1//===--- ContinuationIndenter.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 the continuation indenter.
11///
12//===----------------------------------------------------------------------===//
13
14#include "ContinuationIndenter.h"
15#include "BreakableToken.h"
16#include "FormatInternal.h"
17#include "FormatToken.h"
18#include "WhitespaceManager.h"
19#include "clang/Basic/OperatorPrecedence.h"
20#include "clang/Basic/SourceManager.h"
21#include "clang/Basic/TokenKinds.h"
22#include "clang/Format/Format.h"
23#include "llvm/ADT/StringSet.h"
24#include "llvm/Support/Debug.h"
25#include <optional>
26
27#define DEBUG_TYPE "format-indenter"
28
29namespace clang {
30namespace format {
31
32// Returns true if a TT_SelectorName should be indented when wrapped,
33// false otherwise.
34static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
35 LineType LineType) {
36 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
37}
38
39// Returns true if a binary operator following \p Tok should be unindented when
40// the style permits it.
41static bool shouldUnindentNextOperator(const FormatToken &Tok) {
42 const FormatToken *Previous = Tok.getPreviousNonComment();
43 return Previous && (Previous->getPrecedence() == prec::Assignment ||
44 Previous->isOneOf(K1: tok::kw_return, K2: TT_RequiresClause));
45}
46
47// Returns the length of everything up to the first possible line break after
48// the ), ], } or > matching \c Tok.
49static unsigned getLengthToMatchingParen(const FormatToken &Tok,
50 ArrayRef<ParenState> Stack) {
51 // Normally whether or not a break before T is possible is calculated and
52 // stored in T.CanBreakBefore. Braces, array initializers and text proto
53 // messages like `key: < ... >` are an exception: a break is possible
54 // before a closing brace R if a break was inserted after the corresponding
55 // opening brace. The information about whether or not a break is needed
56 // before a closing brace R is stored in the ParenState field
57 // S.BreakBeforeClosingBrace where S is the state that R closes.
58 //
59 // In order to decide whether there can be a break before encountered right
60 // braces, this implementation iterates over the sequence of tokens and over
61 // the paren stack in lockstep, keeping track of the stack level which visited
62 // right braces correspond to in MatchingStackIndex.
63 //
64 // For example, consider:
65 // L. <- line number
66 // 1. {
67 // 2. {1},
68 // 3. {2},
69 // 4. {{3}}}
70 // ^ where we call this method with this token.
71 // The paren stack at this point contains 3 brace levels:
72 // 0. { at line 1, BreakBeforeClosingBrace: true
73 // 1. first { at line 4, BreakBeforeClosingBrace: false
74 // 2. second { at line 4, BreakBeforeClosingBrace: false,
75 // where there might be fake parens levels in-between these levels.
76 // The algorithm will start at the first } on line 4, which is the matching
77 // brace of the initial left brace and at level 2 of the stack. Then,
78 // examining BreakBeforeClosingBrace: false at level 2, it will continue to
79 // the second } on line 4, and will traverse the stack downwards until it
80 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
81 // false at level 1, it will continue to the third } on line 4 and will
82 // traverse the stack downwards until it finds the matching { on level 0.
83 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
84 // will stop and will use the second } on line 4 to determine the length to
85 // return, as in this example the range will include the tokens: {3}}
86 //
87 // The algorithm will only traverse the stack if it encounters braces, array
88 // initializer squares or text proto angle brackets.
89 if (!Tok.MatchingParen)
90 return 0;
91 FormatToken *End = Tok.MatchingParen;
92 // Maintains a stack level corresponding to the current End token.
93 int MatchingStackIndex = Stack.size() - 1;
94 // Traverses the stack downwards, looking for the level to which LBrace
95 // corresponds. Returns either a pointer to the matching level or nullptr if
96 // LParen is not found in the initial portion of the stack up to
97 // MatchingStackIndex.
98 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
99 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
100 --MatchingStackIndex;
101 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
102 };
103 for (; End->Next; End = End->Next) {
104 if (End->Next->CanBreakBefore)
105 break;
106 if (!End->Next->closesScope())
107 continue;
108 if (End->Next->MatchingParen &&
109 End->Next->MatchingParen->isOneOf(
110 K1: tok::l_brace, K2: TT_ArrayInitializerLSquare, Ks: tok::less)) {
111 const ParenState *State = FindParenState(End->Next->MatchingParen);
112 if (State && State->BreakBeforeClosingBrace)
113 break;
114 }
115 }
116 return End->TotalLength - Tok.TotalLength + 1;
117}
118
119static unsigned getLengthToNextOperator(const FormatToken &Tok) {
120 if (!Tok.NextOperator)
121 return 0;
122 return Tok.NextOperator->TotalLength - Tok.TotalLength;
123}
124
125// Returns \c true if \c Tok is the "." or "->" of a call and starts the next
126// segment of a builder type call.
127static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
128 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
129}
130
131// Returns \c true if \c Token in an alignable binary operator
132static bool isAlignableBinaryOperator(const FormatToken &Token) {
133 // No need to align binary operators that only have two operands.
134 bool HasTwoOperands = Token.OperatorIndex == 0 && !Token.NextOperator;
135 return Token.is(TT: TT_BinaryOperator) && !HasTwoOperands &&
136 Token.getPrecedence() > prec::Conditional &&
137 Token.getPrecedence() < prec::PointerToMember;
138}
139
140// Returns \c true if \c Current starts the next operand in a binary operation.
141static bool startsNextOperand(const FormatToken &Current) {
142 assert(Current.Previous);
143 const auto &Previous = *Current.Previous;
144 return isAlignableBinaryOperator(Token: Previous) && !Current.isTrailingComment();
145}
146
147// Returns the number of operands in the chain containing \c Op.
148// For example, `a && b && c` has 3 operands (and 2 operators).
149static unsigned getChainLength(const FormatToken &Op) {
150 const FormatToken *Last = &Op;
151 while (Last->NextOperator)
152 Last = Last->NextOperator;
153 return Last->OperatorIndex + 2;
154}
155
156// Returns \c true if \c Current is a binary operation that must break.
157static bool mustBreakBinaryOperation(const FormatToken &Current,
158 const FormatStyle &Style) {
159 if (!Current.CanBreakBefore)
160 return false;
161
162 // Determine the operator token: when breaking after the operator,
163 // it is Current.Previous; when breaking before, it is Current itself.
164 bool BreakBefore = Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None;
165 const FormatToken *OpToken = BreakBefore ? &Current : Current.Previous;
166
167 if (!OpToken)
168 return false;
169
170 // Check that this is an alignable binary operator.
171 if (BreakBefore) {
172 if (!isAlignableBinaryOperator(Token: Current))
173 return false;
174 } else if (!startsNextOperand(Current)) {
175 return false;
176 }
177
178 // Look up per-operator rule or fall back to Default.
179 const auto OperatorBreakStyle =
180 Style.BreakBinaryOperations.getStyleForOperator(Kind: OpToken->Tok.getKind());
181 if (OperatorBreakStyle == FormatStyle::BBO_Never)
182 return false;
183
184 // Check MinChainLength: if the chain is too short, don't force a break.
185 const unsigned MinChain =
186 Style.BreakBinaryOperations.getMinChainLengthForOperator(
187 Kind: OpToken->Tok.getKind());
188 return MinChain == 0 || getChainLength(Op: *OpToken) >= MinChain;
189}
190
191static bool opensProtoMessageField(const FormatToken &LessTok,
192 const FormatStyle &Style) {
193 if (LessTok.isNot(Kind: tok::less))
194 return false;
195 return Style.isTextProto() ||
196 (Style.Language == FormatStyle::LK_Proto &&
197 (LessTok.NestingLevel > 0 ||
198 (LessTok.Previous && LessTok.Previous->is(Kind: tok::equal))));
199}
200
201// Returns the delimiter of a raw string literal, or std::nullopt if TokenText
202// is not the text of a raw string literal. The delimiter could be the empty
203// string. For example, the delimiter of R"deli(cont)deli" is deli.
204static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
205 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
206 || !TokenText.starts_with(Prefix: "R\"") || !TokenText.ends_with(Suffix: "\"")) {
207 return std::nullopt;
208 }
209
210 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
211 // size at most 16 by the standard, so the first '(' must be among the first
212 // 19 bytes.
213 size_t LParenPos = TokenText.substr(Start: 0, N: 19).find_first_of(C: '(');
214 if (LParenPos == StringRef::npos)
215 return std::nullopt;
216 StringRef Delimiter = TokenText.substr(Start: 2, N: LParenPos - 2);
217
218 // Check that the string ends in ')Delimiter"'.
