1 //===--- TokenAnnotator.cpp - Format C++ code -----------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file implements a token annotator, i.e. creates
11 /// \c AnnotatedTokens out of \c FormatTokens with required extra information.
12 ///
13 //===----------------------------------------------------------------------===//
14
15 #include "TokenAnnotator.h"
16 #include "FormatToken.h"
17 #include "clang/Basic/SourceManager.h"
18 #include "clang/Basic/TokenKinds.h"
19 #include "llvm/ADT/SmallPtrSet.h"
20 #include "llvm/Support/Debug.h"
21
22 #define DEBUG_TYPE "format-token-annotator"
23
24 namespace clang {
25 namespace format {
26
27 namespace {
28
29 /// Returns \c true if the token can be used as an identifier in
30 /// an Objective-C \c \@selector, \c false otherwise.
31 ///
32 /// Because getFormattingLangOpts() always lexes source code as
33 /// Objective-C++, C++ keywords like \c new and \c delete are
34 /// lexed as tok::kw_*, not tok::identifier, even for Objective-C.
35 ///
36 /// For Objective-C and Objective-C++, both identifiers and keywords
37 /// are valid inside @selector(...) (or a macro which
38 /// invokes @selector(...)). So, we allow treat any identifier or
39 /// keyword as a potential Objective-C selector component.
canBeObjCSelectorComponent(const FormatToken & Tok)40 static bool canBeObjCSelectorComponent(const FormatToken &Tok) {
41 return Tok.Tok.getIdentifierInfo() != nullptr;
42 }
43
44 /// With `Left` being '(', check if we're at either `[...](` or
45 /// `[...]<...>(`, where the [ opens a lambda capture list.
isLambdaParameterList(const FormatToken * Left)46 static bool isLambdaParameterList(const FormatToken *Left) {
47 // Skip <...> if present.
48 if (Left->Previous && Left->Previous->is(tok::greater) &&
49 Left->Previous->MatchingParen &&
50 Left->Previous->MatchingParen->is(TT_TemplateOpener))
51 Left = Left->Previous->MatchingParen;
52
53 // Check for `[...]`.
54 return Left->Previous && Left->Previous->is(tok::r_square) &&
55 Left->Previous->MatchingParen &&
56 Left->Previous->MatchingParen->is(TT_LambdaLSquare);
57 }
58
59 /// Returns \c true if the token is followed by a boolean condition, \c false
60 /// otherwise.
isKeywordWithCondition(const FormatToken & Tok)61 static bool isKeywordWithCondition(const FormatToken &Tok) {
62 return Tok.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while, tok::kw_switch,
63 tok::kw_constexpr, tok::kw_catch);
64 }
65
66 /// A parser that gathers additional information about tokens.
67 ///
68 /// The \c TokenAnnotator tries to match parenthesis and square brakets and
69 /// store a parenthesis levels. It also tries to resolve matching "<" and ">"
70 /// into template parameter lists.
71 class AnnotatingParser {
72 public:
AnnotatingParser(const FormatStyle & Style,AnnotatedLine & Line,const AdditionalKeywords & Keywords)73 AnnotatingParser(const FormatStyle &Style, AnnotatedLine &Line,
74 const AdditionalKeywords &Keywords)
75 : Style(Style), Line(Line), CurrentToken(Line.First), AutoFound(false),
76 Keywords(Keywords) {
77 Contexts.push_back(Context(tok::unknown, 1, /*IsExpression=*/false));
78 resetTokenMetadata(CurrentToken);
79 }
80
81 private:
parseAngle()82 bool parseAngle() {
83 if (!CurrentToken || !CurrentToken->Previous)
84 return false;
85 if (NonTemplateLess.count(CurrentToken->Previous))
86 return false;
87
88 const FormatToken &Previous = *CurrentToken->Previous; // The '<'.
89 if (Previous.Previous) {
90 if (Previous.Previous->Tok.isLiteral())
91 return false;
92 if (Previous.Previous->is(tok::r_paren) && Contexts.size() > 1 &&
93 (!Previous.Previous->MatchingParen ||
94 !Previous.Previous->MatchingParen->is(TT_OverloadedOperatorLParen)))
95 return false;
96 }
97
98 FormatToken *Left = CurrentToken->Previous;
99 Left->ParentBracket = Contexts.back().ContextKind;
100 ScopedContextCreator ContextCreator(*this, tok::less, 12);
101
102 // If this angle is in the context of an expression, we need to be more
103 // hesitant to detect it as opening template parameters.
104 bool InExprContext = Contexts.back().IsExpression;
105
106 Contexts.back().IsExpression = false;
107 // If there's a template keyword before the opening angle bracket, this is a
108 // template parameter, not an argument.
109 Contexts.back().InTemplateArgument =
110 Left->Previous && Left->Previous->Tok.isNot(tok::kw_template);
111
112 if (Style.Language == FormatStyle::LK_Java &&
113 CurrentToken->is(tok::question))
114 next();
115
116 while (CurrentToken) {
117 if (CurrentToken->is(tok::greater)) {
118 // Try to do a better job at looking for ">>" within the condition of
119 // a statement. Conservatively insert spaces between consecutive ">"
120 // tokens to prevent splitting right bitshift operators and potentially
121 // altering program semantics. This check is overly conservative and
122 // will prevent spaces from being inserted in select nested template
123 // parameter cases, but should not alter program semantics.
124 if (CurrentToken->Next && CurrentToken->Next->is(tok::greater) &&
125 Left->ParentBracket != tok::less &&
126 (isKeywordWithCondition(*Line.First) ||
127 CurrentToken->getStartOfNonWhitespace() ==
128 CurrentToken->Next->getStartOfNonWhitespace().getLocWithOffset(
129 -1)))
130 return false;
131 Left->MatchingParen = CurrentToken;
132 CurrentToken->MatchingParen = Left;
133 // In TT_Proto, we must distignuish between:
134 // map<key, value>
135 // msg < item: data >
136 // msg: < item: data >
137 // In TT_TextProto, map<key, value> does not occur.
138 if (Style.Language == FormatStyle::LK_TextProto ||
139 (Style.Language == FormatStyle::LK_Proto && Left->Previous &&
140 Left->Previous->isOneOf(TT_SelectorName, TT_DictLiteral)))
141 CurrentToken->setType(TT_DictLiteral);
142 else
143 CurrentToken->setType(TT_TemplateCloser);
144 next();
145 return true;
146 }
147 if (CurrentToken->is(tok::question) &&
148 Style.Language == FormatStyle::LK_Java) {
149 next();
150 continue;
151 }
152 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
153 (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext &&
154 !Style.isCSharp() && Style.Language != FormatStyle::LK_Proto &&
155 Style.Language != FormatStyle::LK_TextProto))
156 return false;
157 // If a && or || is found and interpreted as a binary operator, this set
158 // of angles is likely part of something like "a < b && c > d". If the
159 // angles are inside an expression, the ||/&& might also be a binary
160 // operator that was misinterpreted because we are parsing template
161 // parameters.
162 // FIXME: This is getting out of hand, write a decent parser.
163 if (CurrentToken->Previous->isOneOf(tok::pipepipe, tok::ampamp) &&
164 CurrentToken->Previous->is(TT_BinaryOperator) &&
165 Contexts[Contexts.size() - 2].IsExpression &&
166 !Line.startsWith(tok::kw_template))
167 return false;
168 updateParameterCount(Left, CurrentToken);
169 if (Style.Language == FormatStyle::LK_Proto) {
170 if (FormatToken *Previous = CurrentToken->getPreviousNonComment()) {
171 if (CurrentToken->is(tok::colon) ||
172 (CurrentToken->isOneOf(tok::l_brace, tok::less) &&
173 Previous->isNot(tok::colon)))
174 Previous->setType(TT_SelectorName);
175 }
176 }
177 if (!consumeToken())
178 return false;
179 }
180 return false;
181 }
182
parseUntouchableParens()183 bool parseUntouchableParens() {
184 while (CurrentToken) {
185 CurrentToken->Finalized = true;
186 switch (CurrentToken->Tok.getKind()) {
187 case tok::l_paren:
188 next();
189 if (!parseUntouchableParens())
190 return false;
191 continue;
192 case tok::r_paren:
193 next();
194 return true;
195 default:
196 // no-op
197 break;
198 }
199 next();
200 }
201 return false;
202 }
203
parseParens(bool LookForDecls=false)204 bool parseParens(bool LookForDecls = false) {
205 if (!CurrentToken)
206 return false;
207 FormatToken *Left = CurrentToken->Previous;
208 assert(Left && "Unknown previous token");
209 FormatToken *PrevNonComment = Left->getPreviousNonComment();
210 Left->ParentBracket = Contexts.back().ContextKind;
211 ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
212
213 // FIXME: This is a bit of a hack. Do better.
214 Contexts.back().ColonIsForRangeExpr =
215 Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
216
217 if (Left->Previous && Left->Previous->is(TT_UntouchableMacroFunc)) {
218 Left->Finalized = true;
219 return parseUntouchableParens();
220 }
221
222 bool StartsObjCMethodExpr = false;
223 if (FormatToken *MaybeSel = Left->Previous) {
224 // @selector( starts a selector.
225 if (MaybeSel->isObjCAtKeyword(tok::objc_selector) && MaybeSel->Previous &&
226 MaybeSel->Previous->is(tok::at)) {
227 StartsObjCMethodExpr = true;
228 }
229 }
230
231 if (Left->is(TT_OverloadedOperatorLParen)) {
232 // Find the previous kw_operator token.
233 FormatToken *Prev = Left;
234 while (!Prev->is(tok::kw_operator)) {
235 Prev = Prev->Previous;
236 assert(Prev && "Expect a kw_operator prior to the OperatorLParen!");
237 }
238
239 // If faced with "a.operator*(argument)" or "a->operator*(argument)",
240 // i.e. the operator is called as a member function,
241 // then the argument must be an expression.
242 bool OperatorCalledAsMemberFunction =
243 Prev->Previous && Prev->Previous->isOneOf(tok::period, tok::arrow);
244 Contexts.back().IsExpression = OperatorCalledAsMemberFunction;
245 } else if (Style.Language == FormatStyle::LK_JavaScript &&
246 (Line.startsWith(Keywords.kw_type, tok::identifier) ||
247 Line.startsWith(tok::kw_export, Keywords.kw_type,
248 tok::identifier))) {
249 // type X = (...);
250 // export type X = (...);
251 Contexts.back().IsExpression = false;
252 } else if (Left->Previous &&
253 (Left->Previous->isOneOf(tok::kw_static_assert, tok::kw_while,
254 tok::l_paren, tok::comma) ||
255 Left->Previous->isIf() ||
256 Left->Previous->is(TT_BinaryOperator))) {
257 // static_assert, if and while usually contain expressions.
258 Contexts.back().IsExpression = true;
259 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
260 (Left->Previous->is(Keywords.kw_function) ||
261 (Left->Previous->endsSequence(tok::identifier,
262 Keywords.kw_function)))) {
263 // function(...) or function f(...)
264 Contexts.back().IsExpression = false;
265 } else if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
266 Left->Previous->is(TT_JsTypeColon)) {
267 // let x: (SomeType);
268 Contexts.back().IsExpression = false;
269 } else if (isLambdaParameterList(Left)) {
270 // This is a parameter list of a lambda expression.
271 Contexts.back().IsExpression = false;
272 } else if (Line.InPPDirective &&
273 (!Left->Previous || !Left->Previous->is(tok::identifier))) {
274 Contexts.back().IsExpression = true;
275 } else if (Contexts[Contexts.size() - 2].CaretFound) {
276 // This is the parameter list of an ObjC block.
277 Contexts.back().IsExpression = false;
278 } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
279 // The first argument to a foreach macro is a declaration.
280 Contexts.back().IsForEachMacro = true;
281 Contexts.back().IsExpression = false;
282 } else if (Left->Previous && Left->Previous->MatchingParen &&
283 Left->Previous->MatchingParen->is(TT_ObjCBlockLParen)) {
284 Contexts.back().IsExpression = false;
285 } else if (!Line.MustBeDeclaration && !Line.InPPDirective) {
286 bool IsForOrCatch =
287 Left->Previous && Left->Previous->isOneOf(tok::kw_for, tok::kw_catch);
288 Contexts.back().IsExpression = !IsForOrCatch;
289 }
290
291 // Infer the role of the l_paren based on the previous token if we haven't
292 // detected one one yet.
293 if (PrevNonComment && Left->is(TT_Unknown)) {
294 if (PrevNonComment->is(tok::kw___attribute)) {
295 Left->setType(TT_AttributeParen);
296 } else if (PrevNonComment->isOneOf(TT_TypenameMacro, tok::kw_decltype,
297 tok::kw_typeof, tok::kw__Atomic,
298 tok::kw___underlying_type)) {
299 Left->setType(TT_TypeDeclarationParen);
300 // decltype() and typeof() usually contain expressions.
301 if (PrevNonComment->isOneOf(tok::kw_decltype, tok::kw_typeof))
302 Contexts.back().IsExpression = true;
303 }
304 }
305
306 if (StartsObjCMethodExpr) {
307 Contexts.back().ColonIsObjCMethodExpr = true;
308 Left->setType(TT_ObjCMethodExpr);
309 }
310
311 // MightBeFunctionType and ProbablyFunctionType are used for
312 // function pointer and reference types as well as Objective-C
313 // block types:
314 //
315 // void (*FunctionPointer)(void);
316 // void (&FunctionReference)(void);
317 // void (^ObjCBlock)(void);
318 bool MightBeFunctionType = !Contexts[Contexts.size() - 2].IsExpression;
319 bool ProbablyFunctionType =
320 CurrentToken->isOneOf(tok::star, tok::amp, tok::caret);
321 bool HasMultipleLines = false;
322 bool HasMultipleParametersOnALine = false;
323 bool MightBeObjCForRangeLoop =
324 Left->Previous && Left->Previous->is(tok::kw_for);
325 FormatToken *PossibleObjCForInToken = nullptr;
326 while (CurrentToken) {
327 // LookForDecls is set when "if (" has been seen. Check for
328 // 'identifier' '*' 'identifier' followed by not '=' -- this
329 // '*' has to be a binary operator but determineStarAmpUsage() will
330 // categorize it as an unary operator, so set the right type here.
331 if (LookForDecls && CurrentToken->Next) {
332 FormatToken *Prev = CurrentToken->getPreviousNonComment();
333 if (Prev) {
334 FormatToken *PrevPrev = Prev->getPreviousNonComment();
335 FormatToken *Next = CurrentToken->Next;
336 if (PrevPrev && PrevPrev->is(tok::identifier) &&
337 Prev->isOneOf(tok::star, tok::amp, tok::ampamp) &&
338 CurrentToken->is(tok::identifier) && Next->isNot(tok::equal)) {
339 Prev->setType(TT_BinaryOperator);
340 LookForDecls = false;
341 }
342 }
343 }
344
345 if (CurrentToken->Previous->is(TT_PointerOrReference) &&
346 CurrentToken->Previous->Previous->isOneOf(tok::l_paren,
347 tok::coloncolon))
348 ProbablyFunctionType = true;
349 if (CurrentToken->is(tok::comma))
350 MightBeFunctionType = false;
351 if (CurrentToken->Previous->is(TT_BinaryOperator))
352 Contexts.back().IsExpression = true;
353 if (CurrentToken->is(tok::r_paren)) {
354 if (MightBeFunctionType && ProbablyFunctionType && CurrentToken->Next &&
355 (CurrentToken->Next->is(tok::l_paren) ||
356 (CurrentToken->Next->is(tok::l_square) && Line.MustBeDeclaration)))
357 Left->setType(Left->Next->is(tok::caret) ? TT_ObjCBlockLParen
358 : TT_FunctionTypeLParen);
359 Left->MatchingParen = CurrentToken;
360 CurrentToken->MatchingParen = Left;
361
362 if (CurrentToken->Next && CurrentToken->Next->is(tok::l_brace) &&
363 Left->Previous && Left->Previous->is(tok::l_paren)) {
364 // Detect the case where macros are used to generate lambdas or
365 // function bodies, e.g.:
366 // auto my_lambda = MACRO((Type *type, int i) { .. body .. });
367 for (FormatToken *Tok = Left; Tok != CurrentToken; Tok = Tok->Next) {
368 if (Tok->is(TT_BinaryOperator) &&
369 Tok->isOneOf(tok::star, tok::amp, tok::ampamp))
370 Tok->setType(TT_PointerOrReference);
371 }
372 }
373
374 if (StartsObjCMethodExpr) {
375 CurrentToken->setType(TT_ObjCMethodExpr);
376 if (Contexts.back().FirstObjCSelectorName) {
377 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
378 Contexts.back().LongestObjCSelectorName;
379 }
380 }
381
382 if (Left->is(TT_AttributeParen))
383 CurrentToken->setType(TT_AttributeParen);
384 if (Left->is(TT_TypeDeclarationParen))
385 CurrentToken->setType(TT_TypeDeclarationParen);
386 if (Left->Previous && Left->Previous->is(TT_JavaAnnotation))
387 CurrentToken->setType(TT_JavaAnnotation);
388 if (Left->Previous && Left->Previous->is(TT_LeadingJavaAnnotation))
389 CurrentToken->setType(TT_LeadingJavaAnnotation);
390 if (Left->Previous && Left->Previous->is(TT_AttributeSquare))
391 CurrentToken->setType(TT_AttributeSquare);
392
393 if (!HasMultipleLines)
394 Left->setPackingKind(PPK_Inconclusive);
395 else if (HasMultipleParametersOnALine)
396 Left->setPackingKind(PPK_BinPacked);
397 else
398 Left->setPackingKind(PPK_OnePerLine);
399
400 next();
401 return true;
402 }
403 if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
404 return false;
405
406 if (CurrentToken->is(tok::l_brace))
407 Left->setType(TT_Unknown); // Not TT_ObjCBlockLParen
408 if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
409 !CurrentToken->Next->HasUnescapedNewline &&
410 !CurrentToken->Next->isTrailingComment())
411 HasMultipleParametersOnALine = true;
412 bool ProbablyFunctionTypeLParen =
413 (CurrentToken->is(tok::l_paren) && CurrentToken->Next &&
414 CurrentToken->Next->isOneOf(tok::star, tok::amp, tok::caret));
415 if ((CurrentToken->Previous->isOneOf(tok::kw_const, tok::kw_auto) ||
416 CurrentToken->Previous->isSimpleTypeSpecifier()) &&
417 !(CurrentToken->is(tok::l_brace) ||
418 (CurrentToken->is(tok::l_paren) && !ProbablyFunctionTypeLParen)))
419 Contexts.back().IsExpression = false;
420 if (CurrentToken->isOneOf(tok::semi, tok::colon)) {
421 MightBeObjCForRangeLoop = false;
422 if (PossibleObjCForInToken) {
423 PossibleObjCForInToken->setType(TT_Unknown);
424 PossibleObjCForInToken = nullptr;
425 }
426 }
427 if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in)) {
428 PossibleObjCForInToken = CurrentToken;
429 PossibleObjCForInToken->setType(TT_ObjCForIn);
430 }
431 // When we discover a 'new', we set CanBeExpression to 'false' in order to
432 // parse the type correctly. Reset that after a comma.
433 if (CurrentToken->is(tok::comma))
434 Contexts.back().CanBeExpression = true;
435
436 FormatToken *Tok = CurrentToken;
437 if (!consumeToken())
438 return false;
439 updateParameterCount(Left, Tok);
440 if (CurrentToken && CurrentToken->HasUnescapedNewline)
441 HasMultipleLines = true;
442 }
443 return false;
444 }
445
isCSharpAttributeSpecifier(const FormatToken & Tok)446 bool isCSharpAttributeSpecifier(const FormatToken &Tok) {
447 if (!Style.isCSharp())
448 return false;
449
450 // `identifier[i]` is not an attribute.
451 if (Tok.Previous && Tok.Previous->is(tok::identifier))
452 return false;
453
454 // Chains of [] in `identifier[i][j][k]` are not attributes.
455 if (Tok.Previous && Tok.Previous->is(tok::r_square)) {
456 auto *MatchingParen = Tok.Previous->MatchingParen;
457 if (!MatchingParen || MatchingParen->is(TT_ArraySubscriptLSquare))
458 return false;
459 }
460
461 const FormatToken *AttrTok = Tok.Next;
462 if (!AttrTok)
463 return false;
464
465 // Just an empty declaration e.g. string [].
466 if (AttrTok->is(tok::r_square))
467 return false;
468
469 // Move along the tokens inbetween the '[' and ']' e.g. [STAThread].
470 while (AttrTok && AttrTok->isNot(tok::r_square)) {
471 AttrTok = AttrTok->Next;
472 }
473
474 if (!AttrTok)
475 return false;
476
477 // Allow an attribute to be the only content of a file.
478 AttrTok = AttrTok->Next;
479 if (!AttrTok)
480 return true;
481
482 // Limit this to being an access modifier that follows.
483 if (AttrTok->isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected,
484 tok::comment, tok::kw_class, tok::kw_static,
485 tok::l_square, Keywords.kw_internal)) {
486 return true;
487 }
488
489 // incase its a [XXX] retval func(....
490 if (AttrTok->Next &&
491 AttrTok->Next->startsSequence(tok::identifier, tok::l_paren))
492 return true;
493
494 return false;
495 }
496
isCpp11AttributeSpecifier(const FormatToken & Tok)497 bool isCpp11AttributeSpecifier(const FormatToken &Tok) {
498 if (!Style.isCpp() || !Tok.startsSequence(tok::l_square, tok::l_square))
499 return false;
500 // The first square bracket is part of an ObjC array literal
501 if (Tok.Previous && Tok.Previous->is(tok::at)) {
502 return false;
503 }
504 const FormatToken *AttrTok = Tok.Next->Next;
505 if (!AttrTok)
506 return false;
507 // C++17 '[[using ns: foo, bar(baz, blech)]]'
508 // We assume nobody will name an ObjC variable 'using'.
509 if (AttrTok->startsSequence(tok::kw_using, tok::identifier, tok::colon))
510 return true;
511 if (AttrTok->isNot(tok::identifier))
512 return false;
513 while (AttrTok && !AttrTok->startsSequence(tok::r_square, tok::r_square)) {
514 // ObjC message send. We assume nobody will use : in a C++11 attribute
515 // specifier parameter, although this is technically valid:
516 // [[foo(:)]].
517 if (AttrTok->is(tok::colon) ||
518 AttrTok->startsSequence(tok::identifier, tok::identifier) ||
519 AttrTok->startsSequence(tok::r_paren, tok::identifier))
520 return false;
521 if (AttrTok->is(tok::ellipsis))
522 return true;
523 AttrTok = AttrTok->Next;
524 }
525 return AttrTok && AttrTok->startsSequence(tok::r_square, tok::r_square);
526 }
527
parseSquare()528 bool parseSquare() {
529 if (!CurrentToken)
530 return false;
531
532 // A '[' could be an index subscript (after an identifier or after
533 // ')' or ']'), it could be the start of an Objective-C method
534 // expression, it could the start of an Objective-C array literal,
535 // or it could be a C++ attribute specifier [[foo::bar]].
536 FormatToken *Left = CurrentToken->Previous;
537 Left->ParentBracket = Contexts.back().ContextKind;
538 FormatToken *Parent = Left->getPreviousNonComment();
539
540 // Cases where '>' is followed by '['.
541 // In C++, this can happen either in array of templates (foo<int>[10])
542 // or when array is a nested template type (unique_ptr<type1<type2>[]>).
543 bool CppArrayTemplates =
544 Style.isCpp() && Parent && Parent->is(TT_TemplateCloser) &&
545 (Contexts.back().CanBeExpression || Contexts.back().IsExpression ||
546 Contexts.back().InTemplateArgument);
547
548 bool IsCpp11AttributeSpecifier = isCpp11AttributeSpecifier(*Left) ||
549 Contexts.back().InCpp11AttributeSpecifier;
550
551 // Treat C# Attributes [STAThread] much like C++ attributes [[...]].
