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);
47 FormatToken *Left = CurrentToken->Previous;
48 Left->ParentBracket = Contexts.back().ContextKind;
49 ScopedContextCreator ContextCreator(*this, tok::less, 10);
51 // If this angle is in the context of an expression, we need to be more
52 // hesitant to detect it as opening template parameters.
53 bool InExprContext = Contexts.back().IsExpression;
55 Contexts.back().IsExpression = false;
56 // If there's a template keyword before the opening angle bracket, this is a
57 // template parameter, not an argument.
58 Contexts.back().InTemplateArgument =
59 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
61 if (Style.Language == FormatStyle::LK_Java &&
62 CurrentToken->is(tok::question))
65 while (CurrentToken) {
66 if (CurrentToken->is(tok::greater)) {
67 Left->MatchingParen = CurrentToken;
68 CurrentToken->MatchingParen = Left;
69 CurrentToken->Type = TT_TemplateCloser;
73 if (CurrentToken->is(tok::question) &&
74 Style.Language == FormatStyle::LK_Java) {
78 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
79 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext))
81 // If a && or || is found and interpreted as a binary operator, this set
82 // of angles is likely part of something like "a < b && c > d". If the
83 // angles are inside an expression, the ||/&& might also be a binary
84 // operator that was misinterpreted because we are parsing template
86 // FIXME: This is getting out of hand, write a decent parser.
87 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
88 CurrentToken->Previous->is(TT_BinaryOperator) &&
89 Contexts[Contexts.size() - 2].IsExpression &&
90 !Line.startsWith(tok::kw_template))
92 updateParameterCount(Left, CurrentToken);
99 bool parseParens(bool LookForDecls = false) {
102 FormatToken *Left = CurrentToken->Previous;
103 Left->ParentBracket = Contexts.back().ContextKind;
104 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
106 // FIXME: This is a bit of a hack. Do better.
107 Contexts.back().ColonIsForRangeExpr =
108 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
110 bool StartsObjCMethodExpr = false;
111 if (CurrentToken->is(tok::caret)) {
112 // (^ can start a block type.
113 Left->Type = TT_ObjCBlockLParen;
114 } else if (FormatToken *MaybeSel = Left->Previous) {
115 // @selector( starts a selector.
116 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
117 MaybeSel->Previous->is(tok::at)) {
118 StartsObjCMethodExpr = true;
122 if (Left->Previous &&
123 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_decltype,
124 tok::kw_if, tok::kw_while, tok::l_paren,
126 Left->Previous->is(TT_BinaryOperator))) {
127 // static_assert, if and while usually contain expressions.
128 Contexts.back().IsExpression = true;
129 } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
130 Left->Previous->MatchingParen &&
131 Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
132 // This is a parameter list of a lambda expression.
133 Contexts.back().IsExpression = false;
134 } else if (Line.InPPDirective &&
136 !Left->Previous->isOneOf(tok::identifier,
137 TT_OverloadedOperator))) {
138 Contexts.back().IsExpression = true;
139 } else if (Contexts[Contexts.size() - 2].CaretFound) {
140 // This is the parameter list of an ObjC block.
141 Contexts.back().IsExpression = false;
142 } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
143 Left->Type = TT_AttributeParen;
144 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
145 // The first argument to a foreach macro is a declaration.
146 Contexts.back().IsForEachMacro = true;
147 Contexts.back().IsExpression = false;
148 } else if (Left->Previous && Left->Previous->MatchingParen &&
149 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
150 Contexts.back().IsExpression = false;
151 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
153 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
154 Contexts.back().IsExpression = !IsForOrCatch;
157 if (StartsObjCMethodExpr) {
158 Contexts.back().ColonIsObjCMethodExpr = true;
159 Left->Type = TT_ObjCMethodExpr;
162 bool MightBeFunctionType = CurrentToken->isOneOf(tok::star, tok::amp) &&
163 !Contexts[Contexts.size() - 2].IsExpression;
164 bool HasMultipleLines = false;
165 bool HasMultipleParametersOnALine = false;
166 bool MightBeObjCForRangeLoop =
167 Left->Previous && Left->Previous->is(tok::kw_for);
168 while (CurrentToken) {
169 // LookForDecls is set when "if (" has been seen. Check for
170 // 'identifier' '*' 'identifier' followed by not '=' -- this
171 // '*' has to be a binary operator but determineStarAmpUsage() will
172 // categorize it as an unary operator, so set the right type here.
173 if (LookForDecls && CurrentToken->Next) {
174 FormatToken *Prev = CurrentToken->getPreviousNonComment();
176 FormatToken *PrevPrev = Prev->getPreviousNonComment();
177 FormatToken *Next = CurrentToken->Next;
178 if (PrevPrev && PrevPrev->is(tok::identifier) &&
179 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
180 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
181 Prev->Type = TT_BinaryOperator;
182 LookForDecls = false;
187 if (CurrentToken->Previous->is(TT_PointerOrReference) &&
188 CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
190 MightBeFunctionType = true;
191 if (CurrentToken->Previous->is(TT_BinaryOperator))
192 Contexts.back().IsExpression = true;
193 if (CurrentToken->is(tok::r_paren)) {
194 if (MightBeFunctionType && CurrentToken->Next &&
195 (CurrentToken->Next->is(tok::l_paren) ||
196 (CurrentToken->Next->is(tok::l_square) &&
197 Line.MustBeDeclaration)))
198 Left->Type = TT_FunctionTypeLParen;
199 Left->MatchingParen = CurrentToken;
200 CurrentToken->MatchingParen = Left;
202 if (StartsObjCMethodExpr) {
203 CurrentToken->Type = TT_ObjCMethodExpr;
204 if (Contexts.back().FirstObjCSelectorName) {
205 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
206 Contexts.back().LongestObjCSelectorName;
210 if (Left->is(TT_AttributeParen))
211 CurrentToken->Type = TT_AttributeParen;
212 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
213 CurrentToken->Type = TT_JavaAnnotation;
214 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
215 CurrentToken->Type = TT_LeadingJavaAnnotation;
217 if (!HasMultipleLines)
218 Left->PackingKind = PPK_Inconclusive;
219 else if (HasMultipleParametersOnALine)
220 Left->PackingKind = PPK_BinPacked;
222 Left->PackingKind = PPK_OnePerLine;
227 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
230 if (CurrentToken->is(tok::l_brace))
231 Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
232 if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
233 !CurrentToken->Next->HasUnescapedNewline &&
234 !CurrentToken->Next->isTrailingComment())
235 HasMultipleParametersOnALine = true;
236 if (CurrentToken->isOneOf(tok::kw_const, tok::kw_auto) ||
237 CurrentToken->isSimpleTypeSpecifier())
238 Contexts.back().IsExpression = false;
239 if (CurrentToken->isOneOf(tok::semi, tok::colon))
240 MightBeObjCForRangeLoop = false;
241 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
242 CurrentToken->Type = TT_ObjCForIn;
243 // When we discover a 'new', we set CanBeExpression to 'false' in order to
244 // parse the type correctly. Reset that after a comma.
245 if (CurrentToken->is(tok::comma))
246 Contexts.back().CanBeExpression = true;
248 FormatToken *Tok = CurrentToken;
251 updateParameterCount(Left, Tok);
252 if (CurrentToken && CurrentToken->HasUnescapedNewline)
253 HasMultipleLines = true;
262 // A '[' could be an index subscript (after an identifier or after
263 // ')' or ']'), it could be the start of an Objective-C method
264 // expression, or it could the start of an Objective-C array literal.
265 FormatToken *Left = CurrentToken->Previous;
266 Left->ParentBracket = Contexts.back().ContextKind;
267 FormatToken *Parent = Left->getPreviousNonComment();
268 bool StartsObjCMethodExpr =
269 Style.Language == FormatStyle::LK_Cpp &&
270 Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
271 CurrentToken->isNot(tok::l_brace) &&
273 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
274 tok::kw_return, tok::kw_throw) ||
275 Parent->isUnaryOperator() ||
276 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
277 getBinOpPrecedence(Parent->Tok.getKind(), true, true) > prec::Unknown);
278 bool ColonFound = false;
280 unsigned BindingIncrease = 1;
281 if (Left->is(TT_Unknown)) {
282 if (StartsObjCMethodExpr) {
283 Left->Type = TT_ObjCMethodExpr;
284 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
285 Contexts.back().ContextKind == tok::l_brace &&
286 Parent->isOneOf(tok::l_brace, tok::comma)) {
287 Left->Type = TT_JsComputedPropertyName;
288 } else if (Style.Language == FormatStyle::LK_Proto ||
290 Parent->isOneOf(tok::at, tok::equal, tok::comma, tok::l_paren,
291 tok::l_square, tok::question, tok::colon,
293 Left->Type = TT_ArrayInitializerLSquare;
295 BindingIncrease = 10;
296 Left->Type = TT_ArraySubscriptLSquare;
300 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
301 Contexts.back().IsExpression = true;
302 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
304 while (CurrentToken) {
305 if (CurrentToken->is(tok::r_square)) {
306 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_paren) &&
307 Left->is(TT_ObjCMethodExpr)) {
308 // An ObjC method call is rarely followed by an open parenthesis.