219 size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
220 if (TokenText[RParenPos] != ')')
221 return std::nullopt;
222 if (!TokenText.substr(Start: RParenPos + 1).starts_with(Prefix: Delimiter))
223 return std::nullopt;
224 return Delimiter;
225}
226
227// Returns the canonical delimiter for \p Language, or the empty string if no
228// canonical delimiter is specified.
229static StringRef
230getCanonicalRawStringDelimiter(const FormatStyle &Style,
231 FormatStyle::LanguageKind Language) {
232 for (const auto &Format : Style.RawStringFormats)
233 if (Format.Language == Language)
234 return StringRef(Format.CanonicalDelimiter);
235 return "";
236}
237
238RawStringFormatStyleManager::RawStringFormatStyleManager(
239 const FormatStyle &CodeStyle) {
240 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
241 std::optional<FormatStyle> LanguageStyle =
242 CodeStyle.GetLanguageStyle(Language: RawStringFormat.Language);
243 if (!LanguageStyle) {
244 FormatStyle PredefinedStyle;
245 if (!getPredefinedStyle(Name: RawStringFormat.BasedOnStyle,
246 Language: RawStringFormat.Language, Style: &PredefinedStyle)) {
247 PredefinedStyle = getLLVMStyle();
248 PredefinedStyle.Language = RawStringFormat.Language;
249 }
250 LanguageStyle = PredefinedStyle;
251 }
252 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
253 for (StringRef Delimiter : RawStringFormat.Delimiters)
254 DelimiterStyle.insert(KV: {Delimiter, *LanguageStyle});
255 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions)
256 EnclosingFunctionStyle.insert(KV: {EnclosingFunction, *LanguageStyle});
257 }
258}
259
260std::optional<FormatStyle>
261RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
262 auto It = DelimiterStyle.find(Key: Delimiter);
263 if (It == DelimiterStyle.end())
264 return std::nullopt;
265 return It->second;
266}
267
268std::optional<FormatStyle>
269RawStringFormatStyleManager::getEnclosingFunctionStyle(
270 StringRef EnclosingFunction) const {
271 auto It = EnclosingFunctionStyle.find(Key: EnclosingFunction);
272 if (It == EnclosingFunctionStyle.end())
273 return std::nullopt;
274 return It->second;
275}
276
277IndentationAndAlignment
278IndentationAndAlignment::addPadding(unsigned Spaces) const {
279 return IndentationAndAlignment(Total + Spaces, IndentedFrom);
280}
281
282IndentationAndAlignment
283IndentationAndAlignment::operator+(unsigned Spaces) const {
284 return IndentationAndAlignment(Total + Spaces, Total);
285}
286
287IndentationAndAlignment
288IndentationAndAlignment::operator-(unsigned Spaces) const {
289 return IndentationAndAlignment(Total - Spaces, Total);
290}
291
292IndentationAndAlignment &IndentationAndAlignment::operator+=(unsigned Spaces) {
293 *this = *this + Spaces;
294 return *this;
295}
296
297IndentationAndAlignment::IndentationAndAlignment(unsigned Total,
298 unsigned IndentedFrom)
299 : Total(Total), IndentedFrom(IndentedFrom) {}
300
301IndentationAndAlignment::IndentationAndAlignment(unsigned Spaces)
302 : Total(Spaces), IndentedFrom(Spaces) {}
303
304bool IndentationAndAlignment::operator<(
305 const IndentationAndAlignment &Other) const {
306 if (Total != Other.Total)
307 return Total < Other.Total;
308 // The sign to use here was decided arbitrarily. This operator is mostly used
309 // when a line's indentation should be the max of 2 things. Using this sign
310 // here makes the program prefer alignment over continuation indentation. That
311 // is, it makes the alignment step that follows prefer to move the line when
312 // aligning the previous line.
313 return IndentedFrom > Other.IndentedFrom;
314}
315
316ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
317 const AdditionalKeywords &Keywords,
318 const SourceManager &SourceMgr,
319 WhitespaceManager &Whitespaces,
320 encoding::Encoding Encoding,
321 bool BinPackInconclusiveFunctions)
322 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
323 Whitespaces(Whitespaces), Encoding(Encoding),
324 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
325 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
326
327LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
328 unsigned FirstStartColumn,
329 const AnnotatedLine *Line,
330 bool DryRun) {
331 LineState State;
332 State.FirstIndent = FirstIndent;
333 if (FirstStartColumn && Line->First->NewlinesBefore == 0)
334 State.Column = FirstStartColumn;
335 else
336 State.Column = FirstIndent;
337 // With preprocessor directive indentation, the line starts on column 0
338 // since it's indented after the hash, but FirstIndent is set to the
339 // preprocessor indent.
340 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
341 (Line->Type == LT_PreprocessorDirective ||
342 Line->Type == LT_ImportStatement)) {
343 State.Column = 0;
344 }
345 State.Line = Line;
346 State.NextToken = Line->First;
347 State.Stack.push_back(Elt: ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
348 /*AvoidBinPacking=*/false,
349 /*NoLineBreak=*/false));
350 State.NoContinuation = false;
351 State.StartOfStringLiteral = 0;
352 State.NoLineBreak = false;
353 State.StartOfLineLevel = 0;
354 State.LowestLevelOnLine = 0;
355 State.IgnoreStackForComparison = false;
356
357 if (Style.isTextProto()) {
358 // We need this in order to deal with the bin packing of text fields at
359 // global scope.
360 auto &CurrentState = State.Stack.back();
361 CurrentState.AvoidBinPacking = true;
362 CurrentState.BreakBeforeParameter = true;
363 CurrentState.AlignColons = false;
364 }
365
366 // The first token has already been indented and thus consumed.
367 moveStateToNextToken(State, DryRun, /*Newline=*/false);
368 return State;
369}
370
371bool ContinuationIndenter::canBreak(const LineState &State) {
372 const FormatToken &Current = *State.NextToken;
373 const FormatToken &Previous = *Current.Previous;
374 const auto &CurrentState = State.Stack.back();
375 assert(&Previous == Current.Previous);
376 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace &&
377 Current.closesBlockOrBlockTypeList(Style))) {
378 return false;
379 }
380 // The opening "{" of a braced list has to be on the same line as the first
381 // element if it is nested in another braced init list or function call.
382 if (!Current.MustBreakBefore && Previous.is(Kind: tok::l_brace) &&
383 Previous.isNot(Kind: TT_DictLiteral) && Previous.is(BBK: BK_BracedInit) &&
384 Previous.Previous &&
385 Previous.Previous->isOneOf(K1: tok::l_brace, K2: tok::l_paren, Ks: tok::comma)) {
386 return false;
387 }
388 // This prevents breaks like:
389 // ...
390 // SomeParameter, OtherParameter).DoSomething(
391 // ...
392 // As they hide "DoSomething" and are generally bad for readability.
393 if (Previous.opensScope() && Previous.isNot(Kind: tok::l_brace) &&
394 State.LowestLevelOnLine < State.StartOfLineLevel &&
395 State.LowestLevelOnLine < Current.NestingLevel) {
396 return false;
397 }
398 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder)
399 return false;
400
401 // Don't create a 'hanging' indent if there are multiple blocks in a single
402 // statement and we are aligning lambda blocks to their signatures.
403 if (Previous.is(Kind: tok::l_brace) && State.Stack.size() > 1 &&
404 State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
405 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) {
406 return Style.isCpp() &&
407 Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope;
408 }
409
410 // Don't break after very short return types (e.g. "void") as that is often
411 // unexpected.
412 if (Current.is(TT: TT_FunctionDeclarationName)) {
413 if (Style.BreakAfterReturnType == FormatStyle::RTBS_None &&
414 State.Column < 6) {
415 return false;
416 }
417
418 if (Style.BreakAfterReturnType == FormatStyle::RTBS_ExceptShortType) {
419 assert(State.Column >= State.FirstIndent);
420 if (State.Column - State.FirstIndent < 6)
421 return false;
422 }
423 }
424
425 // Don't allow breaking before a closing brace of a block-indented braced list
426 // initializer if there isn't already a break.
427 if (Current.is(Kind: tok::r_brace) && Current.MatchingParen &&
428 Current.isBlockIndentedInitRBrace(Style)) {
429 return CurrentState.BreakBeforeClosingBrace;
430 }
431
432 // Check need to break before the right parens if there was a break after
433 // the left parens, which is tracked by BreakBeforeClosingParen.
434 if ((Style.BreakBeforeCloseBracketFunction ||
435 Style.BreakBeforeCloseBracketIf || Style.BreakBeforeCloseBracketLoop ||
436 Style.BreakBeforeCloseBracketSwitch) &&
437 Current.is(Kind: tok::r_paren)) {
438 return CurrentState.BreakBeforeClosingParen;
439 }
440
441 if (Style.BreakBeforeTemplateCloser && Current.is(TT: TT_TemplateCloser))
442 return CurrentState.BreakBeforeClosingAngle;
443
444 // If binary operators are moved to the next line (including commas for some
445 // styles of constructor initializers), that's always ok.
