expressions, prefer breaking before "param:" over
2047 // breaking after it.
2048 if (Right.is(TT_SelectorName))
2050 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
2051 return Line.MightBeFunctionDecl ? 50 : 500;
2053 if (Left.is(tok::l_paren) && InFunctionDecl &&
2054 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
2056 if (Left.is(tok::l_paren) && Left.Previous &&
2057 Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
2059 if (Left.is(tok::equal) && InFunctionDecl)
2061 if (Right.is(tok::r_brace))
2063 if (Left.is(TT_TemplateOpener))
2065 if (Left.opensScope()) {
2066 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
2068 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
2071 if (Left.is(TT_JavaAnnotation))
2074 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous &&
2075 Left.Previous->isLabelString() &&
2076 (Left.NextOperator || Left.OperatorIndex != 0))
2078 if (Right.is(tok::plus) && Left.isLabelString() &&
2079 (Right.NextOperator || Right.OperatorIndex != 0))
2081 if (Left.is(tok::comma))
2083 if (Right.is(tok::lessless) && Left.isLabelString() &&
2084 (Right.NextOperator || Right.OperatorIndex != 1))
2086 if (Right.is(tok::lessless)) {
2087 // Breaking at a << is really cheap.
2088 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0)
2089 // Slightly prefer to break before the first one in log-like statements.
2093 if (Left.is(TT_ConditionalExpr))
2094 return prec::Conditional;
2095 prec::Level Level = Left.getPrecedence();
2096 if (Level == prec::Unknown)
2097 Level = Right.getPrecedence();
2098 if (Level == prec::Assignment)
2099 return Style.PenaltyBreakAssignment;
2100 if (Level != prec::Unknown)
2106 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2107 const FormatToken &Left,
2108 const FormatToken &Right) {
2109 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2111 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
2112 Left.Tok.getObjCKeywordID() == tok::objc_property)
2114 if (Right.is(tok::hashhash))
2115 return Left.is(tok::hash);
2116 if (Left.isOneOf(tok::hashhash, tok::hash))
2117 return Right.is(tok::hash);
2118 if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
2119 return Style.SpaceInEmptyParentheses;
2120 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2121 return (Right.is(TT_CastRParen) ||
2122 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2123 ? Style.SpacesInCStyleCastParentheses
2124 : Style.SpacesInParentheses;
2125 if (Right.isOneOf(tok::semi, tok::comma))
2127 if (Right.is(tok::less) &&
2128 Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)
2130 if (Right.is(tok::less) && Left.is(tok::kw_template))
2131 return Style.SpaceAfterTemplateKeyword;
2132 if (Left.isOneOf(tok::exclaim, tok::tilde))
2134 if (Left.is(tok::at) &&
2135 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2136 tok::numeric_constant, tok::l_paren, tok::l_brace,
2137 tok::kw_true, tok::kw_false))
2139 if (Left.is(tok::colon))
2140 return !Left.is(TT_ObjCMethodExpr);
2141 if (Left.is(tok::coloncolon))
2143 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
2145 if (Right.is(tok::ellipsis))
2146 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2147 Left.Previous->is(tok::kw_case));
2148 if (Left.is(tok::l_square) && Right.is(tok::amp))
2150 if (Right.is(TT_PointerOrReference))
2151 return (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) ||
2152 (Left.Tok.isLiteral() || (Left.is(tok::kw_const) && Left.Previous &&
2153 Left.Previous->is(tok::r_paren)) ||
2154 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
2155 (Style.PointerAlignment != FormatStyle::PAS_Left ||
2156 (Line.IsMultiVariableDeclStmt &&
2157 (Left.NestingLevel == 0 ||
2158 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
2159 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
2160 (!Left.is(TT_PointerOrReference) ||
2161 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2162 !Line.IsMultiVariableDeclStmt)))
2164 if (Left.is(TT_PointerOrReference))
2165 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
2166 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) &&
2167 !Right.is(TT_StartOfName)) ||
2168 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
2169 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
2171 (Style.PointerAlignment != FormatStyle::PAS_Right &&
2172 !Line.IsMultiVariableDeclStmt) &&
2174 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
2175 if (Right.is(tok::star) && Left.is(tok::l_paren))
2177 if (Left.is(tok::l_square))
2178 return (Left.is(TT_ArrayInitializerLSquare) &&
2179 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
2180 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
2181 Right.isNot(tok::r_square));
2182 if (Right.is(tok::r_square))
2183 return Right.MatchingParen &&
2184 ((Style.SpacesInContainerLiterals &&
2185 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
2186 (Style.SpacesInSquareBrackets &&
2187 Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
2188 if (Right.is(tok::l_square) &&
2189 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
2190 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
2192 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
2193 return !Left.Children.empty(); // No spaces in "{}".