309 // FIXME: Do we incorrectly label ":" with this?
310 StartsObjCMethodExpr = false;
311 Left->Type = TT_Unknown;
313 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
314 CurrentToken->Type = TT_ObjCMethodExpr;
315 // determineStarAmpUsage() thinks that '*' '[' is allocating an
316 // array of pointers, but if '[' starts a selector then '*' is a
318 if (Parent && Parent->is(TT_PointerOrReference))
319 Parent->Type = TT_BinaryOperator;
321 Left->MatchingParen = CurrentToken;
322 CurrentToken->MatchingParen = Left;
323 if (Contexts.back().FirstObjCSelectorName) {
324 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
325 Contexts.back().LongestObjCSelectorName;
326 if (Left->BlockParameterCount > 1)
327 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
332 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
334 if (CurrentToken->is(tok::colon)) {
335 if (Left->is(TT_ArraySubscriptLSquare)) {
336 Left->Type = TT_ObjCMethodExpr;
337 StartsObjCMethodExpr = true;
338 Contexts.back().ColonIsObjCMethodExpr = true;
339 if (Parent && Parent->is(tok::r_paren))
340 Parent->Type = TT_CastRParen;
344 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
346 Left->Type = TT_ArrayInitializerLSquare;
347 FormatToken *Tok = CurrentToken;
350 updateParameterCount(Left, Tok);
357 FormatToken *Left = CurrentToken->Previous;
358 Left->ParentBracket = Contexts.back().ContextKind;
360 if (Contexts.back().CaretFound)
361 Left->Type = TT_ObjCBlockLBrace;
362 Contexts.back().CaretFound = false;
364 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
365 Contexts.back().ColonIsDictLiteral = true;
366 if (Left->BlockKind == BK_BracedInit)
367 Contexts.back().IsExpression = true;
369 while (CurrentToken) {
370 if (CurrentToken->is(tok::r_brace)) {
371 Left->MatchingParen = CurrentToken;
372 CurrentToken->MatchingParen = Left;
376 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
378 updateParameterCount(Left, CurrentToken);
379 if (CurrentToken->isOneOf(tok::colon, tok::l_brace)) {
380 FormatToken *Previous = CurrentToken->getPreviousNonComment();
381 if (((CurrentToken->is(tok::colon) &&
382 (!Contexts.back().ColonIsDictLiteral ||
383 Style.Language != FormatStyle::LK_Cpp)) ||
384 Style.Language == FormatStyle::LK_Proto) &&
385 Previous->Tok.getIdentifierInfo())
386 Previous->Type = TT_SelectorName;
387 if (CurrentToken->is(tok::colon) ||
388 Style.Language == FormatStyle::LK_JavaScript)
389 Left->Type = TT_DictLiteral;
398 void updateParameterCount(FormatToken *Left, FormatToken *Current) {
399 if (Current->is(tok::l_brace) && !Current->is(TT_DictLiteral))
400 ++Left->BlockParameterCount;
401 if (Current->is(tok::comma)) {
402 ++Left->ParameterCount;
404 Left->Role.reset(new CommaSeparatedList(Style));
405 Left->Role->CommaFound(Current);
406 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
407 Left->ParameterCount = 1;
411 bool parseConditional() {
412 while (CurrentToken) {
413 if (CurrentToken->is(tok::colon)) {
414 CurrentToken->Type = TT_ConditionalExpr;
424 bool parseTemplateDeclaration() {
425 if (CurrentToken && CurrentToken->is(tok::less)) {
426 CurrentToken->Type = TT_TemplateOpener;
431 CurrentToken->Previous->ClosesTemplateDeclaration = true;
437 bool consumeToken() {
438 FormatToken *Tok = CurrentToken;
440 switch (Tok->Tok.getKind()) {
443 if (!Tok->Previous && Line.MustBeDeclaration)
444 Tok->Type = TT_ObjCMethodSpecifier;
449 // Colons from ?: are handled in parseConditional().
450 if (Style.Language == FormatStyle::LK_JavaScript) {
451 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
452 (Contexts.size() == 1 && // switch/case labels
453 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
454 Contexts.back().ContextKind == tok::l_paren || // function params
455 Contexts.back().ContextKind == tok::l_square || // array type
456 (Contexts.size() == 1 &&
457 Line.MustBeDeclaration)) { // method/property declaration
458 Tok->Type = TT_JsTypeColon;
462 if (Contexts.back().ColonIsDictLiteral ||
463 Style.Language == FormatStyle::LK_Proto) {
464 Tok->Type = TT_DictLiteral;
465 } else if (Contexts.back().ColonIsObjCMethodExpr ||
466 Line.startsWith(TT_ObjCMethodSpecifier)) {
467 Tok->Type = TT_ObjCMethodExpr;
468 Tok->Previous->Type = TT_SelectorName;
469 if (Tok->Previous->ColumnWidth >
470 Contexts.back().LongestObjCSelectorName) {
471 Contexts.back().LongestObjCSelectorName = Tok->Previous->ColumnWidth;
473 if (!Contexts.back().FirstObjCSelectorName)
474 Contexts.back().FirstObjCSelectorName = Tok->Previous;
475 } else if (Contexts.back().ColonIsForRangeExpr) {
476 Tok->Type = TT_RangeBasedForLoopColon;
477 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
478 Tok->Type = TT_BitFieldColon;
479 } else if (Contexts.size() == 1 &&
480 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
481 if (Tok->Previous->is(tok::r_paren))
482 Tok->Type = TT_CtorInitializerColon;
484 Tok->Type = TT_InheritanceColon;
485 } else if (Tok->Previous->is(tok::identifier) && Tok->Next &&
486 Tok->Next->isOneOf(tok::r_paren, tok::comma)) {
487 // This handles a special macro in ObjC code where selectors including
488 // the colon are passed as macro arguments.
489 Tok->Type = TT_ObjCMethodExpr;
490 } else if (Contexts.back().ContextKind == tok::l_paren) {
491 Tok->Type = TT_InlineASMColon;
496 if (CurrentToken && CurrentToken->is(tok::l_paren)) {
498 if (!parseParens(/*LookForDecls=*/true))
503 Contexts.back().ColonIsForRangeExpr = true;
509 // When faced with 'operator()()', the kw_operator handler incorrectly
510 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
511 // the first two parens OverloadedOperators and the second l_paren an
512 // OverloadedOperatorLParen.
514 Tok->Previous->is(tok::r_paren) &&
515 Tok->Previous->MatchingParen &&
516 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
517 Tok->Previous->Type = TT_OverloadedOperator;
518 Tok->Previous->MatchingParen->Type = TT_OverloadedOperator;
519 Tok->Type = TT_OverloadedOperatorLParen;
524 if (Line.MustBeDeclaration && Contexts.size() == 1 &&
525 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
527 !Tok->Previous->isOneOf(tok::kw_decltype, tok::kw___attribute,
528 TT_LeadingJavaAnnotation)))
529 Line.MightBeFunctionDecl = true;
540 if (!NonTemplateLess.count(Tok) &&
542 (!Tok->Previous->Tok.isLiteral() &&
543 !(Tok->Previous->is(tok::r_paren) && Contexts.size() > 1))) &&
545 Tok->Type = TT_TemplateOpener;
547 Tok->Type = TT_BinaryOperator;
548 NonTemplateLess.insert(Tok);
557 // Lines can start with '}'.
562 Tok->Type = TT_BinaryOperator;
564 case tok::kw_operator:
565 while (CurrentToken &&
566 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
567 if (CurrentToken->isOneOf(tok::star, tok::amp))
568 CurrentToken->Type = TT_PointerOrReference;
570 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator))
571 CurrentToken->Previous->Type = TT_OverloadedOperator;
574 CurrentToken->Type = TT_OverloadedOperatorLParen;
575 if (CurrentToken->Previous->is(TT_BinaryOperator))
576 CurrentToken->Previous->Type = TT_OverloadedOperator;
580 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
581 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
583 // Question marks before semicolons, colons, etc. indicate optional
584 // types (fields, parameters), e.g.
585 // function(x?: string, y?) {...}
587 Tok->Type = TT_JsTypeOptionalQuestion;
590 // Declarations cannot be conditional expressions, this can only be part
591 // of a type declaration.
592 if (Line.MustBeDeclaration &&
593 Style.Language == FormatStyle::LK_JavaScript)
597 case tok::kw_template:
598 parseTemplateDeclaration();
601 if (Contexts.back().InCtorInitializer)
602 Tok->Type = TT_CtorInitializerComma;
603 else if (Contexts.back().FirstStartOfName &&
604 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
605 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
606 Line.IsMultiVariableDeclStmt = true;
608 if (Contexts.back().IsForEachMacro)
609 Contexts.back().IsExpression = true;
617 void parseIncludeDirective() {
618 if (CurrentToken && CurrentToken->is(tok::less)) {
620 while (CurrentToken) {
621 if (CurrentToken->isNot(tok::comment) || CurrentToken->Next)
622 CurrentToken->Type = TT_ImplicitStringLiteral;
628 void parseWarningOrError() {
630 // We still want to format the whitespace left of the first token of the
633 while (CurrentToken) {
634 CurrentToken->Type = TT_ImplicitStringLiteral;
640 next(); // Consume "pragma".