446 if (Current.isNoneOf(Ks: TT_BinaryOperator, Ks: tok::comma) &&
447 // Allow breaking opening brace of lambdas (when passed as function
448 // arguments) to a new line when BeforeLambdaBody brace wrapping is
449 // enabled.
450 (!Style.BraceWrapping.BeforeLambdaBody ||
451 Current.isNot(Kind: TT_LambdaLBrace)) &&
452 CurrentState.NoLineBreakInOperand) {
453 return false;
454 }
455
456 if (Previous.is(Kind: tok::l_square) && Previous.is(TT: TT_ObjCMethodExpr))
457 return false;
458
459 if (Current.is(TT: TT_ConditionalExpr) && Previous.is(Kind: tok::r_paren) &&
460 Previous.MatchingParen && Previous.MatchingParen->Previous &&
461 Previous.MatchingParen->Previous->MatchingParen &&
462 Previous.MatchingParen->Previous->MatchingParen->is(TT: TT_LambdaLBrace)) {
463 // We have a lambda within a conditional expression, allow breaking here.
464 assert(Previous.MatchingParen->Previous->is(tok::r_brace));
465 return true;
466 }
467
468 return !State.NoLineBreak && !CurrentState.NoLineBreak;
469}
470
471bool ContinuationIndenter::mustBreak(const LineState &State) {
472 const FormatToken &Current = *State.NextToken;
473 const FormatToken &Previous = *Current.Previous;
474 const auto &CurrentState = State.Stack.back();
475 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore &&
476 Current.is(TT: TT_LambdaLBrace) && Previous.isNot(Kind: TT_LineComment)) {
477 auto LambdaBodyLength = getLengthToMatchingParen(Tok: Current, Stack: State.Stack);
478 return LambdaBodyLength > getColumnLimit(State);
479 }
480 if (Current.MustBreakBefore ||
481 (Current.is(TT: TT_InlineASMColon) &&
482 (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always ||
483 (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_OnlyMultiline &&
484 Style.ColumnLimit > 0)))) {
485 return true;
486 }
487 if (CurrentState.BreakBeforeClosingBrace &&
488 (Current.closesBlockOrBlockTypeList(Style) ||
489 (Current.is(Kind: tok::r_brace) && Current.MatchingParen &&
490 Current.isBlockIndentedInitRBrace(Style)))) {
491 return true;
492 }
493 if (CurrentState.BreakBeforeClosingParen && Current.is(Kind: tok::r_paren))
494 return true;
495 if (CurrentState.BreakBeforeClosingAngle && Current.is(TT: TT_TemplateCloser))
496 return true;
497 if (Style.Language == FormatStyle::LK_ObjC &&
498 Style.ObjCBreakBeforeNestedBlockParam &&
499 Current.ObjCSelectorNameParts > 1 &&
500 Current.startsSequence(K1: TT_SelectorName, Tokens: tok::colon, Tokens: tok::caret)) {
501 return true;
502 }
503 // Avoid producing inconsistent states by requiring breaks where they are not
504 // permitted for C# generic type constraints.
505 if (CurrentState.IsCSharpGenericTypeConstraint &&
506 Previous.isNot(Kind: TT_CSharpGenericTypeConstraintComma)) {
507 return false;
508 }
509 if ((startsNextParameter(Current, Style) || Previous.is(Kind: tok::semi) ||
510 (Previous.is(TT: TT_TemplateCloser) && Current.is(TT: TT_StartOfName) &&
511 State.Line->First->isNot(Kind: TT_AttributeLSquare) && Style.isCpp() &&
512 // FIXME: This is a temporary workaround for the case where clang-format
513 // sets BreakBeforeParameter to avoid bin packing and this creates a
514 // completely unnecessary line break after a template type that isn't
515 // line-wrapped.
516 (Previous.NestingLevel == 1 ||
517 Style.BinPackParameters == FormatStyle::BPPS_BinPack)) ||
518 (Style.BreakBeforeTernaryOperators && Current.is(TT: TT_ConditionalExpr) &&
519 Previous.isNot(Kind: tok::question)) ||
520 (!Style.BreakBeforeTernaryOperators &&
521 Previous.is(TT: TT_ConditionalExpr))) &&
522 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
523 Current.isNoneOf(Ks: tok::r_paren, Ks: tok::r_brace)) {
524 return true;
525 }
526 if (CurrentState.IsChainedConditional &&
527 ((Style.BreakBeforeTernaryOperators && Current.is(TT: TT_ConditionalExpr) &&
528 Current.is(Kind: tok::colon)) ||
529 (!Style.BreakBeforeTernaryOperators && Previous.is(TT: TT_ConditionalExpr) &&
530 Previous.is(Kind: tok::colon)))) {
531 return true;
532 }
533 if (((Previous.is(TT: TT_DictLiteral) && Previous.is(Kind: tok::l_brace)) ||
534 (Previous.is(TT: TT_ArrayInitializerLSquare) &&
535 Previous.ParameterCount > 1) ||
536 opensProtoMessageField(LessTok: Previous, Style)) &&
537 Style.ColumnLimit > 0 &&
538 getLengthToMatchingParen(Tok: Previous, Stack: State.Stack) + State.Column - 1 >
539 getColumnLimit(State)) {
540 return true;
541 }
542
543 const FormatToken &BreakConstructorInitializersToken =
544 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
545 ? Previous
546 : Current;
547 if (Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterComma &&
548 BreakConstructorInitializersToken.is(TT: TT_CtorInitializerColon) &&
549 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
550 getColumnLimit(State) ||
551 CurrentState.BreakBeforeParameter) &&
552 ((!Current.isTrailingComment() && Style.ColumnLimit > 0) ||
553 Current.NewlinesBefore > 0)) {
554 return true;
555 }
556
557 if (Current.is(TT: TT_ObjCMethodExpr) && Previous.isNot(Kind: TT_SelectorName) &&
558 State.Line->startsWith(Tokens: TT_ObjCMethodSpecifier)) {
559 return true;
560 }
561 if (Current.is(TT: TT_SelectorName) && Previous.isNot(Kind: tok::at) &&
562 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
563 (Style.ObjCBreakBeforeNestedBlockParam ||
564 !Current.startsSequence(K1: TT_SelectorName, Tokens: tok::colon, Tokens: tok::caret))) {
565 return true;
566 }
567
568 unsigned NewLineColumn = getNewLineColumn(State).Total;
569 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
570 State.Column + getLengthToNextOperator(Tok: Current) > Style.ColumnLimit &&
571 (State.Column > NewLineColumn ||
572 Current.NestingLevel < State.StartOfLineLevel)) {
573 return true;
574 }
575
576 if (startsSegmentOfBuilderTypeCall(Tok: Current) &&
577 (CurrentState.CallContinuation != 0 ||
578 CurrentState.BreakBeforeParameter) &&
579 // JavaScript is treated different here as there is a frequent pattern:
580 // SomeFunction(function() {
581 // ...
582 // }.bind(...));
583 // FIXME: We should find a more generic solution to this problem.
584 !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
585 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
586 return true;
587 }
588
589 // If the template declaration spans multiple lines, force wrap before the
590 // function/class declaration.
591 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
592 Current.CanBreakBefore) {
593 return true;
594 }
595
596 if (State.Line->First->isNot(Kind: tok::kw_enum) && State.Column <= NewLineColumn)
597 return false;
598
599 if (Style.AlwaysBreakBeforeMultilineStrings &&
600 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
601 Previous.is(Kind: tok::comma) || Current.NestingLevel < 2) &&
602 Previous.isNoneOf(Ks: tok::kw_return, Ks: tok::lessless, Ks: tok::at,
603 Ks: Keywords.kw_dollar) &&
604 Previous.isNoneOf(Ks: TT_InlineASMColon, Ks: TT_ConditionalExpr) &&
605 nextIsMultilineString(State)) {
606 return true;
607 }
608
609 // Using CanBreakBefore here and below takes care of the decision whether the
610 // current style uses wrapping before or after operators for the given
611 // operator.
612 if (Previous.is(TT: TT_BinaryOperator) && Current.CanBreakBefore) {
613 const auto PreviousPrecedence = Previous.getPrecedence();
614 if (PreviousPrecedence != prec::Assignment &&
615 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
616 const bool LHSIsBinaryExpr =
617 Previous.Previous && Previous.Previous->EndsBinaryExpression;
618 if (LHSIsBinaryExpr)
619 return true;
620 // If we need to break somewhere inside the LHS of a binary expression, we
621 // should also break after the operator. Otherwise, the formatting would
622 // hide the operator precedence, e.g. in:
623 // if (aaaaaaaaaaaaaa ==
624 // bbbbbbbbbbbbbb && c) {..