552 bool IsCSharpAttributeSpecifier =
553 isCSharpAttributeSpecifier(*Left) ||
554 Contexts.back().InCSharpAttributeSpecifier;
555
556 bool InsideInlineASM = Line.startsWith(tok::kw_asm);
557 bool IsCppStructuredBinding = Left->isCppStructuredBinding(Style);
558 bool StartsObjCMethodExpr =
559 !IsCppStructuredBinding && !InsideInlineASM && !CppArrayTemplates &&
560 Style.isCpp() && !IsCpp11AttributeSpecifier &&
561 !IsCSharpAttributeSpecifier && Contexts.back().CanBeExpression &&
562 Left->isNot(TT_LambdaLSquare) &&
563 !CurrentToken->isOneOf(tok::l_brace, tok::r_square) &&
564 (!Parent ||
565 Parent->isOneOf(tok::colon, tok::l_square, tok::l_paren,
566 tok::kw_return, tok::kw_throw) ||
567 Parent->isUnaryOperator() ||
568 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
569 Parent->isOneOf(TT_ObjCForIn, TT_CastRParen) ||
570 (getBinOpPrecedence(Parent->Tok.getKind(), true, true) >
571 prec::Unknown));
572 bool ColonFound = false;
573
574 unsigned BindingIncrease = 1;
575 if (IsCppStructuredBinding) {
576 Left->setType(TT_StructuredBindingLSquare);
577 } else if (Left->is(TT_Unknown)) {
578 if (StartsObjCMethodExpr) {
579 Left->setType(TT_ObjCMethodExpr);
580 } else if (InsideInlineASM) {
581 Left->setType(TT_InlineASMSymbolicNameLSquare);
582 } else if (IsCpp11AttributeSpecifier) {
583 Left->setType(TT_AttributeSquare);
584 } else if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
585 Contexts.back().ContextKind == tok::l_brace &&
586 Parent->isOneOf(tok::l_brace, tok::comma)) {
587 Left->setType(TT_JsComputedPropertyName);
588 } else if (Style.isCpp() && Contexts.back().ContextKind == tok::l_brace &&
589 Parent && Parent->isOneOf(tok::l_brace, tok::comma)) {
590 Left->setType(TT_DesignatedInitializerLSquare);
591 } else if (IsCSharpAttributeSpecifier) {
592 Left->setType(TT_AttributeSquare);
593 } else if (CurrentToken->is(tok::r_square) && Parent &&
594 Parent->is(TT_TemplateCloser)) {
595 Left->setType(TT_ArraySubscriptLSquare);
596 } else if (Style.Language == FormatStyle::LK_Proto ||
597 Style.Language == FormatStyle::LK_TextProto) {
598 // Square braces in LK_Proto can either be message field attributes:
599 //
600 // optional Aaa aaa = 1 [
601 // (aaa) = aaa
602 // ];
603 //
604 // extensions 123 [
605 // (aaa) = aaa
606 // ];
607 //
608 // or text proto extensions (in options):
609 //
610 // option (Aaa.options) = {
611 // [type.type/type] {
612 // key: value
613 // }
614 // }
615 //
616 // or repeated fields (in options):
617 //
618 // option (Aaa.options) = {
619 // keys: [ 1, 2, 3 ]
620 // }
621 //
622 // In the first and the third case we want to spread the contents inside
623 // the square braces; in the second we want to keep them inline.
624 Left->setType(TT_ArrayInitializerLSquare);
625 if (!Left->endsSequence(tok::l_square, tok::numeric_constant,
626 tok::equal) &&
627 !Left->endsSequence(tok::l_square, tok::numeric_constant,
628 tok::identifier) &&
629 !Left->endsSequence(tok::l_square, tok::colon, TT_SelectorName)) {
630 Left->setType(TT_ProtoExtensionLSquare);
631 BindingIncrease = 10;
632 }
633 } else if (!CppArrayTemplates && Parent &&
634 Parent->isOneOf(TT_BinaryOperator, TT_TemplateCloser, tok::at,
635 tok::comma, tok::l_paren, tok::l_square,
636 tok::question, tok::colon, tok::kw_return,
637 // Should only be relevant to JavaScript:
638 tok::kw_default)) {
639 Left->setType(TT_ArrayInitializerLSquare);
640 } else {
641 BindingIncrease = 10;
642 Left->setType(TT_ArraySubscriptLSquare);
643 }
644 }
645
646 ScopedContextCreator ContextCreator(*this, tok::l_square, BindingIncrease);
647 Contexts.back().IsExpression = true;
648 if (Style.Language == FormatStyle::LK_JavaScript && Parent &&
649 Parent->is(TT_JsTypeColon))
650 Contexts.back().IsExpression = false;
651
652 Contexts.back().ColonIsObjCMethodExpr = StartsObjCMethodExpr;
653 Contexts.back().InCpp11AttributeSpecifier = IsCpp11AttributeSpecifier;
654 Contexts.back().InCSharpAttributeSpecifier = IsCSharpAttributeSpecifier;
655
656 while (CurrentToken) {
657 if (CurrentToken->is(tok::r_square)) {
658 if (IsCpp11AttributeSpecifier)
659 CurrentToken->setType(TT_AttributeSquare);
660 if (IsCSharpAttributeSpecifier)
661 CurrentToken->setType(TT_AttributeSquare);
662 else if (((CurrentToken->Next &&
663 CurrentToken->Next->is(tok::l_paren)) ||
664 (CurrentToken->Previous &&
665 CurrentToken->Previous->Previous == Left)) &&
666 Left->is(TT_ObjCMethodExpr)) {
667 // An ObjC method call is rarely followed by an open parenthesis. It
668 // also can't be composed of just one token, unless it's a macro that
669 // will be expanded to more tokens.
670 // FIXME: Do we incorrectly label ":" with this?
671 StartsObjCMethodExpr = false;
672 Left->setType(TT_Unknown);
673 }
674 if (StartsObjCMethodExpr && CurrentToken->Previous != Left) {
675 CurrentToken->setType(TT_ObjCMethodExpr);
676 // If we haven't seen a colon yet, make sure the last identifier
677 // before the r_square is tagged as a selector name component.
678 if (!ColonFound && CurrentToken->Previous &&
679 CurrentToken->Previous->is(TT_Unknown) &&
680 canBeObjCSelectorComponent(*CurrentToken->Previous))
681 CurrentToken->Previous->setType(TT_SelectorName);
682 // determineStarAmpUsage() thinks that '*' '[' is allocating an
683 // array of pointers, but if '[' starts a selector then '*' is a
684 // binary operator.
685 if (Parent && Parent->is(TT_PointerOrReference))
686 Parent->setType(TT_BinaryOperator);
687 }
688 // An arrow after an ObjC method expression is not a lambda arrow.
689 if (CurrentToken->getType() == TT_ObjCMethodExpr &&
690 CurrentToken->Next && CurrentToken->Next->is(TT_LambdaArrow))
691 CurrentToken->Next->setType(TT_Unknown);
692 Left->MatchingParen = CurrentToken;
693 CurrentToken->MatchingParen = Left;
694 // FirstObjCSelectorName is set when a colon is found. This does
695 // not work, however, when the method has no parameters.
696 // Here, we set FirstObjCSelectorName when the end of the method call is
697 // reached, in case it was not set already.
698 if (!Contexts.back().FirstObjCSelectorName) {
699 FormatToken *Previous = CurrentToken->getPreviousNonComment();
700 if (Previous && Previous->is(TT_SelectorName)) {
701 Previous->ObjCSelectorNameParts = 1;
702 Contexts.back().FirstObjCSelectorName = Previous;
703 }
704 } else {
705 Left->ParameterCount =
706 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
707 }
708 if (Contexts.back().FirstObjCSelectorName) {
709 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
710 Contexts.back().LongestObjCSelectorName;
711 if (Left->BlockParameterCount > 1)
712 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName = 0;
713 }
714 next();
715 return true;
716 }
717 if (CurrentToken->isOneOf(tok::r_paren, tok::r_brace))
718 return false;
719 if (CurrentToken->is(tok::colon)) {
720 if (IsCpp11AttributeSpecifier &&
721 CurrentToken->endsSequence(tok::colon, tok::identifier,
722 tok::kw_using)) {
723 // Remember that this is a [[using ns: foo]] C++ attribute, so we
724 // don't add a space before the colon (unlike other colons).
725 CurrentToken->setType(TT_AttributeColon);
726 } else if (Left->isOneOf(TT_ArraySubscriptLSquare,
727 TT_DesignatedInitializerLSquare)) {
728 Left->setType(TT_ObjCMethodExpr);
729 StartsObjCMethodExpr = true;
730 Contexts.back().ColonIsObjCMethodExpr = true;
731 if (Parent && Parent->is(tok::r_paren))
732 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
733 Parent->setType(TT_CastRParen);
734 }
735 ColonFound = true;
736 }
737 if (CurrentToken->is(tok::comma) && Left->is(TT_ObjCMethodExpr) &&
738 !ColonFound)
739 Left->setType(TT_ArrayInitializerLSquare);
740 FormatToken *Tok = CurrentToken;
741 if (!consumeToken())
742 return false;
743 updateParameterCount(Left, Tok);
744 }
745 return false;
746 }
747
couldBeInStructArrayInitializer() const748 bool couldBeInStructArrayInitializer() const {
749 if (Contexts.size() < 2)
750 return false;
751 // We want to back up no more then 2 context levels i.e.
752 // . { { <-
753 const auto End = std::next(Contexts.rbegin(), 2);
754 auto Last = Contexts.rbegin();
755 unsigned Depth = 0;
756 for (; Last != End; ++Last) {
757 if (Last->ContextKind == tok::l_brace)
758 ++Depth;
759 }
760 return Depth == 2 && Last->ContextKind != tok::l_brace;
761 }
762
parseBrace()763 bool parseBrace() {
764 if (CurrentToken) {
765 FormatToken *Left = CurrentToken->Previous;
766 Left->ParentBracket = Contexts.back().ContextKind;
767
768 if (Contexts.back().CaretFound)
769 Left->setType(TT_ObjCBlockLBrace);
770 Contexts.back().CaretFound = false;
771
772 ScopedContextCreator ContextCreator(*this, tok::l_brace, 1);
773 Contexts.back().ColonIsDictLiteral = true;
774 if (Left->is(BK_BracedInit))
775 Contexts.back().IsExpression = true;
776 if (Style.Language == FormatStyle::LK_JavaScript && Left->Previous &&
777 Left->Previous->is(TT_JsTypeColon))
778 Contexts.back().IsExpression = false;
779
780 unsigned CommaCount = 0;
781 while (CurrentToken) {
782 if (CurrentToken->is(tok::r_brace)) {
783 Left->MatchingParen = CurrentToken;
784 CurrentToken->MatchingParen = Left;
785 if (Style.AlignArrayOfStructures != FormatStyle::AIAS_None) {
786 if (Left->ParentBracket == tok::l_brace &&
787 couldBeInStructArrayInitializer() && CommaCount > 0) {
788 Contexts.back().InStructArrayInitializer = true;
789 }
790 }
791 next();
792 return true;
793 }
794 if (CurrentToken->isOneOf(tok::r_paren, tok::r_square))
795 return false;
796 updateParameterCount(Left, CurrentToken);
797 if (CurrentToken->isOneOf(tok::colon, tok::l_brace, tok::less)) {
798 FormatToken *Previous = CurrentToken->getPreviousNonComment();
799 if (Previous->is(TT_JsTypeOptionalQuestion))
800 Previous = Previous->getPreviousNonComment();
801 if ((CurrentToken->is(tok::colon) &&
802 (!Contexts.back().ColonIsDictLiteral || !Style.isCpp())) ||
803 Style.Language == FormatStyle::LK_Proto ||
804 Style.Language == FormatStyle::LK_TextProto) {
805 Left->setType(TT_DictLiteral);
806 if (Previous->Tok.getIdentifierInfo() ||
807 Previous->is(tok::string_literal))
808 Previous->setType(TT_SelectorName);
809 }
810 if (CurrentToken->is(tok::colon) ||
811 Style.Language == FormatStyle::LK_JavaScript)
812 Left->setType(TT_DictLiteral);
813 }
814 if (CurrentToken->is(tok::comma)) {
815 if (Style.Language == FormatStyle::LK_JavaScript)
816 Left->setType(TT_DictLiteral);
817 ++CommaCount;
818 }
819 if (!consumeToken())
820 return false;
821 }
822 }
823 return true;
824 }
825
updateParameterCount(FormatToken * Left,FormatToken * Current)826 void updateParameterCount(FormatToken *Left, FormatToken *Current) {
827 // For ObjC methods, the number of parameters is calculated differently as
828 // method declarations have a different structure (the parameters are not
829 // inside a bracket scope).
830 if (Current->is(tok::l_brace) && Current->is(BK_Block))
831 ++Left->BlockParameterCount;
832 if (Current->is(tok::comma)) {
833 ++Left->ParameterCount;
834 if (!Left->Role)
835 Left->Role.reset(new CommaSeparatedList(Style));
836 Left->Role->CommaFound(Current);
837 } else if (Left->ParameterCount == 0 && Current->isNot(tok::comment)) {
838 Left->ParameterCount = 1;
839 }
840 }
841
parseConditional()842 bool parseConditional() {
843 while (CurrentToken) {
844 if (CurrentToken->is(tok::colon)) {
845 CurrentToken->setType(TT_ConditionalExpr);
846 next();
847 return true;
848 }
849 if (!consumeToken())
850 return false;
851 }
852 return false;
853 }
854
parseTemplateDeclaration()855 bool parseTemplateDeclaration() {
856 if (CurrentToken && CurrentToken->is(tok::less)) {
857 CurrentToken->setType(TT_TemplateOpener);
858 next();
859 if (!parseAngle())
860 return false;
861 if (CurrentToken)
862 CurrentToken->Previous->ClosesTemplateDeclaration = true;
863 return true;
864 }
865 return false;
866 }
867
consumeToken()868 bool consumeToken() {
869 FormatToken *Tok = CurrentToken;
870 next();
871 switch (Tok->Tok.getKind()) {
872 case tok::plus:
873 case tok::minus:
874 if (!Tok->Previous && Line.MustBeDeclaration)
875 Tok->setType(TT_ObjCMethodSpecifier);
876 break;
877 case tok::colon:
878 if (!Tok->Previous)
879 return false;
880 // Colons from ?: are handled in parseConditional().
881 if (Style.Language == FormatStyle::LK_JavaScript) {
882 if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
883 (Contexts.size() == 1 && // switch/case labels
884 !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
885 Contexts.back().ContextKind == tok::l_paren || // function params
886 Contexts.back().ContextKind == tok::l_square || // array type
887 (!Contexts.back().IsExpression &&
888 Contexts.back().ContextKind == tok::l_brace) || // object type
889 (Contexts.size() == 1 &&
890 Line.MustBeDeclaration)) { // method/property declaration
891 Contexts.back().IsExpression = false;
892 Tok->setType(TT_JsTypeColon);
893 break;
894 }
895 } else if (Style.isCSharp()) {
896 if (Contexts.back().InCSharpAttributeSpecifier) {
897 Tok->setType(TT_AttributeColon);
898 break;
899 }
900 if (Contexts.back().ContextKind == tok::l_paren) {
901 Tok->setType(TT_CSharpNamedArgumentColon);
902 break;
903 }
904 }
905 if (Contexts.back().ColonIsDictLiteral ||
906 Style.Language == FormatStyle::LK_Proto ||
907 Style.Language == FormatStyle::LK_TextProto) {
908 Tok->setType(TT_DictLiteral);
909 if (Style.Language == FormatStyle::LK_TextProto) {
910 if (FormatToken *Previous = Tok->getPreviousNonComment())
911 Previous->setType(TT_SelectorName);
912 }
913 } else if (Contexts.back().ColonIsObjCMethodExpr ||
914 Line.startsWith(TT_ObjCMethodSpecifier)) {
915 Tok->setType(TT_ObjCMethodExpr);
916 const FormatToken *BeforePrevious = Tok->Previous->Previous;
917 // Ensure we tag all identifiers in method declarations as
918 // TT_SelectorName.
919 bool UnknownIdentifierInMethodDeclaration =
920 Line.startsWith(TT_ObjCMethodSpecifier) &&
921 Tok->Previous->is(tok::identifier) && Tok->Previous->is(TT_Unknown);
922 if (!BeforePrevious ||
923 // FIXME(bug 36976): ObjC return types shouldn't use TT_CastRParen.
924 !(BeforePrevious->is(TT_CastRParen) ||
925 (BeforePrevious->is(TT_ObjCMethodExpr) &&
926 BeforePrevious->is(tok::colon))) ||
927 BeforePrevious->is(tok::r_square) ||
928 Contexts.back().LongestObjCSelectorName == 0 ||
929 UnknownIdentifierInMethodDeclaration) {
930 Tok->Previous->setType(TT_SelectorName);
931 if (!Contexts.back().FirstObjCSelectorName)
932 Contexts.back().FirstObjCSelectorName = Tok->Previous;
933 else if (Tok->Previous->ColumnWidth >
934 Contexts.back().LongestObjCSelectorName)
935 Contexts.back().LongestObjCSelectorName =
936 Tok->Previous->ColumnWidth;
937 Tok->Previous->ParameterIndex =
938 Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
939 ++Contexts.back().FirstObjCSelectorName->ObjCSelectorNameParts;
940 }
941 } else if (Contexts.back().ColonIsForRangeExpr) {
942 Tok->setType(TT_RangeBasedForLoopColon);
943 } else if (CurrentToken && CurrentToken->is(tok::numeric_constant)) {
944 Tok->setType(TT_BitFieldColon);
945 } else if (Contexts.size() == 1 &&
946 !Line.First->isOneOf(tok::kw_enum, tok::kw_case,
947 tok::kw_default)) {
948 FormatToken *Prev = Tok->getPreviousNonComment();
949 if (Prev->isOneOf(tok::r_paren, tok::kw_noexcept))
950 Tok->setType(TT_CtorInitializerColon);
951 else if (Prev->is(tok::kw_try)) {
952 // Member initializer list within function try block.
953 FormatToken *PrevPrev = Prev->getPreviousNonComment();
954 if (PrevPrev && PrevPrev->isOneOf(tok::r_paren, tok::kw_noexcept))
955 Tok->setType(TT_CtorInitializerColon);
956 } else
957 Tok->setType(TT_InheritanceColon);
958 } else if (canBeObjCSelectorComponent(*Tok->Previous) && Tok->Next &&
959 (Tok->Next->isOneOf(tok::r_paren, tok::comma) ||
960 (canBeObjCSelectorComponent(*Tok->Next) && Tok->Next->Next &&
961 Tok->Next->Next->is(tok::colon)))) {
962 // This handles a special macro in ObjC code where selectors including
963 // the colon are passed as macro arguments.
964 Tok->setType(TT_ObjCMethodExpr);
965 } else if (Contexts.back().ContextKind == tok::l_paren) {
966 Tok->setType(TT_InlineASMColon);
967 }
968 break;
969 case tok::pipe:
970 case tok::amp:
971 // | and & in declarations/type expressions represent union and
972 // intersection types, respectively.
973 if (Style.Language == FormatStyle::LK_JavaScript &&
974 !Contexts.back().IsExpression)
975 Tok->setType(TT_JsTypeOperator);
976 break;
977 case tok::kw_if:
978 case tok::kw_while:
979 if (Tok->is(tok::kw_if) && CurrentToken &&
980 CurrentToken->isOneOf(tok::kw_constexpr, tok::identifier))
981 next();
982 if (CurrentToken && CurrentToken->is(tok::l_paren)) {
983 next();
984 if (!parseParens(/*LookForDecls=*/true))
985 return false;
986 }
987 break;
988 case tok::kw_for:
989 if (Style.Language == FormatStyle::LK_JavaScript) {
990 // x.for and {for: ...}
991 if ((Tok->Previous && Tok->Previous->is(tok::period)) ||
992 (Tok->Next && Tok->Next->is(tok::colon)))
993 break;
994 // JS' for await ( ...
995 if (CurrentToken && CurrentToken->is(Keywords.kw_await))
996 next();
997 }
998 Contexts.back().ColonIsForRangeExpr = true;
999 next();
1000 if (!parseParens())
1001 return false;
1002 break;
1003 case tok::l_paren:
1004 // When faced with 'operator()()', the kw_operator handler incorrectly
1005 // marks the first l_paren as a OverloadedOperatorLParen. Here, we make
1006 // the first two parens OverloadedOperators and the second l_paren an
1007 // OverloadedOperatorLParen.
1008 if (Tok->Previous && Tok->Previous->is(tok::r_paren) &&
1009 Tok->Previous->MatchingParen &&
1010 Tok->Previous->MatchingParen->is(TT_OverloadedOperatorLParen)) {
1011 Tok->Previous->setType(TT_OverloadedOperator);
1012 Tok->Previous->MatchingParen->setType(TT_OverloadedOperator);
1013 Tok->setType(TT_OverloadedOperatorLParen);
1014 }
1015
1016 if (!parseParens())
1017 return false;
1018 if (Line.MustBeDeclaration && Contexts.size() == 1 &&
1019 !Contexts.back().IsExpression && !Line.startsWith(TT_ObjCProperty) &&
1020 !Tok->is(TT_TypeDeclarationParen) &&
1021 (!Tok->Previous || !Tok->Previous->isOneOf(tok::kw___attribute,
1022 TT_LeadingJavaAnnotation)))
1023 Line.MightBeFunctionDecl = true;
1024 break;
1025 case tok::l_square:
1026 if (!parseSquare())
1027 return false;
1028 break;
1029 case tok::l_brace:
1030 if (Style.Language == FormatStyle::LK_TextProto) {
1031 FormatToken *Previous = Tok->getPreviousNonComment();
1032 if (Previous && Previous->getType() != TT_DictLiteral)
1033 Previous->setType(TT_SelectorName);
1034 }
1035 if (!parseBrace())
1036 return false;
1037 break;
1038 case tok::less:
1039 if (parseAngle()) {
1040 Tok->setType(TT_TemplateOpener);
1041 // In TT_Proto, we must distignuish between:
1042 // map<key, value>
1043 // msg < item: data >
1044 // msg: < item: data >
1045 // In TT_TextProto, map<key, value> does not occur.
1046 if (Style.Language == FormatStyle::LK_TextProto ||
1047 (Style.Language == FormatStyle::LK_Proto && Tok->Previous &&
1048 Tok->Previous->isOneOf(TT_SelectorName, TT_DictLiteral))) {
1049 Tok->setType(TT_DictLiteral);
1050 FormatToken *Previous = Tok->getPreviousNonComment();
1051 if (Previous && Previous->getType() != TT_DictLiteral)
1052 Previous->setType(TT_SelectorName);
1053 }
1054 } else {
1055 Tok->setType(TT_BinaryOperator);
1056 NonTemplateLess.insert(Tok);
1057 CurrentToken = Tok;
1058 next();
1059 }
1060 break;
1061 case tok::r_paren:
1062 case tok::r_square:
1063 return false;
1064 case tok::r_brace:
1065 // Lines can start with '}'.
1066 if (Tok->Previous)
1067 return false;
1068 break;
1069 case tok::greater:
1070 if (Style.Language != FormatStyle::LK_TextProto)
1071 Tok->setType(TT_BinaryOperator);
1072 if (Tok->Previous && Tok->Previous->is(TT_TemplateCloser))
1073 Tok->SpacesRequiredBefore = 1;
1074 break;
1075 case tok::kw_operator:
1076 if (Style.Language == FormatStyle::LK_TextProto ||
1077 Style.Language == FormatStyle::LK_Proto)
1078 break;
1079 while (CurrentToken &&
1080 !CurrentToken->isOneOf(tok::l_paren, tok::semi, tok::r_paren)) {
1081 if (CurrentToken->isOneOf(tok::star, tok::amp))
1082 CurrentToken->setType(TT_PointerOrReference);
1083 consumeToken();
1084 if (CurrentToken && CurrentToken->is(tok::comma) &&
1085 CurrentToken->Previous->isNot(tok::kw_operator))
1086 break;
1087 if (CurrentToken && CurrentToken->Previous->isOneOf(
1088 TT_BinaryOperator, TT_UnaryOperator, tok::comma,
1089 tok::star, tok::arrow, tok::amp, tok::ampamp))
1090 CurrentToken->Previous->setType(TT_OverloadedOperator);
1091 }
1092 if (CurrentToken && CurrentToken->is(tok::l_paren))
1093 CurrentToken->setType(TT_OverloadedOperatorLParen);
1094 if (CurrentToken && CurrentToken->Previous->is(TT_BinaryOperator))
1095 CurrentToken->Previous->setType(TT_OverloadedOperator);
1096 break;
1097 case tok::question:
1098 if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
1099 Tok->Next->isOneOf(tok::semi, tok::comma, tok::colon, tok::r_paren,
1100 tok::r_brace)) {
1101 // Question marks before semicolons, colons, etc. indicate optional
1102 // types (fields, parameters), e.g.