642 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
643 bool IsMark = CurrentToken->is(Keywords.kw_mark);
644 next(); // Consume "mark".
645 next(); // Consume first token (so we fix leading whitespace).
646 while (CurrentToken) {
647 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
648 CurrentToken->Type = TT_ImplicitStringLiteral;
654 LineType parsePreprocessorDirective() {
655 LineType Type = LT_PreprocessorDirective;
659 if (CurrentToken->Tok.is(tok::numeric_constant)) {
660 CurrentToken->SpacesRequiredBefore = 1;
663 // Hashes in the middle of a line can lead to any strange token
665 if (!CurrentToken->Tok.getIdentifierInfo())
667 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
668 case tok::pp_include:
669 case tok::pp_include_next:
672 parseIncludeDirective();
673 Type = LT_ImportStatement;
676 case tok::pp_warning:
677 parseWarningOrError();
684 Contexts.back().IsExpression = true;
696 LineType parseLine() {
697 NonTemplateLess.clear();
698 if (CurrentToken->is(tok::hash))
699 return parsePreprocessorDirective();
701 // Directly allow to 'import <string-literal>' to support protocol buffer
702 // definitions (code.google.com/p/protobuf) or missing "#" (either way we
703 // should not break the line).
704 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
705 if ((Style.Language == FormatStyle::LK_Java &&
706 CurrentToken->is(Keywords.kw_package)) ||
707 (Info && Info->getPPKeywordID() == tok::pp_import &&
708 CurrentToken->Next &&
709 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
712 parseIncludeDirective();
713 return LT_ImportStatement;
716 // If this line starts and ends in '<' and '>', respectively, it is likely
717 // part of "#define <a/b.h>".
718 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
719 parseIncludeDirective();
720 return LT_ImportStatement;
723 // In .proto files, top-level options are very similar to import statements
724 // and should not be line-wrapped.
725 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
726 CurrentToken->is(Keywords.kw_option)) {
728 if (CurrentToken && CurrentToken->is(tok::identifier))
729 return LT_ImportStatement;
732 bool KeywordVirtualFound = false;
733 bool ImportStatement = false;
734 while (CurrentToken) {
735 if (CurrentToken->is(tok::kw_virtual))
736 KeywordVirtualFound = true;
737 if (isImportStatement(*CurrentToken))
738 ImportStatement = true;
742 if (KeywordVirtualFound)
743 return LT_VirtualFunctionDecl;
745 return LT_ImportStatement;
747 if (Line.startsWith(TT_ObjCMethodSpecifier)) {
748 if (Contexts.back().FirstObjCSelectorName)
749 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
750 Contexts.back().LongestObjCSelectorName;
751 return LT_ObjCMethodDecl;
758 bool isImportStatement(const FormatToken &Tok) {
759 // FIXME: Closure-library specific stuff should not be hard-coded but be
761 return Style.Language == FormatStyle::LK_JavaScript &&
762 Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
763 Tok.Next->Next && (Tok.Next->Next->TokenText == "module" ||
764 Tok.Next->Next->TokenText == "provide" ||
765 Tok.Next->Next->TokenText == "require" ||
766 Tok.Next->Next->TokenText == "setTestOnly") &&
767 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
770 void resetTokenMetadata(FormatToken *Token) {
774 // Reset token type in case we have already looked at it and then
775 // recovered from an error (e.g. failure to find the matching >).
776 if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
777 TT_FunctionLBrace, TT_ImplicitStringLiteral,
778 TT_InlineASMBrace, TT_JsFatArrow, TT_LambdaArrow,
780 CurrentToken->Type = TT_Unknown;
781 CurrentToken->Role.reset();
782 CurrentToken->MatchingParen = nullptr;
783 CurrentToken->FakeLParens.clear();
784 CurrentToken->FakeRParens = 0;
789 CurrentToken->NestingLevel = Contexts.size() - 1;
790 CurrentToken->BindingStrength = Contexts.back().BindingStrength;
791 modifyContext(*CurrentToken);
792 determineTokenType(*CurrentToken);
793 CurrentToken = CurrentToken->Next;
796 resetTokenMetadata(CurrentToken);
799 /// \brief A struct to hold information valid in a specific context, e.g.
800 /// a pair of parenthesis.
802 Context(tok::TokenKind ContextKind, unsigned BindingStrength,
804 : ContextKind(ContextKind), BindingStrength(BindingStrength),
805 IsExpression(IsExpression) {}
807 tok::TokenKind ContextKind;
808 unsigned BindingStrength;
810 unsigned LongestObjCSelectorName = 0;
811 bool ColonIsForRangeExpr = false;
812 bool ColonIsDictLiteral = false;
813 bool ColonIsObjCMethodExpr = false;
814 FormatToken *FirstObjCSelectorName = nullptr;
815 FormatToken *FirstStartOfName = nullptr;
816 bool CanBeExpression = true;
817 bool InTemplateArgument = false;
818 bool InCtorInitializer = false;
819 bool CaretFound = false;
820 bool IsForEachMacro = false;
823 /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
824 /// of each instance.
825 struct ScopedContextCreator {
828 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
831 P.Contexts.push_back(Context(ContextKind,
832 P.Contexts.back().BindingStrength + Increase,
833 P.Contexts.back().IsExpression));
836 ~ScopedContextCreator() { P.Contexts.pop_back(); }
839 void modifyContext(const FormatToken &Current) {
840 if (Current.getPrecedence() == prec::Assignment &&
841 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
842 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
843 Contexts.back().IsExpression = true;
844 if (!Line.startsWith(TT_UnaryOperator)) {
845 for (FormatToken *Previous = Current.Previous;
846 Previous && Previous->Previous &&
847 !Previous->Previous->isOneOf(tok::comma, tok::semi);
848 Previous = Previous->Previous) {
849 if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
850 Previous = Previous->MatchingParen;
854 if (Previous->opensScope())
856 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
857 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
858 Previous->Previous && Previous->Previous->isNot(tok::equal))
859 Previous->Type = TT_PointerOrReference;
862 } else if (Current.is(tok::lessless) &&
863 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
864 Contexts.back().IsExpression = true;
865 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
866 Contexts.back().IsExpression = true;
867 } else if (Current.is(TT_TrailingReturnArrow)) {
868 Contexts.back().IsExpression = false;
869 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
870 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
871 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
872 for (FormatToken *Previous = Current.Previous;
873 Previous && Previous->isOneOf(tok::star, tok::amp);
874 Previous = Previous->Previous)
875 Previous->Type = TT_PointerOrReference;
876 if (Line.MustBeDeclaration)
877 Contexts.back().IsExpression = Contexts.front().InCtorInitializer;
878 } else if (Current.Previous &&
879 Current.Previous->is(TT_CtorInitializerColon)) {
880 Contexts.back().IsExpression = true;
881 Contexts.back().InCtorInitializer = true;
882 } else if (Current.is(tok::kw_new)) {
883 Contexts.back().CanBeExpression = false;
884 } else if (Current.isOneOf(tok::semi, tok::exclaim)) {
885 // This should be the condition or increment in a for-loop.
886 Contexts.back().IsExpression = true;
890 void determineTokenType(FormatToken &Current) {
891 if (!Current.is(TT_Unknown))
892 // The token type is already known.
895 // Line.MightBeFunctionDecl can only be true after the parentheses of a
896 // function declaration have been found. In this case, 'Current' is a
897 // trailing token of this declaration and thus cannot be a name.
898 if (Current.is(Keywords.kw_instanceof)) {
899 Current.Type = TT_BinaryOperator;
900 } else if (isStartOfName(Current) &&
901 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
902 Contexts.back().FirstStartOfName = &Current;
903 Current.Type = TT_StartOfName;
904 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
906 } else if (Current.is(tok::arrow) &&
907 Style.Language == FormatStyle::LK_Java) {
908 Current.Type = TT_LambdaArrow;
909 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
910 Current.NestingLevel == 0) {
911 Current.Type = TT_TrailingReturnArrow;
912 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
914 determineStarAmpUsage(Current, Contexts.back().CanBeExpression &&
915 Contexts.back().IsExpression,
916 Contexts.back().InTemplateArgument);
917 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
918 Current.Type = determinePlusMinusCaretUsage(Current);
919 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
920 Contexts.back().CaretFound = true;
921 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
922 Current.Type = determineIncrementUsage(Current);
923 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
924 Current.Type = TT_UnaryOperator;
925 } else if (Current.is(tok::question)) {
926 if (Style.Language == FormatStyle::LK_JavaScript &&
927 Line.MustBeDeclaration) {
928 // In JavaScript, `interface X { foo?(): bar; }` is an optional method
929 // on the interface, not a ternary expression.
930 Current.Type = TT_JsTypeOptionalQuestion;
932 Current.Type = TT_ConditionalExpr;
934 } else if (Current.isBinaryOperator() &&
935 (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
936 Current.Type = TT_BinaryOperator;
937 } else if (Current.is(tok::comment)) {
938 if (Current.TokenText.startswith("/*")) {
939 if (Current.TokenText.endswith("*/"))
940 Current.Type = TT_BlockComment;
942 // The lexer has for some reason determined a comment here. But we
943 // cannot really handle it, if it isn't properly terminated.