625 // For comparisons, we only apply this rule, if the LHS is a binary
626 // expression itself as otherwise, the line breaks seem superfluous.
627 // We need special cases for ">>" which we have split into two ">" while
628 // lexing in order to make template parsing easier.
629 const bool IsComparison =
630 (PreviousPrecedence == prec::Relational ||
631 PreviousPrecedence == prec::Equality ||
632 PreviousPrecedence == prec::Spaceship) &&
633 Previous.Previous &&
634 Previous.Previous->isNot(Kind: TT_BinaryOperator); // For >>.
635 if (!IsComparison)
636 return true;
637 }
638 } else if (Current.is(TT: TT_BinaryOperator) && Current.CanBreakBefore &&
639 Current.getPrecedence() != prec::Assignment &&
640 CurrentState.BreakBeforeParameter) {
641 return true;
642 }
643
644 // Same as above, but for the first "<<" operator.
645 if (Current.is(Kind: tok::lessless) && Current.isNot(Kind: TT_OverloadedOperator) &&
646 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
647 return true;
648 }
649
650 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
651 // Always break after "template <...>"(*) and leading annotations. This is
652 // only for cases where the entire line does not fit on a single line as a
653 // different LineFormatter would be used otherwise.
654 // *: Except when another option interferes with that, like concepts.
655 if (Previous.ClosesTemplateDeclaration) {
656 if (Current.is(Kind: tok::kw_concept)) {
657 switch (Style.BreakBeforeConceptDeclarations) {
658 case FormatStyle::BBCDS_Allowed:
659 break;
660 case FormatStyle::BBCDS_Always:
661 return true;
662 case FormatStyle::BBCDS_Never:
663 return false;
664 }
665 }
666 if (Current.is(TT: TT_RequiresClause)) {
667 switch (Style.RequiresClausePosition) {
668 case FormatStyle::RCPS_SingleLine:
669 case FormatStyle::RCPS_WithPreceding:
670 return false;
671 default:
672 return true;
673 }
674 }
675 return Style.BreakTemplateDeclarations != FormatStyle::BTDS_No &&
676 (Style.BreakTemplateDeclarations != FormatStyle::BTDS_Leave ||
677 Current.NewlinesBefore > 0);
678 }
679 if (Previous.is(TT: TT_FunctionAnnotationRParen) &&
680 State.Line->Type != LT_PreprocessorDirective) {
681 return true;
682 }
683 if (Previous.is(TT: TT_LeadingJavaAnnotation) && Current.isNot(Kind: tok::l_paren) &&
684 Current.isNot(Kind: TT_LeadingJavaAnnotation)) {
685 return true;
686 }
687 }
688
689 if (Style.isJavaScript() && Previous.is(Kind: tok::r_paren) &&
690 Previous.is(TT: TT_JavaAnnotation)) {
691 // Break after the closing parenthesis of TypeScript decorators before
692 // functions, getters and setters.
693 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
694 "function"};
695 if (BreakBeforeDecoratedTokens.contains(key: Current.TokenText))
696 return true;
697 }
698
699 if (Current.is(TT: TT_FunctionDeclarationName) &&
700 !State.Line->ReturnTypeWrapped &&
701 // Don't break before a C# function when no break after return type.
702 (!Style.isCSharp() ||
703 Style.BreakAfterReturnType > FormatStyle::RTBS_ExceptShortType) &&
704 // Don't always break between a JavaScript `function` and the function
705 // name.
706 !Style.isJavaScript() && Previous.isNot(Kind: tok::kw_template) &&
707 CurrentState.BreakBeforeParameter) {
708 for (const auto *Tok = &Previous; Tok; Tok = Tok->Previous) {
709 if (Tok->is(TT: TT_LineComment))
710 return false;
711 if (Tok->is(TT: TT_TemplateCloser)) {
712 Tok = Tok->MatchingParen;
713 if (!Tok)
714 return false;
715 }
716 if (Tok->FirstAfterPPLine)
717 return false;
718 }
719
720 return true;
721 }
722
723 // The following could be precomputed as they do not depend on the state.
724 // However, as they should take effect only if the UnwrappedLine does not fit
725 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
726 if (Style.ColumnLimit != 0 && Previous.is(BBK: BK_Block) &&
727 Previous.is(Kind: tok::l_brace) &&
728 Current.isNoneOf(Ks: tok::r_brace, Ks: tok::comment)) {
729 return true;
730 }
731
732 if (Current.is(Kind: tok::lessless) &&
733 ((Previous.is(Kind: tok::identifier) && Previous.TokenText == "endl") ||
734 (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with(Suffix: "\\n\"") ||
735 Previous.TokenText == "\'\\n\'")))) {
736 return true;
737 }
738
739 if (Previous.is(TT: TT_BlockComment) && Previous.IsMultiline)
740 return true;
741
742 if (State.NoContinuation)
743 return true;
744
745 return false;
746}
747
748unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
749 bool DryRun,
750 unsigned ExtraSpaces) {
751 const FormatToken &Current = *State.NextToken;
752 assert(State.NextToken->Previous);
753 const FormatToken &Previous = *State.NextToken->Previous;
754
755 assert(!State.Stack.empty());
756 State.NoContinuation = false;
757
758 if (Current.is(TT: TT_ImplicitStringLiteral) &&
759 (!Previous.Tok.getIdentifierInfo() ||
760 Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
761 tok::pp_not_keyword)) {
762 unsigned EndColumn =
763 SourceMgr.getSpellingColumnNumber(Loc: Current.WhitespaceRange.getEnd());
764 if (Current.LastNewlineOffset != 0) {
765 // If there is a newline within this token, the final column will solely
766 // determined by the current end column.
767 State.Column = EndColumn;
768 } else {
769 unsigned StartColumn =
770 SourceMgr.getSpellingColumnNumber(Loc: Current.WhitespaceRange.getBegin());
771 assert(EndColumn >= StartColumn);
772 State.Column += EndColumn - StartColumn;
773 }
774 moveStateToNextToken(State, DryRun, /*Newline=*/false);
775 return 0;
776 }
777
778 unsigned Penalty = 0;
779 if (Newline)
780 Penalty = addTokenOnNewLine(State, DryRun);
781 else
782 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
783
784 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
785}
786
787void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
788 unsigned ExtraSpaces) {
789 FormatToken &Current = *State.NextToken;
790 assert(State.NextToken->Previous);
791 const FormatToken &Previous = *State.NextToken->Previous;
792 auto &CurrentState = State.Stack.back();
793
794 // Deal with lambda arguments in C++. The aim here is to ensure that we don't
795 // over-indent lambda function bodies when lambdas are passed as arguments to
796 // function calls. We do this by ensuring that either all arguments (including
797 // any lambdas) go on the same line as the function call, or we break before
798 // the first argument.
799 auto DisallowLineBreaks = [&] {
800 if (!Style.isCpp() ||
801 Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope) {
802 return false;
803 }
804
805 // For example, `/*Newline=*/false`.
806 if (Previous.is(TT: TT_BlockComment) && Current.SpacesRequiredBefore == 0)
807 return false;
808
809 if (Current.isOneOf(K1: tok::comment, K2: tok::l_paren, Ks: TT_LambdaLSquare))
810 return false;
811
812 const auto *Prev = Current.getPreviousNonComment();
813 if (!Prev || Prev->isNot(Kind: tok::l_paren))
814 return false;
815
816 if (Prev->BlockParameterCount == 0)
817 return false;
818
819 // Multiple lambdas in the same function call.
820 if (Prev->BlockParameterCount > 1)
821 return true;
822
823 // A lambda followed by another arg.
824 if (!Prev->Role)
825 return false;
826
827 const auto *Comma = Prev->Role->lastComma();
828 if (!Comma)
829 return false;
830
831 const auto *Next = Comma->getNextNonComment();
832 return Next && Next->isNoneOf(Ks: TT_LambdaLSquare, Ks: tok::l_brace, Ks: tok::caret);
833 };
834
835 if (DisallowLineBreaks())
836 State.NoLineBreak = true;
837
838 if (Current.is(Kind: tok::equal) &&
839 (State.Line->First->is(Kind: tok::kw_for) || Current.NestingLevel == 0) &&
840 CurrentState.VariablePos == 0 &&
841 (!Previous.Previous ||
842 Previous.Previous->isNot(Kind: TT_DesignatedInitializerPeriod))) {
843 CurrentState.VariablePos = State.Column;
844 // Move over * and & if they are bound to the variable name.