2194 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
2195 (Right.is(tok::r_brace) && Right.MatchingParen &&
2196 Right.MatchingParen->BlockKind != BK_Block))
2197 return !Style.Cpp11BracedListStyle;
2198 if (Left.is(TT_BlockComment))
2199 return !Left.TokenText.endswith("=*/");
2200 if (Right.is(tok::l_paren)) {
2201 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
2203 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
2204 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
2205 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
2206 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
2208 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
2209 tok::kw_new, tok::kw_delete) &&
2210 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
2211 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
2212 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
2213 Left.is(tok::r_paren)) &&
2214 Line.Type != LT_PreprocessorDirective);
2216 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
2218 if (Right.is(TT_UnaryOperator))
2219 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
2220 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
2221 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
2223 Left.isSimpleTypeSpecifier()) &&
2224 Right.is(tok::l_brace) && Right.getNextNonComment() &&
2225 Right.BlockKind != BK_Block)
2227 if (Left.is(tok::period) || Right.is(tok::period))
2229 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
2231 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
2232 Left.MatchingParen->Previous &&
2233 Left.MatchingParen->Previous->is(tok::period))
2234 // A.<B>DoSomething();
2236 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
2241 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
2242 const FormatToken &Right) {
2243 const FormatToken &Left = *Right.Previous;
2244 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
2245 return true; // Never ever merge two identifiers.
2246 if (Style.isCpp()) {
2247 if (Left.is(tok::kw_operator))
2248 return Right.is(tok::coloncolon);
2249 } else if (Style.Language == FormatStyle::LK_Proto) {
2250 if (Right.is(tok::period) &&
2251 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
2252 Keywords.kw_repeated, Keywords.kw_extend))
2254 if (Right.is(tok::l_paren) &&
2255 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
2257 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2258 if (Left.is(TT_JsFatArrow))
2261 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) &&
2262 Left.Previous && Left.Previous->is(tok::kw_for))
2264 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) &&
2265 Right.MatchingParen) {
2266 const FormatToken *Next = Right.MatchingParen->getNextNonComment();
2267 // An async arrow function, for example: `x = async () => foo();`,
2268 // as opposed to calling a function called async: `x = async();`
2269 if (Next && Next->is(TT_JsFatArrow))
2272 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2273 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2275 if (Left.is(tok::identifier) && Right.is(TT_TemplateString))
2277 if (Right.is(tok::star) &&
2278 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2280 if (Right.isOneOf(tok::l_brace, tok::l_square) &&
2281 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
2283 // JS methods can use some keywords as names (e.g. `delete()`).
2284 if (Right.is(tok::l_paren) && Line.MustBeDeclaration &&
2285 Left.Tok.getIdentifierInfo())
2287 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
2289 // "of" is only a keyword if it appears after another identifier
2290 // (e.g. as "const x of y" in a for loop).
2291 (Left.is(Keywords.kw_of) && Left.Previous &&
2292 Left.Previous->Tok.getIdentifierInfo())) &&
2293 (!Left.Previous || !Left.Previous->is(tok::period)))
2295 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
2296 Left.Previous->is(tok::period) && Right.is(tok::l_paren))
2298 if (Left.is(Keywords.kw_as) &&
2299 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
2301 if (Left.is(tok::kw_default) && Left.Previous &&
2302 Left.Previous->is(tok::kw_export))
2304 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2306 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2308 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
2310 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2311 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2313 if (Left.is(tok::ellipsis))
2315 if (Left.is(TT_TemplateCloser) &&
2316 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2317 Keywords.kw_implements, Keywords.kw_extends))
2318 // Type assertions ('<type>expr') are not followed by whitespace. Other
2319 // locations that should have whitespace following are identified by the
2320 // above set of follower tokens.