944 Current.Tok.setKind(tok::unknown);
946 Current.Type = TT_LineComment;
948 } else if (Current.is(tok::r_paren)) {
949 if (rParenEndsCast(Current))
950 Current.Type = TT_CastRParen;
951 if (Current.MatchingParen && Current.Next &&
952 !Current.Next->isBinaryOperator() &&
953 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace))
954 if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
955 if (BeforeParen->is(tok::identifier) &&
956 BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
957 (!BeforeParen->Previous ||
958 BeforeParen->Previous->ClosesTemplateDeclaration))
959 Current.Type = TT_FunctionAnnotationRParen;
960 } else if (Current.is(tok::at) && Current.Next) {
961 if (Current.Next->isStringLiteral()) {
962 Current.Type = TT_ObjCStringLiteral;
964 switch (Current.Next->Tok.getObjCKeywordID()) {
965 case tok::objc_interface:
966 case tok::objc_implementation:
967 case tok::objc_protocol:
968 Current.Type = TT_ObjCDecl;
970 case tok::objc_property:
971 Current.Type = TT_ObjCProperty;
977 } else if (Current.is(tok::period)) {
978 FormatToken *PreviousNoComment = Current.getPreviousNonComment();
979 if (PreviousNoComment &&
980 PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
981 Current.Type = TT_DesignatedInitializerPeriod;
982 else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
983 Current.Previous->isOneOf(TT_JavaAnnotation,
984 TT_LeadingJavaAnnotation)) {
985 Current.Type = Current.Previous->Type;
987 } else if (Current.isOneOf(tok::identifier, tok::kw_const) &&
989 !Current.Previous->isOneOf(tok::equal, tok::at) &&
990 Line.MightBeFunctionDecl && Contexts.size() == 1) {
991 // Line.MightBeFunctionDecl can only be true after the parentheses of a
992 // function declaration have been found.
993 Current.Type = TT_TrailingAnnotation;
994 } else if ((Style.Language == FormatStyle::LK_Java ||
995 Style.Language == FormatStyle::LK_JavaScript) &&
997 if (Current.Previous->is(tok::at) &&
998 Current.isNot(Keywords.kw_interface)) {
999 const FormatToken &AtToken = *Current.Previous;
1000 const FormatToken *Previous = AtToken.getPreviousNonComment();
1001 if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1002 Current.Type = TT_LeadingJavaAnnotation;
1004 Current.Type = TT_JavaAnnotation;
1005 } else if (Current.Previous->is(tok::period) &&
1006 Current.Previous->isOneOf(TT_JavaAnnotation,
1007 TT_LeadingJavaAnnotation)) {
1008 Current.Type = Current.Previous->Type;
1013 /// \brief Take a guess at whether \p Tok starts a name of a function or
1014 /// variable declaration.
1016 /// This is a heuristic based on whether \p Tok is an identifier following
1017 /// something that is likely a type.
1018 bool isStartOfName(const FormatToken &Tok) {
1019 if (Tok.isNot(tok::identifier) || !Tok.Previous)
1022 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof))
1025 // Skip "const" as it does not have an influence on whether this is a name.
1026 FormatToken *PreviousNotConst = Tok.Previous;
1027 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1028 PreviousNotConst = PreviousNotConst->Previous;
1030 if (!PreviousNotConst)
1033 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1034 PreviousNotConst->Previous &&
1035 PreviousNotConst->Previous->is(tok::hash);
1037 if (PreviousNotConst->is(TT_TemplateCloser))
1038 return PreviousNotConst && PreviousNotConst->MatchingParen &&
1039 PreviousNotConst->MatchingParen->Previous &&
1040 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1041 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1043 if (PreviousNotConst->is(tok::r_paren) && PreviousNotConst->MatchingParen &&
1044 PreviousNotConst->MatchingParen->Previous &&
1045 PreviousNotConst->MatchingParen->Previous->is(tok::kw_decltype))
1048 return (!IsPPKeyword &&
1049 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1050 PreviousNotConst->is(TT_PointerOrReference) ||
1051 PreviousNotConst->isSimpleTypeSpecifier();
1054 /// \brief Determine whether ')' is ending a cast.
1055 bool rParenEndsCast(const FormatToken &Tok) {
1056 // C-style casts are only used in C++ and Java.
1057 if (Style.Language != FormatStyle::LK_Cpp &&
1058 Style.Language != FormatStyle::LK_Java)
1061 // Empty parens aren't casts and there are no casts at the end of the line.
1062 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1065 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1067 // If there is an opening parenthesis left of the current parentheses,
1068 // look past it as these might be chained casts.
1069 if (LeftOfParens->is(tok::r_paren)) {
1070 if (!LeftOfParens->MatchingParen ||
1071 !LeftOfParens->MatchingParen->Previous)
1073 LeftOfParens = LeftOfParens->MatchingParen->Previous;
1076 // If there is an identifier (or with a few exceptions a keyword) right
1077 // before the parentheses, this is unlikely to be a cast.
1078 if (LeftOfParens->Tok.getIdentifierInfo() &&
1079 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1083 // Certain other tokens right before the parentheses are also signals that
1084 // this cannot be a cast.
1085 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1090 if (Tok.Next->is(tok::question))
1093 // As Java has no function types, a "(" after the ")" likely means that this
1095 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1098 // If a (non-string) literal follows, this is likely a cast.
1099 if (Tok.Next->isNot(tok::string_literal) &&
1100 (Tok.Next->Tok.isLiteral() ||
1101 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1104 // Heuristically try to determine whether the parentheses contain a type.
1105 bool ParensAreType =
1107 Tok.Previous->isOneOf(TT_PointerOrReference, TT_TemplateCloser) ||
1108 Tok.Previous->isSimpleTypeSpecifier();
1109 bool ParensCouldEndDecl =
1110 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1111 if (ParensAreType && !ParensCouldEndDecl)
1114 // At this point, we heuristically assume that there are no casts at the
1115 // start of the line. We assume that we have found most cases where there
1116 // are by the logic above, e.g. "(void)x;".
1120 // If the following token is an identifier, this is a cast. All cases where
1121 // this can be something else are handled above.
1122 if (Tok.Next->is(tok::identifier))
1125 if (!Tok.Next->Next)
1128 // If the next token after the parenthesis is a unary operator, assume
1129 // that this is cast, unless there are unexpected tokens inside the
1132 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1133 if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1134 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1136 // Search for unexpected tokens.
1137 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1138 Prev = Prev->Previous) {
1139 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1145 /// \brief Return the type of the given token assuming it is * or &.
1146 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1147 bool InTemplateArgument) {
1148 if (Style.Language == FormatStyle::LK_JavaScript)
1149 return TT_BinaryOperator;
1151 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1153 return TT_UnaryOperator;
1155 const FormatToken *NextToken = Tok.getNextNonComment();
1157 NextToken->isOneOf(tok::arrow, Keywords.kw_final,
1158 Keywords.kw_override) ||
1159 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
1160 return TT_PointerOrReference;
1162 if (PrevToken->is(tok::coloncolon))
1163 return TT_PointerOrReference;
1165 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
1166 tok::comma, tok::semi, tok::kw_return, tok::colon,
1167 tok::equal, tok::kw_delete, tok::kw_sizeof) ||
1168 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
1169 TT_UnaryOperator, TT_CastRParen))
1170 return TT_UnaryOperator;
1172 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
1173 return TT_PointerOrReference;
1174 if (NextToken->is(tok::kw_operator) && !IsExpression)
1175 return TT_PointerOrReference;
1176 if (NextToken->isOneOf(tok::comma, tok::semi))
1177 return TT_PointerOrReference;
1179 if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
1180 PrevToken->MatchingParen->Previous &&
1181 PrevToken->MatchingParen->Previous->isOneOf(tok::kw_typeof,
1183 return TT_PointerOrReference;
1185 if (PrevToken->Tok.isLiteral() ||
1186 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
1187 tok::kw_false, tok::r_brace) ||
1188 NextToken->Tok.isLiteral() ||
1189 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
1190 NextToken->isUnaryOperator() ||
1191 // If we know we're in a template argument, there are no named
1192 // declarations. Thus, having an identifier on the right-hand side
1193 // indicates a binary operator.
1194 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
1195 return TT_BinaryOperator;
1197 // "&&(" is quite unlikely to be two successive unary "&".
1198 if (Tok.is(tok::ampamp) && NextToken && NextToken->is(tok::l_paren))
1199 return TT_BinaryOperator;
1201 // This catches some cases where evaluation order is used as control flow:
1203 const FormatToken *NextNextToken = NextToken->getNextNonComment();
1204 if (NextNextToken && NextNextToken->is(tok::arrow))
1205 return TT_BinaryOperator;
1207 // It is very unlikely that we are going to find a pointer or reference type
1208 // definition on the RHS of an assignment.
1209 if (IsExpression && !Contexts.back().CaretFound)
1210 return TT_BinaryOperator;
1212 return TT_PointerOrReference;
1215 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
1216 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1217 if (!PrevToken || PrevToken->is(TT_CastRParen))
1218 return TT_UnaryOperator;
1220 // Use heuristics to recognize unary operators.