845 const FormatToken *Tok = &Previous;
846 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
847 CurrentState.VariablePos -= Tok->ColumnWidth;
848 if (Tok->SpacesRequiredBefore != 0)
849 break;
850 Tok = Tok->Previous;
851 }
852 if (Previous.PartOfMultiVariableDeclStmt)
853 CurrentState.LastSpace = CurrentState.VariablePos;
854 }
855
856 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
857
858 // Indent preprocessor directives after the hash if required.
859 int PPColumnCorrection = 0;
860 if (&Previous == State.Line->First && Previous.is(Kind: tok::hash) &&
861 (State.Line->Type == LT_PreprocessorDirective ||
862 State.Line->Type == LT_ImportStatement)) {
863 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) {
864 Spaces += State.FirstIndent;
865
866 // For preprocessor indent with tabs, State.Column will be 1 because of
867 // the hash. This causes second-level indents onward to have an extra
868 // space after the tabs. We avoid this misalignment by subtracting 1 from
869 // the column value passed to replaceWhitespace().
870 if (Style.UseTab != FormatStyle::UT_Never)
871 PPColumnCorrection = -1;
872 } else if (Style.IndentPPDirectives == FormatStyle::PPDIS_Leave) {
873 Spaces += Current.OriginalColumn - Previous.OriginalColumn - 1;
874 }
875 }
876
877 if (!DryRun) {
878 const bool ContinuePPDirective =
879 State.Line->InMacroBody && Current.isNot(Kind: TT_LineComment);
880 Whitespaces.replaceWhitespace(Tok&: Current, /*Newlines=*/0, Spaces,
881 StartOfTokenColumn: State.Column + Spaces + PPColumnCorrection,
882 /*IsAligned=*/false, InPPDirective: ContinuePPDirective);
883 }
884
885 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
886 // declaration unless there is multiple inheritance.
887 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
888 Current.is(TT: TT_InheritanceColon)) {
889 CurrentState.NoLineBreak = true;
890 }
891 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
892 Previous.is(TT: TT_InheritanceColon)) {
893 CurrentState.NoLineBreak = true;
894 }
895
896 if (Current.is(TT: TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
897 unsigned MinIndent =
898 std::max(a: State.FirstIndent + Style.ContinuationIndentWidth,
899 b: CurrentState.Indent.Total);
900 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
901 if (Current.LongestObjCSelectorName == 0)
902 CurrentState.AlignColons = false;
903 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
904 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
905 else
906 CurrentState.ColonPos = FirstColonPos;
907 }
908
909 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
910 // parenthesis by disallowing any further line breaks if there is no line
911 // break after the opening parenthesis. Don't break if it doesn't conserve
912 // columns.
913 auto IsOpeningBracket = [&](const FormatToken &Tok) {
914 auto IsStartOfBracedList = [&]() {
915 return Tok.is(Kind: tok::l_brace) && Tok.isNot(Kind: BK_Block) &&
916 Style.Cpp11BracedListStyle != FormatStyle::BLS_Block;
917 };
918 if (IsStartOfBracedList())
919 return Style.BreakAfterOpenBracketBracedList;
920 if (Tok.isNoneOf(Ks: tok::l_paren, Ks: TT_TemplateOpener, Ks: tok::l_square))
921 return false;
922 if (!Tok.Previous)
923 return true;
924 if (Tok.Previous->isIf())
925 return Style.BreakAfterOpenBracketIf;
926 if (Tok.Previous->isLoop(Style))
927 return Style.BreakAfterOpenBracketLoop;
928 if (Tok.Previous->is(Kind: tok::kw_switch))
929 return Style.BreakAfterOpenBracketSwitch;
930 if (Style.BreakAfterOpenBracketFunction) {
931 return !Tok.Previous->is(TT: TT_CastRParen) &&
932 !(Style.isJavaScript() && Tok.is(II: Keywords.kw_await));
933 }
934 return false;
935 };
936 auto IsFunctionCallParen = [](const FormatToken &Tok) {
937 return Tok.is(Kind: tok::l_paren) && Tok.ParameterCount > 0 && Tok.Previous &&
938 Tok.Previous->is(Kind: tok::identifier);
939 };
940 auto IsInTemplateString = [this](const FormatToken &Tok, bool NestBlocks) {
941 if (!Style.isJavaScript())
942 return false;
943 for (const auto *Prev = &Tok; Prev; Prev = Prev->Previous) {
944 if (Prev->is(TT: TT_TemplateString) && Prev->opensScope())
945 return true;
946 if (Prev->opensScope() && !NestBlocks)
947 return false;
948 if (Prev->is(TT: TT_TemplateString) && Prev->closesScope())
949 return false;
950 }
951 return false;
952 };
953 // Identifies simple (no expression) one-argument function calls.
954 auto StartsSimpleOneArgList = [&](const FormatToken &TokAfterLParen) {
955 assert(TokAfterLParen.isNot(tok::comment) || TokAfterLParen.Next);
956 const auto &Tok =
957 TokAfterLParen.is(Kind: tok::comment) ? *TokAfterLParen.Next : TokAfterLParen;
958 if (!Tok.FakeLParens.empty() && Tok.FakeLParens.back() > prec::Unknown)
959 return false;
960 // Nested calls that involve `new` expressions also look like simple
961 // function calls, eg:
962 // - foo(new Bar())
963 // - foo(::new Bar())
964 if (Tok.is(Kind: tok::kw_new) || Tok.startsSequence(K1: tok::coloncolon, Tokens: tok::kw_new))
965 return true;
966 if (Tok.is(TT: TT_UnaryOperator) ||
967 (Style.isJavaScript() &&
968 Tok.isOneOf(K1: tok::ellipsis, K2: Keywords.kw_await))) {
969 return true;
970 }
971 const auto *Previous = TokAfterLParen.Previous;
972 assert(Previous); // IsOpeningBracket(Previous)
973 if (Previous->Previous &&
974 (Previous->Previous->isIf() || Previous->Previous->isLoop(Style) ||
975 Previous->Previous->is(Kind: tok::kw_switch))) {
976 return false;
977 }
978 if (Previous->isNoneOf(Ks: TT_FunctionDeclarationLParen,
979 Ks: TT_LambdaDefinitionLParen) &&
980 !IsFunctionCallParen(*Previous)) {
981 return true;
982 }
983 if (IsOpeningBracket(Tok) || IsInTemplateString(Tok, true))
984 return true;
985 const auto *Next = Tok.Next;
986 return !Next || Next->isMemberAccess() ||
987 Next->is(TT: TT_FunctionDeclarationLParen) || IsFunctionCallParen(*Next);
988 };
989 if (IsOpeningBracket(Previous) &&
990 State.Column > getNewLineColumn(State).Total &&
991 // Don't do this for simple (no expressions) one-argument function calls
992 // as that feels like needlessly wasting whitespace, e.g.:
993 //
994 // caaaaaaaaaaaall(
995 // caaaaaaaaaaaall(
996 // caaaaaaaaaaaall(
997 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
998 // or
999 // caaaaaaaaaaaaaaaaaaaaal(
1000 // new SomethingElseeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee());
1001 !StartsSimpleOneArgList(Current)) {
1002 CurrentState.NoLineBreak = true;
1003 }
1004
1005 if (Previous.is(TT: TT_TemplateString) && Previous.opensScope())
1006 CurrentState.NoLineBreak = true;
1007
1008 // Align following lines within parentheses / brackets if configured.
1009 // Note: This doesn't apply to macro expansion lines, which are MACRO( , , )
1010 // with args as children of the '(' and ',' tokens. It does not make sense to
1011 // align the commas with the opening paren.