1103 // function(x?: string, y?) {...}
1104 // class X { y?; }
1105 Tok->setType(TT_JsTypeOptionalQuestion);
1106 break;
1107 }
1108 // Declarations cannot be conditional expressions, this can only be part
1109 // of a type declaration.
1110 if (Line.MustBeDeclaration && !Contexts.back().IsExpression &&
1111 Style.Language == FormatStyle::LK_JavaScript)
1112 break;
1113 if (Style.isCSharp()) {
1114 // `Type?)`, `Type?>`, `Type? name;` and `Type? name =` can only be
1115 // nullable types.
1116 // Line.MustBeDeclaration will be true for `Type? name;`.
1117 if ((!Contexts.back().IsExpression && Line.MustBeDeclaration) ||
1118 (Tok->Next && Tok->Next->isOneOf(tok::r_paren, tok::greater)) ||
1119 (Tok->Next && Tok->Next->is(tok::identifier) && Tok->Next->Next &&
1120 Tok->Next->Next->is(tok::equal))) {
1121 Tok->setType(TT_CSharpNullable);
1122 break;
1123 }
1124 }
1125 parseConditional();
1126 break;
1127 case tok::kw_template:
1128 parseTemplateDeclaration();
1129 break;
1130 case tok::comma:
1131 if (Contexts.back().InCtorInitializer)
1132 Tok->setType(TT_CtorInitializerComma);
1133 else if (Contexts.back().InInheritanceList)
1134 Tok->setType(TT_InheritanceComma);
1135 else if (Contexts.back().FirstStartOfName &&
1136 (Contexts.size() == 1 || Line.startsWith(tok::kw_for))) {
1137 Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
1138 Line.IsMultiVariableDeclStmt = true;
1139 }
1140 if (Contexts.back().IsForEachMacro)
1141 Contexts.back().IsExpression = true;
1142 break;
1143 case tok::identifier:
1144 if (Tok->isOneOf(Keywords.kw___has_include,
1145 Keywords.kw___has_include_next)) {
1146 parseHasInclude();
1147 }
1148 if (Style.isCSharp() && Tok->is(Keywords.kw_where) && Tok->Next &&
1149 Tok->Next->isNot(tok::l_paren)) {
1150 Tok->setType(TT_CSharpGenericTypeConstraint);
1151 parseCSharpGenericTypeConstraint();
1152 }
1153 break;
1154 default:
1155 break;
1156 }
1157 return true;
1158 }
1159
parseCSharpGenericTypeConstraint()1160 void parseCSharpGenericTypeConstraint() {
1161 int OpenAngleBracketsCount = 0;
1162 while (CurrentToken) {
1163 if (CurrentToken->is(tok::less)) {
1164 // parseAngle is too greedy and will consume the whole line.
1165 CurrentToken->setType(TT_TemplateOpener);
1166 ++OpenAngleBracketsCount;
1167 next();
1168 } else if (CurrentToken->is(tok::greater)) {
1169 CurrentToken->setType(TT_TemplateCloser);
1170 --OpenAngleBracketsCount;
1171 next();
1172 } else if (CurrentToken->is(tok::comma) && OpenAngleBracketsCount == 0) {
1173 // We allow line breaks after GenericTypeConstraintComma's
1174 // so do not flag commas in Generics as GenericTypeConstraintComma's.
1175 CurrentToken->setType(TT_CSharpGenericTypeConstraintComma);
1176 next();
1177 } else if (CurrentToken->is(Keywords.kw_where)) {
1178 CurrentToken->setType(TT_CSharpGenericTypeConstraint);
1179 next();
1180 } else if (CurrentToken->is(tok::colon)) {
1181 CurrentToken->setType(TT_CSharpGenericTypeConstraintColon);
1182 next();
1183 } else {
1184 next();
1185 }
1186 }
1187 }
1188
parseIncludeDirective()1189 void parseIncludeDirective() {
1190 if (CurrentToken && CurrentToken->is(tok::less)) {
1191 next();
1192 while (CurrentToken) {
1193 // Mark tokens up to the trailing line comments as implicit string
1194 // literals.
1195 if (CurrentToken->isNot(tok::comment) &&
1196 !CurrentToken->TokenText.startswith("//"))
1197 CurrentToken->setType(TT_ImplicitStringLiteral);
1198 next();
1199 }
1200 }
1201 }
1202
parseWarningOrError()1203 void parseWarningOrError() {
1204 next();
1205 // We still want to format the whitespace left of the first token of the
1206 // warning or error.
1207 next();
1208 while (CurrentToken) {
1209 CurrentToken->setType(TT_ImplicitStringLiteral);
1210 next();
1211 }
1212 }
1213
parsePragma()1214 void parsePragma() {
1215 next(); // Consume "pragma".
1216 if (CurrentToken &&
1217 CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
1218 bool IsMark = CurrentToken->is(Keywords.kw_mark);
1219 next(); // Consume "mark".
1220 next(); // Consume first token (so we fix leading whitespace).
1221 while (CurrentToken) {
1222 if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
1223 CurrentToken->setType(TT_ImplicitStringLiteral);
1224 next();
1225 }
1226 }
1227 }
1228
parseHasInclude()1229 void parseHasInclude() {
1230 if (!CurrentToken || !CurrentToken->is(tok::l_paren))
1231 return;
1232 next(); // '('
1233 parseIncludeDirective();
1234 next(); // ')'
1235 }
1236
parsePreprocessorDirective()1237 LineType parsePreprocessorDirective() {
1238 bool IsFirstToken = CurrentToken->IsFirst;
1239 LineType Type = LT_PreprocessorDirective;
1240 next();
1241 if (!CurrentToken)
1242 return Type;
1243
1244 if (Style.Language == FormatStyle::LK_JavaScript && IsFirstToken) {
1245 // JavaScript files can contain shebang lines of the form:
1246 // #!/usr/bin/env node
1247 // Treat these like C++ #include directives.
1248 while (CurrentToken) {
1249 // Tokens cannot be comments here.
1250 CurrentToken->setType(TT_ImplicitStringLiteral);
1251 next();
1252 }
1253 return LT_ImportStatement;
1254 }
1255
1256 if (CurrentToken->Tok.is(tok::numeric_constant)) {
1257 CurrentToken->SpacesRequiredBefore = 1;
1258 return Type;
1259 }
1260 // Hashes in the middle of a line can lead to any strange token
1261 // sequence.
1262 if (!CurrentToken->Tok.getIdentifierInfo())
1263 return Type;
1264 switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
1265 case tok::pp_include:
1266 case tok::pp_include_next:
1267 case tok::pp_import:
1268 next();
1269 parseIncludeDirective();
1270 Type = LT_ImportStatement;
1271 break;
1272 case tok::pp_error:
1273 case tok::pp_warning:
1274 parseWarningOrError();
1275 break;
1276 case tok::pp_pragma:
1277 parsePragma();
1278 break;
1279 case tok::pp_if:
1280 case tok::pp_elif:
1281 Contexts.back().IsExpression = true;
1282 next();
1283 parseLine();
1284 break;
1285 default:
1286 break;
1287 }
1288 while (CurrentToken) {
1289 FormatToken *Tok = CurrentToken;
1290 next();
1291 if (Tok->is(tok::l_paren))
1292 parseParens();
1293 else if (Tok->isOneOf(Keywords.kw___has_include,
1294 Keywords.kw___has_include_next))
1295 parseHasInclude();
1296 }
1297 return Type;
1298 }
1299
1300 public:
parseLine()1301 LineType parseLine() {
1302 if (!CurrentToken)
1303 return LT_Invalid;
1304 NonTemplateLess.clear();
1305 if (CurrentToken->is(tok::hash))
1306 return parsePreprocessorDirective();
1307
1308 // Directly allow to 'import <string-literal>' to support protocol buffer
1309 // definitions (github.com/google/protobuf) or missing "#" (either way we
1310 // should not break the line).
1311 IdentifierInfo *Info = CurrentToken->Tok.getIdentifierInfo();
1312 if ((Style.Language == FormatStyle::LK_Java &&
1313 CurrentToken->is(Keywords.kw_package)) ||
1314 (Info && Info->getPPKeywordID() == tok::pp_import &&
1315 CurrentToken->Next &&
1316 CurrentToken->Next->isOneOf(tok::string_literal, tok::identifier,
1317 tok::kw_static))) {
1318 next();
1319 parseIncludeDirective();
1320 return LT_ImportStatement;
1321 }
1322
1323 // If this line starts and ends in '<' and '>', respectively, it is likely
1324 // part of "#define <a/b.h>".
1325 if (CurrentToken->is(tok::less) && Line.Last->is(tok::greater)) {
1326 parseIncludeDirective();
1327 return LT_ImportStatement;
1328 }
1329
1330 // In .proto files, top-level options and package statements are very
1331 // similar to import statements and should not be line-wrapped.
1332 if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
1333 CurrentToken->isOneOf(Keywords.kw_option, Keywords.kw_package)) {
1334 next();
1335 if (CurrentToken && CurrentToken->is(tok::identifier)) {
1336 while (CurrentToken)
1337 next();
1338 return LT_ImportStatement;
1339 }
1340 }
1341
1342 bool KeywordVirtualFound = false;
1343 bool ImportStatement = false;
1344
1345 // import {...} from '...';
1346 if (Style.Language == FormatStyle::LK_JavaScript &&
1347 CurrentToken->is(Keywords.kw_import))
1348 ImportStatement = true;
1349
1350 while (CurrentToken) {
1351 if (CurrentToken->is(tok::kw_virtual))
1352 KeywordVirtualFound = true;
1353 if (Style.Language == FormatStyle::LK_JavaScript) {
1354 // export {...} from '...';
1355 // An export followed by "from 'some string';" is a re-export from
1356 // another module identified by a URI and is treated as a
1357 // LT_ImportStatement (i.e. prevent wraps on it for long URIs).
1358 // Just "export {...};" or "export class ..." should not be treated as
1359 // an import in this sense.
1360 if (Line.First->is(tok::kw_export) &&
1361 CurrentToken->is(Keywords.kw_from) && CurrentToken->Next &&
1362 CurrentToken->Next->isStringLiteral())
1363 ImportStatement = true;
1364 if (isClosureImportStatement(*CurrentToken))
1365 ImportStatement = true;
1366 }
1367 if (!consumeToken())
1368 return LT_Invalid;
1369 }
1370 if (KeywordVirtualFound)
1371 return LT_VirtualFunctionDecl;
1372 if (ImportStatement)
1373 return LT_ImportStatement;
1374
1375 if (Line.startsWith(TT_ObjCMethodSpecifier)) {
1376 if (Contexts.back().FirstObjCSelectorName)
1377 Contexts.back().FirstObjCSelectorName->LongestObjCSelectorName =
1378 Contexts.back().LongestObjCSelectorName;
1379 return LT_ObjCMethodDecl;
1380 }
1381
1382 for (const auto &ctx : Contexts) {
1383 if (ctx.InStructArrayInitializer) {
1384 return LT_ArrayOfStructInitializer;
1385 }
1386 }
1387
1388 return LT_Other;
1389 }
1390
1391 private:
isClosureImportStatement(const FormatToken & Tok)1392 bool isClosureImportStatement(const FormatToken &Tok) {
1393 // FIXME: Closure-library specific stuff should not be hard-coded but be
1394 // configurable.
1395 return Tok.TokenText == "goog" && Tok.Next && Tok.Next->is(tok::period) &&
1396 Tok.Next->Next &&
1397 (Tok.Next->Next->TokenText == "module" ||
1398 Tok.Next->Next->TokenText == "provide" ||
1399 Tok.Next->Next->TokenText == "require" ||
1400 Tok.Next->Next->TokenText == "requireType" ||
1401 Tok.Next->Next->TokenText == "forwardDeclare") &&
1402 Tok.Next->Next->Next && Tok.Next->Next->Next->is(tok::l_paren);
1403 }
1404
resetTokenMetadata(FormatToken * Token)1405 void resetTokenMetadata(FormatToken *Token) {
1406 if (!Token)
1407 return;
1408
1409 // Reset token type in case we have already looked at it and then
1410 // recovered from an error (e.g. failure to find the matching >).
1411 if (!CurrentToken->isOneOf(
1412 TT_LambdaLSquare, TT_LambdaLBrace, TT_AttributeMacro, TT_IfMacro,
1413 TT_ForEachMacro, TT_TypenameMacro, TT_FunctionLBrace,
1414 TT_ImplicitStringLiteral, TT_InlineASMBrace, TT_FatArrow,
1415 TT_LambdaArrow, TT_NamespaceMacro, TT_OverloadedOperator,
1416 TT_RegexLiteral, TT_TemplateString, TT_ObjCStringLiteral,
1417 TT_UntouchableMacroFunc, TT_ConstraintJunctions,
1418 TT_StatementAttributeLikeMacro))
1419 CurrentToken->setType(TT_Unknown);
1420 CurrentToken->Role.reset();
1421 CurrentToken->MatchingParen = nullptr;
1422 CurrentToken->FakeLParens.clear();
1423 CurrentToken->FakeRParens = 0;
1424 }
1425
next()1426 void next() {
1427 if (CurrentToken) {
1428 CurrentToken->NestingLevel = Contexts.size() - 1;
1429 CurrentToken->BindingStrength = Contexts.back().BindingStrength;
1430 modifyContext(*CurrentToken);
1431 determineTokenType(*CurrentToken);
1432 CurrentToken = CurrentToken->Next;
1433 }
1434
1435 resetTokenMetadata(CurrentToken);
1436 }
1437
1438 /// A struct to hold information valid in a specific context, e.g.
1439 /// a pair of parenthesis.
1440 struct Context {
Contextclang::format::__anonfa1d464f0111::AnnotatingParser::Context1441 Context(tok::TokenKind ContextKind, unsigned BindingStrength,
1442 bool IsExpression)
1443 : ContextKind(ContextKind), BindingStrength(BindingStrength),
1444 IsExpression(IsExpression) {}
1445
1446 tok::TokenKind ContextKind;
1447 unsigned BindingStrength;
1448 bool IsExpression;
1449 unsigned LongestObjCSelectorName = 0;
1450 bool ColonIsForRangeExpr = false;
1451 bool ColonIsDictLiteral = false;
1452 bool ColonIsObjCMethodExpr = false;
1453 FormatToken *FirstObjCSelectorName = nullptr;
1454 FormatToken *FirstStartOfName = nullptr;
1455 bool CanBeExpression = true;
1456 bool InTemplateArgument = false;
1457 bool InCtorInitializer = false;
1458 bool InInheritanceList = false;
1459 bool CaretFound = false;
1460 bool IsForEachMacro = false;
1461 bool InCpp11AttributeSpecifier = false;
1462 bool InCSharpAttributeSpecifier = false;
1463 bool InStructArrayInitializer = false;
1464 };
1465
1466 /// Puts a new \c Context onto the stack \c Contexts for the lifetime
1467 /// of each instance.
1468 struct ScopedContextCreator {
1469 AnnotatingParser &P;
1470
ScopedContextCreatorclang::format::__anonfa1d464f0111::AnnotatingParser::ScopedContextCreator1471 ScopedContextCreator(AnnotatingParser &P, tok::TokenKind ContextKind,
1472 unsigned Increase)
1473 : P(P) {
1474 P.Contexts.push_back(Context(ContextKind,
1475 P.Contexts.back().BindingStrength + Increase,
1476 P.Contexts.back().IsExpression));
1477 }
1478
~ScopedContextCreatorclang::format::__anonfa1d464f0111::AnnotatingParser::ScopedContextCreator1479 ~ScopedContextCreator() {
1480 if (P.Style.AlignArrayOfStructures != FormatStyle::AIAS_None) {
1481 if (P.Contexts.back().InStructArrayInitializer) {
1482 P.Contexts.pop_back();
1483 P.Contexts.back().InStructArrayInitializer = true;
1484 return;
1485 }
1486 }
1487 P.Contexts.pop_back();
1488 }
1489 };
1490
modifyContext(const FormatToken & Current)1491 void modifyContext(const FormatToken &Current) {
1492 if (Current.getPrecedence() == prec::Assignment &&
1493 !Line.First->isOneOf(tok::kw_template, tok::kw_using, tok::kw_return) &&
1494 // Type aliases use `type X = ...;` in TypeScript and can be exported
1495 // using `export type ...`.
1496 !(Style.Language == FormatStyle::LK_JavaScript &&
1497 (Line.startsWith(Keywords.kw_type, tok::identifier) ||
1498 Line.startsWith(tok::kw_export, Keywords.kw_type,
1499 tok::identifier))) &&
1500 (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
1501 Contexts.back().IsExpression = true;
1502 if (!Line.startsWith(TT_UnaryOperator)) {
1503 for (FormatToken *Previous = Current.Previous;
1504 Previous && Previous->Previous &&
1505 !Previous->Previous->isOneOf(tok::comma, tok::semi);
1506 Previous = Previous->Previous) {
1507 if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
1508 Previous = Previous->MatchingParen;
1509 if (!Previous)
1510 break;
1511 }
1512 if (Previous->opensScope())
1513 break;
1514 if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
1515 Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
1516 Previous->Previous && Previous->Previous->isNot(tok::equal))
1517 Previous->setType(TT_PointerOrReference);
1518 }
1519 }
1520 } else if (Current.is(tok::lessless) &&
1521 (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
1522 Contexts.back().IsExpression = true;
1523 } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
1524 Contexts.back().IsExpression = true;
1525 } else if (Current.is(TT_TrailingReturnArrow)) {
1526 Contexts.back().IsExpression = false;
1527 } else if (Current.is(TT_LambdaArrow) || Current.is(Keywords.kw_assert)) {
1528 Contexts.back().IsExpression = Style.Language == FormatStyle::LK_Java;
1529 } else if (Current.Previous &&
1530 Current.Previous->is(TT_CtorInitializerColon)) {
1531 Contexts.back().IsExpression = true;
1532 Contexts.back().InCtorInitializer = true;
1533 } else if (Current.Previous && Current.Previous->is(TT_InheritanceColon)) {
1534 Contexts.back().InInheritanceList = true;
1535 } else if (Current.isOneOf(tok::r_paren, tok::greater, tok::comma)) {
1536 for (FormatToken *Previous = Current.Previous;
1537 Previous && Previous->isOneOf(tok::star, tok::amp);
1538 Previous = Previous->Previous)
1539 Previous->setType(TT_PointerOrReference);
1540 if (Line.MustBeDeclaration && !Contexts.front().InCtorInitializer)
1541 Contexts.back().IsExpression = false;
1542 } else if (Current.is(tok::kw_new)) {
1543 Contexts.back().CanBeExpression = false;
1544 } else if (Current.is(tok::semi) ||
1545 (Current.is(tok::exclaim) && Current.Previous &&
1546 !Current.Previous->is(tok::kw_operator))) {
1547 // This should be the condition or increment in a for-loop.
1548 // But not operator !() (can't use TT_OverloadedOperator here as its not
1549 // been annotated yet).
1550 Contexts.back().IsExpression = true;
1551 }
1552 }
1553
untilMatchingParen(FormatToken * Current)1554 static FormatToken *untilMatchingParen(FormatToken *Current) {
1555 // Used when `MatchingParen` is not yet established.
1556 int ParenLevel = 0;
1557 while (Current) {
1558 if (Current->is(tok::l_paren))
1559 ParenLevel++;
1560 if (Current->is(tok::r_paren))
1561 ParenLevel--;
1562 if (ParenLevel < 1)
1563 break;
1564 Current = Current->Next;
1565 }
1566 return Current;
1567 }
1568
isDeductionGuide(FormatToken & Current)1569 static bool isDeductionGuide(FormatToken &Current) {
1570 // Look for a deduction guide template<T> A(...) -> A<...>;
1571 if (Current.Previous && Current.Previous->is(tok::r_paren) &&
1572 Current.startsSequence(tok::arrow, tok::identifier, tok::less)) {
1573 // Find the TemplateCloser.
1574 FormatToken *TemplateCloser = Current.Next->Next;
1575 int NestingLevel = 0;
1576 while (TemplateCloser) {
1577 // Skip over an expressions in parens A<(3 < 2)>;
1578 if (TemplateCloser->is(tok::l_paren)) {
1579 // No Matching Paren yet so skip to matching paren
1580 TemplateCloser = untilMatchingParen(TemplateCloser);
1581 }
1582 if (TemplateCloser->is(tok::less))
1583 NestingLevel++;
1584 if (TemplateCloser->is(tok::greater))
1585 NestingLevel--;
1586 if (NestingLevel < 1)
1587 break;
1588 TemplateCloser = TemplateCloser->Next;
1589 }
1590 // Assuming we have found the end of the template ensure its followed
1591 // with a semi-colon.
1592 if (TemplateCloser && TemplateCloser->Next &&
1593 TemplateCloser->Next->is(tok::semi) &&
1594 Current.Previous->MatchingParen) {
1595 // Determine if the identifier `A` prior to the A<..>; is the same as
1596 // prior to the A(..)
1597 FormatToken *LeadingIdentifier =
1598 Current.Previous->MatchingParen->Previous;
1599
1600 // Differentiate a deduction guide by seeing the
1601 // > of the template prior to the leading identifier.
1602 if (LeadingIdentifier) {
1603 FormatToken *PriorLeadingIdentifier = LeadingIdentifier->Previous;
1604 // Skip back past explicit decoration
1605 if (PriorLeadingIdentifier &&
1606 PriorLeadingIdentifier->is(tok::kw_explicit))
1607 PriorLeadingIdentifier = PriorLeadingIdentifier->Previous;
1608
1609 return (PriorLeadingIdentifier &&
1610 PriorLeadingIdentifier->is(TT_TemplateCloser) &&
1611 LeadingIdentifier->TokenText == Current.Next->TokenText);
1612 }
1613 }
1614 }
1615 return false;
1616 }
1617
determineTokenType(FormatToken & Current)1618 void determineTokenType(FormatToken &Current) {
1619 if (!Current.is(TT_Unknown))
1620 // The token type is already known.
1621 return;
1622
1623 if ((Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp()) &&
1624 Current.is(tok::exclaim)) {
1625 if (Current.Previous) {
1626 bool IsIdentifier =
1627 Style.Language == FormatStyle::LK_JavaScript
1628 ? Keywords.IsJavaScriptIdentifier(
1629 *Current.Previous, /* AcceptIdentifierName= */ true)
1630 : Current.Previous->is(tok::identifier);
1631 if (IsIdentifier ||
1632 Current.Previous->isOneOf(
1633 tok::kw_namespace, tok::r_paren, tok::r_square, tok::r_brace,
1634 tok::kw_false, tok::kw_true, Keywords.kw_type, Keywords.kw_get,
1635 Keywords.kw_set) ||
1636 Current.Previous->Tok.isLiteral()) {
1637 Current.setType(TT_NonNullAssertion);
1638 return;
1639 }
1640 }
1641 if (Current.Next &&
1642 Current.Next->isOneOf(TT_BinaryOperator, Keywords.kw_as)) {
1643 Current.setType(TT_NonNullAssertion);
1644 return;
1645 }
1646 }
1647
1648 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1649 // function declaration have been found. In this case, 'Current' is a
1650 // trailing token of this declaration and thus cannot be a name.
1651 if (Current.is(Keywords.kw_instanceof)) {
1652 Current.setType(TT_BinaryOperator);
1653 } else if (isStartOfName(Current) &&
1654 (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {
1655 Contexts.back().FirstStartOfName = &Current;
1656 Current.setType(TT_StartOfName);
1657 } else if (Current.is(tok::semi)) {
1658 // Reset FirstStartOfName after finding a semicolon so that a for loop
1659 // with multiple increment statements is not confused with a for loop
1660 // having multiple variable declarations.
1661 Contexts.back().FirstStartOfName = nullptr;
1662 } else if (Current.isOneOf(tok::kw_auto, tok::kw___auto_type)) {
1663 AutoFound = true;
1664 } else if (Current.is(tok::arrow) &&
1665 Style.Language == FormatStyle::LK_Java) {
1666 Current.setType(TT_LambdaArrow);
1667 } else if (Current.is(tok::arrow) && AutoFound && Line.MustBeDeclaration &&
1668 Current.NestingLevel == 0 &&
1669 !Current.Previous->is(tok::kw_operator)) {
1670 // not auto operator->() -> xxx;
1671 Current.setType(TT_TrailingReturnArrow);
1672 } else if (Current.is(tok::arrow) && Current.Previous &&
1673 Current.Previous->is(tok::r_brace)) {
1674 // Concept implicit conversion contraint needs to be treated like
1675 // a trailing return type ... } -> <type>.
1676 Current.setType(TT_TrailingReturnArrow);
1677 } else if (isDeductionGuide(Current)) {
1678 // Deduction guides trailing arrow " A(...) -> A<T>;".