2322 if (Right.is(TT_JsNonNullAssertion))
2324 if (Left.is(TT_JsNonNullAssertion) && Right.is(Keywords.kw_as))
2325 return true; // "x! as string"
2326 } else if (Style.Language == FormatStyle::LK_Java) {
2327 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2329 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2330 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2331 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2332 tok::kw_protected) ||
2333 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2334 Keywords.kw_native)) &&
2335 Right.is(TT_TemplateOpener))
2338 if (Left.is(TT_ImplicitStringLiteral))
2339 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2340 if (Line.Type == LT_ObjCMethodDecl) {
2341 if (Left.is(TT_ObjCMethodSpecifier))
2343 if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2344 // Don't space between ')' and <id>
2347 if (Line.Type == LT_ObjCProperty &&
2348 (Right.is(tok::equal) || Left.is(tok::equal)))
2351 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2352 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2354 if (Right.is(TT_OverloadedOperatorLParen))
2355 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2356 if (Left.is(tok::comma))
2358 if (Right.is(tok::comma))
2360 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2362 if (Right.is(tok::colon)) {
2363 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2364 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2366 if (Right.is(TT_ObjCMethodExpr))
2368 if (Left.is(tok::question))
2370 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2372 if (Right.is(TT_DictLiteral))
2373 return Style.SpacesInContainerLiterals;
2376 if (Left.is(TT_UnaryOperator))
2377 return Right.is(TT_BinaryOperator);
2379 // If the next token is a binary operator or a selector name, we have
2380 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2381 if (Left.is(TT_CastRParen))
2382 return Style.SpaceAfterCStyleCast ||
2383 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2385 if (Left.is(tok::greater) && Right.is(tok::greater))
2386 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2387 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2388 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2389 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2391 if (!Style.SpaceBeforeAssignmentOperators &&
2392 Right.getPrecedence() == prec::Assignment)
2394 if (Right.is(tok::coloncolon) && Left.is(tok::identifier))
2395 // Generally don't remove existing spaces between an identifier and "::".
2396 // The identifier might actually be a macro name such as ALWAYS_INLINE. If
2397 // this turns out to be too lenient, add analysis of the identifier itself.
2398 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2399 if (Right.is(tok::coloncolon) && !Left.isOneOf(tok::l_brace, tok::comment))
2400 return (Left.is(TT_TemplateOpener) &&
2401 Style.Standard == FormatStyle::LS_Cpp03) ||
2402 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square,
2403 tok::kw___super, TT_TemplateCloser, TT_TemplateOpener));
2404 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2405 return Style.SpacesInAngles;
2406 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2407 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2408 !Right.is(tok::r_paren)))
2410 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2411 Right.isNot(TT_FunctionTypeLParen))
2412 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2413 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2414 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2416 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2417 Line.startsWith(tok::hash))
2419 if (Right.is(TT_TrailingUnaryOperator))
2421 if (Left.is(TT_RegexLiteral))
2423 return spaceRequiredBetween(Line, Left, Right);
2426 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2427 static bool isAllmanBrace(const FormatToken &Tok) {
2428 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2429 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2432 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2433 const FormatToken &Right) {
2434 const FormatToken &Left = *Right.Previous;
2435 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2438 if (Style.Language == FormatStyle::LK_JavaScript) {
2439 // FIXME: This might apply to other languages and token kinds.
2440 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
2441 Left.Previous->is(tok::string_literal))
2443 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2444 Left.Previous && Left.Previous->is(tok::equal) &&
2445 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2447 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2449 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2450 // Object literals on the top level of a file are treated as "enum-style".
2451 // Each key/value pair is put on a separate line, instead of bin-packing.
2453 if (Left.is(tok::l_brace) && Line.Level == 0 &&
2454 (Line.startsWith(tok::kw_enum) ||
2455 Line.startsWith(tok::kw_export, tok::kw_enum)))
2456 // JavaScript top-level enum key/value pairs are put on separate lines
2457 // instead of bin-packing.
2459 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2460 !Left.Children.empty())
2461 // Support AllowShortFunctionsOnASingleLine for JavaScript.
2462 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2463 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2464 (Left.NestingLevel == 0 && Line.Level == 0 &&
2465 Style.AllowShortFunctionsOnASingleLine ==
2466 FormatStyle::SFS_Inline);
2467 } else if (Style.Language == FormatStyle::LK_Java) {
2468 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2469 Right.Next->is(tok::string_literal))
2471 } else if (Style.Language == FormatStyle::LK_Cpp ||
2472 Style.Language == FormatStyle::LK_ObjC ||
2473 Style.Language == FormatStyle::LK_Proto) {
2474 if (Left.isStringLiteral() && Right.isStringLiteral())
2478 // If the last token before a '}', ']', or ')' is a comma or a trailing
2479 // comment, the intention is to insert a line break after it in order to make
2480 // shuffling around entries easier. Import statements, especially in
2481 // JavaScript, can be an exception to this rule.