1012 if (Style.AlignAfterOpenBracket &&
1013 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
1014 Previous.isNoneOf(Ks: TT_ObjCMethodExpr, Ks: TT_RequiresClause,
1015 Ks: TT_TableGenDAGArgOpener,
1016 Ks: TT_TableGenDAGArgOpenerToBreak) &&
1017 !(Current.MacroParent && Previous.MacroParent) &&
1018 (Current.isNot(Kind: TT_LineComment) ||
1019 (Previous.is(BBK: BK_BracedInit) &&
1020 (Style.Cpp11BracedListStyle != FormatStyle::BLS_FunctionCall ||
1021 !Previous.Previous ||
1022 Previous.Previous->isNoneOf(Ks: tok::identifier, Ks: tok::l_paren,
1023 Ks: BK_BracedInit))) ||
1024 Previous.is(TT: TT_VerilogMultiLineListLParen)) &&
1025 !IsInTemplateString(Current, false)) {
1026 CurrentState.Indent = State.Column + Spaces;
1027 CurrentState.IsAligned = true;
1028 }
1029 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
1030 CurrentState.NoLineBreak = true;
1031 if (mustBreakBinaryOperation(Current, Style))
1032 CurrentState.NoLineBreak = true;
1033
1034 if (startsSegmentOfBuilderTypeCall(Tok: Current) &&
1035 State.Column > getNewLineColumn(State).Total) {
1036 CurrentState.ContainsUnwrappedBuilder = true;
1037 }
1038
1039 if (Current.is(TT: TT_LambdaArrow) && Style.isJava())
1040 CurrentState.NoLineBreak = true;
1041 if (Current.isMemberAccess() && Previous.is(Kind: tok::r_paren) &&
1042 (Previous.MatchingParen &&
1043 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
1044 // If there is a function call with long parameters, break before trailing
1045 // calls. This prevents things like:
1046 // EXPECT_CALL(SomeLongParameter).Times(
1047 // 2);
1048 // We don't want to do this for short parameters as they can just be
1049 // indexes.
1050 CurrentState.NoLineBreak = true;
1051 }
1052
1053 // Don't allow the RHS of an operator to be split over multiple lines unless
1054 // there is a line-break right after the operator.
1055 // Exclude relational operators, as there, it is always more desirable to
1056 // have the LHS 'left' of the RHS.
1057 const FormatToken *P = Current.getPreviousNonComment();
1058 if (Current.isNot(Kind: tok::comment) && P &&
1059 (P->isOneOf(K1: TT_BinaryOperator, K2: tok::comma) ||
1060 (P->is(TT: TT_ConditionalExpr) && P->is(Kind: tok::colon))) &&
1061 P->isNoneOf(Ks: TT_OverloadedOperator, Ks: TT_CtorInitializerComma) &&
1062 P->getPrecedence() != prec::Assignment &&
1063 P->getPrecedence() != prec::Relational &&
1064 P->getPrecedence() != prec::Spaceship) {
1065 bool BreakBeforeOperator =
1066 P->MustBreakBefore || P->is(Kind: tok::lessless) ||
1067 (P->is(TT: TT_BinaryOperator) &&
1068 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
1069 (P->is(TT: TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
1070 // Don't do this if there are only two operands. In these cases, there is
1071 // always a nice vertical separation between them and the extra line break
1072 // does not help.
1073 bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator &&
1074 P->isNot(Kind: TT_ConditionalExpr);
1075 if ((!BreakBeforeOperator &&
1076 !(HasTwoOperands &&
1077 Style.AlignOperands != FormatStyle::OAS_DontAlign)) ||
1078 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
1079 CurrentState.NoLineBreakInOperand = true;
1080 }
1081 }
1082
1083 State.Column += Spaces;
1084 if (Current.isNot(Kind: tok::comment) && Previous.is(Kind: tok::l_paren) &&
1085 Previous.Previous &&
1086 (Previous.Previous->is(Kind: tok::kw_for) || Previous.Previous->isIf())) {
1087 // Treat the condition inside an if as if it was a second function
1088 // parameter, i.e. let nested calls have a continuation indent.
1089 CurrentState.LastSpace = State.Column;
1090 CurrentState.NestedBlockIndent = State.Column;
1091 } else if (Current.isNoneOf(Ks: tok::comment, Ks: tok::caret) &&
1092 ((Previous.is(Kind: tok::comma) &&
1093 Previous.isNot(Kind: TT_OverloadedOperator)) ||
1094 (Previous.is(Kind: tok::colon) && Previous.is(TT: TT_ObjCMethodExpr)))) {
1095 CurrentState.LastSpace = State.Column;
1096 } else if (Previous.is(TT: TT_CtorInitializerColon) &&
1097 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
1098 Style.BreakConstructorInitializers ==
1099 FormatStyle::BCIS_AfterColon) {
1100 CurrentState.Indent = State.Column;
1101 CurrentState.LastSpace = State.Column;
1102 } else if (Previous.isOneOf(K1: TT_ConditionalExpr, K2: TT_CtorInitializerColon)) {
1103 CurrentState.LastSpace = State.Column;
1104 } else if (Previous.is(TT: TT_BinaryOperator) &&
1105 ((Previous.getPrecedence() != prec::Assignment &&
1106 (Previous.isNot(Kind: tok::lessless) || Previous.OperatorIndex != 0 ||
1107 Previous.NextOperator)) ||
1108 Current.StartsBinaryExpression)) {
1109 // Indent relative to the RHS of the expression unless this is a simple
1110 // assignment without binary expression on the RHS.
1111 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
1112 CurrentState.LastSpace = State.Column;
1113 } else if (Previous.is(TT: TT_InheritanceColon)) {
1114 CurrentState.Indent = State.Column;
1115 CurrentState.LastSpace = State.Column;
1116 } else if (Current.is(TT: TT_CSharpGenericTypeConstraintColon)) {
1117 CurrentState.ColonPos = State.Column;
1118 } else if (Previous.opensScope()) {
1119 // If a function has a trailing call, indent all parameters from the
1120 // opening parenthesis. This avoids confusing indents like:
1121 // OuterFunction(InnerFunctionCall( // break
1122 // ParameterToInnerFunction)) // break
1123 // .SecondInnerFunctionCall();
1124 if (Previous.MatchingParen) {
1125 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
1126 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
1127 State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
1128 CurrentState.LastSpace = State.Column;
1129 }
1130 }
1131 }
1132}
1133
1134unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
1135 bool DryRun) {
1136 FormatToken &Current = *State.NextToken;
1137 assert(State.NextToken->Previous);
1138 const FormatToken &Previous = *State.NextToken->Previous;
1139 auto &CurrentState = State.Stack.back();
1140
1141 // Extra penalty that needs to be added because of the way certain line
1142 // breaks are chosen.
1143 unsigned Penalty = 0;
1144
1145 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1146 const FormatToken *NextNonComment = Previous.getNextNonComment();
1147 if (!NextNonComment)
1148 NextNonComment = &Current;
1149 // The first line break on any NestingLevel causes an extra penalty in order
1150 // prefer similar line breaks.
1151 if (!CurrentState.ContainsLineBreak)
1152 Penalty += 15;
1153 CurrentState.ContainsLineBreak = true;
1154
1155 Penalty += State.NextToken->SplitPenalty;
1156
1157 // Breaking before the first "<<" is generally not desirable if the LHS is
1158 // short. Also always add the penalty if the LHS is split over multiple lines
1159 // to avoid unnecessary line breaks that just work around this penalty.
1160 if (NextNonComment->is(Kind: tok::lessless) && CurrentState.FirstLessLess == 0 &&
1161 (State.Column <= Style.ColumnLimit / 3 ||
1162 CurrentState.BreakBeforeParameter)) {
1163 Penalty += Style.PenaltyBreakFirstLessLess;
1164 }
1165
1166 const auto [TotalColumn, IndentedFromColumn] = getNewLineColumn(State);
1167 State.Column = TotalColumn;
1168
1169 // Add Penalty proportional to amount of whitespace away from FirstColumn
1170 // This tends to penalize several lines that are far-right indented,
1171 // and prefers a line-break prior to such a block, e.g:
1172 //
1173 // Constructor() :
1174 // member(value), looooooooooooooooong_member(
1175 // looooooooooong_call(param_1, param_2, param_3))
1176 // would then become
1177 // Constructor() :
1178 // member(value),
1179 // looooooooooooooooong_member(
1180 // looooooooooong_call(param_1, param_2, param_3))
1181 if (State.Column > State.FirstIndent) {
1182 Penalty +=
1183 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
1184 }
1185
1186 // Indent nested blocks relative to this column, unless in a very specific
1187 // JavaScript special case where:
1188 //
1189 // var loooooong_name =
1190 // function() {
1191 // // code
1192 // }
1193 //
1194 // is common and should be formatted like a free-standing function. The same
1195 // goes for wrapping before the lambda return type arrow.