1679 Current.setType(TT_TrailingReturnArrow);
1680 } else if (Current.isOneOf(tok::star, tok::amp, tok::ampamp)) {
1681 Current.setType(determineStarAmpUsage(
1682 Current,
1683 Contexts.back().CanBeExpression && Contexts.back().IsExpression,
1684 Contexts.back().InTemplateArgument));
1685 } else if (Current.isOneOf(tok::minus, tok::plus, tok::caret)) {
1686 Current.setType(determinePlusMinusCaretUsage(Current));
1687 if (Current.is(TT_UnaryOperator) && Current.is(tok::caret))
1688 Contexts.back().CaretFound = true;
1689 } else if (Current.isOneOf(tok::minusminus, tok::plusplus)) {
1690 Current.setType(determineIncrementUsage(Current));
1691 } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
1692 Current.setType(TT_UnaryOperator);
1693 } else if (Current.is(tok::question)) {
1694 if (Style.Language == FormatStyle::LK_JavaScript &&
1695 Line.MustBeDeclaration && !Contexts.back().IsExpression) {
1696 // In JavaScript, `interface X { foo?(): bar; }` is an optional method
1697 // on the interface, not a ternary expression.
1698 Current.setType(TT_JsTypeOptionalQuestion);
1699 } else {
1700 Current.setType(TT_ConditionalExpr);
1701 }
1702 } else if (Current.isBinaryOperator() &&
1703 (!Current.Previous || Current.Previous->isNot(tok::l_square)) &&
1704 (!Current.is(tok::greater) &&
1705 Style.Language != FormatStyle::LK_TextProto)) {
1706 Current.setType(TT_BinaryOperator);
1707 } else if (Current.is(tok::comment)) {
1708 if (Current.TokenText.startswith("/*")) {
1709 if (Current.TokenText.endswith("*/"))
1710 Current.setType(TT_BlockComment);
1711 else
1712 // The lexer has for some reason determined a comment here. But we
1713 // cannot really handle it, if it isn't properly terminated.
1714 Current.Tok.setKind(tok::unknown);
1715 } else {
1716 Current.setType(TT_LineComment);
1717 }
1718 } else if (Current.is(tok::r_paren)) {
1719 if (rParenEndsCast(Current))
1720 Current.setType(TT_CastRParen);
1721 if (Current.MatchingParen && Current.Next &&
1722 !Current.Next->isBinaryOperator() &&
1723 !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace,
1724 tok::comma, tok::period, tok::arrow,
1725 tok::coloncolon))
1726 if (FormatToken *AfterParen = Current.MatchingParen->Next) {
1727 // Make sure this isn't the return type of an Obj-C block declaration
1728 if (AfterParen->Tok.isNot(tok::caret)) {
1729 if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
1730 if (BeforeParen->is(tok::identifier) &&
1731 !BeforeParen->is(TT_TypenameMacro) &&
1732 BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
1733 (!BeforeParen->Previous ||
1734 BeforeParen->Previous->ClosesTemplateDeclaration))
1735 Current.setType(TT_FunctionAnnotationRParen);
1736 }
1737 }
1738 } else if (Current.is(tok::at) && Current.Next &&
1739 Style.Language != FormatStyle::LK_JavaScript &&
1740 Style.Language != FormatStyle::LK_Java) {
1741 // In Java & JavaScript, "@..." is a decorator or annotation. In ObjC, it
1742 // marks declarations and properties that need special formatting.
1743 switch (Current.Next->Tok.getObjCKeywordID()) {
1744 case tok::objc_interface:
1745 case tok::objc_implementation:
1746 case tok::objc_protocol:
1747 Current.setType(TT_ObjCDecl);
1748 break;
1749 case tok::objc_property:
1750 Current.setType(TT_ObjCProperty);
1751 break;
1752 default:
1753 break;
1754 }
1755 } else if (Current.is(tok::period)) {
1756 FormatToken *PreviousNoComment = Current.getPreviousNonComment();
1757 if (PreviousNoComment &&
1758 PreviousNoComment->isOneOf(tok::comma, tok::l_brace))
1759 Current.setType(TT_DesignatedInitializerPeriod);
1760 else if (Style.Language == FormatStyle::LK_Java && Current.Previous &&
1761 Current.Previous->isOneOf(TT_JavaAnnotation,
1762 TT_LeadingJavaAnnotation)) {
1763 Current.setType(Current.Previous->getType());
1764 }
1765 } else if (canBeObjCSelectorComponent(Current) &&
1766 // FIXME(bug 36976): ObjC return types shouldn't use
1767 // TT_CastRParen.
1768 Current.Previous && Current.Previous->is(TT_CastRParen) &&
1769 Current.Previous->MatchingParen &&
1770 Current.Previous->MatchingParen->Previous &&
1771 Current.Previous->MatchingParen->Previous->is(
1772 TT_ObjCMethodSpecifier)) {
1773 // This is the first part of an Objective-C selector name. (If there's no
1774 // colon after this, this is the only place which annotates the identifier
1775 // as a selector.)
1776 Current.setType(TT_SelectorName);
1777 } else if (Current.isOneOf(tok::identifier, tok::kw_const, tok::kw_noexcept,
1778 tok::kw_requires) &&
1779 Current.Previous &&
1780 !Current.Previous->isOneOf(tok::equal, tok::at) &&
1781 Line.MightBeFunctionDecl && Contexts.size() == 1) {
1782 // Line.MightBeFunctionDecl can only be true after the parentheses of a
1783 // function declaration have been found.
1784 Current.setType(TT_TrailingAnnotation);
1785 } else if ((Style.Language == FormatStyle::LK_Java ||
1786 Style.Language == FormatStyle::LK_JavaScript) &&
1787 Current.Previous) {
1788 if (Current.Previous->is(tok::at) &&
1789 Current.isNot(Keywords.kw_interface)) {
1790 const FormatToken &AtToken = *Current.Previous;
1791 const FormatToken *Previous = AtToken.getPreviousNonComment();
1792 if (!Previous || Previous->is(TT_LeadingJavaAnnotation))
1793 Current.setType(TT_LeadingJavaAnnotation);
1794 else
1795 Current.setType(TT_JavaAnnotation);
1796 } else if (Current.Previous->is(tok::period) &&
1797 Current.Previous->isOneOf(TT_JavaAnnotation,
1798 TT_LeadingJavaAnnotation)) {
1799 Current.setType(Current.Previous->getType());
1800 }
1801 }
1802 }
1803
1804 /// Take a guess at whether \p Tok starts a name of a function or
1805 /// variable declaration.
1806 ///
1807 /// This is a heuristic based on whether \p Tok is an identifier following
1808 /// something that is likely a type.
isStartOfName(const FormatToken & Tok)1809 bool isStartOfName(const FormatToken &Tok) {
1810 if (Tok.isNot(tok::identifier) || !Tok.Previous)
1811 return false;
1812
1813 if (Tok.Previous->isOneOf(TT_LeadingJavaAnnotation, Keywords.kw_instanceof,
1814 Keywords.kw_as))
1815 return false;
1816 if (Style.Language == FormatStyle::LK_JavaScript &&
1817 Tok.Previous->is(Keywords.kw_in))
1818 return false;
1819
1820 // Skip "const" as it does not have an influence on whether this is a name.
1821 FormatToken *PreviousNotConst = Tok.getPreviousNonComment();
1822 while (PreviousNotConst && PreviousNotConst->is(tok::kw_const))
1823 PreviousNotConst = PreviousNotConst->getPreviousNonComment();
1824
1825 if (!PreviousNotConst)
1826 return false;
1827
1828 bool IsPPKeyword = PreviousNotConst->is(tok::identifier) &&
1829 PreviousNotConst->Previous &&
1830 PreviousNotConst->Previous->is(tok::hash);
1831
1832 if (PreviousNotConst->is(TT_TemplateCloser))
1833 return PreviousNotConst && PreviousNotConst->MatchingParen &&
1834 PreviousNotConst->MatchingParen->Previous &&
1835 PreviousNotConst->MatchingParen->Previous->isNot(tok::period) &&
1836 PreviousNotConst->MatchingParen->Previous->isNot(tok::kw_template);
1837
1838 if (PreviousNotConst->is(tok::r_paren) &&
1839 PreviousNotConst->is(TT_TypeDeclarationParen))
1840 return true;
1841
1842 return (!IsPPKeyword &&
1843 PreviousNotConst->isOneOf(tok::identifier, tok::kw_auto)) ||
1844 PreviousNotConst->is(TT_PointerOrReference) ||
1845 PreviousNotConst->isSimpleTypeSpecifier();
1846 }
1847
1848 /// Determine whether ')' is ending a cast.
rParenEndsCast(const FormatToken & Tok)1849 bool rParenEndsCast(const FormatToken &Tok) {
1850 // C-style casts are only used in C++, C# and Java.
1851 if (!Style.isCSharp() && !Style.isCpp() &&
1852 Style.Language != FormatStyle::LK_Java)
1853 return false;
1854
1855 // Empty parens aren't casts and there are no casts at the end of the line.
1856 if (Tok.Previous == Tok.MatchingParen || !Tok.Next || !Tok.MatchingParen)
1857 return false;
1858
1859 FormatToken *LeftOfParens = Tok.MatchingParen->getPreviousNonComment();
1860 if (LeftOfParens) {
1861 // If there is a closing parenthesis left of the current parentheses,
1862 // look past it as these might be chained casts.
1863 if (LeftOfParens->is(tok::r_paren)) {
1864 if (!LeftOfParens->MatchingParen ||
1865 !LeftOfParens->MatchingParen->Previous)
1866 return false;
1867 LeftOfParens = LeftOfParens->MatchingParen->Previous;
1868 }
1869
1870 // If there is an identifier (or with a few exceptions a keyword) right
1871 // before the parentheses, this is unlikely to be a cast.
1872 if (LeftOfParens->Tok.getIdentifierInfo() &&
1873 !LeftOfParens->isOneOf(Keywords.kw_in, tok::kw_return, tok::kw_case,
1874 tok::kw_delete))
1875 return false;
1876
1877 // Certain other tokens right before the parentheses are also signals that
1878 // this cannot be a cast.
1879 if (LeftOfParens->isOneOf(tok::at, tok::r_square, TT_OverloadedOperator,
1880 TT_TemplateCloser, tok::ellipsis))
1881 return false;
1882 }
1883
1884 if (Tok.Next->is(tok::question))
1885 return false;
1886
1887 // `foreach((A a, B b) in someList)` should not be seen as a cast.
1888 if (Tok.Next->is(Keywords.kw_in) && Style.isCSharp())
1889 return false;
1890
1891 // Functions which end with decorations like volatile, noexcept are unlikely
1892 // to be casts.
1893 if (Tok.Next->isOneOf(tok::kw_noexcept, tok::kw_volatile, tok::kw_const,
1894 tok::kw_requires, tok::kw_throw, tok::arrow,
1895 Keywords.kw_override, Keywords.kw_final) ||
1896 isCpp11AttributeSpecifier(*Tok.Next))
1897 return false;
1898
1899 // As Java has no function types, a "(" after the ")" likely means that this
1900 // is a cast.
1901 if (Style.Language == FormatStyle::LK_Java && Tok.Next->is(tok::l_paren))
1902 return true;
1903
1904 // If a (non-string) literal follows, this is likely a cast.
1905 if (Tok.Next->isNot(tok::string_literal) &&
1906 (Tok.Next->Tok.isLiteral() ||
1907 Tok.Next->isOneOf(tok::kw_sizeof, tok::kw_alignof)))
1908 return true;
1909
1910 // Heuristically try to determine whether the parentheses contain a type.
1911 auto IsQualifiedPointerOrReference = [](FormatToken *T) {
1912 // This is used to handle cases such as x = (foo *const)&y;
1913 assert(!T->isSimpleTypeSpecifier() && "Should have already been checked");
1914 // Strip trailing qualifiers such as const or volatile when checking
1915 // whether the parens could be a cast to a pointer/reference type.
1916 while (T) {
1917 if (T->is(TT_AttributeParen)) {
1918 // Handle `x = (foo *__attribute__((foo)))&v;`:
1919 if (T->MatchingParen && T->MatchingParen->Previous &&
1920 T->MatchingParen->Previous->is(tok::kw___attribute)) {
1921 T = T->MatchingParen->Previous->Previous;
1922 continue;
1923 }
1924 } else if (T->is(TT_AttributeSquare)) {
1925 // Handle `x = (foo *[[clang::foo]])&v;`:
1926 if (T->MatchingParen && T->MatchingParen->Previous) {
1927 T = T->MatchingParen->Previous;
1928 continue;
1929 }
1930 } else if (T->canBePointerOrReferenceQualifier()) {
1931 T = T->Previous;
1932 continue;
1933 }
1934 break;
1935 }
1936 return T && T->is(TT_PointerOrReference);
1937 };
1938 bool ParensAreType =
1939 !Tok.Previous ||
1940 Tok.Previous->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) ||
1941 Tok.Previous->isSimpleTypeSpecifier() ||
1942 IsQualifiedPointerOrReference(Tok.Previous);
1943 bool ParensCouldEndDecl =
1944 Tok.Next->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
1945 if (ParensAreType && !ParensCouldEndDecl)
1946 return true;
1947
1948 // At this point, we heuristically assume that there are no casts at the
1949 // start of the line. We assume that we have found most cases where there
1950 // are by the logic above, e.g. "(void)x;".
1951 if (!LeftOfParens)
1952 return false;
1953
1954 // Certain token types inside the parentheses mean that this can't be a
1955 // cast.
1956 for (const FormatToken *Token = Tok.MatchingParen->Next; Token != &Tok;
1957 Token = Token->Next)
1958 if (Token->is(TT_BinaryOperator))
1959 return false;
1960
1961 // If the following token is an identifier or 'this', this is a cast. All
1962 // cases where this can be something else are handled above.
1963 if (Tok.Next->isOneOf(tok::identifier, tok::kw_this))
1964 return true;
1965
1966 // Look for a cast `( x ) (`.
1967 if (Tok.Next->is(tok::l_paren) && Tok.Previous && Tok.Previous->Previous) {
1968 if (Tok.Previous->is(tok::identifier) &&
1969 Tok.Previous->Previous->is(tok::l_paren))
1970 return true;
1971 }
1972
1973 if (!Tok.Next->Next)
1974 return false;
1975
1976 // If the next token after the parenthesis is a unary operator, assume
1977 // that this is cast, unless there are unexpected tokens inside the
1978 // parenthesis.
1979 bool NextIsUnary =
1980 Tok.Next->isUnaryOperator() || Tok.Next->isOneOf(tok::amp, tok::star);
1981 if (!NextIsUnary || Tok.Next->is(tok::plus) ||
1982 !Tok.Next->Next->isOneOf(tok::identifier, tok::numeric_constant))
1983 return false;
1984 // Search for unexpected tokens.
1985 for (FormatToken *Prev = Tok.Previous; Prev != Tok.MatchingParen;
1986 Prev = Prev->Previous) {
1987 if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
1988 return false;
1989 }
1990 return true;
1991 }
1992
1993 /// Return the type of the given token assuming it is * or &.
determineStarAmpUsage(const FormatToken & Tok,bool IsExpression,bool InTemplateArgument)1994 TokenType determineStarAmpUsage(const FormatToken &Tok, bool IsExpression,
1995 bool InTemplateArgument) {
1996 if (Style.Language == FormatStyle::LK_JavaScript)
1997 return TT_BinaryOperator;
1998
1999 // && in C# must be a binary operator.
2000 if (Style.isCSharp() && Tok.is(tok::ampamp))
2001 return TT_BinaryOperator;
2002
2003 const FormatToken *PrevToken = Tok.getPreviousNonComment();
2004 if (!PrevToken)
2005 return TT_UnaryOperator;
2006
2007 const FormatToken *NextToken = Tok.getNextNonComment();
2008 if (!NextToken ||
2009 NextToken->isOneOf(tok::arrow, tok::equal, tok::kw_noexcept) ||
2010 NextToken->canBePointerOrReferenceQualifier() ||
2011 (NextToken->is(tok::l_brace) && !NextToken->getNextNonComment()))
2012 return TT_PointerOrReference;
2013
2014 if (PrevToken->is(tok::coloncolon))
2015 return TT_PointerOrReference;
2016
2017 if (PrevToken->is(tok::r_paren) && PrevToken->is(TT_TypeDeclarationParen))
2018 return TT_PointerOrReference;
2019
2020 if (PrevToken->isOneOf(tok::l_paren, tok::l_square, tok::l_brace,
2021 tok::comma, tok::semi, tok::kw_return, tok::colon,
2022 tok::kw_co_return, tok::kw_co_await,
2023 tok::kw_co_yield, tok::equal, tok::kw_delete,
2024 tok::kw_sizeof, tok::kw_throw) ||
2025 PrevToken->isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
2026 TT_UnaryOperator, TT_CastRParen))
2027 return TT_UnaryOperator;
2028
2029 if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
2030 return TT_PointerOrReference;
2031 if (NextToken->is(tok::kw_operator) && !IsExpression)
2032 return TT_PointerOrReference;
2033 if (NextToken->isOneOf(tok::comma, tok::semi))
2034 return TT_PointerOrReference;
2035
2036 if (PrevToken->Tok.isLiteral() ||
2037 PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::kw_true,
2038 tok::kw_false, tok::r_brace) ||
2039 NextToken->Tok.isLiteral() ||
2040 NextToken->isOneOf(tok::kw_true, tok::kw_false) ||
2041 NextToken->isUnaryOperator() ||
2042 // If we know we're in a template argument, there are no named
2043 // declarations. Thus, having an identifier on the right-hand side
2044 // indicates a binary operator.
2045 (InTemplateArgument && NextToken->Tok.isAnyIdentifier()))
2046 return TT_BinaryOperator;
2047
2048 // "&&(" is quite unlikely to be two successive unary "&".
2049 if (Tok.is(tok::ampamp) && NextToken->is(tok::l_paren))
2050 return TT_BinaryOperator;
2051
2052 // This catches some cases where evaluation order is used as control flow:
2053 // aaa && aaa->f();
2054 if (NextToken->Tok.isAnyIdentifier()) {
2055 const FormatToken *NextNextToken = NextToken->getNextNonComment();
2056 if (NextNextToken && NextNextToken->is(tok::arrow))
2057 return TT_BinaryOperator;
2058 }
2059
2060 // It is very unlikely that we are going to find a pointer or reference type
2061 // definition on the RHS of an assignment.
2062 if (IsExpression && !Contexts.back().CaretFound)
2063 return TT_BinaryOperator;
2064
2065 return TT_PointerOrReference;
2066 }
2067
determinePlusMinusCaretUsage(const FormatToken & Tok)2068 TokenType determinePlusMinusCaretUsage(const FormatToken &Tok) {
2069 const FormatToken *PrevToken = Tok.getPreviousNonComment();
2070 if (!PrevToken)
2071 return TT_UnaryOperator;
2072
2073 if (PrevToken->isOneOf(TT_CastRParen, TT_UnaryOperator))
2074 // This must be a sequence of leading unary operators.
2075 return TT_UnaryOperator;
2076
2077 // Use heuristics to recognize unary operators.
2078 if (PrevToken->isOneOf(tok::equal, tok::l_paren, tok::comma, tok::l_square,
2079 tok::question, tok::colon, tok::kw_return,
2080 tok::kw_case, tok::at, tok::l_brace, tok::kw_throw,
2081 tok::kw_co_return, tok::kw_co_yield))
2082 return TT_UnaryOperator;
2083
2084 // There can't be two consecutive binary operators.
2085 if (PrevToken->is(TT_BinaryOperator))
2086 return TT_UnaryOperator;
2087
2088 // Fall back to marking the token as binary operator.
2089 return TT_BinaryOperator;
2090 }
2091
2092 /// Determine whether ++/-- are pre- or post-increments/-decrements.
determineIncrementUsage(const FormatToken & Tok)2093 TokenType determineIncrementUsage(const FormatToken &Tok) {
2094 const FormatToken *PrevToken = Tok.getPreviousNonComment();
2095 if (!PrevToken || PrevToken->is(TT_CastRParen))
2096 return TT_UnaryOperator;
2097 if (PrevToken->isOneOf(tok::r_paren, tok::r_square, tok::identifier))
2098 return TT_TrailingUnaryOperator;
2099
2100 return TT_UnaryOperator;
2101 }
2102
2103 SmallVector<Context, 8> Contexts;
2104
2105 const FormatStyle &Style;
2106 AnnotatedLine &Line;
2107 FormatToken *CurrentToken;
2108 bool AutoFound;
2109 const AdditionalKeywords &Keywords;
2110
2111 // Set of "<" tokens that do not open a template parameter list. If parseAngle
2112 // determines that a specific token can't be a template opener, it will make
2113 // same decision irrespective of the decisions for tokens leading up to it.
2114 // Store this information to prevent this from causing exponential runtime.
2115 llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
2116 };
2117
2118 static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
2119 static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
2120
2121 /// Parses binary expressions by inserting fake parenthesis based on
2122 /// operator precedence.
2123 class ExpressionParser {
2124 public:
ExpressionParser(const FormatStyle & Style,const AdditionalKeywords & Keywords,AnnotatedLine & Line)2125 ExpressionParser(const FormatStyle &Style, const AdditionalKeywords &Keywords,
2126 AnnotatedLine &Line)
2127 : Style(Style), Keywords(Keywords), Current(Line.First) {}
2128
2129 /// Parse expressions with the given operator precedence.
parse(int Precedence=0)2130 void parse(int Precedence = 0) {
2131 // Skip 'return' and ObjC selector colons as they are not part of a binary
2132 // expression.
2133 while (Current && (Current->is(tok::kw_return) ||
2134 (Current->is(tok::colon) &&
2135 Current->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))))
2136 next();
2137
2138 if (!Current || Precedence > PrecedenceArrowAndPeriod)
2139 return;
2140
2141 // Conditional expressions need to be parsed separately for proper nesting.
2142 if (Precedence == prec::Conditional) {
2143 parseConditionalExpr();
2144 return;
2145 }
2146
2147 // Parse unary operators, which all have a higher precedence than binary
2148 // operators.
2149 if (Precedence == PrecedenceUnaryOperator) {
2150 parseUnaryOperator();
2151 return;
2152 }
2153
2154 FormatToken *Start = Current;
2155 FormatToken *LatestOperator = nullptr;
2156 unsigned OperatorIndex = 0;
2157
2158 while (Current) {
2159 // Consume operators with higher precedence.
2160 parse(Precedence + 1);
2161
2162 int CurrentPrecedence = getCurrentPrecedence();
2163
2164 if (Current && Current->is(TT_SelectorName) &&
2165 Precedence == CurrentPrecedence) {
2166 if (LatestOperator)
2167 addFakeParenthesis(Start, prec::Level(Precedence));
2168 Start = Current;
2169 }
2170
2171 // At the end of the line or when an operator with higher precedence is
2172 // found, insert fake parenthesis and return.
2173 if (!Current ||
2174 (Current->closesScope() &&
2175 (Current->MatchingParen || Current->is(TT_TemplateString))) ||
2176 (CurrentPrecedence != -1 && CurrentPrecedence < Precedence) ||
2177 (CurrentPrecedence == prec::Conditional &&
2178 Precedence == prec::Assignment && Current->is(tok::colon))) {
2179 break;
2180 }
2181
2182 // Consume scopes: (), [], <> and {}
2183 if (Current->opensScope()) {
2184 // In fragment of a JavaScript template string can look like '}..${' and
2185 // thus close a scope and open a new one at the same time.
2186 while (Current && (!Current->closesScope() || Current->opensScope())) {
2187 next();
2188 parse();
2189 }
2190 next();
2191 } else {
2192 // Operator found.
2193 if (CurrentPrecedence == Precedence) {
2194 if (LatestOperator)
2195 LatestOperator->NextOperator = Current;
2196 LatestOperator = Current;
2197 Current->OperatorIndex = OperatorIndex;
2198 ++OperatorIndex;
2199 }
2200 next(/*SkipPastLeadingComments=*/Precedence > 0);
2201 }
2202 }
2203
2204 if (LatestOperator && (Current || Precedence > 0)) {
2205 // LatestOperator->LastOperator = true;
2206 if (Precedence == PrecedenceArrowAndPeriod) {
2207 // Call expressions don't have a binary operator precedence.