1221 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
1222 tok::question, tok::colon, tok::kw_return,
1223 tok::kw_case, tok::at, tok::l_brace))
1224 return TT_UnaryOperator;
1226 // There can't be two consecutive binary operators.
1227 if (PrevToken->is(TT_BinaryOperator))
1228 return TT_UnaryOperator;
1230 // Fall back to marking the token as binary operator.
1231 return TT_BinaryOperator;
1234 /// \brief Determine whether ++/-- are pre- or post-increments/-decrements.
1235 TokenType determineIncrementUsage(const FormatToken &Tok) {
1236 const FormatToken *PrevToken = Tok.getPreviousNonComment();
1237 if (!PrevToken || PrevToken->is(TT_CastRParen))
1238 return TT_UnaryOperator;
1239 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
1240 return TT_TrailingUnaryOperator;
1242 return TT_UnaryOperator;
1245 SmallVector<Context, 8> Contexts;
1247 const FormatStyle &Style;
1248 AnnotatedLine &Line;
1249 FormatToken *CurrentToken;
1251 const AdditionalKeywords &Keywords;
1253 // Set of "<" tokens that do not open a template parameter list. If parseAngle
1254 // determines that a specific token can't be a template opener, it will make
1255 // same decision irrespective of the decisions for tokens leading up to it.
1256 // Store this information to prevent this from causing exponential runtime.
1257 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
1260 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
1261 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
1263 /// \brief Parses binary expressions by inserting fake parenthesis based on
1264 /// operator precedence.
1265 class ExpressionParser {
1267 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
1268 AnnotatedLine &Line)
1269 : Style(Style), Keywords(Keywords), Current(Line.First) {}
1271 /// \brief Parse expressions with the given operatore precedence.
1272 void parse(int Precedence = 0) {
1273 // Skip 'return' and ObjC selector colons as they are not part of a binary
1275 while (Current && (Current->is(tok::kw_return) ||
1276 (Current->is(tok::colon) &&
1277 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
1280 if (!Current || Precedence > PrecedenceArrowAndPeriod)
1283 // Conditional expressions need to be parsed separately for proper nesting.
1284 if (Precedence == prec::Conditional) {
1285 parseConditionalExpr();
1289 // Parse unary operators, which all have a higher precedence than binary
1291 if (Precedence == PrecedenceUnaryOperator) {
1292 parseUnaryOperator();
1296 FormatToken *Start = Current;
1297 FormatToken *LatestOperator = nullptr;
1298 unsigned OperatorIndex = 0;
1301 // Consume operators with higher precedence.
1302 parse(Precedence + 1);
1304 int CurrentPrecedence = getCurrentPrecedence();
1306 if (Current && Current->is(TT_SelectorName) &&
1307 Precedence == CurrentPrecedence) {
1309 addFakeParenthesis(Start, prec::Level(Precedence));
1313 // At the end of the line or when an operator with higher precedence is
1314 // found, insert fake parenthesis and return.
1315 if (!Current || (Current->closesScope() && Current->MatchingParen) ||
1316 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
1317 (CurrentPrecedence == prec::Conditional &&
1318 Precedence == prec::Assignment && Current->is(tok::colon))) {
1322 // Consume scopes: (), [], <> and {}
1323 if (Current->opensScope()) {
1324 while (Current && !Current->closesScope()) {
1331 if (CurrentPrecedence == Precedence) {
1332 LatestOperator = Current;
1333 Current->OperatorIndex = OperatorIndex;
1336 next(/*SkipPastLeadingComments=*/Precedence > 0);
1340 if (LatestOperator && (Current || Precedence > 0)) {
1341 LatestOperator->LastOperator = true;
1342 if (Precedence == PrecedenceArrowAndPeriod) {
1343 // Call expressions don't have a binary operator precedence.
1344 addFakeParenthesis(Start, prec::Unknown);
1346 addFakeParenthesis(Start, prec::Level(Precedence));
1352 /// \brief Gets the precedence (+1) of the given token for binary operators
1353 /// and other tokens that we treat like binary operators.
1354 int getCurrentPrecedence() {
1356 const FormatToken *NextNonComment = Current->getNextNonComment();
1357 if (Current->is(TT_ConditionalExpr))
1358 return prec::Conditional;
1359 if (NextNonComment && NextNonComment->is(tok::colon) &&
1360 NextNonComment->is(TT_DictLiteral))
1362 if (Current->is(TT_LambdaArrow))
1364 if (Current->is(TT_JsFatArrow))
1365 return prec::Assignment;
1366 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName,
1367 TT_JsComputedPropertyName) ||
1368 (Current->is(tok::comment) && NextNonComment &&
1369 NextNonComment->is(TT_SelectorName)))
1371 if (Current->is(TT_RangeBasedForLoopColon))
1373 if ((Style.Language == FormatStyle::LK_Java ||
1374 Style.Language == FormatStyle::LK_JavaScript) &&
1375 Current->is(Keywords.kw_instanceof))
1376 return prec::Relational;
1377 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
1378 return Current->getPrecedence();
1379 if (Current->isOneOf(tok::period, tok::arrow))
1380 return PrecedenceArrowAndPeriod;
1381 if (Style.Language == FormatStyle::LK_Java &&
1382 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
1383 Keywords.kw_throws))
1389 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
1390 Start->FakeLParens.push_back(Precedence);
1391 if (Precedence > prec::Unknown)
1392 Start->StartsBinaryExpression = true;
1394 FormatToken *Previous = Current->Previous;
1395 while (Previous->is(tok::comment) && Previous->Previous)
1396 Previous = Previous->Previous;
1397 ++Previous->FakeRParens;
1398 if (Precedence > prec::Unknown)
1399 Previous->EndsBinaryExpression = true;
1403 /// \brief Parse unary operator expressions and surround them with fake
1404 /// parentheses if appropriate.
1405 void parseUnaryOperator() {
1406 if (!Current || Current->isNot(TT_UnaryOperator)) {
1407 parse(PrecedenceArrowAndPeriod);
1411 FormatToken *Start = Current;
1413 parseUnaryOperator();
1415 // The actual precedence doesn't matter.
1416 addFakeParenthesis(Start, prec::Unknown);
1419 void parseConditionalExpr() {
1420 while (Current && Current->isTrailingComment()) {
1423 FormatToken *Start = Current;
1424 parse(prec::LogicalOr);
1425 if (!Current || !Current->is(tok::question))
1428 parse(prec::Assignment);
1429 if (!Current || Current->isNot(TT_ConditionalExpr))
1432 parse(prec::Assignment);
1433 addFakeParenthesis(Start, prec::Conditional);
1436 void next(bool SkipPastLeadingComments = true) {
1438 Current = Current->Next;
1440 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
1441 Current->isTrailingComment())
1442 Current = Current->Next;
1445 const FormatStyle &Style;
1446 const AdditionalKeywords &Keywords;
1447 FormatToken *Current;
1450 } // end anonymous namespace
1452 void TokenAnnotator::setCommentLineLevels(
1453 SmallVectorImpl<AnnotatedLine *> &Lines) {
1454 const AnnotatedLine *NextNonCommentLine = nullptr;
1455 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
1458 if (NextNonCommentLine && (*I)->First->is(tok::comment) &&
1459 (*I)->First->Next == nullptr)
1460 (*I)->Level = NextNonCommentLine->Level;
1462 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
1464 setCommentLineLevels((*I)->Children);
1468 void TokenAnnotator::annotate(AnnotatedLine &Line) {
1469 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1470 E = Line.Children.end();
1474 AnnotatingParser Parser(Style, Line, Keywords);
1475 Line.Type = Parser.parseLine();
1476 if (Line.Type == LT_Invalid)
1479 ExpressionParser ExprParser(Style, Keywords, Line);
1482 if (Line.startsWith(TT_ObjCMethodSpecifier))
1483 Line.Type = LT_ObjCMethodDecl;
1484 else if (Line.startsWith(TT_ObjCDecl))
1485 Line.Type = LT_ObjCDecl;
1486 else if (Line.startsWith(TT_ObjCProperty))
1487 Line.Type = LT_ObjCProperty;
1489 Line.First->SpacesRequiredBefore = 1;
1490 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
1493 // This function heuristically determines whether 'Current' starts the name of a
1494 // function declaration.
1495 static bool isFunctionDeclarationName(const FormatToken &Current) {
1496 auto skipOperatorName = [](const FormatToken* Next) -> const FormatToken* {
1497 for (; Next; Next = Next->Next) {
1498 if (Next->is(TT_OverloadedOperatorLParen))
1500 if (Next->is(TT_OverloadedOperator))
1502 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
1503 // For 'new[]' and 'delete[]'.