1196 if (Current.isNot(Kind: TT_LambdaArrow) &&
1197 (!Style.isJavaScript() || Current.NestingLevel != 0 ||
1198 !PreviousNonComment || PreviousNonComment->isNot(Kind: tok::equal) ||
1199 Current.isNoneOf(Ks: Keywords.kw_async, Ks: Keywords.kw_function))) {
1200 CurrentState.NestedBlockIndent = State.Column;
1201 }
1202
1203 if (NextNonComment->isMemberAccess()) {
1204 if (CurrentState.CallContinuation == 0)
1205 CurrentState.CallContinuation = State.Column;
1206 } else if (NextNonComment->is(TT: TT_SelectorName)) {
1207 if (!CurrentState.ObjCSelectorNameFound) {
1208 if (NextNonComment->LongestObjCSelectorName == 0) {
1209 CurrentState.AlignColons = false;
1210 } else {
1211 CurrentState.ColonPos =
1212 (shouldIndentWrappedSelectorName(Style, LineType: State.Line->Type)
1213 ? std::max(a: CurrentState.Indent.Total,
1214 b: State.FirstIndent + Style.ContinuationIndentWidth)
1215 : CurrentState.Indent.Total) +
1216 std::max(a: NextNonComment->LongestObjCSelectorName,
1217 b: NextNonComment->ColumnWidth);
1218 }
1219 } else if (CurrentState.AlignColons &&
1220 CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
1221 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
1222 }
1223 } else if (PreviousNonComment && PreviousNonComment->is(Kind: tok::colon) &&
1224 PreviousNonComment->isOneOf(K1: TT_ObjCMethodExpr, K2: TT_DictLiteral)) {
1225 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
1226 // method expression, the block should be aligned to the line starting it,
1227 // e.g.:
1228 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
1229 // ^(int *i) {
1230 // // ...
1231 // }];
1232 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
1233 // when we consume all of the "}"'s FakeRParens at the "{".
1234 if (State.Stack.size() > 1) {
1235 State.Stack[State.Stack.size() - 2].LastSpace =
1236 std::max(a: CurrentState.LastSpace, b: CurrentState.Indent.Total) +
1237 Style.ContinuationIndentWidth;
1238 }
1239 }
1240
1241 switch (Style.BreakInheritanceList) {
1242 case FormatStyle::BILS_BeforeColon:
1243 case FormatStyle::BILS_AfterComma:
1244 CurrentState.IsAligned = CurrentState.IsAligned ||
1245 Current.is(TT: TT_InheritanceColon) ||
1246 Previous.is(TT: TT_InheritanceComma);
1247 break;
1248 case FormatStyle::BILS_BeforeComma:
1249 CurrentState.IsAligned =
1250 CurrentState.IsAligned ||
1251 Current.isOneOf(K1: TT_InheritanceColon, K2: TT_InheritanceComma);
1252 break;
1253 case FormatStyle::BILS_AfterColon:
1254 CurrentState.IsAligned =
1255 CurrentState.IsAligned ||
1256 Previous.isOneOf(K1: TT_InheritanceColon, K2: TT_InheritanceComma);
1257 break;
1258 }
1259
1260 if ((PreviousNonComment &&
1261 PreviousNonComment->isOneOf(K1: tok::comma, K2: tok::semi) &&
1262 !CurrentState.AvoidBinPacking) ||
1263 Previous.is(TT: TT_BinaryOperator)) {
1264 CurrentState.BreakBeforeParameter = false;
1265 }
1266 if (PreviousNonComment &&
1267 (PreviousNonComment->isOneOf(K1: TT_TemplateCloser, K2: TT_JavaAnnotation) ||
1268 PreviousNonComment->ClosesRequiresClause) &&
1269 Current.NestingLevel == 0) {
1270 CurrentState.BreakBeforeParameter = false;
1271 }
1272 if (NextNonComment->is(Kind: tok::question) ||
1273 (PreviousNonComment && PreviousNonComment->is(Kind: tok::question))) {
1274 CurrentState.BreakBeforeParameter = true;
1275 }
1276 if (Current.is(TT: TT_BinaryOperator) && Current.CanBreakBefore) {
1277 CurrentState.BreakBeforeParameter = false;
1278 CurrentState.IsAligned = true;
1279 }
1280
1281 if (!DryRun) {
1282 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
1283 if (Current.is(Kind: tok::r_brace) && Current.MatchingParen &&
1284 // Only strip trailing empty lines for l_braces that have children, i.e.
1285 // for function expressions (lambdas, arrows, etc).
1286 !Current.MatchingParen->Children.empty()) {
1287 // lambdas and arrow functions are expressions, thus their r_brace is not
1288 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
1289 // about removing empty lines on closing blocks. Special case them here.
1290 MaxEmptyLinesToKeep = 1;
1291 }
1292 const unsigned Newlines =
1293 std::max(a: 1u, b: std::min(a: Current.NewlinesBefore, b: MaxEmptyLinesToKeep));
1294 const bool ContinuePPDirective = State.Line->InPPDirective &&
1295 State.Line->Type != LT_ImportStatement &&
1296 Current.isNot(Kind: TT_LineComment);
1297 Whitespaces.replaceWhitespace(Tok&: Current, Newlines, Spaces: State.Column, StartOfTokenColumn: State.Column,
1298 IsAligned: CurrentState.IsAligned, InPPDirective: ContinuePPDirective,
1299 IndentedFromColumn);
1300 }
1301
1302 if (!Current.isTrailingComment())
1303 CurrentState.LastSpace = State.Column;
1304 if (Current.is(Kind: tok::lessless)) {
1305 // If we are breaking before a "<<", we always want to indent relative to
1306 // RHS. This is necessary only for "<<", as we special-case it and don't
1307 // always indent relative to the RHS.
1308 CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1309 }
1310
1311 State.StartOfLineLevel = Current.NestingLevel;
1312 State.LowestLevelOnLine = Current.NestingLevel;
1313
1314 // Any break on this level means that the parent level has been broken
1315 // and we need to avoid bin packing there.
1316 bool NestedBlockSpecialCase =
1317 (!Style.isCpp() && Current.is(Kind: tok::r_brace) && State.Stack.size() > 1 &&
1318 State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1319 (Style.Language == FormatStyle::LK_ObjC && Current.is(Kind: tok::r_brace) &&
1320 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1321 // Do not force parameter break for statements with requires expressions.
1322 NestedBlockSpecialCase =
1323 NestedBlockSpecialCase ||
1324 (Current.MatchingParen &&
1325 Current.MatchingParen->is(TT: TT_RequiresExpressionLBrace));
1326 if (!NestedBlockSpecialCase) {
1327 auto ParentLevelIt = std::next(x: State.Stack.rbegin());
1328 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1329 Current.MatchingParen && Current.MatchingParen->is(TT: TT_LambdaLBrace)) {
1330 // If the first character on the new line is a lambda's closing brace, the
1331 // stack still contains that lambda's parenthesis. As such, we need to
1332 // recurse further down the stack than usual to find the parenthesis level
1333 // containing the lambda, which is where we want to set
1334 // BreakBeforeParameter.
1335 //
1336 // We specifically special case "OuterScope"-formatted lambdas here
1337 // because, when using that setting, breaking before the parameter
1338 // directly following the lambda is particularly unsightly. However, when
1339 // "OuterScope" is not set, the logic to find the parent parenthesis level
1340 // still appears to be sometimes incorrect. It has not been fixed yet
1341 // because it would lead to significant changes in existing behaviour.
1342 //
1343 // TODO: fix the non-"OuterScope" case too.
1344 auto FindCurrentLevel = [&](const auto &It) {
1345 return std::find_if(It, State.Stack.rend(), [](const auto &PState) {
1346 return PState.Tok != nullptr; // Ignore fake parens.
1347 });
1348 };
1349 auto MaybeIncrement = [&](const auto &It) {
1350 return It != State.Stack.rend() ? std::next(It) : It;
1351 };
1352 auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin());
1353 auto LevelContainingLambdaIt =
1354 FindCurrentLevel(MaybeIncrement(LambdaLevelIt));
1355 ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt);
1356 }
1357 for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I)
1358 I->BreakBeforeParameter = true;
1359 }
1360
1361 if (PreviousNonComment &&
1362 PreviousNonComment->isNoneOf(Ks: tok::comma, Ks: tok::colon, Ks: tok::semi) &&
1363 ((PreviousNonComment->isNot(Kind: TT_TemplateCloser) &&
1364 !PreviousNonComment->ClosesRequiresClause) ||
1365 Current.NestingLevel != 0) &&
1366 PreviousNonComment->isNoneOf(
1367 Ks: TT_BinaryOperator, Ks: TT_FunctionAnnotationRParen, Ks: TT_JavaAnnotation,
1368 Ks: TT_LeadingJavaAnnotation) &&
1369 Current.isNot(Kind: TT_BinaryOperator) && !PreviousNonComment->opensScope() &&
1370 // We don't want to enforce line breaks for subsequent arguments just
1371 // because we have been forced to break before a lambda body.
1372 (!Style.BraceWrapping.BeforeLambdaBody ||
1373 Current.isNot(Kind: TT_LambdaLBrace))) {
1374 CurrentState.BreakBeforeParameter = true;
1375 }
1376
1377 // If we break after { or the [ of an array initializer, we should also break
1378 // before the corresponding } or ].