2208 addFakeParenthesis(Start, prec::Unknown);
2209 } else {
2210 addFakeParenthesis(Start, prec::Level(Precedence));
2211 }
2212 }
2213 }
2214
2215 private:
2216 /// Gets the precedence (+1) of the given token for binary operators
2217 /// and other tokens that we treat like binary operators.
getCurrentPrecedence()2218 int getCurrentPrecedence() {
2219 if (Current) {
2220 const FormatToken *NextNonComment = Current->getNextNonComment();
2221 if (Current->is(TT_ConditionalExpr))
2222 return prec::Conditional;
2223 if (NextNonComment && Current->is(TT_SelectorName) &&
2224 (NextNonComment->isOneOf(TT_DictLiteral, TT_JsTypeColon) ||
2225 ((Style.Language == FormatStyle::LK_Proto ||
2226 Style.Language == FormatStyle::LK_TextProto) &&
2227 NextNonComment->is(tok::less))))
2228 return prec::Assignment;
2229 if (Current->is(TT_JsComputedPropertyName))
2230 return prec::Assignment;
2231 if (Current->is(TT_LambdaArrow))
2232 return prec::Comma;
2233 if (Current->is(TT_FatArrow))
2234 return prec::Assignment;
2235 if (Current->isOneOf(tok::semi, TT_InlineASMColon, TT_SelectorName) ||
2236 (Current->is(tok::comment) && NextNonComment &&
2237 NextNonComment->is(TT_SelectorName)))
2238 return 0;
2239 if (Current->is(TT_RangeBasedForLoopColon))
2240 return prec::Comma;
2241 if ((Style.Language == FormatStyle::LK_Java ||
2242 Style.Language == FormatStyle::LK_JavaScript) &&
2243 Current->is(Keywords.kw_instanceof))
2244 return prec::Relational;
2245 if (Style.Language == FormatStyle::LK_JavaScript &&
2246 Current->isOneOf(Keywords.kw_in, Keywords.kw_as))
2247 return prec::Relational;
2248 if (Current->is(TT_BinaryOperator) || Current->is(tok::comma))
2249 return Current->getPrecedence();
2250 if (Current->isOneOf(tok::period, tok::arrow))
2251 return PrecedenceArrowAndPeriod;
2252 if ((Style.Language == FormatStyle::LK_Java ||
2253 Style.Language == FormatStyle::LK_JavaScript) &&
2254 Current->isOneOf(Keywords.kw_extends, Keywords.kw_implements,
2255 Keywords.kw_throws))
2256 return 0;
2257 }
2258 return -1;
2259 }
2260
addFakeParenthesis(FormatToken * Start,prec::Level Precedence)2261 void addFakeParenthesis(FormatToken *Start, prec::Level Precedence) {
2262 Start->FakeLParens.push_back(Precedence);
2263 if (Precedence > prec::Unknown)
2264 Start->StartsBinaryExpression = true;
2265 if (Current) {
2266 FormatToken *Previous = Current->Previous;
2267 while (Previous->is(tok::comment) && Previous->Previous)
2268 Previous = Previous->Previous;
2269 ++Previous->FakeRParens;
2270 if (Precedence > prec::Unknown)
2271 Previous->EndsBinaryExpression = true;
2272 }
2273 }
2274
2275 /// Parse unary operator expressions and surround them with fake
2276 /// parentheses if appropriate.
parseUnaryOperator()2277 void parseUnaryOperator() {
2278 llvm::SmallVector<FormatToken *, 2> Tokens;
2279 while (Current && Current->is(TT_UnaryOperator)) {
2280 Tokens.push_back(Current);
2281 next();
2282 }
2283 parse(PrecedenceArrowAndPeriod);
2284 for (FormatToken *Token : llvm::reverse(Tokens))
2285 // The actual precedence doesn't matter.
2286 addFakeParenthesis(Token, prec::Unknown);
2287 }
2288
parseConditionalExpr()2289 void parseConditionalExpr() {
2290 while (Current && Current->isTrailingComment()) {
2291 next();
2292 }
2293 FormatToken *Start = Current;
2294 parse(prec::LogicalOr);
2295 if (!Current || !Current->is(tok::question))
2296 return;
2297 next();
2298 parse(prec::Assignment);
2299 if (!Current || Current->isNot(TT_ConditionalExpr))
2300 return;
2301 next();
2302 parse(prec::Assignment);
2303 addFakeParenthesis(Start, prec::Conditional);
2304 }
2305
next(bool SkipPastLeadingComments=true)2306 void next(bool SkipPastLeadingComments = true) {
2307 if (Current)
2308 Current = Current->Next;
2309 while (Current &&
2310 (Current->NewlinesBefore == 0 || SkipPastLeadingComments) &&
2311 Current->isTrailingComment())
2312 Current = Current->Next;
2313 }
2314
2315 const FormatStyle &Style;
2316 const AdditionalKeywords &Keywords;
2317 FormatToken *Current;
2318 };
2319
2320 } // end anonymous namespace
2321
setCommentLineLevels(SmallVectorImpl<AnnotatedLine * > & Lines)2322 void TokenAnnotator::setCommentLineLevels(
2323 SmallVectorImpl<AnnotatedLine *> &Lines) {
2324 const AnnotatedLine *NextNonCommentLine = nullptr;
2325 for (SmallVectorImpl<AnnotatedLine *>::reverse_iterator I = Lines.rbegin(),
2326 E = Lines.rend();
2327 I != E; ++I) {
2328 bool CommentLine = true;
2329 for (const FormatToken *Tok = (*I)->First; Tok; Tok = Tok->Next) {
2330 if (!Tok->is(tok::comment)) {
2331 CommentLine = false;
2332 break;
2333 }
2334 }
2335
2336 // If the comment is currently aligned with the line immediately following
2337 // it, that's probably intentional and we should keep it.
2338 if (NextNonCommentLine && CommentLine &&
2339 NextNonCommentLine->First->NewlinesBefore <= 1 &&
2340 NextNonCommentLine->First->OriginalColumn ==
2341 (*I)->First->OriginalColumn) {
2342 // Align comments for preprocessor lines with the # in column 0 if
2343 // preprocessor lines are not indented. Otherwise, align with the next
2344 // line.
2345 (*I)->Level =
2346 (Style.IndentPPDirectives != FormatStyle::PPDIS_BeforeHash &&
2347 (NextNonCommentLine->Type == LT_PreprocessorDirective ||
2348 NextNonCommentLine->Type == LT_ImportStatement))
2349 ? 0
2350 : NextNonCommentLine->Level;
2351 } else {
2352 NextNonCommentLine = (*I)->First->isNot(tok::r_brace) ? (*I) : nullptr;
2353 }
2354
2355 setCommentLineLevels((*I)->Children);
2356 }
2357 }
2358
maxNestingDepth(const AnnotatedLine & Line)2359 static unsigned maxNestingDepth(const AnnotatedLine &Line) {
2360 unsigned Result = 0;
2361 for (const auto *Tok = Line.First; Tok != nullptr; Tok = Tok->Next)
2362 Result = std::max(Result, Tok->NestingLevel);
2363 return Result;
2364 }
2365
annotate(AnnotatedLine & Line)2366 void TokenAnnotator::annotate(AnnotatedLine &Line) {
2367 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
2368 E = Line.Children.end();
2369 I != E; ++I) {
2370 annotate(**I);
2371 }
2372 AnnotatingParser Parser(Style, Line, Keywords);
2373 Line.Type = Parser.parseLine();
2374
2375 // With very deep nesting, ExpressionParser uses lots of stack and the
2376 // formatting algorithm is very slow. We're not going to do a good job here
2377 // anyway - it's probably generated code being formatted by mistake.
2378 // Just skip the whole line.
2379 if (maxNestingDepth(Line) > 50)
2380 Line.Type = LT_Invalid;
2381
2382 if (Line.Type == LT_Invalid)
2383 return;
2384
2385 ExpressionParser ExprParser(Style, Keywords, Line);
2386 ExprParser.parse();
2387
2388 if (Line.startsWith(TT_ObjCMethodSpecifier))
2389 Line.Type = LT_ObjCMethodDecl;
2390 else if (Line.startsWith(TT_ObjCDecl))
2391 Line.Type = LT_ObjCDecl;
2392 else if (Line.startsWith(TT_ObjCProperty))
2393 Line.Type = LT_ObjCProperty;
2394
2395 Line.First->SpacesRequiredBefore = 1;
2396 Line.First->CanBreakBefore = Line.First->MustBreakBefore;
2397 }
2398
2399 // This function heuristically determines whether 'Current' starts the name of a
2400 // function declaration.
isFunctionDeclarationName(bool IsCpp,const FormatToken & Current,const AnnotatedLine & Line)2401 static bool isFunctionDeclarationName(bool IsCpp, const FormatToken &Current,
2402 const AnnotatedLine &Line) {
2403 auto skipOperatorName = [](const FormatToken *Next) -> const FormatToken * {
2404 for (; Next; Next = Next->Next) {
2405 if (Next->is(TT_OverloadedOperatorLParen))
2406 return Next;
2407 if (Next->is(TT_OverloadedOperator))
2408 continue;
2409 if (Next->isOneOf(tok::kw_new, tok::kw_delete)) {
2410 // For 'new[]' and 'delete[]'.
2411 if (Next->Next &&
2412 Next->Next->startsSequence(tok::l_square, tok::r_square))
2413 Next = Next->Next->Next;
2414 continue;
2415 }
2416 if (Next->startsSequence(tok::l_square, tok::r_square)) {
2417 // For operator[]().
2418 Next = Next->Next;
2419 continue;
2420 }
2421 if ((Next->isSimpleTypeSpecifier() || Next->is(tok::identifier)) &&
2422 Next->Next && Next->Next->isOneOf(tok::star, tok::amp, tok::ampamp)) {
2423 // For operator void*(), operator char*(), operator Foo*().
2424 Next = Next->Next;
2425 continue;
2426 }
2427 if (Next->is(TT_TemplateOpener) && Next->MatchingParen) {
2428 Next = Next->MatchingParen;
2429 continue;
2430 }
2431
2432 break;
2433 }
2434 return nullptr;
2435 };
2436
2437 // Find parentheses of parameter list.
2438 const FormatToken *Next = Current.Next;
2439 if (Current.is(tok::kw_operator)) {
2440 if (Current.Previous && Current.Previous->is(tok::coloncolon))
2441 return false;
2442 Next = skipOperatorName(Next);
2443 } else {
2444 if (!Current.is(TT_StartOfName) || Current.NestingLevel != 0)
2445 return false;
2446 for (; Next; Next = Next->Next) {
2447 if (Next->is(TT_TemplateOpener)) {
2448 Next = Next->MatchingParen;
2449 } else if (Next->is(tok::coloncolon)) {
2450 Next = Next->Next;
2451 if (!Next)
2452 return false;
2453 if (Next->is(tok::kw_operator)) {
2454 Next = skipOperatorName(Next->Next);
2455 break;
2456 }
2457 if (!Next->is(tok::identifier))
2458 return false;
2459 } else if (Next->is(tok::l_paren)) {
2460 break;
2461 } else {
2462 return false;
2463 }
2464 }
2465 }
2466
2467 // Check whether parameter list can belong to a function declaration.
2468 if (!Next || !Next->is(tok::l_paren) || !Next->MatchingParen)
2469 return false;
2470 // If the lines ends with "{", this is likely an function definition.
2471 if (Line.Last->is(tok::l_brace))
2472 return true;
2473 if (Next->Next == Next->MatchingParen)
2474 return true; // Empty parentheses.
2475 // If there is an &/&& after the r_paren, this is likely a function.
2476 if (Next->MatchingParen->Next &&
2477 Next->MatchingParen->Next->is(TT_PointerOrReference))
2478 return true;
2479
2480 // Check for K&R C function definitions (and C++ function definitions with
2481 // unnamed parameters), e.g.:
2482 // int f(i)
2483 // {
2484 // return i + 1;
2485 // }
2486 // bool g(size_t = 0, bool b = false)
2487 // {
2488 // return !b;
2489 // }
2490 if (IsCpp && Next->Next && Next->Next->is(tok::identifier) &&
2491 !Line.endsWith(tok::semi))
2492 return true;
2493
2494 for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
2495 Tok = Tok->Next) {
2496 if (Tok->is(TT_TypeDeclarationParen))
2497 return true;
2498 if (Tok->isOneOf(tok::l_paren, TT_TemplateOpener) && Tok->MatchingParen) {
2499 Tok = Tok->MatchingParen;
2500 continue;
2501 }
2502 if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
2503 Tok->isOneOf(TT_PointerOrReference, TT_StartOfName, tok::ellipsis))
2504 return true;
2505 if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
2506 Tok->Tok.isLiteral())
2507 return false;
2508 }
2509 return false;
2510 }
2511
mustBreakForReturnType(const AnnotatedLine & Line) const2512 bool TokenAnnotator::mustBreakForReturnType(const AnnotatedLine &Line) const {
2513 assert(Line.MightBeFunctionDecl);
2514
2515 if ((Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_TopLevel ||
2516 Style.AlwaysBreakAfterReturnType ==
2517 FormatStyle::RTBS_TopLevelDefinitions) &&
2518 Line.Level > 0)
2519 return false;
2520
2521 switch (Style.AlwaysBreakAfterReturnType) {
2522 case FormatStyle::RTBS_None:
2523 return false;
2524 case FormatStyle::RTBS_All:
2525 case FormatStyle::RTBS_TopLevel:
2526 return true;
2527 case FormatStyle::RTBS_AllDefinitions:
2528 case FormatStyle::RTBS_TopLevelDefinitions:
2529 return Line.mightBeFunctionDefinition();
2530 }
2531
2532 return false;
2533 }
2534
calculateFormattingInformation(AnnotatedLine & Line)2535 void TokenAnnotator::calculateFormattingInformation(AnnotatedLine &Line) {
2536 for (SmallVectorImpl<AnnotatedLine *>::iterator I = Line.Children.begin(),
2537 E = Line.Children.end();
2538 I != E; ++I) {
2539 calculateFormattingInformation(**I);
2540 }
2541
2542 Line.First->TotalLength =
2543 Line.First->IsMultiline ? Style.ColumnLimit
2544 : Line.FirstStartColumn + Line.First->ColumnWidth;
2545 FormatToken *Current = Line.First->Next;
2546 bool InFunctionDecl = Line.MightBeFunctionDecl;
2547 bool AlignArrayOfStructures =
2548 (Style.AlignArrayOfStructures != FormatStyle::AIAS_None &&
2549 Line.Type == LT_ArrayOfStructInitializer);
2550 if (AlignArrayOfStructures)
2551 calculateArrayInitializerColumnList(Line);
2552
2553 while (Current) {
2554 if (isFunctionDeclarationName(Style.isCpp(), *Current, Line))
2555 Current->setType(TT_FunctionDeclarationName);
2556 if (Current->is(TT_LineComment)) {
2557 if (Current->Previous->is(BK_BracedInit) &&
2558 Current->Previous->opensScope())
2559 Current->SpacesRequiredBefore =
2560 (Style.Cpp11BracedListStyle && !Style.SpacesInParentheses) ? 0 : 1;
2561 else
2562 Current->SpacesRequiredBefore = Style.SpacesBeforeTrailingComments;
2563
2564 // If we find a trailing comment, iterate backwards to determine whether
2565 // it seems to relate to a specific parameter. If so, break before that
2566 // parameter to avoid changing the comment's meaning. E.g. don't move 'b'
2567 // to the previous line in:
2568 // SomeFunction(a,
2569 // b, // comment
2570 // c);
2571 if (!Current->HasUnescapedNewline) {
2572 for (FormatToken *Parameter = Current->Previous; Parameter;
2573 Parameter = Parameter->Previous) {
2574 if (Parameter->isOneOf(tok::comment, tok::r_brace))
2575 break;
2576 if (Parameter->Previous && Parameter->Previous->is(tok::comma)) {
2577 if (!Parameter->Previous->is(TT_CtorInitializerComma) &&
2578 Parameter->HasUnescapedNewline)
2579 Parameter->MustBreakBefore = true;
2580 break;
2581 }
2582 }
2583 }
2584 } else if (Current->SpacesRequiredBefore == 0 &&
2585 spaceRequiredBefore(Line, *Current)) {
2586 Current->SpacesRequiredBefore = 1;
2587 }
2588
2589 Current->MustBreakBefore =
2590 Current->MustBreakBefore || mustBreakBefore(Line, *Current);
2591
2592 if (!Current->MustBreakBefore && InFunctionDecl &&
2593 Current->is(TT_FunctionDeclarationName))
2594 Current->MustBreakBefore = mustBreakForReturnType(Line);
2595
2596 Current->CanBreakBefore =
2597 Current->MustBreakBefore || canBreakBefore(Line, *Current);
2598 unsigned ChildSize = 0;
2599 if (Current->Previous->Children.size() == 1) {
2600 FormatToken &LastOfChild = *Current->Previous->Children[0]->Last;
2601 ChildSize = LastOfChild.isTrailingComment() ? Style.ColumnLimit
2602 : LastOfChild.TotalLength + 1;
2603 }
2604 const FormatToken *Prev = Current->Previous;
2605 if (Current->MustBreakBefore || Prev->Children.size() > 1 ||
2606 (Prev->Children.size() == 1 &&
2607 Prev->Children[0]->First->MustBreakBefore) ||
2608 Current->IsMultiline)
2609 Current->TotalLength = Prev->TotalLength + Style.ColumnLimit;
2610 else
2611 Current->TotalLength = Prev->TotalLength + Current->ColumnWidth +
2612 ChildSize + Current->SpacesRequiredBefore;
2613
2614 if (Current->is(TT_CtorInitializerColon))
2615 InFunctionDecl = false;
2616
2617 // FIXME: Only calculate this if CanBreakBefore is true once static
2618 // initializers etc. are sorted out.
2619 // FIXME: Move magic numbers to a better place.
2620
2621 // Reduce penalty for aligning ObjC method arguments using the colon
2622 // alignment as this is the canonical way (still prefer fitting everything
2623 // into one line if possible). Trying to fit a whole expression into one
2624 // line should not force other line breaks (e.g. when ObjC method
2625 // expression is a part of other expression).
2626 Current->SplitPenalty = splitPenalty(Line, *Current, InFunctionDecl);
2627 if (Style.Language == FormatStyle::LK_ObjC &&
2628 Current->is(TT_SelectorName) && Current->ParameterIndex > 0) {
2629 if (Current->ParameterIndex == 1)
2630 Current->SplitPenalty += 5 * Current->BindingStrength;
2631 } else {
2632 Current->SplitPenalty += 20 * Current->BindingStrength;
2633 }
2634
2635 Current = Current->Next;
2636 }
2637
2638 calculateUnbreakableTailLengths(Line);
2639 unsigned IndentLevel = Line.Level;
2640 for (Current = Line.First; Current != nullptr; Current = Current->Next) {
2641 if (Current->Role)
2642 Current->Role->precomputeFormattingInfos(Current);
2643 if (Current->MatchingParen &&
2644 Current->MatchingParen->opensBlockOrBlockTypeList(Style)) {
2645 assert(IndentLevel > 0);
2646 --IndentLevel;
2647 }
2648 Current->IndentLevel = IndentLevel;
2649 if (Current->opensBlockOrBlockTypeList(Style))
2650 ++IndentLevel;
2651 }
2652
2653 LLVM_DEBUG({ printDebugInfo(Line); });
2654 }
2655
calculateUnbreakableTailLengths(AnnotatedLine & Line)2656 void TokenAnnotator::calculateUnbreakableTailLengths(AnnotatedLine &Line) {
2657 unsigned UnbreakableTailLength = 0;
2658 FormatToken *Current = Line.Last;
2659 while (Current) {
2660 Current->UnbreakableTailLength = UnbreakableTailLength;
2661 if (Current->CanBreakBefore ||
2662 Current->isOneOf(tok::comment, tok::string_literal)) {
2663 UnbreakableTailLength = 0;
2664 } else {
2665 UnbreakableTailLength +=
2666 Current->ColumnWidth + Current->SpacesRequiredBefore;
2667 }
2668 Current = Current->Previous;
2669 }
2670 }
2671
calculateArrayInitializerColumnList(AnnotatedLine & Line)2672 void TokenAnnotator::calculateArrayInitializerColumnList(AnnotatedLine &Line) {
2673 if (Line.First == Line.Last) {
2674 return;
2675 }
2676 auto *CurrentToken = Line.First;
2677 CurrentToken->ArrayInitializerLineStart = true;
2678 unsigned Depth = 0;
2679 while (CurrentToken != nullptr && CurrentToken != Line.Last) {
2680 if (CurrentToken->is(tok::l_brace)) {
2681 CurrentToken->IsArrayInitializer = true;
2682 if (CurrentToken->Next != nullptr)
2683 CurrentToken->Next->MustBreakBefore = true;
2684 CurrentToken =
2685 calculateInitializerColumnList(Line, CurrentToken->Next, Depth + 1);
2686 } else {
2687 CurrentToken = CurrentToken->Next;
2688 }
2689 }
2690 }
2691
calculateInitializerColumnList(AnnotatedLine & Line,FormatToken * CurrentToken,unsigned Depth)2692 FormatToken *TokenAnnotator::calculateInitializerColumnList(
2693 AnnotatedLine &Line, FormatToken *CurrentToken, unsigned Depth) {
2694 while (CurrentToken != nullptr && CurrentToken != Line.Last) {
2695 if (CurrentToken->is(tok::l_brace))
2696 ++Depth;
2697 else if (CurrentToken->is(tok::r_brace))
2698 --Depth;
2699 if (Depth == 2 && CurrentToken->isOneOf(tok::l_brace, tok::comma)) {
2700 CurrentToken = CurrentToken->Next;
2701 if (CurrentToken == nullptr)
2702 break;
2703 CurrentToken->StartsColumn = true;
2704 CurrentToken = CurrentToken->Previous;
2705 }
2706 CurrentToken = CurrentToken->Next;
2707 }
2708 return CurrentToken;
2709 }
2710
splitPenalty(const AnnotatedLine & Line,const FormatToken & Tok,bool InFunctionDecl)2711 unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
2712 const FormatToken &Tok,
2713 bool InFunctionDecl) {
2714 const FormatToken &Left = *Tok.Previous;
2715 const FormatToken &Right = Tok;
2716
2717 if (Left.is(tok::semi))
2718 return 0;
2719
2720 if (Style.Language == FormatStyle::LK_Java) {
2721 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_throws))
2722 return 1;
2723 if (Right.is(Keywords.kw_implements))
2724 return 2;
2725 if (Left.is(tok::comma) && Left.NestingLevel == 0)
2726 return 3;
2727 } else if (Style.Language == FormatStyle::LK_JavaScript) {
2728 if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
2729 return 100;
2730 if (Left.is(TT_JsTypeColon))
2731 return 35;
2732 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
2733 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
2734 return 100;
2735 // Prefer breaking call chains (".foo") over empty "{}", "[]" or "()".
2736 if (Left.opensScope() && Right.closesScope())
2737 return 200;
2738 }
2739
2740 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
2741 return 1;
2742 if (Right.is(tok::l_square)) {
2743 if (Style.Language == FormatStyle::LK_Proto)
2744 return 1;
2745 if (Left.is(tok::r_square))
2746 return 200;
2747 // Slightly prefer formatting local lambda definitions like functions.
2748 if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
2749 return 35;
2750 if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
2751 TT_ArrayInitializerLSquare,
2752 TT_DesignatedInitializerLSquare, TT_AttributeSquare))
2753 return 500;
2754 }
2755
2756 if (Left.is(tok::coloncolon) ||
2757 (Right.is(tok::period) && Style.Language == FormatStyle::LK_Proto))
2758 return 500;
2759 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
2760 Right.is(tok::kw_operator)) {
2761 if (Line.startsWith(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
2762 return 3;
2763 if (Left.is(TT_StartOfName))
2764 return 110;
2765 if (InFunctionDecl && Right.NestingLevel == 0)
2766 return Style.PenaltyReturnTypeOnItsOwnLine;
2767 return 200;
2768 }
2769 if (Right.is(TT_PointerOrReference))
2770 return 190;
2771 if (Right.is(TT_LambdaArrow))
2772 return 110;
2773 if (Left.is(tok::equal) && Right.is(tok::l_brace))
2774 return 160;
2775 if (Left.is(TT_CastRParen))
2776 return 100;
2777 if (Left.isOneOf(tok::kw_class, tok::kw_struct))
2778 return 5000;
2779 if (Left.is(tok::comment))
2780 return 1000;
2781
2782 if (Left.isOneOf(TT_RangeBasedForLoopColon, TT_InheritanceColon,
2783 TT_CtorInitializerColon))
2784 return 2;
2785
2786 if (Right.isMemberAccess()) {
2787 // Breaking before the "./->" of a chained call/member access is reasonably
2788 // cheap, as formatting those with one call per line is generally
2789 // desirable. In particular, it should be cheaper to break before the call
2790 // than it is to break inside a call's parameters, which could lead to weird
2791 // "hanging" indents. The exception is the very last "./->" to support this
2792 // frequent pattern:
2793 //
2794 // aaaaaaaa.aaaaaaaa.bbbbbbb().ccccccccccccccccccccc(
2795 // dddddddd);
2796 //
2797 // which might otherwise be blown up onto many lines. Here, clang-format
2798 // won't produce "hanging" indents anyway as there is no other trailing
2799 // call.