2482 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) {
2483 const FormatToken *BeforeClosingBrace = nullptr;
2484 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
2485 (Style.Language == FormatStyle::LK_JavaScript &&
2486 Left.is(tok::l_paren))) &&
2487 Left.BlockKind != BK_Block && Left.MatchingParen)
2488 BeforeClosingBrace = Left.MatchingParen->Previous;
2489 else if (Right.MatchingParen &&
2490 (Right.MatchingParen->isOneOf(tok::l_brace,
2491 TT_ArrayInitializerLSquare) ||
2492 (Style.Language == FormatStyle::LK_JavaScript &&
2493 Right.MatchingParen->is(tok::l_paren))))
2494 BeforeClosingBrace = &Left;
2495 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2496 BeforeClosingBrace->isTrailingComment()))
2500 if (Right.is(tok::comment))
2501 return Left.BlockKind != BK_BracedInit &&
2502 Left.isNot(TT_CtorInitializerColon) &&
2503 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2504 if (Left.isTrailingComment())
2506 if (Right.Previous->IsUnterminatedLiteral)
2508 if (Right.is(tok::lessless) && Right.Next &&
2509 Right.Previous->is(tok::string_literal) &&
2510 Right.Next->is(tok::string_literal))
2512 if (Right.Previous->ClosesTemplateDeclaration &&
2513 Right.Previous->MatchingParen &&
2514 Right.Previous->MatchingParen->NestingLevel == 0 &&
2515 Style.AlwaysBreakTemplateDeclarations)
2517 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2518 Style.BreakConstructorInitializersBeforeComma &&
2519 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2521 // Break only if we have multiple inheritance.
2522 if (Style.BreakBeforeInheritanceComma &&
2523 Right.is(TT_InheritanceComma))
2525 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2526 // Raw string literals are special wrt. line breaks. The author has made a
2527 // deliberate choice and might have aligned the contents of the string
2528 // literal accordingly. Thus, we try keep existing line breaks.
2529 return Right.NewlinesBefore > 0;
2530 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2531 Style.Language == FormatStyle::LK_Proto)
2532 // Don't put enums onto single lines in protocol buffers.
2534 if (Right.is(TT_InlineASMBrace))
2535 return Right.HasUnescapedNewline;
2536 if (isAllmanBrace(Left) || isAllmanBrace(Right))
2537 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2538 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2539 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2540 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2543 if ((Style.Language == FormatStyle::LK_Java ||
2544 Style.Language == FormatStyle::LK_JavaScript) &&
2545 Left.is(TT_LeadingJavaAnnotation) &&
2546 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2547 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2553 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2554 const FormatToken &Right) {
2555 const FormatToken &Left = *Right.Previous;
2557 // Language-specific stuff.
2558 if (Style.Language == FormatStyle::LK_Java) {
2559 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2560 Keywords.kw_implements))
2562 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2563 Keywords.kw_implements))
2565 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2566 const FormatToken *NonComment = Right.getPreviousNonComment();
2568 NonComment->isOneOf(
2569 tok::kw_return, tok::kw_continue, tok::kw_break, tok::kw_throw,
2570 Keywords.kw_interface, Keywords.kw_type, tok::kw_static,
2571 tok::kw_public, tok::kw_private, tok::kw_protected,
2572 Keywords.kw_abstract, Keywords.kw_get, Keywords.kw_set))
2573 return false; // Otherwise automatic semicolon insertion would trigger.
2574 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2576 if (Left.is(TT_JsTypeColon))
2578 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2580 if (Left.is(Keywords.kw_in))
2581 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
2582 if (Right.is(Keywords.kw_in))
2583 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
2584 if (Right.is(Keywords.kw_as))
2585 return false; // must not break before as in 'x as type' casts
2586 if (Left.is(Keywords.kw_as))
2588 if (Left.is(TT_JsNonNullAssertion))
2590 if (Left.is(Keywords.kw_declare) &&
2591 Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
2592 Keywords.kw_function, tok::kw_class, tok::kw_enum,
2593 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
2594 Keywords.kw_let, tok::kw_const))
2595 // See grammar for 'declare' statements at:
2596 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
2598 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
2599 Right.isOneOf(tok::identifier, tok::string_literal))
2600 return false; // must not break in "module foo { ...}"
2601 if (Right.is(TT_TemplateString) && Right.closesScope())
2603 if (Left.is(TT_TemplateString) && Left.opensScope())
2607 if (Left.is(tok::at))
2609 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2611 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2612 return !Right.is(tok::l_paren);
2613 if (Right.is(TT_PointerOrReference))
2614 return Line.IsMultiVariableDeclStmt ||
2615 (Style.PointerAlignment == FormatStyle::PAS_Right &&
2616 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2617 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2618 Right.is(tok::kw_operator))
2620 if (Left.is(TT_PointerOrReference))
2622 if (Right.isTrailingComment())
2623 // We rely on MustBreakBefore being set correctly here as we should not
2624 // change the "binding" behavior of a comment.