1504 if (Next->Next && Next->Next->is(tok::l_square) &&
1505 Next->Next->Next && Next->Next->Next->is(tok::r_square))
1506 Next = Next->Next->Next;
1515 const FormatToken *Next = Current.Next;
1516 if (Current.is(tok::kw_operator)) {
1517 if (Current.Previous && Current.Previous->is(tok::coloncolon))
1519 Next = skipOperatorName(Next);
1521 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
1523 for (; Next; Next = Next->Next) {
1524 if (Next->is(TT_TemplateOpener)) {
1525 Next = Next->MatchingParen;
1526 } else if (Next->is(tok::coloncolon)) {
1530 if (Next->is(tok::kw_operator)) {
1531 Next = skipOperatorName(Next->Next);
1534 if (!Next->is(tok::identifier))
1536 } else if (Next->is(tok::l_paren)) {
1544 if (!Next || !Next->is(tok::l_paren))
1546 if (Next->Next == Next->MatchingParen)
1548 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
1550 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
1551 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName))
1553 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
1554 Tok->Tok.isLiteral())
1560 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
1561 assert(Line.MightBeFunctionDecl);
1563 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
1564 Style.AlwaysBreakAfterReturnType ==
1565 FormatStyle::RTBS_TopLevelDefinitions) &&
1569 switch (Style.AlwaysBreakAfterReturnType) {
1570 case FormatStyle::RTBS_None:
1572 case FormatStyle::RTBS_All:
1573 case FormatStyle::RTBS_TopLevel:
1575 case FormatStyle::RTBS_AllDefinitions:
1576 case FormatStyle::RTBS_TopLevelDefinitions:
1577 return Line.mightBeFunctionDefinition();
1583 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
1584 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
1585 E = Line.Children.end();
1587 calculateFormattingInformation(**I);
1590 Line.First->TotalLength =
1591 Line.First->IsMultiline ? Style.ColumnLimit : Line.First->ColumnWidth;
1592 if (!Line.First->Next)
1594 FormatToken *Current = Line.First->Next;
1595 bool InFunctionDecl = Line.MightBeFunctionDecl;
1597 if (isFunctionDeclarationName(*Current))
1598 Current->Type = TT_FunctionDeclarationName;
1599 if (Current->is(TT_LineComment)) {
1600 if (Current->Previous->BlockKind == BK_BracedInit &&
1601 Current->Previous->opensScope())
1602 Current->SpacesRequiredBefore = Style.Cpp11BracedListStyle ? 0 : 1;
1604 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
1606 // If we find a trailing comment, iterate backwards to determine whether
1607 // it seems to relate to a specific parameter. If so, break before that
1608 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
1609 // to the previous line in:
1613 if (!Current->HasUnescapedNewline) {
1614 for (FormatToken *Parameter = Current->Previous; Parameter;
1615 Parameter = Parameter->Previous) {
1616 if (Parameter->isOneOf(tok::comment, tok::r_brace))
1618 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
1619 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
1620 Parameter->HasUnescapedNewline)
1621 Parameter->MustBreakBefore = true;
1626 } else if (Current->SpacesRequiredBefore == 0 &&
1627 spaceRequiredBefore(Line, *Current)) {
1628 Current->SpacesRequiredBefore = 1;
1631 Current->MustBreakBefore =
1632 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
1634 if (!Current->MustBreakBefore && InFunctionDecl &&
1635 Current->is(TT_FunctionDeclarationName))
1636 Current->MustBreakBefore = mustBreakForReturnType(Line);
1638 Current->CanBreakBefore =
1639 Current->MustBreakBefore || canBreakBefore(Line, *Current);
1640 unsigned ChildSize = 0;
1641 if (Current->Previous->Children.size() == 1) {
1642 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
1643 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
1644 : LastOfChild.TotalLength + 1;
1646 const FormatToken *Prev = Current->Previous;
1647 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
1648 (Prev->Children.size() == 1 &&
1649 Prev->Children[0]->First->MustBreakBefore) ||
1650 Current->IsMultiline)
1651 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
1653 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
1654 ChildSize + Current->SpacesRequiredBefore;
1656 if (Current->is(TT_CtorInitializerColon))
1657 InFunctionDecl = false;
1659 // FIXME: Only calculate this if CanBreakBefore is true once static
1660 // initializers etc. are sorted out.
1661 // FIXME: Move magic numbers to a better place.
1662 Current->SplitPenalty = 20 * Current->BindingStrength +
1663 splitPenalty(Line, *Current, InFunctionDecl);
1665 Current = Current->Next;
1668 calculateUnbreakableTailLengths(Line);
1669 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
1671 Current->Role->precomputeFormattingInfos(Current);
1674 DEBUG({ printDebugInfo(Line); });
1677 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
1678 unsigned UnbreakableTailLength = 0;
1679 FormatToken *Current = Line.Last;
1681 Current->UnbreakableTailLength = UnbreakableTailLength;
1682 if (Current->CanBreakBefore ||
1683 Current->isOneOf(tok::comment, tok::string_literal)) {
1684 UnbreakableTailLength = 0;
1686 UnbreakableTailLength +=
1687 Current->ColumnWidth + Current->SpacesRequiredBefore;
1689 Current = Current->Previous;
1693 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
1694 const FormatToken &Tok,
1695 bool InFunctionDecl) {
1696 const FormatToken &Left = *Tok.Previous;
1697 const FormatToken &Right = Tok;
1699 if (Left.is(tok::semi))
1702 if (Style.Language == FormatStyle::LK_Java) {
1703 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
1705 if (Right.is(Keywords.kw_implements))
1707 if (Left.is(tok::comma) && Left.NestingLevel == 0)
1709 } else if (Style.Language == FormatStyle::LK_JavaScript) {
1710 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
1712 if (Left.is(TT_JsTypeColon))
1716 if (Left.is(tok::comma) || (Right.is(tok::identifier) && Right.Next &&
1717 Right.Next->is(TT_DictLiteral)))
1719 if (Right.is(tok::l_square)) {
1720 if (Style.Language == FormatStyle::LK_Proto)
1722 if (Left.is(tok::r_square))
1724 // Slightly prefer formatting local lambda definitions like functions.
1725 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
1727 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
1728 TT_ArrayInitializerLSquare))
1732 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
1733 Right.is(tok::kw_operator)) {
1734 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
1736 if (Left.is(TT_StartOfName))
1738 if (InFunctionDecl && Right.NestingLevel == 0)
1739 return Style.PenaltyReturnTypeOnItsOwnLine;
1742 if (Right.is(TT_PointerOrReference))
1744 if (Right.is(TT_LambdaArrow))
1746 if (Left.is(tok::equal) && Right.is(tok::l_brace))
1748 if (Left.is(TT_CastRParen))
1750 if (Left.is(tok::coloncolon) ||
1751 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
1753 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
1756 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon))
1759 if (Right.isMemberAccess()) {
1760 // Breaking before the "./->" of a chained call/member access is reasonably
1761 // cheap, as formatting those with one call per line is generally
1762 // desirable. In particular, it should be cheaper to break before the call
1763 // than it is to break inside a call's parameters, which could lead to weird
1764 // "hanging" indents. The exception is the very last "./->" to support this
1765 // frequent pattern:
1767 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
1770 // which might otherwise be blown up onto many lines. Here, clang-format
1771 // won't produce "hanging" indents anyway as there is no other trailing
1773 return Right.LastOperator ? 150 : 40;
1776 if (Right.is(TT_TrailingAnnotation) &&
1777 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
1778 // Moving trailing annotations to the next line is fine for ObjC method
1780 if (Line.startsWith(TT_ObjCMethodSpecifier))
1782 // Generally, breaking before a trailing annotation is bad unless it is
1783 // function-like. It seems to be especially preferable to keep standard
1784 // annotations (i.e. "const", "final" and "override") on the same line.
1785 // Use a slightly higher penalty after ")" so that annotations like
1786 // "const override" are kept together.
1787 bool is_short_annotation = Right.TokenText.size() < 10;
1788 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
1791 // In for-loops, prefer breaking at ',' and ';'.
1792 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
1795 // In Objective-C method expressions, prefer breaking before "param:" over
1796 // breaking after it.
1797 if (Right.is(TT_SelectorName))
1799 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
1800 return Line.MightBeFunctionDecl ? 50 : 500;
1802 if (Left.is(tok::l_paren) && InFunctionDecl &&
1803 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1805 if (Left.is(tok::l_paren) && Left.Previous &&
1806 Left.Previous->isOneOf(tok::kw_if, tok::kw_for))
1808 if (Left.is(tok::equal) && InFunctionDecl)
1810 if (Right.is(tok::r_brace))
1812 if (Left.is(TT_TemplateOpener))
1814 if (Left.opensScope()) {
1815 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign)
1817 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
1820 if (Left.is(TT_JavaAnnotation))
1823 if (Right.is(tok::lessless)) {
1824 if (Left.is(tok::string_literal) &&
1825 (!Right.LastOperator || Right.OperatorIndex != 1)) {
1826 StringRef Content = Left.TokenText;
1827 if (Content.startswith("\""))
1828 Content = Content.drop_front(1);
1829 if (Content.endswith("\""))
1830 Content = Content.drop_back(1);
1831 Content = Content.trim();
1832 if (Content.size() > 1 &&
1833 (Content.back() == ':' || Content.back() == '='))
1836 return 1; // Breaking at a << is really cheap.