2800 //
2801 // Also apply higher penalty is not a call as that might lead to a wrapping
2802 // like:
2803 //
2804 // aaaaaaa
2805 // .aaaaaaaaa.bbbbbbbb(cccccccc);
2806 return !Right.NextOperator || !Right.NextOperator->Previous->closesScope()
2807 ? 150
2808 : 35;
2809 }
2810
2811 if (Right.is(TT_TrailingAnnotation) &&
2812 (!Right.Next || Right.Next->isNot(tok::l_paren))) {
2813 // Moving trailing annotations to the next line is fine for ObjC method
2814 // declarations.
2815 if (Line.startsWith(TT_ObjCMethodSpecifier))
2816 return 10;
2817 // Generally, breaking before a trailing annotation is bad unless it is
2818 // function-like. It seems to be especially preferable to keep standard
2819 // annotations (i.e. "const", "final" and "override") on the same line.
2820 // Use a slightly higher penalty after ")" so that annotations like
2821 // "const override" are kept together.
2822 bool is_short_annotation = Right.TokenText.size() < 10;
2823 return (Left.is(tok::r_paren) ? 100 : 120) + (is_short_annotation ? 50 : 0);
2824 }
2825
2826 // In for-loops, prefer breaking at ',' and ';'.
2827 if (Line.startsWith(tok::kw_for) && Left.is(tok::equal))
2828 return 4;
2829
2830 // In Objective-C method expressions, prefer breaking before "param:" over
2831 // breaking after it.
2832 if (Right.is(TT_SelectorName))
2833 return 0;
2834 if (Left.is(tok::colon) && Left.is(TT_ObjCMethodExpr))
2835 return Line.MightBeFunctionDecl ? 50 : 500;
2836
2837 // In Objective-C type declarations, avoid breaking after the category's
2838 // open paren (we'll prefer breaking after the protocol list's opening
2839 // angle bracket, if present).
2840 if (Line.Type == LT_ObjCDecl && Left.is(tok::l_paren) && Left.Previous &&
2841 Left.Previous->isOneOf(tok::identifier, tok::greater))
2842 return 500;
2843
2844 if (Left.is(tok::l_paren) && InFunctionDecl &&
2845 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
2846 return 100;
2847 if (Left.is(tok::l_paren) && Left.Previous &&
2848 (Left.Previous->is(tok::kw_for) || Left.Previous->isIf()))
2849 return 1000;
2850 if (Left.is(tok::equal) && InFunctionDecl)
2851 return 110;
2852 if (Right.is(tok::r_brace))
2853 return 1;
2854 if (Left.is(TT_TemplateOpener))
2855 return 100;
2856 if (Left.opensScope()) {
2857 // If we aren't aligning after opening parens/braces we can always break
2858 // here unless the style does not want us to place all arguments on the
2859 // next line.
2860 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_DontAlign &&
2861 (Left.ParameterCount <= 1 || Style.AllowAllArgumentsOnNextLine))
2862 return 0;
2863 if (Left.is(tok::l_brace) && !Style.Cpp11BracedListStyle)
2864 return 19;
2865 return Left.ParameterCount > 1 ? Style.PenaltyBreakBeforeFirstCallParameter
2866 : 19;
2867 }
2868 if (Left.is(TT_JavaAnnotation))
2869 return 50;
2870
2871 if (Left.is(TT_UnaryOperator))
2872 return 60;
2873 if (Left.isOneOf(tok::plus, tok::comma) && Left.Previous &&
2874 Left.Previous->isLabelString() &&
2875 (Left.NextOperator || Left.OperatorIndex != 0))
2876 return 50;
2877 if (Right.is(tok::plus) && Left.isLabelString() &&
2878 (Right.NextOperator || Right.OperatorIndex != 0))
2879 return 25;
2880 if (Left.is(tok::comma))
2881 return 1;
2882 if (Right.is(tok::lessless) && Left.isLabelString() &&
2883 (Right.NextOperator || Right.OperatorIndex != 1))
2884 return 25;
2885 if (Right.is(tok::lessless)) {
2886 // Breaking at a << is really cheap.
2887 if (!Left.is(tok::r_paren) || Right.OperatorIndex > 0)
2888 // Slightly prefer to break before the first one in log-like statements.
2889 return 2;
2890 return 1;
2891 }
2892 if (Left.ClosesTemplateDeclaration)
2893 return Style.PenaltyBreakTemplateDeclaration;
2894 if (Left.is(TT_ConditionalExpr))
2895 return prec::Conditional;
2896 prec::Level Level = Left.getPrecedence();
2897 if (Level == prec::Unknown)
2898 Level = Right.getPrecedence();
2899 if (Level == prec::Assignment)
2900 return Style.PenaltyBreakAssignment;
2901 if (Level != prec::Unknown)
2902 return Level;
2903
2904 return 3;
2905 }
2906
spaceRequiredBeforeParens(const FormatToken & Right) const2907 bool TokenAnnotator::spaceRequiredBeforeParens(const FormatToken &Right) const {
2908 return Style.SpaceBeforeParens == FormatStyle::SBPO_Always ||
2909 (Style.SpaceBeforeParens == FormatStyle::SBPO_NonEmptyParentheses &&
2910 Right.ParameterCount > 0);
2911 }
2912
spaceRequiredBetween(const AnnotatedLine & Line,const FormatToken & Left,const FormatToken & Right)2913 bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
2914 const FormatToken &Left,
2915 const FormatToken &Right) {
2916 if (Left.is(tok::kw_return) && Right.isNot(tok::semi))
2917 return true;
2918 if (Style.isJson() && Left.is(tok::string_literal) && Right.is(tok::colon))
2919 return false;
2920 if (Left.is(Keywords.kw_assert) && Style.Language == FormatStyle::LK_Java)
2921 return true;
2922 if (Style.ObjCSpaceAfterProperty && Line.Type == LT_ObjCProperty &&
2923 Left.Tok.getObjCKeywordID() == tok::objc_property)
2924 return true;
2925 if (Right.is(tok::hashhash))
2926 return Left.is(tok::hash);
2927 if (Left.isOneOf(tok::hashhash, tok::hash))
2928 return Right.is(tok::hash);
2929 if ((Left.is(tok::l_paren) && Right.is(tok::r_paren)) ||
2930 (Left.is(tok::l_brace) && Left.isNot(BK_Block) &&
2931 Right.is(tok::r_brace) && Right.isNot(BK_Block)))
2932 return Style.SpaceInEmptyParentheses;
2933 if (Style.SpacesInConditionalStatement) {
2934 if (Left.is(tok::l_paren) && Left.Previous &&
2935 isKeywordWithCondition(*Left.Previous))
2936 return true;
2937 if (Right.is(tok::r_paren) && Right.MatchingParen &&
2938 Right.MatchingParen->Previous &&
2939 isKeywordWithCondition(*Right.MatchingParen->Previous))
2940 return true;
2941 }
2942
2943 // requires ( or requires(
2944 if (Right.is(tok::l_paren) && Left.is(tok::kw_requires))
2945 return spaceRequiredBeforeParens(Right);
2946 // requires clause Concept1<T> && Concept2<T>
2947 if (Left.is(TT_ConstraintJunctions) && Right.is(tok::identifier))
2948 return true;
2949
2950 if (Left.is(tok::l_paren) || Right.is(tok::r_paren))
2951 return (Right.is(TT_CastRParen) ||
2952 (Left.MatchingParen && Left.MatchingParen->is(TT_CastRParen)))
2953 ? Style.SpacesInCStyleCastParentheses
2954 : Style.SpacesInParentheses;
2955 if (Right.isOneOf(tok::semi, tok::comma))
2956 return false;
2957 if (Right.is(tok::less) && Line.Type == LT_ObjCDecl) {
2958 bool IsLightweightGeneric = Right.MatchingParen &&
2959 Right.MatchingParen->Next &&
2960 Right.MatchingParen->Next->is(tok::colon);
2961 return !IsLightweightGeneric && Style.ObjCSpaceBeforeProtocolList;
2962 }
2963 if (Right.is(tok::less) && Left.is(tok::kw_template))
2964 return Style.SpaceAfterTemplateKeyword;
2965 if (Left.isOneOf(tok::exclaim, tok::tilde))
2966 return false;
2967 if (Left.is(tok::at) &&
2968 Right.isOneOf(tok::identifier, tok::string_literal, tok::char_constant,
2969 tok::numeric_constant, tok::l_paren, tok::l_brace,
2970 tok::kw_true, tok::kw_false))
2971 return false;
2972 if (Left.is(tok::colon))
2973 return !Left.is(TT_ObjCMethodExpr);
2974 if (Left.is(tok::coloncolon))
2975 return false;
2976 if (Left.is(tok::less) || Right.isOneOf(tok::greater, tok::less)) {
2977 if (Style.Language == FormatStyle::LK_TextProto ||
2978 (Style.Language == FormatStyle::LK_Proto &&
2979 (Left.is(TT_DictLiteral) || Right.is(TT_DictLiteral)))) {
2980 // Format empty list as `<>`.
2981 if (Left.is(tok::less) && Right.is(tok::greater))
2982 return false;
2983 return !Style.Cpp11BracedListStyle;
2984 }
2985 return false;
2986 }
2987 if (Right.is(tok::ellipsis))
2988 return Left.Tok.isLiteral() || (Left.is(tok::identifier) && Left.Previous &&
2989 Left.Previous->is(tok::kw_case));
2990 if (Left.is(tok::l_square) && Right.is(tok::amp))
2991 return Style.SpacesInSquareBrackets;
2992 if (Right.is(TT_PointerOrReference)) {
2993 if (Left.is(tok::r_paren) && Line.MightBeFunctionDecl) {
2994 if (!Left.MatchingParen)
2995 return true;
2996 FormatToken *TokenBeforeMatchingParen =
2997 Left.MatchingParen->getPreviousNonComment();
2998 if (!TokenBeforeMatchingParen || !Left.is(TT_TypeDeclarationParen))
2999 return true;
3000 }
3001 // Add a space if the previous token is a pointer qualifer or the closing
3002 // parenthesis of __attribute__(()) expression and the style requires spaces
3003 // after pointer qualifiers.
3004 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_After ||
3005 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) &&
3006 (Left.is(TT_AttributeParen) || Left.canBePointerOrReferenceQualifier()))
3007 return true;
3008 return (
3009 Left.Tok.isLiteral() ||
3010 (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
3011 (getTokenPointerOrReferenceAlignment(Right) != FormatStyle::PAS_Left ||
3012 (Line.IsMultiVariableDeclStmt &&
3013 (Left.NestingLevel == 0 ||
3014 (Left.NestingLevel == 1 && Line.First->is(tok::kw_for)))))));
3015 }
3016 if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
3017 (!Left.is(TT_PointerOrReference) ||
3018 (getTokenPointerOrReferenceAlignment(Left) != FormatStyle::PAS_Right &&
3019 !Line.IsMultiVariableDeclStmt)))
3020 return true;
3021 if (Left.is(TT_PointerOrReference)) {
3022 // Add a space if the next token is a pointer qualifer and the style
3023 // requires spaces before pointer qualifiers.
3024 if ((Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Before ||
3025 Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both) &&
3026 Right.canBePointerOrReferenceQualifier())
3027 return true;
3028 return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
3029 (Right.isOneOf(Keywords.kw_override, Keywords.kw_final) &&
3030 !Right.is(TT_StartOfName)) ||
3031 (Right.is(tok::l_brace) && Right.is(BK_Block)) ||
3032 (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
3033 tok::l_paren) &&
3034 (getTokenPointerOrReferenceAlignment(Left) !=
3035 FormatStyle::PAS_Right &&
3036 !Line.IsMultiVariableDeclStmt) &&
3037 Left.Previous &&
3038 !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon,
3039 tok::l_square));
3040 }
3041 // Ensure right pointer alignement with ellipsis e.g. int *...P
3042 if (Left.is(tok::ellipsis) && Left.Previous &&
3043 Left.Previous->isOneOf(tok::star, tok::amp, tok::ampamp))
3044 return Style.PointerAlignment != FormatStyle::PAS_Right;
3045
3046 if (Right.is(tok::star) && Left.is(tok::l_paren))
3047 return false;
3048 if (Left.is(tok::star) && Right.isOneOf(tok::star, tok::amp, tok::ampamp))
3049 return false;
3050 if (Right.isOneOf(tok::star, tok::amp, tok::ampamp)) {
3051 const FormatToken *Previous = &Left;
3052 while (Previous && !Previous->is(tok::kw_operator)) {
3053 if (Previous->is(tok::identifier) || Previous->isSimpleTypeSpecifier()) {
3054 Previous = Previous->getPreviousNonComment();
3055 continue;
3056 }
3057 if (Previous->is(TT_TemplateCloser) && Previous->MatchingParen) {
3058 Previous = Previous->MatchingParen->getPreviousNonComment();
3059 continue;
3060 }
3061 if (Previous->is(tok::coloncolon)) {
3062 Previous = Previous->getPreviousNonComment();
3063 continue;
3064 }
3065 break;
3066 }
3067 // Space between the type and the * in:
3068 // operator void*()
3069 // operator char*()
3070 // operator void const*()
3071 // operator void volatile*()
3072 // operator /*comment*/ const char*()
3073 // operator volatile /*comment*/ char*()
3074 // operator Foo*()
3075 // operator C<T>*()
3076 // operator std::Foo*()
3077 // operator C<T>::D<U>*()
3078 // dependent on PointerAlignment style.
3079 if (Previous) {
3080 if (Previous->endsSequence(tok::kw_operator))
3081 return (Style.PointerAlignment != FormatStyle::PAS_Left);
3082 if (Previous->is(tok::kw_const) || Previous->is(tok::kw_volatile))
3083 return (Style.PointerAlignment != FormatStyle::PAS_Left) ||
3084 (Style.SpaceAroundPointerQualifiers ==
3085 FormatStyle::SAPQ_After) ||
3086 (Style.SpaceAroundPointerQualifiers == FormatStyle::SAPQ_Both);
3087 }
3088 }
3089 const auto SpaceRequiredForArrayInitializerLSquare =
3090 [](const FormatToken &LSquareTok, const FormatStyle &Style) {
3091 return Style.SpacesInContainerLiterals ||
3092 ((Style.Language == FormatStyle::LK_Proto ||
3093 Style.Language == FormatStyle::LK_TextProto) &&
3094 !Style.Cpp11BracedListStyle &&
3095 LSquareTok.endsSequence(tok::l_square, tok::colon,
3096 TT_SelectorName));
3097 };
3098 if (Left.is(tok::l_square))
3099 return (Left.is(TT_ArrayInitializerLSquare) && Right.isNot(tok::r_square) &&
3100 SpaceRequiredForArrayInitializerLSquare(Left, Style)) ||
3101 (Left.isOneOf(TT_ArraySubscriptLSquare, TT_StructuredBindingLSquare,
3102 TT_LambdaLSquare) &&
3103 Style.SpacesInSquareBrackets && Right.isNot(tok::r_square));
3104 if (Right.is(tok::r_square))
3105 return Right.MatchingParen &&
3106 ((Right.MatchingParen->is(TT_ArrayInitializerLSquare) &&
3107 SpaceRequiredForArrayInitializerLSquare(*Right.MatchingParen,
3108 Style)) ||
3109 (Style.SpacesInSquareBrackets &&
3110 Right.MatchingParen->isOneOf(TT_ArraySubscriptLSquare,
3111 TT_StructuredBindingLSquare,
3112 TT_LambdaLSquare)) ||
3113 Right.MatchingParen->is(TT_AttributeParen));
3114 if (Right.is(tok::l_square) &&
3115 !Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare,
3116 TT_DesignatedInitializerLSquare,
3117 TT_StructuredBindingLSquare, TT_AttributeSquare) &&
3118 !Left.isOneOf(tok::numeric_constant, TT_DictLiteral) &&
3119 !(!Left.is(tok::r_square) && Style.SpaceBeforeSquareBrackets &&
3120 Right.is(TT_ArraySubscriptLSquare)))
3121 return false;
3122 if (Left.is(tok::l_brace) && Right.is(tok::r_brace))
3123 return !Left.Children.empty(); // No spaces in "{}".
3124 if ((Left.is(tok::l_brace) && Left.isNot(BK_Block)) ||
3125 (Right.is(tok::r_brace) && Right.MatchingParen &&
3126 Right.MatchingParen->isNot(BK_Block)))
3127 return Style.Cpp11BracedListStyle ? Style.SpacesInParentheses : true;
3128 if (Left.is(TT_BlockComment))
3129 // No whitespace in x(/*foo=*/1), except for JavaScript.
3130 return Style.Language == FormatStyle::LK_JavaScript ||
3131 !Left.TokenText.endswith("=*/");
3132
3133 // Space between template and attribute.
3134 // e.g. template <typename T> [[nodiscard]] ...
3135 if (Left.is(TT_TemplateCloser) && Right.is(TT_AttributeSquare))
3136 return true;
3137 if (Right.is(tok::l_paren)) {
3138 if ((Left.is(tok::r_paren) && Left.is(TT_AttributeParen)) ||
3139 (Left.is(tok::r_square) && Left.is(TT_AttributeSquare)))
3140 return true;
3141 if (Style.SpaceBeforeParens ==
3142 FormatStyle::SBPO_ControlStatementsExceptControlMacros &&
3143 Left.is(TT_ForEachMacro))
3144 return false;
3145 if (Style.SpaceBeforeParens ==
3146 FormatStyle::SBPO_ControlStatementsExceptControlMacros &&
3147 Left.is(TT_IfMacro))
3148 return false;
3149 return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
3150 (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
3151 (Left.isOneOf(tok::pp_elif, tok::kw_for, tok::kw_while,
3152 tok::kw_switch, tok::kw_case, TT_ForEachMacro,
3153 TT_ObjCForIn) ||
3154 Left.isIf(Line.Type != LT_PreprocessorDirective) ||
3155 (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
3156 tok::kw_new, tok::kw_delete) &&
3157 (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
3158 (spaceRequiredBeforeParens(Right) &&
3159 (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() ||
3160 Left.is(tok::r_paren) || Left.isSimpleTypeSpecifier() ||
3161 (Left.is(tok::r_square) && Left.MatchingParen &&
3162 Left.MatchingParen->is(TT_LambdaLSquare))) &&
3163 Line.Type != LT_PreprocessorDirective);
3164 }
3165 if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
3166 return false;
3167 if (Right.is(TT_UnaryOperator))
3168 return !Left.isOneOf(tok::l_paren, tok::l_square, tok::at) &&
3169 (Left.isNot(tok::colon) || Left.isNot(TT_ObjCMethodExpr));
3170 if ((Left.isOneOf(tok::identifier, tok::greater, tok::r_square,
3171 tok::r_paren) ||
3172 Left.isSimpleTypeSpecifier()) &&
3173 Right.is(tok::l_brace) && Right.getNextNonComment() &&
3174 Right.isNot(BK_Block))
3175 return false;
3176 if (Left.is(tok::period) || Right.is(tok::period))
3177 return false;
3178 if (Right.is(tok::hash) && Left.is(tok::identifier) && Left.TokenText == "L")
3179 return false;
3180 if (Left.is(TT_TemplateCloser) && Left.MatchingParen &&
3181 Left.MatchingParen->Previous &&
3182 (Left.MatchingParen->Previous->is(tok::period) ||
3183 Left.MatchingParen->Previous->is(tok::coloncolon)))
3184 // Java call to generic function with explicit type:
3185 // A.<B<C<...>>>DoSomething();
3186 // A::<B<C<...>>>DoSomething(); // With a Java 8 method reference.
3187 return false;
3188 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_square))
3189 return false;
3190 if (Left.is(tok::l_brace) && Left.endsSequence(TT_DictLiteral, tok::at))
3191 // Objective-C dictionary literal -> no space after opening brace.
3192 return false;
3193 if (Right.is(tok::r_brace) && Right.MatchingParen &&
3194 Right.MatchingParen->endsSequence(TT_DictLiteral, tok::at))
3195 // Objective-C dictionary literal -> no space before closing brace.
3196 return false;
3197 if (Right.getType() == TT_TrailingAnnotation &&
3198 Right.isOneOf(tok::amp, tok::ampamp) &&
3199 Left.isOneOf(tok::kw_const, tok::kw_volatile) &&
3200 (!Right.Next || Right.Next->is(tok::semi)))
3201 // Match const and volatile ref-qualifiers without any additional
3202 // qualifiers such as
3203 // void Fn() const &;
3204 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left;
3205 return true;
3206 }
3207
spaceRequiredBefore(const AnnotatedLine & Line,const FormatToken & Right)3208 bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
3209 const FormatToken &Right) {
3210 const FormatToken &Left = *Right.Previous;
3211 auto HasExistingWhitespace = [&Right]() {
3212 return Right.WhitespaceRange.getBegin() != Right.WhitespaceRange.getEnd();
3213 };
3214 if (Right.Tok.getIdentifierInfo() && Left.Tok.getIdentifierInfo())
3215 return true; // Never ever merge two identifiers.
3216 if (Style.isCpp()) {
3217 if (Left.is(tok::kw_operator))
3218 return Right.is(tok::coloncolon);
3219 if (Right.is(tok::l_brace) && Right.is(BK_BracedInit) &&
3220 !Left.opensScope() && Style.SpaceBeforeCpp11BracedList)
3221 return true;
3222 } else if (Style.Language == FormatStyle::LK_Proto ||
3223 Style.Language == FormatStyle::LK_TextProto) {
3224 if (Right.is(tok::period) &&
3225 Left.isOneOf(Keywords.kw_optional, Keywords.kw_required,
3226 Keywords.kw_repeated, Keywords.kw_extend))
3227 return true;
3228 if (Right.is(tok::l_paren) &&
3229 Left.isOneOf(Keywords.kw_returns, Keywords.kw_option))
3230 return true;
3231 if (Right.isOneOf(tok::l_brace, tok::less) && Left.is(TT_SelectorName))
3232 return true;
3233 // Slashes occur in text protocol extension syntax: [type/type] { ... }.
3234 if (Left.is(tok::slash) || Right.is(tok::slash))
3235 return false;
3236 if (Left.MatchingParen &&
3237 Left.MatchingParen->is(TT_ProtoExtensionLSquare) &&
3238 Right.isOneOf(tok::l_brace, tok::less))
3239 return !Style.Cpp11BracedListStyle;
3240 // A percent is probably part of a formatting specification, such as %lld.
3241 if (Left.is(tok::percent))
3242 return false;
3243 // Preserve the existence of a space before a percent for cases like 0x%04x
3244 // and "%d %d"
3245 if (Left.is(tok::numeric_constant) && Right.is(tok::percent))
3246 return HasExistingWhitespace();
3247 } else if (Style.isJson()) {
3248 if (Right.is(tok::colon))
3249 return false;
3250 } else if (Style.isCSharp()) {
3251 // Require spaces around '{' and before '}' unless they appear in
3252 // interpolated strings. Interpolated strings are merged into a single token
3253 // so cannot have spaces inserted by this function.
3254
3255 // No space between 'this' and '['
3256 if (Left.is(tok::kw_this) && Right.is(tok::l_square))
3257 return false;
3258
3259 // No space between 'new' and '('
3260 if (Left.is(tok::kw_new) && Right.is(tok::l_paren))
3261 return false;
3262
3263 // Space before { (including space within '{ {').
3264 if (Right.is(tok::l_brace))
3265 return true;
3266
3267 // Spaces inside braces.
3268 if (Left.is(tok::l_brace) && Right.isNot(tok::r_brace))
3269 return true;
3270
3271 if (Left.isNot(tok::l_brace) && Right.is(tok::r_brace))
3272 return true;
3273
3274 // Spaces around '=>'.