2625 // The first comment in a braced lists is always interpreted as belonging to
2626 // the first list element. Otherwise, it should be placed outside of the
2628 return Left.BlockKind == BK_BracedInit;
2629 if (Left.is(tok::question) && Right.is(tok::colon))
2631 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2632 return Style.BreakBeforeTernaryOperators;
2633 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2634 return !Style.BreakBeforeTernaryOperators;
2635 if (Right.is(TT_InheritanceColon))
2637 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
2638 Left.isNot(TT_SelectorName))
2640 if (Right.is(tok::colon) &&
2641 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2643 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr))
2645 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2646 Right.Next->is(TT_ObjCMethodExpr)))
2647 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2648 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2650 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2652 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2653 TT_OverloadedOperator))
2655 if (Left.is(TT_RangeBasedForLoopColon))
2657 if (Right.is(TT_RangeBasedForLoopColon))
2659 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
2661 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2662 Left.is(tok::kw_operator))
2664 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2665 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2667 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2669 if (Left.is(tok::l_paren) && Left.Previous &&
2670 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2672 if (Right.is(TT_ImplicitStringLiteral))
2675 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2677 if (Right.is(tok::r_square) && Right.MatchingParen &&
2678 Right.MatchingParen->is(TT_LambdaLSquare))
2681 // We only break before r_brace if there was a corresponding break before
2682 // the l_brace, which is tracked by BreakBeforeClosingBrace.
2683 if (Right.is(tok::r_brace))
2684 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2686 // Allow breaking after a trailing annotation, e.g. after a method
2688 if (Left.is(TT_TrailingAnnotation))
2689 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2690 tok::less, tok::coloncolon);
2692 if (Right.is(tok::kw___attribute))
2695 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2698 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2701 if (Left.is(TT_CtorInitializerComma) &&
2702 Style.BreakConstructorInitializersBeforeComma)
2704 if (Right.is(TT_CtorInitializerComma) &&
2705 Style.BreakConstructorInitializersBeforeComma)
2707 if (Left.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma)
2709 if (Right.is(TT_InheritanceComma) && Style.BreakBeforeInheritanceComma)
2711 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2712 (Left.is(tok::less) && Right.is(tok::less)))
2714 if (Right.is(TT_BinaryOperator) &&
2715 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2716 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2717 Right.getPrecedence() != prec::Assignment))
2719 if (Left.is(TT_ArrayInitializerLSquare))
2721 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2723 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2724 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2725 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2726 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2727 Left.getPrecedence() == prec::Assignment))
2729 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2730 tok::kw_class, tok::kw_struct, tok::comment) ||
2731 Right.isMemberAccess() ||
2732 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2733 tok::colon, tok::l_square, tok::at) ||
2734 (Left.is(tok::r_paren) &&
2735 Right.isOneOf(tok::identifier, tok::kw_const)) ||
2736 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) ||
2737 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser));
2740 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2741 llvm::errs() << "AnnotatedTokens:\n";
2742 const FormatToken *Tok = Line.First;
2744 llvm::errs() << " M=" << Tok->MustBreakBefore
2745 << " C=" << Tok->CanBreakBefore
2746 << " T=" << getTokenTypeName(Tok->Type)
2747 << " S=" << Tok->SpacesRequiredBefore
2748 << " B=" << Tok->BlockParameterCount
2749 << " BK=" << Tok->BlockKind
2750 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2751 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2753 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2754 llvm::errs() << Tok->FakeLParens[i] << "/";
2755 llvm::errs() << " FakeRParens=" << Tok->FakeRParens;
2756 llvm::errs() << " Text='" << Tok->TokenText << "'\n";
2758 assert(Tok == Line.Last);
2761 llvm::errs() << "----\n";
2764 } // namespace format
2765 } // namespace clang