1838 if (Left.is(TT_ConditionalExpr))
1839 return prec::Conditional;
1840 prec::Level Level = Left.getPrecedence();
1841 if (Level != prec::Unknown)
1843 Level = Right.getPrecedence();
1844 if (Level != prec::Unknown)
1850 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
1851 const FormatToken &Left,
1852 const FormatToken &Right) {
1853 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
1855 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
1856 Left.Tok.getObjCKeywordID() == tok::objc_property)
1858 if (Right.is(tok::hashhash))
1859 return Left.is(tok::hash);
1860 if (Left.isOneOf(tok::hashhash, tok::hash))
1861 return Right.is(tok::hash);
1862 if (Left.is(tok::l_paren) && Right.is(tok::r_paren))
1863 return Style.SpaceInEmptyParentheses;
1864 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
1865 return (Right.is(TT_CastRParen) ||
1866 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
1867 ? Style.SpacesInCStyleCastParentheses
1868 : Style.SpacesInParentheses;
1869 if (Right.isOneOf(tok::semi, tok::comma))
1871 if (Right.is(tok::less) &&
1872 (Left.is(tok::kw_template) ||
1873 (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)))
1875 if (Left.isOneOf(tok::exclaim, tok::tilde))
1877 if (Left.is(tok::at) &&
1878 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
1879 tok::numeric_constant, tok::l_paren, tok::l_brace,
1880 tok::kw_true, tok::kw_false))
1882 if (Left.is(tok::colon))
1883 return !Left.is(TT_ObjCMethodExpr);
1884 if (Left.is(tok::coloncolon))
1886 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less))
1888 if (Right.is(tok::ellipsis))
1889 return Left.Tok.isLiteral();
1890 if (Left.is(tok::l_square) && Right.is(tok::amp))
1892 if (Right.is(TT_PointerOrReference))
1893 return (Left.is(tok::r_paren) && Left.MatchingParen &&
1894 (Left.MatchingParen->is(TT_OverloadedOperatorLParen) ||
1895 (Left.MatchingParen->Previous &&
1896 Left.MatchingParen->Previous->is(TT_FunctionDeclarationName)))) ||
1897 (Left.Tok.isLiteral() ||
1898 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
1899 (Style.PointerAlignment != FormatStyle::PAS_Left ||
1900 Line.IsMultiVariableDeclStmt)));
1901 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
1902 (!Left.is(TT_PointerOrReference) ||
1903 (Style.PointerAlignment != FormatStyle::PAS_Right &&
1904 !Line.IsMultiVariableDeclStmt)))
1906 if (Left.is(TT_PointerOrReference))
1907 return Right.Tok.isLiteral() ||
1908 Right.isOneOf(TT_BlockComment, Keywords.kw_final,
1909 Keywords.kw_override) ||
1910 (Right.is(tok::l_brace) && Right.BlockKind == BK_Block) ||
1911 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
1913 (Style.PointerAlignment != FormatStyle::PAS_Right &&
1914 !Line.IsMultiVariableDeclStmt) &&
1916 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
1917 if (Right.is(tok::star) && Left.is(tok::l_paren))
1919 if (Left.is(tok::l_square))
1920 return (Left.is(TT_ArrayInitializerLSquare) &&
1921 Style.SpacesInContainerLiterals && Right.isNot(tok::r_square)) ||
1922 (Left.is(TT_ArraySubscriptLSquare) && Style.SpacesInSquareBrackets &&
1923 Right.isNot(tok::r_square));
1924 if (Right.is(tok::r_square))
1925 return Right.MatchingParen &&
1926 ((Style.SpacesInContainerLiterals &&
1927 Right.MatchingParen->is(TT_ArrayInitializerLSquare)) ||
1928 (Style.SpacesInSquareBrackets &&
1929 Right.MatchingParen->is(TT_ArraySubscriptLSquare)));
1930 if (Right.is(tok::l_square) &&
1931 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare) &&
1932 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral))
1934 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
1935 return !Left.Children.empty(); // No spaces in "{}".
1936 if ((Left.is(tok::l_brace) && Left.BlockKind != BK_Block) ||
1937 (Right.is(tok::r_brace) && Right.MatchingParen &&
1938 Right.MatchingParen->BlockKind != BK_Block))
1939 return !Style.Cpp11BracedListStyle;
1940 if (Left.is(TT_BlockComment))
1941 return !Left.TokenText.endswith("=*/");
1942 if (Right.is(tok::l_paren)) {
1943 if (Left.is(tok::r_paren) && Left.is(TT_AttributeParen))
1945 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
1946 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
1947 (Left.isOneOf(tok::kw_if, tok::pp_elif, tok::kw_for, tok::kw_while,
1948 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
1950 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
1951 tok::kw_new, tok::kw_delete) &&
1952 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
1953 (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
1954 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
1955 Left.is(tok::r_paren)) &&
1956 Line.Type != LT_PreprocessorDirective);
1958 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
1960 if (Right.is(TT_UnaryOperator))
1961 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
1962 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
1963 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
1965 Left.isSimpleTypeSpecifier()) &&
1966 Right.is(tok::l_brace) && Right.getNextNonComment() &&
1967 Right.BlockKind != BK_Block)
1969 if (Left.is(tok::period) || Right.is(tok::period))
1971 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
1973 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
1974 Left.MatchingParen->Previous &&
1975 Left.MatchingParen->Previous->is(tok::period))
1976 // A.<B>DoSomething();
1978 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
1983 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
1984 const FormatToken &Right) {
1985 const FormatToken &Left = *Right.Previous;
1986 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
1987 return true; // Never ever merge two identifiers.
1988 if (Style.Language == FormatStyle::LK_Cpp) {
1989 if (Left.is(tok::kw_operator))
1990 return Right.is(tok::coloncolon);
1991 } else if (Style.Language == FormatStyle::LK_Proto) {
1992 if (Right.is(tok::period) &&
1993 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
1994 Keywords.kw_repeated, Keywords.kw_extend))
1996 if (Right.is(tok::l_paren) &&
1997 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
1999 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2000 if (Left.isOneOf(Keywords.kw_let, Keywords.kw_var, TT_JsFatArrow,
2003 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
2005 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
2007 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
2008 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
2010 if (Left.is(tok::ellipsis))
2012 if (Left.is(TT_TemplateCloser) &&
2013 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
2014 Keywords.kw_implements, Keywords.kw_extends))
2015 // Type assertions ('<type>expr') are not followed by whitespace. Other
2016 // locations that should have whitespace following are identified by the
2017 // above set of follower tokens.
2019 } else if (Style.Language == FormatStyle::LK_Java) {
2020 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
2022 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
2023 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
2024 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
2025 tok::kw_protected) ||
2026 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
2027 Keywords.kw_native)) &&
2028 Right.is(TT_TemplateOpener))
2031 if (Left.is(TT_ImplicitStringLiteral))
2032 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
2033 if (Line.Type == LT_ObjCMethodDecl) {
2034 if (Left.is(TT_ObjCMethodSpecifier))
2036 if (Left.is(tok::r_paren) && Right.is(tok::identifier))
2037 // Don't space between ')' and <id>
2040 if (Line.Type == LT_ObjCProperty &&
2041 (Right.is(tok::equal) || Left.is(tok::equal)))
2044 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
2045 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
2047 if (Left.is(tok::comma))
2049 if (Right.is(tok::comma))
2051 if (Right.isOneOf(TT_CtorInitializerColon, TT_ObjCBlockLParen))
2053 if (Right.is(TT_OverloadedOperatorLParen))
2054 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2055 if (Right.is(tok::colon)) {
2056 if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
2057 !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
2059 if (Right.is(TT_ObjCMethodExpr))
2061 if (Left.is(tok::question))
2063 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
2065 if (Right.is(TT_DictLiteral))
2066 return Style.SpacesInContainerLiterals;
2069 if (Left.is(TT_UnaryOperator))
2070 return Right.is(TT_BinaryOperator);
2072 // If the next token is a binary operator or a selector name, we have
2073 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
2074 if (Left.is(TT_CastRParen))
2075 return Style.SpaceAfterCStyleCast ||
2076 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
2078 if (Left.is(tok::greater) && Right.is(tok::greater))
2079 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
2080 (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
2081 if (Right.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
2082 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar))
2084 if (!Style.SpaceBeforeAssignmentOperators &&
2085 Right.getPrecedence() == prec::Assignment)
2087 if (Right.is(tok::coloncolon) && Left.isNot(tok::l_brace))
2088 return (Left.is(TT_TemplateOpener) &&
2089 Style.Standard == FormatStyle::LS_Cpp03) ||
2090 !(Left.isOneOf(tok::identifier, tok::l_paren, tok::r_paren) ||
2091 Left.isOneOf(TT_TemplateCloser, TT_TemplateOpener));
2092 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
2093 return Style.SpacesInAngles;
2094 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
2095 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
2096 !Right.is(tok::r_paren)))
2098 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
2099 Right.isNot(TT_FunctionTypeLParen))
2100 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
2101 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
2102 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
2104 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
2105 Line.startsWith(tok::hash))
2107 if (Right.is(TT_TrailingUnaryOperator))
2109 if (Left.is(TT_RegexLiteral))
2111 return spaceRequiredBetween(Line, Left, Right);
2114 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
2115 static bool isAllmanBrace(const FormatToken &Tok) {
2116 return Tok.is(tok::l_brace) && Tok.BlockKind == BK_Block &&
2117 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral);
2120 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
2121 const FormatToken &Right) {
2122 const FormatToken &Left = *Right.Previous;
2123 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
2126 if (Style.Language == FormatStyle::LK_JavaScript) {
2127 // FIXME: This might apply to other languages and token kinds.