3275 if (Left.is(TT_FatArrow) || Right.is(TT_FatArrow))
3276 return true;
3277
3278 // No spaces around attribute target colons
3279 if (Left.is(TT_AttributeColon) || Right.is(TT_AttributeColon))
3280 return false;
3281
3282 // space between type and variable e.g. Dictionary<string,string> foo;
3283 if (Left.is(TT_TemplateCloser) && Right.is(TT_StartOfName))
3284 return true;
3285
3286 // spaces inside square brackets.
3287 if (Left.is(tok::l_square) || Right.is(tok::r_square))
3288 return Style.SpacesInSquareBrackets;
3289
3290 // No space before ? in nullable types.
3291 if (Right.is(TT_CSharpNullable))
3292 return false;
3293
3294 // No space before null forgiving '!'.
3295 if (Right.is(TT_NonNullAssertion))
3296 return false;
3297
3298 // No space between consecutive commas '[,,]'.
3299 if (Left.is(tok::comma) && Right.is(tok::comma))
3300 return false;
3301
3302 // space after var in `var (key, value)`
3303 if (Left.is(Keywords.kw_var) && Right.is(tok::l_paren))
3304 return true;
3305
3306 // space between keywords and paren e.g. "using ("
3307 if (Right.is(tok::l_paren))
3308 if (Left.isOneOf(tok::kw_using, Keywords.kw_async, Keywords.kw_when,
3309 Keywords.kw_lock))
3310 return Style.SpaceBeforeParens == FormatStyle::SBPO_ControlStatements ||
3311 spaceRequiredBeforeParens(Right);
3312
3313 // space between method modifier and opening parenthesis of a tuple return
3314 // type
3315 if (Left.isOneOf(tok::kw_public, tok::kw_private, tok::kw_protected,
3316 tok::kw_virtual, tok::kw_extern, tok::kw_static,
3317 Keywords.kw_internal, Keywords.kw_abstract,
3318 Keywords.kw_sealed, Keywords.kw_override,
3319 Keywords.kw_async, Keywords.kw_unsafe) &&
3320 Right.is(tok::l_paren))
3321 return true;
3322 } else if (Style.Language == FormatStyle::LK_JavaScript) {
3323 if (Left.is(TT_FatArrow))
3324 return true;
3325 // for await ( ...
3326 if (Right.is(tok::l_paren) && Left.is(Keywords.kw_await) && Left.Previous &&
3327 Left.Previous->is(tok::kw_for))
3328 return true;
3329 if (Left.is(Keywords.kw_async) && Right.is(tok::l_paren) &&
3330 Right.MatchingParen) {
3331 const FormatToken *Next = Right.MatchingParen->getNextNonComment();
3332 // An async arrow function, for example: `x = async () => foo();`,
3333 // as opposed to calling a function called async: `x = async();`
3334 if (Next && Next->is(TT_FatArrow))
3335 return true;
3336 }
3337 if ((Left.is(TT_TemplateString) && Left.TokenText.endswith("${")) ||
3338 (Right.is(TT_TemplateString) && Right.TokenText.startswith("}")))
3339 return false;
3340 // In tagged template literals ("html`bar baz`"), there is no space between
3341 // the tag identifier and the template string.
3342 if (Keywords.IsJavaScriptIdentifier(Left,
3343 /* AcceptIdentifierName= */ false) &&
3344 Right.is(TT_TemplateString))
3345 return false;
3346 if (Right.is(tok::star) &&
3347 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield))
3348 return false;
3349 if (Right.isOneOf(tok::l_brace, tok::l_square) &&
3350 Left.isOneOf(Keywords.kw_function, Keywords.kw_yield,
3351 Keywords.kw_extends, Keywords.kw_implements))
3352 return true;
3353 if (Right.is(tok::l_paren)) {
3354 // JS methods can use some keywords as names (e.g. `delete()`).
3355 if (Line.MustBeDeclaration && Left.Tok.getIdentifierInfo())
3356 return false;
3357 // Valid JS method names can include keywords, e.g. `foo.delete()` or
3358 // `bar.instanceof()`. Recognize call positions by preceding period.
3359 if (Left.Previous && Left.Previous->is(tok::period) &&
3360 Left.Tok.getIdentifierInfo())
3361 return false;
3362 // Additional unary JavaScript operators that need a space after.
3363 if (Left.isOneOf(tok::kw_throw, Keywords.kw_await, Keywords.kw_typeof,
3364 tok::kw_void))
3365 return true;
3366 }
3367 // `foo as const;` casts into a const type.
3368 if (Left.endsSequence(tok::kw_const, Keywords.kw_as)) {
3369 return false;
3370 }
3371 if ((Left.isOneOf(Keywords.kw_let, Keywords.kw_var, Keywords.kw_in,
3372 tok::kw_const) ||
3373 // "of" is only a keyword if it appears after another identifier
3374 // (e.g. as "const x of y" in a for loop), or after a destructuring
3375 // operation (const [x, y] of z, const {a, b} of c).
3376 (Left.is(Keywords.kw_of) && Left.Previous &&
3377 (Left.Previous->Tok.is(tok::identifier) ||
3378 Left.Previous->isOneOf(tok::r_square, tok::r_brace)))) &&
3379 (!Left.Previous || !Left.Previous->is(tok::period)))
3380 return true;
3381 if (Left.isOneOf(tok::kw_for, Keywords.kw_as) && Left.Previous &&
3382 Left.Previous->is(tok::period) && Right.is(tok::l_paren))
3383 return false;
3384 if (Left.is(Keywords.kw_as) &&
3385 Right.isOneOf(tok::l_square, tok::l_brace, tok::l_paren))
3386 return true;
3387 if (Left.is(tok::kw_default) && Left.Previous &&
3388 Left.Previous->is(tok::kw_export))
3389 return true;
3390 if (Left.is(Keywords.kw_is) && Right.is(tok::l_brace))
3391 return true;
3392 if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
3393 return false;
3394 if (Left.is(TT_JsTypeOperator) || Right.is(TT_JsTypeOperator))
3395 return false;
3396 if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
3397 Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
3398 return false;
3399 if (Left.is(tok::ellipsis))
3400 return false;
3401 if (Left.is(TT_TemplateCloser) &&
3402 !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
3403 Keywords.kw_implements, Keywords.kw_extends))
3404 // Type assertions ('<type>expr') are not followed by whitespace. Other
3405 // locations that should have whitespace following are identified by the
3406 // above set of follower tokens.
3407 return false;
3408 if (Right.is(TT_NonNullAssertion))
3409 return false;
3410 if (Left.is(TT_NonNullAssertion) &&
3411 Right.isOneOf(Keywords.kw_as, Keywords.kw_in))
3412 return true; // "x! as string", "x! in y"
3413 } else if (Style.Language == FormatStyle::LK_Java) {
3414 if (Left.is(tok::r_square) && Right.is(tok::l_brace))
3415 return true;
3416 if (Left.is(Keywords.kw_synchronized) && Right.is(tok::l_paren))
3417 return Style.SpaceBeforeParens != FormatStyle::SBPO_Never;
3418 if ((Left.isOneOf(tok::kw_static, tok::kw_public, tok::kw_private,
3419 tok::kw_protected) ||
3420 Left.isOneOf(Keywords.kw_final, Keywords.kw_abstract,
3421 Keywords.kw_native)) &&
3422 Right.is(TT_TemplateOpener))
3423 return true;
3424 }
3425 if (Left.is(TT_ImplicitStringLiteral))
3426 return HasExistingWhitespace();
3427 if (Line.Type == LT_ObjCMethodDecl) {
3428 if (Left.is(TT_ObjCMethodSpecifier))
3429 return true;
3430 if (Left.is(tok::r_paren) && canBeObjCSelectorComponent(Right))
3431 // Don't space between ')' and <id> or ')' and 'new'. 'new' is not a
3432 // keyword in Objective-C, and '+ (instancetype)new;' is a standard class
3433 // method declaration.
3434 return false;
3435 }
3436 if (Line.Type == LT_ObjCProperty &&
3437 (Right.is(tok::equal) || Left.is(tok::equal)))
3438 return false;
3439
3440 if (Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow) ||
3441 Left.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow))
3442 return true;
3443 if (Right.is(TT_OverloadedOperatorLParen))
3444 return spaceRequiredBeforeParens(Right);
3445 if (Left.is(tok::comma))
3446 return true;
3447 if (Right.is(tok::comma))
3448 return false;
3449 if (Right.is(TT_ObjCBlockLParen))
3450 return true;
3451 if (Right.is(TT_CtorInitializerColon))
3452 return Style.SpaceBeforeCtorInitializerColon;
3453 if (Right.is(TT_InheritanceColon) && !Style.SpaceBeforeInheritanceColon)
3454 return false;
3455 if (Right.is(TT_RangeBasedForLoopColon) &&
3456 !Style.SpaceBeforeRangeBasedForLoopColon)
3457 return false;
3458 if (Left.is(TT_BitFieldColon))
3459 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both ||
3460 Style.BitFieldColonSpacing == FormatStyle::BFCS_After;
3461 if (Right.is(tok::colon)) {
3462 if (Line.First->isOneOf(tok::kw_default, tok::kw_case))
3463 return Style.SpaceBeforeCaseColon;
3464 if (!Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
3465 return false;
3466 if (Right.is(TT_ObjCMethodExpr))
3467 return false;
3468 if (Left.is(tok::question))
3469 return false;
3470 if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
3471 return false;
3472 if (Right.is(TT_DictLiteral))
3473 return Style.SpacesInContainerLiterals;
3474 if (Right.is(TT_AttributeColon))
3475 return false;
3476 if (Right.is(TT_CSharpNamedArgumentColon))
3477 return false;
3478 if (Right.is(TT_BitFieldColon))
3479 return Style.BitFieldColonSpacing == FormatStyle::BFCS_Both ||
3480 Style.BitFieldColonSpacing == FormatStyle::BFCS_Before;
3481 return true;
3482 }
3483 // Do not merge "- -" into "--".
3484 if ((Left.isOneOf(tok::minus, tok::minusminus) &&
3485 Right.isOneOf(tok::minus, tok::minusminus)) ||
3486 (Left.isOneOf(tok::plus, tok::plusplus) &&
3487 Right.isOneOf(tok::plus, tok::plusplus)))
3488 return true;
3489 if (Left.is(TT_UnaryOperator)) {
3490 if (!Right.is(tok::l_paren)) {
3491 // The alternative operators for ~ and ! are "compl" and "not".
3492 // If they are used instead, we do not want to combine them with
3493 // the token to the right, unless that is a left paren.
3494 if (Left.is(tok::exclaim) && Left.TokenText == "not")
3495 return true;
3496 if (Left.is(tok::tilde) && Left.TokenText == "compl")
3497 return true;
3498 // Lambda captures allow for a lone &, so "&]" needs to be properly
3499 // handled.
3500 if (Left.is(tok::amp) && Right.is(tok::r_square))
3501 return Style.SpacesInSquareBrackets;
3502 }
3503 return (Style.SpaceAfterLogicalNot && Left.is(tok::exclaim)) ||
3504 Right.is(TT_BinaryOperator);
3505 }
3506
3507 // If the next token is a binary operator or a selector name, we have
3508 // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
3509 if (Left.is(TT_CastRParen))
3510 return Style.SpaceAfterCStyleCast ||
3511 Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
3512
3513 auto ShouldAddSpacesInAngles = [this, &HasExistingWhitespace]() {
3514 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Always)
3515 return true;
3516 if (this->Style.SpacesInAngles == FormatStyle::SIAS_Leave)
3517 return HasExistingWhitespace();
3518 return false;
3519 };
3520
3521 if (Left.is(tok::greater) && Right.is(tok::greater)) {
3522 if (Style.Language == FormatStyle::LK_TextProto ||
3523 (Style.Language == FormatStyle::LK_Proto && Left.is(TT_DictLiteral)))
3524 return !Style.Cpp11BracedListStyle;
3525 return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
3526 ((Style.Standard < FormatStyle::LS_Cpp11) ||
3527 ShouldAddSpacesInAngles());
3528 }
3529 if (Right.isOneOf(tok::arrow, tok::arrowstar, tok::periodstar) ||
3530 Left.isOneOf(tok::arrow, tok::period, tok::arrowstar, tok::periodstar) ||
3531 (Right.is(tok::period) && Right.isNot(TT_DesignatedInitializerPeriod)))
3532 return false;
3533 if (!Style.SpaceBeforeAssignmentOperators && Left.isNot(TT_TemplateCloser) &&
3534 Right.getPrecedence() == prec::Assignment)
3535 return false;
3536 if (Style.Language == FormatStyle::LK_Java && Right.is(tok::coloncolon) &&
3537 (Left.is(tok::identifier) || Left.is(tok::kw_this)))
3538 return false;
3539 if (Right.is(tok::coloncolon) && Left.is(tok::identifier))
3540 // Generally don't remove existing spaces between an identifier and "::".
3541 // The identifier might actually be a macro name such as ALWAYS_INLINE. If
3542 // this turns out to be too lenient, add analysis of the identifier itself.
3543 return HasExistingWhitespace();
3544 if (Right.is(tok::coloncolon) &&
3545 !Left.isOneOf(tok::l_brace, tok::comment, tok::l_paren))
3546 // Put a space between < and :: in vector< ::std::string >
3547 return (Left.is(TT_TemplateOpener) &&
3548 ((Style.Standard < FormatStyle::LS_Cpp11) ||
3549 ShouldAddSpacesInAngles())) ||
3550 !(Left.isOneOf(tok::l_paren, tok::r_paren, tok::l_square,
3551 tok::kw___super, TT_TemplateOpener,
3552 TT_TemplateCloser)) ||
3553 (Left.is(tok::l_paren) && Style.SpacesInParentheses);
3554 if ((Left.is(TT_TemplateOpener)) != (Right.is(TT_TemplateCloser)))
3555 return ShouldAddSpacesInAngles();
3556 // Space before TT_StructuredBindingLSquare.
3557 if (Right.is(TT_StructuredBindingLSquare))
3558 return !Left.isOneOf(tok::amp, tok::ampamp) ||
3559 getTokenReferenceAlignment(Left) != FormatStyle::PAS_Right;
3560 // Space before & or && following a TT_StructuredBindingLSquare.
3561 if (Right.Next && Right.Next->is(TT_StructuredBindingLSquare) &&
3562 Right.isOneOf(tok::amp, tok::ampamp))
3563 return getTokenReferenceAlignment(Right) != FormatStyle::PAS_Left;
3564 if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
3565 (Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) &&
3566 !Right.is(tok::r_paren)))
3567 return true;
3568 if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
3569 Right.isNot(TT_FunctionTypeLParen))
3570 return spaceRequiredBeforeParens(Right);
3571 if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
3572 Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
3573 return false;
3574 if (Right.is(tok::less) && Left.isNot(tok::l_paren) &&
3575 Line.startsWith(tok::hash))
3576 return true;
3577 if (Right.is(TT_TrailingUnaryOperator))
3578 return false;
3579 if (Left.is(TT_RegexLiteral))
3580 return false;
3581 return spaceRequiredBetween(Line, Left, Right);
3582 }
3583
3584 // Returns 'true' if 'Tok' is a brace we'd want to break before in Allman style.
isAllmanBrace(const FormatToken & Tok)3585 static bool isAllmanBrace(const FormatToken &Tok) {
3586 return Tok.is(tok::l_brace) && Tok.is(BK_Block) &&
3587 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_LambdaLBrace, TT_DictLiteral);
3588 }
3589
3590 // Returns 'true' if 'Tok' is an function argument.
IsFunctionArgument(const FormatToken & Tok)3591 static bool IsFunctionArgument(const FormatToken &Tok) {
3592 return Tok.MatchingParen && Tok.MatchingParen->Next &&
3593 Tok.MatchingParen->Next->isOneOf(tok::comma, tok::r_paren);
3594 }
3595
3596 static bool
isItAnEmptyLambdaAllowed(const FormatToken & Tok,FormatStyle::ShortLambdaStyle ShortLambdaOption)3597 isItAnEmptyLambdaAllowed(const FormatToken &Tok,
3598 FormatStyle::ShortLambdaStyle ShortLambdaOption) {
3599 return Tok.Children.empty() && ShortLambdaOption != FormatStyle::SLS_None;
3600 }
3601
isAllmanLambdaBrace(const FormatToken & Tok)3602 static bool isAllmanLambdaBrace(const FormatToken &Tok) {
3603 return (Tok.is(tok::l_brace) && Tok.is(BK_Block) &&
3604 !Tok.isOneOf(TT_ObjCBlockLBrace, TT_DictLiteral));
3605 }
3606
3607 // Returns the first token on the line that is not a comment.
getFirstNonComment(const AnnotatedLine & Line)3608 static const FormatToken *getFirstNonComment(const AnnotatedLine &Line) {
3609 const FormatToken *Next = Line.First;
3610 if (!Next)
3611 return Next;
3612 if (Next->is(tok::comment))
3613 Next = Next->getNextNonComment();
3614 return Next;
3615 }
3616
mustBreakBefore(const AnnotatedLine & Line,const FormatToken & Right)3617 bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
3618 const FormatToken &Right) {
3619 const FormatToken &Left = *Right.Previous;
3620 if (Right.NewlinesBefore > 1 && Style.MaxEmptyLinesToKeep > 0)
3621 return true;
3622
3623 if (Style.isCSharp()) {
3624 if (Right.is(TT_CSharpNamedArgumentColon) ||
3625 Left.is(TT_CSharpNamedArgumentColon))
3626 return false;
3627 if (Right.is(TT_CSharpGenericTypeConstraint))
3628 return true;
3629
3630 // Break after C# [...] and before public/protected/private/internal.
3631 if (Left.is(TT_AttributeSquare) && Left.is(tok::r_square) &&
3632 (Right.isAccessSpecifier(/*ColonRequired=*/false) ||
3633 Right.is(Keywords.kw_internal)))
3634 return true;
3635 // Break between ] and [ but only when there are really 2 attributes.
3636 if (Left.is(TT_AttributeSquare) && Right.is(TT_AttributeSquare) &&
3637 Left.is(tok::r_square) && Right.is(tok::l_square))
3638 return true;
3639
3640 } else if (Style.Language == FormatStyle::LK_JavaScript) {
3641 // FIXME: This might apply to other languages and token kinds.
3642 if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
3643 Left.Previous->is(tok::string_literal))
3644 return true;
3645 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
3646 Left.Previous && Left.Previous->is(tok::equal) &&
3647 Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
3648 tok::kw_const) &&
3649 // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
3650 // above.
3651 !Line.First->isOneOf(Keywords.kw_var, Keywords.kw_let))
3652 // Object literals on the top level of a file are treated as "enum-style".
3653 // Each key/value pair is put on a separate line, instead of bin-packing.
3654 return true;
3655 if (Left.is(tok::l_brace) && Line.Level == 0 &&
3656 (Line.startsWith(tok::kw_enum) ||
3657 Line.startsWith(tok::kw_const, tok::kw_enum) ||
3658 Line.startsWith(tok::kw_export, tok::kw_enum) ||
3659 Line.startsWith(tok::kw_export, tok::kw_const, tok::kw_enum)))
3660 // JavaScript top-level enum key/value pairs are put on separate lines
3661 // instead of bin-packing.
3662 return true;
3663 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous &&
3664 Left.Previous->is(TT_FatArrow)) {
3665 // JS arrow function (=> {...}).
3666 switch (Style.AllowShortLambdasOnASingleLine) {
3667 case FormatStyle::SLS_All:
3668 return false;
3669 case FormatStyle::SLS_None:
3670 return true;
3671 case FormatStyle::SLS_Empty:
3672 return !Left.Children.empty();
3673 case FormatStyle::SLS_Inline:
3674 // allow one-lining inline (e.g. in function call args) and empty arrow
3675 // functions.
3676 return (Left.NestingLevel == 0 && Line.Level == 0) &&
3677 !Left.Children.empty();
3678 }
3679 llvm_unreachable("Unknown FormatStyle::ShortLambdaStyle enum");
3680 }
3681
3682 if (Right.is(tok::r_brace) && Left.is(tok::l_brace) &&
3683 !Left.Children.empty())
3684 // Support AllowShortFunctionsOnASingleLine for JavaScript.
3685 return Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_None ||
3686 Style.AllowShortFunctionsOnASingleLine == FormatStyle::SFS_Empty ||
3687 (Left.NestingLevel == 0 && Line.Level == 0 &&
3688 Style.AllowShortFunctionsOnASingleLine &
3689 FormatStyle::SFS_InlineOnly);
3690 } else if (Style.Language == FormatStyle::LK_Java) {
3691 if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
3692 Right.Next->is(tok::string_literal))
3693 return true;
3694 } else if (Style.Language == FormatStyle::LK_Cpp ||
3695 Style.Language == FormatStyle::LK_ObjC ||
3696 Style.Language == FormatStyle::LK_Proto ||
3697 Style.Language == FormatStyle::LK_TableGen ||
3698 Style.Language == FormatStyle::LK_TextProto) {
3699 if (Left.isStringLiteral() && Right.isStringLiteral())
3700 return true;
3701 }
3702
3703 // Basic JSON newline processing.
3704 if (Style.isJson()) {
3705 // Always break after a JSON record opener.
3706 // {
3707 // }
3708 if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace))
3709 return true;
3710 // Always break after a JSON array opener.
3711 // [
3712 // ]
3713 if (Left.is(TT_ArrayInitializerLSquare) && Left.is(tok::l_square) &&
3714 !Right.is(tok::r_square))
3715 return true;
3716 // Always break afer successive entries.
3717 // 1,
3718 // 2
3719 if (Left.is(tok::comma))
3720 return true;
3721 }
3722
3723 // If the last token before a '}', ']', or ')' is a comma or a trailing
3724 // comment, the intention is to insert a line break after it in order to make
3725 // shuffling around entries easier. Import statements, especially in
3726 // JavaScript, can be an exception to this rule.
3727 if (Style.JavaScriptWrapImports || Line.Type != LT_ImportStatement) {
3728 const FormatToken *BeforeClosingBrace = nullptr;
3729 if ((Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
3730 (Style.Language == FormatStyle::LK_JavaScript &&
3731 Left.is(tok::l_paren))) &&
3732 Left.isNot(BK_Block) && Left.MatchingParen)
3733 BeforeClosingBrace = Left.MatchingParen->Previous;
3734 else if (Right.MatchingParen &&
3735 (Right.MatchingParen->isOneOf(tok::l_brace,
3736 TT_ArrayInitializerLSquare) ||
3737 (Style.Language == FormatStyle::LK_JavaScript &&
3738 Right.MatchingParen->is(tok::l_paren))))
3739 BeforeClosingBrace = &Left;
3740 if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
3741 BeforeClosingBrace->isTrailingComment()))
3742 return true;
3743 }
3744
3745 if (Right.is(tok::comment))
3746 return Left.isNot(BK_BracedInit) && Left.isNot(TT_CtorInitializerColon) &&
3747 (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
3748 if (Left.isTrailingComment())
3749 return true;
3750 if (Right.Previous->IsUnterminatedLiteral)
3751 return true;
3752 if (Right.is(tok::lessless) && Right.Next &&
3753 Right.Previous->is(tok::string_literal) &&
3754 Right.Next->is(tok::string_literal))
3755 return true;
3756 // Can break after template<> declaration
3757 if (Right.Previous->ClosesTemplateDeclaration &&
3758 Right.Previous->MatchingParen &&
3759 Right.Previous->MatchingParen->NestingLevel == 0) {
3760 // Put concepts on the next line e.g.
3761 // template<typename T>
3762 // concept ...
3763 if (Right.is(tok::kw_concept))
3764 return Style.BreakBeforeConceptDeclarations;
3765 return (Style.AlwaysBreakTemplateDeclarations == FormatStyle::BTDS_Yes);
3766 }
3767 if (Right.is(TT_CtorInitializerComma) &&
3768 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma &&
3769 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
3770 return true;
3771 if (Right.is(TT_CtorInitializerColon) &&
3772 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma &&
3773 !Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
3774 return true;
3775 // Break only if we have multiple inheritance.
3776 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
3777 Right.is(TT_InheritanceComma))
3778 return true;
3779 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterComma &&
3780 Left.is(TT_InheritanceComma))
3781 return true;
3782 if (Right.is(tok::string_literal) && Right.TokenText.startswith("R\""))
3783 // Multiline raw string literals are special wrt. line breaks. The author
3784 // has made a deliberate choice and might have aligned the contents of the
3785 // string literal accordingly. Thus, we try keep existing line breaks.
3786 return Right.IsMultiline && Right.NewlinesBefore > 0;
3787 if ((Right.Previous->is(tok::l_brace) ||
3788 (Right.Previous->is(tok::less) && Right.Previous->Previous &&
3789 Right.Previous->Previous->is(tok::equal))) &&
3790 Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) {
3791 // Don't put enums or option definitions onto single lines in protocol
3792 // buffers.
3793 return true;
3794 }
3795 if (Right.is(TT_InlineASMBrace))
3796 return Right.HasUnescapedNewline;
3797
3798 if (isAllmanBrace(Left) || isAllmanBrace(Right)) {
3799 auto FirstNonComment = getFirstNonComment(Line);
3800 bool AccessSpecifier =
3801 FirstNonComment &&
3802 FirstNonComment->isOneOf(Keywords.kw_internal, tok::kw_public,
3803 tok::kw_private, tok::kw_protected);
3804
3805 if (Style.BraceWrapping.AfterEnum) {
3806 if (Line.startsWith(tok::kw_enum) ||
3807 Line.startsWith(tok::kw_typedef, tok::kw_enum))
3808 return true;
3809 // Ensure BraceWrapping for `public enum A {`.
3810 if (AccessSpecifier && FirstNonComment->Next &&
3811 FirstNonComment->Next->is(tok::kw_enum))
3812 return true;
3813 }
3814
3815 // Ensure BraceWrapping for `public interface A {`.
3816 if (Style.BraceWrapping.AfterClass &&
3817 ((AccessSpecifier && FirstNonComment->Next &&
3818 FirstNonComment->Next->is(Keywords.kw_interface)) ||
3819 Line.startsWith(Keywords.kw_interface)))
3820 return true;
3821
3822 return (Line.startsWith(tok::kw_class) && Style.BraceWrapping.AfterClass) ||
3823 (Line.startsWith(tok::kw_struct) && Style.BraceWrapping.AfterStruct);
3824 }
3825
3826 if (Left.is(TT_ObjCBlockLBrace) &&
3827 Style.AllowShortBlocksOnASingleLine == FormatStyle::SBS_Never)
3828 return true;
3829
3830 if (Left.is(TT_LambdaLBrace)) {
3831 if (IsFunctionArgument(Left) &&
3832 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline)
3833 return false;
3834
3835 if (Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_None ||
3836 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Inline ||
3837 (!Left.Children.empty() &&
3838 Style.AllowShortLambdasOnASingleLine == FormatStyle::SLS_Empty))
3839 return true;
3840 }
3841
3842 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace) &&
3843 Left.isOneOf(tok::star, tok::amp, tok::ampamp, TT_TemplateCloser)) {
3844 return true;
3845 }
3846
3847 // Put multiple Java annotation on a new line.
3848 if ((Style.Language == FormatStyle::LK_Java ||
3849 Style.Language == FormatStyle::LK_JavaScript) &&
3850 Left.is(TT_LeadingJavaAnnotation) &&
3851 Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
3852 (Line.Last->is(tok::l_brace) || Style.BreakAfterJavaFieldAnnotations))
3853 return true;
3854
3855 if (Right.is(TT_ProtoExtensionLSquare))
3856 return true;
3857
3858 // In text proto instances if a submessage contains at least 2 entries and at
3859 // least one of them is a submessage, like A { ... B { ... } ... },
3860 // put all of the entries of A on separate lines by forcing the selector of
3861 // the submessage B to be put on a newline.
3862 //
3863 // Example: these can stay on one line:
3864 // a { scalar_1: 1 scalar_2: 2 }
3865 // a { b { key: value } }
3866 //
3867 // and these entries need to be on a new line even if putting them all in one
3868 // line is under the column limit:
3869 // a {
3870 // scalar: 1
3871 // b { key: value }
3872 // }
3873 //
3874 // We enforce this by breaking before a submessage field that has previous
3875 // siblings, *and* breaking before a field that follows a submessage field.
3876 //
3877 // Be careful to exclude the case [proto.ext] { ... } since the `]` is
3878 // the TT_SelectorName there, but we don't want to break inside the brackets.
3879 //
3880 // Another edge case is @submessage { key: value }, which is a common
3881 // substitution placeholder. In this case we want to keep `@` and `submessage`
3882 // together.
3883 //
3884 // We ensure elsewhere that extensions are always on their own line.
3885 if ((Style.Language == FormatStyle::LK_Proto ||
3886 Style.Language == FormatStyle::LK_TextProto) &&
3887 Right.is(TT_SelectorName) && !Right.is(tok::r_square) && Right.Next) {
3888 // Keep `@submessage` together in:
3889 // @submessage { key: value }
3890 if (Right.Previous && Right.Previous->is(tok::at))
3891 return false;
3892 // Look for the scope opener after selector in cases like:
3893 // selector { ...
3894 // selector: { ...
3895 // selector: @base { ...
3896 FormatToken *LBrace = Right.Next;
3897 if (LBrace && LBrace->is(tok::colon)) {
3898 LBrace = LBrace->Next;
3899 if (LBrace && LBrace->is(tok::at)) {
3900 LBrace = LBrace->Next;
3901 if (LBrace)
3902 LBrace = LBrace->Next;
3903 }
3904 }
3905 if (LBrace &&
3906 // The scope opener is one of {, [, <:
3907 // selector { ... }
3908 // selector [ ... ]
3909 // selector < ... >
3910 //
3911 // In case of selector { ... }, the l_brace is TT_DictLiteral.
3912 // In case of an empty selector {}, the l_brace is not TT_DictLiteral,
3913 // so we check for immediately following r_brace.
3914 ((LBrace->is(tok::l_brace) &&
3915 (LBrace->is(TT_DictLiteral) ||
3916 (LBrace->Next && LBrace->Next->is(tok::r_brace)))) ||
3917 LBrace->is(TT_ArrayInitializerLSquare) || LBrace->is(tok::less))) {
3918 // If Left.ParameterCount is 0, then this submessage entry is not the
3919 // first in its parent submessage, and we want to break before this entry.
3920 // If Left.ParameterCount is greater than 0, then its parent submessage
3921 // might contain 1 or more entries and we want to break before this entry
3922 // if it contains at least 2 entries. We deal with this case later by
3923 // detecting and breaking before the next entry in the parent submessage.
3924 if (Left.ParameterCount == 0)
3925 return true;
3926 // However, if this submessage is the first entry in its parent
3927 // submessage, Left.ParameterCount might be 1 in some cases.
3928 // We deal with this case later by detecting an entry
3929 // following a closing paren of this submessage.
3930 }
3931
3932 // If this is an entry immediately following a submessage, it will be
3933 // preceded by a closing paren of that submessage, like in:
3934 // left---. .---right
3935 // v v
3936 // sub: { ... } key: value
3937 // If there was a comment between `}` an `key` above, then `key` would be
3938 // put on a new line anyways.
3939 if (Left.isOneOf(tok::r_brace, tok::greater, tok::r_square))
3940 return true;
3941 }
3942
3943 // Deal with lambda arguments in C++ - we want consistent line breaks whether
3944 // they happen to be at arg0, arg1 or argN. The selection is a bit nuanced
3945 // as aggressive line breaks are placed when the lambda is not the last arg.
3946 if ((Style.Language == FormatStyle::LK_Cpp ||
3947 Style.Language == FormatStyle::LK_ObjC) &&
3948 Left.is(tok::l_paren) && Left.BlockParameterCount > 0 &&
3949 !Right.isOneOf(tok::l_paren, TT_LambdaLSquare)) {
3950 // Multiple lambdas in the same function call force line breaks.
3951 if (Left.BlockParameterCount > 1)
3952 return true;
3953
3954 // A lambda followed by another arg forces a line break.
3955 if (!Left.Role)
3956 return false;
3957 auto Comma = Left.Role->lastComma();
3958 if (!Comma)
3959 return false;
3960 auto Next = Comma->getNextNonComment();
3961 if (!Next)
3962 return false;
3963 if (!Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret))
3964 return true;
3965 }
3966
3967 return false;
3968 }
3969
canBreakBefore(const AnnotatedLine & Line,const FormatToken & Right)3970 bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
3971 const FormatToken &Right) {
3972 const FormatToken &Left = *Right.Previous;
3973 // Language-specific stuff.
3974 if (Style.isCSharp()) {
3975 if (Left.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon) ||
3976 Right.isOneOf(TT_CSharpNamedArgumentColon, TT_AttributeColon))
3977 return false;
3978 // Only break after commas for generic type constraints.
3979 if (Line.First->is(TT_CSharpGenericTypeConstraint))
3980 return Left.is(TT_CSharpGenericTypeConstraintComma);
3981 // Keep nullable operators attached to their identifiers.
3982 if (Right.is(TT_CSharpNullable)) {
3983 return false;
3984 }
3985 } else if (Style.Language == FormatStyle::LK_Java) {
3986 if (Left.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
3987 Keywords.kw_implements))
3988 return false;
3989 if (Right.isOneOf(Keywords.kw_throws, Keywords.kw_extends,
3990 Keywords.kw_implements))
3991 return true;
3992 } else if (Style.Language == FormatStyle::LK_JavaScript) {
3993 const FormatToken *NonComment = Right.getPreviousNonComment();
3994 if (NonComment &&
3995 NonComment->isOneOf(
3996 tok::kw_return, Keywords.kw_yield, tok::kw_continue, tok::kw_break,
3997 tok::kw_throw, Keywords.kw_interface, Keywords.kw_type,
3998 tok::kw_static, tok::kw_public, tok::kw_private, tok::kw_protected,
3999 Keywords.kw_readonly, Keywords.kw_abstract, Keywords.kw_get,
4000 Keywords.kw_set, Keywords.kw_async, Keywords.kw_await))
4001 return false; // Otherwise automatic semicolon insertion would trigger.
4002 if (Right.NestingLevel == 0 &&
4003 (Left.Tok.getIdentifierInfo() ||
4004 Left.isOneOf(tok::r_square, tok::r_paren)) &&
4005 Right.isOneOf(tok::l_square, tok::l_paren))
4006 return false; // Otherwise automatic semicolon insertion would trigger.
4007 if (NonComment && NonComment->is(tok::identifier) &&
4008 NonComment->TokenText == "asserts")
4009 return false;
4010 if (Left.is(TT_FatArrow) && Right.is(tok::l_brace))
4011 return false;
4012 if (Left.is(TT_JsTypeColon))
4013 return true;
4014 // Don't wrap between ":" and "!" of a strict prop init ("field!: type;").
4015 if (Left.is(tok::exclaim) && Right.is(tok::colon))
4016 return false;
4017 // Look for is type annotations like:
4018 // function f(): a is B { ... }
4019 // Do not break before is in these cases.
4020 if (Right.is(Keywords.kw_is)) {
4021 const FormatToken *Next = Right.getNextNonComment();
4022 // If `is` is followed by a colon, it's likely that it's a dict key, so
4023 // ignore it for this check.
4024 // For example this is common in Polymer:
4025 // Polymer({
4026 // is: 'name',
4027 // ...
4028 // });
4029 if (!Next || !Next->is(tok::colon))
4030 return false;
4031 }
4032 if (Left.is(Keywords.kw_in))
4033 return Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None;
4034 if (Right.is(Keywords.kw_in))
4035 return Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
4036 if (Right.is(Keywords.kw_as))
4037 return false; // must not break before as in 'x as type' casts
4038 if (Right.isOneOf(Keywords.kw_extends, Keywords.kw_infer)) {
4039 // extends and infer can appear as keywords in conditional types:
4040 // https://www.typescriptlang.org/docs/handbook/release-notes/typescript-2-8.html#conditional-types
4041 // do not break before them, as the expressions are subject to ASI.
4042 return false;
4043 }
4044 if (Left.is(Keywords.kw_as))
4045 return true;
4046 if (Left.is(TT_NonNullAssertion))
4047 return true;
4048 if (Left.is(Keywords.kw_declare) &&
4049 Right.isOneOf(Keywords.kw_module, tok::kw_namespace,
4050 Keywords.kw_function, tok::kw_class, tok::kw_enum,
4051 Keywords.kw_interface, Keywords.kw_type, Keywords.kw_var,
4052 Keywords.kw_let, tok::kw_const))
4053 // See grammar for 'declare' statements at:
4054 // https://github.com/Microsoft/TypeScript/blob/master/doc/spec.md#A.10
4055 return false;
4056 if (Left.isOneOf(Keywords.kw_module, tok::kw_namespace) &&
4057 Right.isOneOf(tok::identifier, tok::string_literal))
4058 return false; // must not break in "module foo { ...}"
4059 if (Right.is(TT_TemplateString) && Right.closesScope())
4060 return false;
4061 // Don't split tagged template literal so there is a break between the tag
4062 // identifier and template string.
4063 if (Left.is(tok::identifier) && Right.is(TT_TemplateString)) {
4064 return false;
4065 }
4066 if (Left.is(TT_TemplateString) && Left.opensScope())
4067 return true;
4068 }
4069
4070 if (Left.is(tok::at))
4071 return false;
4072 if (Left.Tok.getObjCKeywordID() == tok::objc_interface)
4073 return false;
4074 if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
4075 return !Right.is(tok::l_paren);
4076 if (Right.is(TT_PointerOrReference))
4077 return Line.IsMultiVariableDeclStmt ||
4078 (getTokenPointerOrReferenceAlignment(Right) ==
4079 FormatStyle::PAS_Right &&
4080 (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
4081 if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
4082 Right.is(tok::kw_operator))
4083 return true;
4084 if (Left.is(TT_PointerOrReference))
4085 return false;
4086 if (Right.isTrailingComment())
4087 // We rely on MustBreakBefore being set correctly here as we should not
4088 // change the "binding" behavior of a comment.
4089 // The first comment in a braced lists is always interpreted as belonging to
4090 // the first list element. Otherwise, it should be placed outside of the
4091 // list.
4092 return Left.is(BK_BracedInit) ||
4093 (Left.is(TT_CtorInitializerColon) &&
4094 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon);
4095 if (Left.is(tok::question) && Right.is(tok::colon))
4096 return false;
4097 if (Right.is(TT_ConditionalExpr) || Right.is(tok::question))
4098 return Style.BreakBeforeTernaryOperators;
4099 if (Left.is(TT_ConditionalExpr) || Left.is(tok::question))
4100 return !Style.BreakBeforeTernaryOperators;
4101 if (Left.is(TT_InheritanceColon))
4102 return Style.BreakInheritanceList == FormatStyle::BILS_AfterColon;
4103 if (Right.is(TT_InheritanceColon))
4104 return Style.BreakInheritanceList != FormatStyle::BILS_AfterColon;
4105 if (Right.is(TT_ObjCMethodExpr) && !Right.is(tok::r_square) &&
4106 Left.isNot(TT_SelectorName))
4107 return true;
4108
4109 if (Right.is(tok::colon) &&
4110 !Right.isOneOf(TT_CtorInitializerColon, TT_InlineASMColon))
4111 return false;
4112 if (Left.is(tok::colon) && Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) {
4113 if (Style.Language == FormatStyle::LK_Proto ||
4114 Style.Language == FormatStyle::LK_TextProto) {
4115 if (!Style.AlwaysBreakBeforeMultilineStrings && Right.isStringLiteral())
4116 return false;
4117 // Prevent cases like:
4118 //
4119 // submessage:
4120 // { key: valueeeeeeeeeeee }
4121 //
4122 // when the snippet does not fit into one line.
4123 // Prefer:
4124 //
4125 // submessage: {
4126 // key: valueeeeeeeeeeee
4127 // }
4128 //
4129 // instead, even if it is longer by one line.
4130 //
4131 // Note that this allows allows the "{" to go over the column limit
4132 // when the column limit is just between ":" and "{", but that does
4133 // not happen too often and alternative formattings in this case are
4134 // not much better.
4135 //
4136 // The code covers the cases:
4137 //
4138 // submessage: { ... }
4139 // submessage: < ... >
4140 // repeated: [ ... ]
4141 if (((Right.is(tok::l_brace) || Right.is(tok::less)) &&
4142 Right.is(TT_DictLiteral)) ||
4143 Right.is(TT_ArrayInitializerLSquare))
4144 return false;
4145 }
4146 return true;
4147 }
4148 if (Right.is(tok::r_square) && Right.MatchingParen &&
4149 Right.MatchingParen->is(TT_ProtoExtensionLSquare))
4150 return false;
4151 if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
4152 Right.Next->is(TT_ObjCMethodExpr)))
4153 return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
4154 if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
4155 return true;
4156 if (Left.ClosesTemplateDeclaration || Left.is(TT_FunctionAnnotationRParen))
4157 return true;
4158 if (Right.isOneOf(TT_RangeBasedForLoopColon, TT_OverloadedOperatorLParen,
4159 TT_OverloadedOperator))
4160 return false;
4161 if (Left.is(TT_RangeBasedForLoopColon))
4162 return true;
4163 if (Right.is(TT_RangeBasedForLoopColon))
4164 return false;
4165 if (Left.is(TT_TemplateCloser) && Right.is(TT_TemplateOpener))
4166 return true;
4167 if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
4168 Left.is(tok::kw_operator))
4169 return false;
4170 if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
4171 Line.Type == LT_VirtualFunctionDecl && Left.NestingLevel == 0)
4172 return false;
4173 if (Left.is(tok::equal) && Right.is(tok::l_brace) &&
4174 !Style.Cpp11BracedListStyle)
4175 return false;
4176 if (Left.is(tok::l_paren) &&
4177 Left.isOneOf(TT_AttributeParen, TT_TypeDeclarationParen))
4178 return false;
4179 if (Left.is(tok::l_paren) && Left.Previous &&
4180 (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
4181 return false;
4182 if (Right.is(TT_ImplicitStringLiteral))
4183 return false;
4184
4185 if (Right.is(tok::r_paren) || Right.is(TT_TemplateCloser))
4186 return false;
4187 if (Right.is(tok::r_square) && Right.MatchingParen &&
4188 Right.MatchingParen->is(TT_LambdaLSquare))
4189 return false;
4190
4191 // We only break before r_brace if there was a corresponding break before
4192 // the l_brace, which is tracked by BreakBeforeClosingBrace.
4193 if (Right.is(tok::r_brace))
4194 return Right.MatchingParen && Right.MatchingParen->is(BK_Block);
4195
4196 // Allow breaking after a trailing annotation, e.g. after a method
4197 // declaration.
4198 if (Left.is(TT_TrailingAnnotation))
4199 return !Right.isOneOf(tok::l_brace, tok::semi, tok::equal, tok::l_paren,
4200 tok::less, tok::coloncolon);
4201
4202 if (Right.is(tok::kw___attribute) ||
4203 (Right.is(tok::l_square) && Right.is(TT_AttributeSquare)))
4204 return !Left.is(TT_AttributeSquare);
4205
4206 if (Left.is(tok::identifier) && Right.is(tok::string_literal))
4207 return true;
4208
4209 if (Right.is(tok::identifier) && Right.Next && Right.Next->is(TT_DictLiteral))
4210 return true;
4211
4212 if (Left.is(TT_CtorInitializerColon))
4213 return Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
4214 if (Right.is(TT_CtorInitializerColon))
4215 return Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon;
4216 if (Left.is(TT_CtorInitializerComma) &&
4217 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma)
4218 return false;
4219 if (Right.is(TT_CtorInitializerComma) &&
4220 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma)
4221 return true;
4222 if (Left.is(TT_InheritanceComma) &&
4223 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma)
4224 return false;
4225 if (Right.is(TT_InheritanceComma) &&
4226 Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma)
4227 return true;
4228 if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
4229 (Left.is(tok::less) && Right.is(tok::less)))
4230 return false;
4231 if (Right.is(TT_BinaryOperator) &&
4232 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
4233 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_All ||
4234 Right.getPrecedence() != prec::Assignment))
4235 return true;
4236 if (Left.is(TT_ArrayInitializerLSquare))
4237 return true;
4238 if (Right.is(tok::kw_typename) && Left.isNot(tok::kw_const))
4239 return true;
4240 if ((Left.isBinaryOperator() || Left.is(TT_BinaryOperator)) &&
4241 !Left.isOneOf(tok::arrowstar, tok::lessless) &&
4242 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_All &&
4243 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None ||
4244 Left.getPrecedence() == prec::Assignment))
4245 return true;
4246 if ((Left.is(TT_AttributeSquare) && Right.is(tok::l_square)) ||
4247 (Left.is(tok::r_square) && Right.is(TT_AttributeSquare)))
4248 return false;
4249
4250 auto ShortLambdaOption = Style.AllowShortLambdasOnASingleLine;
4251 if (Style.BraceWrapping.BeforeLambdaBody && Right.is(TT_LambdaLBrace)) {
4252 if (isAllmanLambdaBrace(Left))
4253 return !isItAnEmptyLambdaAllowed(Left, ShortLambdaOption);
4254 if (isAllmanLambdaBrace(Right))
4255 return !isItAnEmptyLambdaAllowed(Right, ShortLambdaOption);
4256 }
4257
4258 return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
4259 tok::kw_class, tok::kw_struct, tok::comment) ||
4260 Right.isMemberAccess() ||
4261 Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
4262 tok::colon, tok::l_square, tok::at) ||
4263 (Left.is(tok::r_paren) &&
4264 Right.isOneOf(tok::identifier, tok::kw_const)) ||
4265 (Left.is(tok::l_paren) && !Right.is(tok::r_paren)) ||
4266 (Left.is(TT_TemplateOpener) && !Right.is(TT_TemplateCloser));
4267 }
4268
printDebugInfo(const AnnotatedLine & Line)4269 void TokenAnnotator::printDebugInfo(const AnnotatedLine &Line) {
4270 llvm::errs() << "AnnotatedTokens(L=" << Line.Level << "):\n";
4271 const FormatToken *Tok = Line.First;
4272 while (Tok) {
4273 llvm::errs() << " M=" << Tok->MustBreakBefore
4274 << " C=" << Tok->CanBreakBefore
4275 << " T=" << getTokenTypeName(Tok->getType())
4276 << " S=" << Tok->SpacesRequiredBefore
4277 << " F=" << Tok->Finalized << " B=" << Tok->BlockParameterCount
4278 << " BK=" << Tok->getBlockKind() << " P=" << Tok->SplitPenalty
4279 << " Name=" << Tok->Tok.getName() << " L=" << Tok->TotalLength
4280 << " PPK=" << Tok->getPackingKind() << " FakeLParens=";
4281 for (unsigned i = 0, e = Tok->FakeLParens.size(); i != e; ++i)
4282 llvm::errs() << Tok->FakeLParens[i] << "/";
4283 llvm::errs() << " FakeRParens=" << Tok->FakeRParens;
4284 llvm::errs() << " II=" << Tok->Tok.getIdentifierInfo();
4285 llvm::errs() << " Text='" << Tok->TokenText << "'\n";
4286 if (!Tok->Next)
4287 assert(Tok == Line.Last);
4288 Tok = Tok->Next;
4289 }
4290 llvm::errs() << "----\n";
4291 }
4292
4293 FormatStyle::PointerAlignmentStyle
getTokenReferenceAlignment(const FormatToken & Reference)4294 TokenAnnotator::getTokenReferenceAlignment(const FormatToken &Reference) {
4295 assert(Reference.isOneOf(tok::amp, tok::ampamp));
4296 switch (Style.ReferenceAlignment) {
4297 case FormatStyle::RAS_Pointer:
4298 return Style.PointerAlignment;
4299 case FormatStyle::RAS_Left:
4300 return FormatStyle::PAS_Left;
4301 case FormatStyle::RAS_Right:
4302 return FormatStyle::PAS_Right;
4303 case FormatStyle::RAS_Middle:
4304 return FormatStyle::PAS_Middle;
4305 }
4306 assert(0); //"Unhandled value of ReferenceAlignment"
4307 return Style.PointerAlignment;
4308 }
4309
4310 FormatStyle::PointerAlignmentStyle
getTokenPointerOrReferenceAlignment(const FormatToken & PointerOrReference)4311 TokenAnnotator::getTokenPointerOrReferenceAlignment(
4312 const FormatToken &PointerOrReference) {
4313 if (PointerOrReference.isOneOf(tok::amp, tok::ampamp)) {
4314 switch (Style.ReferenceAlignment) {
4315 case FormatStyle::RAS_Pointer:
4316 return Style.PointerAlignment;
4317 case FormatStyle::RAS_Left:
4318 return FormatStyle::PAS_Left;
4319 case FormatStyle::RAS_Right:
4320 return FormatStyle::PAS_Right;
4321 case FormatStyle::RAS_Middle:
4322 return FormatStyle::PAS_Middle;
4323 }
4324 }
4325 assert(PointerOrReference.is(tok::star));
4326 return Style.PointerAlignment;
4327 }
4328
4329 } // namespace format
4330 } // namespace clang
4331