2128 if (Right.is(tok::char_constant) && Left.is(tok::plus) && Left.Previous &&
2129 Left.Previous->is(tok::char_constant))
2131 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
2132 Left.Previous && Left.Previous->is(tok::equal) &&
2133 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
2135 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
2137 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
2138 // Object literals on the top level of a file are treated as "enum-style".
2139 // Each key/value pair is put on a separate line, instead of bin-packing.
2141 if (Left.is(tok::l_brace) && Line.Level == 0 &&
2142 (Line.startsWith(tok::kw_enum) ||
2143 Line.startsWith(tok::kw_export, tok::kw_enum)))
2144 // JavaScript top-level enum key/value pairs are put on separate lines
2145 // instead of bin-packing.
2147 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
2148 !Left.Children.empty())
2149 // Support AllowShortFunctionsOnASingleLine for JavaScript.
2150 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
2151 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
2152 (Left.NestingLevel == 0 && Line.Level == 0 &&
2153 Style.AllowShortFunctionsOnASingleLine ==
2154 FormatStyle::SFS_Inline);
2155 } else if (Style.Language == FormatStyle::LK_Java) {
2156 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
2157 Right.Next->is(tok::string_literal))
2161 // If the last token before a '}' is a comma or a trailing comment, the
2162 // intention is to insert a line break after it in order to make shuffling
2163 // around entries easier.
2164 const FormatToken *BeforeClosingBrace = nullptr;
2165 if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
2166 Left.BlockKind != BK_Block && Left.MatchingParen)
2167 BeforeClosingBrace = Left.MatchingParen->Previous;
2168 else if (Right.MatchingParen &&
2169 Right.MatchingParen->isOneOf(tok::l_brace,
2170 TT_ArrayInitializerLSquare))
2171 BeforeClosingBrace = &Left;
2172 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
2173 BeforeClosingBrace->isTrailingComment()))
2176 if (Right.is(tok::comment))
2177 return Left.BlockKind != BK_BracedInit &&
2178 Left.isNot(TT_CtorInitializerColon) &&
2179 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
2180 if (Left.isTrailingComment())
2182 if (Left.isStringLiteral() &&
2183 (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral)))
2185 if (Right.Previous->IsUnterminatedLiteral)
2187 if (Right.is(tok::lessless) && Right.Next &&
2188 Right.Previous->is(tok::string_literal) &&
2189 Right.Next->is(tok::string_literal))
2191 if (Right.Previous->ClosesTemplateDeclaration &&
2192 Right.Previous->MatchingParen &&
2193 Right.Previous->MatchingParen->NestingLevel == 0 &&
2194 Style.AlwaysBreakTemplateDeclarations)
2196 if ((Right.isOneOf(TT_CtorInitializerComma, TT_CtorInitializerColon)) &&
2197 Style.BreakConstructorInitializersBeforeComma &&
2198 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
2200 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
2201 // Raw string literals are special wrt. line breaks. The author has made a
2202 // deliberate choice and might have aligned the contents of the string
2203 // literal accordingly. Thus, we try keep existing line breaks.
2204 return Right.NewlinesBefore > 0;
2205 if (Right.Previous->is(tok::l_brace) && Right.NestingLevel == 1 &&
2206 Style.Language == FormatStyle::LK_Proto)
2207 // Don't put enums onto single lines in protocol buffers.
2209 if (Right.is(TT_InlineASMBrace))
2210 return Right.HasUnescapedNewline;
2211 if (isAllmanBrace(Left) || isAllmanBrace(Right))
2212 return (Line.startsWith(tok::kw_enum) && Style.BraceWrapping.AfterEnum) ||
2213 (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
2214 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
2215 if (Style.Language == FormatStyle::LK_Proto && Left.isNot(tok::l_brace) &&
2216 Right.is(TT_SelectorName))
2218 if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
2221 if ((Style.Language == FormatStyle::LK_Java ||
2222 Style.Language == FormatStyle::LK_JavaScript) &&
2223 Left.is(TT_LeadingJavaAnnotation) &&
2224 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
2225 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
2231 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
2232 const FormatToken &Right) {
2233 const FormatToken &Left = *Right.Previous;
2235 // Language-specific stuff.
2236 if (Style.Language == FormatStyle::LK_Java) {
2237 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2238 Keywords.kw_implements))
2240 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
2241 Keywords.kw_implements))
2243 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2244 if (Left.is(TT_JsFatArrow) && Right.is(tok::l_brace))
2246 if (Left.is(TT_JsTypeColon))
2248 if (Right.NestingLevel == 0 && Right.is(Keywords.kw_is))
2252 if (Left.is(tok::at))
2254 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
2256 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
2257 return !Right.is(tok::l_paren);
2258 if (Right.is(TT_PointerOrReference))
2259 return Line.IsMultiVariableDeclStmt ||
2260 (Style.PointerAlignment == FormatStyle::PAS_Right &&
2261 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
2262 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2263 Right.is(tok::kw_operator))
2265 if (Left.is(TT_PointerOrReference))
2267 if (Right.isTrailingComment())
2268 // We rely on MustBreakBefore being set correctly here as we should not
2269 // change the "binding" behavior of a comment.
2270 // The first comment in a braced lists is always interpreted as belonging to
2271 // the first list element. Otherwise, it should be placed outside of the
2273 return Left.BlockKind == BK_BracedInit;
2274 if (Left.is(tok::question) && Right.is(tok::colon))
2276 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
2277 return Style.BreakBeforeTernaryOperators;
2278 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
2279 return !Style.BreakBeforeTernaryOperators;
2280 if (Right.is(TT_InheritanceColon))
2282 if (Right.is(tok::colon) &&
2283 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
2285 if (Left.is(tok::colon) && (Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))
2287 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
2288 Right.Next->is(TT_ObjCMethodExpr)))
2289 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
2290 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
2292 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
2294 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
2295 TT_OverloadedOperator))
2297 if (Left.is(TT_RangeBasedForLoopColon))
2299 if (Right.is(TT_RangeBasedForLoopColon))
2301 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
2302 Left.is(tok::kw_operator))
2304 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
2305 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
2307 if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
2309 if (Left.is(tok::l_paren) && Left.Previous &&
2310 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
2312 if (Right.is(TT_ImplicitStringLiteral))
2315 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
2317 if (Right.is(tok::r_square) && Right.MatchingParen &&
2318 Right.MatchingParen->is(TT_LambdaLSquare))
2321 // We only break before r_brace if there was a corresponding break before
2322 // the l_brace, which is tracked by BreakBeforeClosingBrace.
2323 if (Right.is(tok::r_brace))
2324 return Right.MatchingParen && Right.MatchingParen->BlockKind == BK_Block;
2326 // Allow breaking after a trailing annotation, e.g. after a method
2328 if (Left.is(TT_TrailingAnnotation))
2329 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
2330 tok::less, tok::coloncolon);
2332 if (Right.is(tok::kw___attribute))
2335 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
2338 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2341 if (Left.is(TT_CtorInitializerComma) &&
2342 Style.BreakConstructorInitializersBeforeComma)
2344 if (Right.is(TT_CtorInitializerComma) &&
2345 Style.BreakConstructorInitializersBeforeComma)
2347 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
2348 (Left.is(tok::less) && Right.is(tok::less)))
2350 if (Right.is(TT_BinaryOperator) &&
2351 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
2352 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
2353 Right.getPrecedence() != prec::Assignment))
2355 if (Left.is(TT_ArrayInitializerLSquare))
2357 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
2359 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
2360 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
2361 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
2362 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
2363 Left.getPrecedence() == prec::Assignment))
2365 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
2366 tok::kw_class, tok::kw_struct) ||
2367 Right.isMemberAccess() ||
2368 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
2369 tok::colon, tok::l_square, tok::at) ||
2370 (Left.is(tok::r_paren) &&
2371 Right.isOneOf(tok::identifier, tok::kw_const)) ||
2372 (Left.is(tok::l_paren) && !Right.is(tok::r_paren));
2375 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
2376 llvm::errs() << "AnnotatedTokens:\n";
2377 const FormatToken *Tok = Line.First;
2379 llvm::errs() << " M=" << Tok->MustBreakBefore
2380 << " C=" << Tok->CanBreakBefore
2381 << " T=" << getTokenTypeName(Tok->Type)
2382 << " S=" << Tok->SpacesRequiredBefore
2383 << " B=" << Tok->BlockParameterCount
2384 << " P=" << Tok->SplitPenalty << " Name=" << Tok->Tok.getName()
2385 << " L=" << Tok->TotalLength << " PPK=" << Tok->PackingKind
2387 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
2388 llvm::errs() << Tok->FakeLParens[i] << "/";
2389 llvm::errs() << " FakeRParens=" << Tok->FakeRParens << "\n";
2391 assert(Tok == Line.Last);
2394 llvm::errs() << "----\n";
2397 } // namespace format
2398 } // namespace clang