1379 if (PreviousNonComment &&
1380 (PreviousNonComment->isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) ||
1381 opensProtoMessageField(LessTok: *PreviousNonComment, Style))) {
1382 CurrentState.BreakBeforeClosingBrace = true;
1383 }
1384
1385 if (PreviousNonComment && PreviousNonComment->is(Kind: tok::l_paren)) {
1386 if (auto Previous = PreviousNonComment->Previous) {
1387 if (Previous->isIf()) {
1388 CurrentState.BreakBeforeClosingParen = Style.BreakBeforeCloseBracketIf;
1389 } else if (Previous->isLoop(Style)) {
1390 CurrentState.BreakBeforeClosingParen =
1391 Style.BreakBeforeCloseBracketLoop;
1392 } else if (Previous->is(Kind: tok::kw_switch)) {
1393 CurrentState.BreakBeforeClosingParen =
1394 Style.BreakBeforeCloseBracketSwitch;
1395 } else {
1396 CurrentState.BreakBeforeClosingParen =
1397 Style.BreakBeforeCloseBracketFunction;
1398 }
1399 }
1400 }
1401
1402 if (PreviousNonComment && PreviousNonComment->is(TT: TT_TemplateOpener))
1403 CurrentState.BreakBeforeClosingAngle = Style.BreakBeforeTemplateCloser;
1404
1405 if (CurrentState.AvoidBinPacking) {
1406 // If we are breaking after '(', '{', '<', or this is the break after a ':'
1407 // to start a member initializer list in a constructor, this should not
1408 // be considered bin packing unless the relevant AllowAll option is false or
1409 // this is a dict/object literal.
1410 bool PreviousIsBreakingCtorInitializerColon =
1411 PreviousNonComment && PreviousNonComment->is(TT: TT_CtorInitializerColon) &&
1412 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
1413 bool AllowAllConstructorInitializersOnNextLine =
1414 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine ||
1415 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLineOnly;
1416 if ((Previous.isNoneOf(Ks: tok::l_paren, Ks: tok::l_brace, Ks: TT_BinaryOperator) &&
1417 !PreviousIsBreakingCtorInitializerColon) ||
1418 (!Style.AllowAllParametersOfDeclarationOnNextLine &&
1419 State.Line->MustBeDeclaration) ||
1420 (!Style.AllowAllArgumentsOnNextLine &&
1421 !State.Line->MustBeDeclaration) ||
1422 (!AllowAllConstructorInitializersOnNextLine &&
1423 PreviousIsBreakingCtorInitializerColon) ||
1424 Previous.is(TT: TT_DictLiteral)) {
1425 CurrentState.BreakBeforeParameter = true;
1426 }
1427
1428 // If we are breaking after a ':' to start a member initializer list,
1429 // and we allow all arguments on the next line, we should not break
1430 // before the next parameter.
1431 if (PreviousIsBreakingCtorInitializerColon &&
1432 AllowAllConstructorInitializersOnNextLine) {
1433 CurrentState.BreakBeforeParameter = false;
1434 }
1435 }
1436
1437 if (mustBreakBinaryOperation(Current, Style))
1438 CurrentState.BreakBeforeParameter = true;
1439
1440 return Penalty;
1441}
1442
1443IndentationAndAlignment
1444ContinuationIndenter::getNewLineColumn(const LineState &State) {
1445 if (!State.NextToken || !State.NextToken->Previous)
1446 return 0;
1447
1448 FormatToken &Current = *State.NextToken;
1449 const auto &CurrentState = State.Stack.back();
1450
1451 if (CurrentState.IsCSharpGenericTypeConstraint &&
1452 Current.isNot(Kind: TT_CSharpGenericTypeConstraint)) {
1453 return CurrentState.ColonPos + 2;
1454 }
1455
1456 const FormatToken &Previous = *Current.Previous;
1457 // If we are continuing an expression, we want to use the continuation indent.
1458 const auto ContinuationIndent =
1459 std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1460 b: CurrentState.Indent) +
1461 Style.ContinuationIndentWidth;
1462 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1463 const FormatToken *NextNonComment = Previous.getNextNonComment();
1464 if (!NextNonComment)
1465 NextNonComment = &Current;
1466
1467 // Java specific bits.
1468 if (Style.isJava() &&
1469 Current.isOneOf(K1: Keywords.kw_implements, K2: Keywords.kw_extends)) {
1470 return std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1471 b: CurrentState.Indent + Style.ContinuationIndentWidth);
1472 }
1473
1474 // Indentation of the statement following a Verilog case label is taken care
1475 // of in moveStateToNextToken.
1476 if (Style.isVerilog() && PreviousNonComment &&
1477 Keywords.isVerilogEndOfLabel(Tok: *PreviousNonComment)) {
1478 return State.FirstIndent;
1479 }
1480
1481 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths &&
1482 State.Line->First->is(Kind: tok::kw_enum)) {
1483 return IndentationAndAlignment(Style.IndentWidth *
1484 State.Line->First->IndentLevel) +
1485 Style.IndentWidth;
1486 }
1487
1488 if (Style.BraceWrapping.BeforeLambdaBody &&
1489 Style.BraceWrapping.IndentBraces && Current.is(TT: TT_LambdaLBrace)) {
1490 const auto From = Style.LambdaBodyIndentation == FormatStyle::LBI_Signature
1491 ? CurrentState.Indent
1492 : State.FirstIndent;
1493 return From + Style.IndentWidth;
1494 }
1495
1496 if ((NextNonComment->is(Kind: tok::l_brace) && NextNonComment->is(BBK: BK_Block)) ||
1497 (Style.isVerilog() && Keywords.isVerilogBegin(Tok: *NextNonComment))) {
1498 if (Current.NestingLevel == 0 ||
1499 (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1500 State.NextToken->is(TT: TT_LambdaLBrace))) {
1501 return State.FirstIndent;
1502 }
1503 return CurrentState.Indent;
1504 }
1505 if (Current.is(TT: TT_LambdaArrow) &&
1506 Previous.isOneOf(K1: tok::kw_noexcept, K2: tok::kw_mutable, Ks: tok::kw_constexpr,
1507 Ks: tok::kw_consteval, Ks: tok::kw_static,
1508 Ks: TT_AttributeRSquare)) {
1509 return ContinuationIndent;
1510 }
1511 if ((Current.isOneOf(K1: tok::r_brace, K2: tok::r_square) ||
1512 (Current.is(Kind: tok::greater) && (Style.isProto() || Style.isTableGen()))) &&
1513 State.Stack.size() > 1) {
1514 if (Current.closesBlockOrBlockTypeList(Style))
1515 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1516 if (Current.MatchingParen && Current.MatchingParen->is(BBK: BK_BracedInit))
1517 return State.Stack[State.Stack.size() - 2].LastSpace;
1518 return State.FirstIndent;
1519 }
1520 // Indent a closing parenthesis at the previous level if followed by a semi,
1521 // const, or opening brace. This allows indentations such as:
1522 // foo(
1523 // a,
1524 // );
1525 // int Foo::getter(
1526 // //
1527 // ) const {
1528 // return foo;
1529 // }
1530 // function foo(
1531 // a,
1532 // ) {
1533 // code(); //
1534 // }
1535 if (Current.is(Kind: tok::r_paren) && State.Stack.size() > 1 &&
1536 (!Current.Next ||
1537 Current.Next->isOneOf(K1: tok::semi, K2: tok::kw_const, Ks: tok::l_brace))) {
1538 return State.Stack[State.Stack.size() - 2].LastSpace;
1539 }
1540 // When DAGArg closer exists top of line, it should be aligned in the similar
1541 // way as function call above.
1542 if (Style.isTableGen() && Current.is(TT: TT_TableGenDAGArgCloser) &&
1543 State.Stack.size() > 1) {
1544 return State.Stack[State.Stack.size() - 2].LastSpace;
1545 }
1546 if (Style.BreakBeforeCloseBracketBracedList && Current.is(Kind: tok::r_brace) &&
1547 Current.MatchingParen && Current.MatchingParen->is(BBK: BK_BracedInit) &&
1548 State.Stack.size() > 1) {
1549 return State.Stack[State.Stack.size() - 2].LastSpace;
1550 }
1551 if ((Style.BreakBeforeCloseBracketFunction ||
1552 Style.BreakBeforeCloseBracketIf || Style.BreakBeforeCloseBracketLoop ||
1553 Style.BreakBeforeCloseBracketSwitch) &&
1554 Current.is(Kind: tok::r_paren) && State.Stack.size() > 1) {
1555 return State.Stack[State.Stack.size() - 2].LastSpace;
1556 }
1557 if (Style.BreakBeforeTemplateCloser && Current.is(TT: TT_TemplateCloser) &&
1558 State.Stack.size() > 1) {
1559 return State.Stack[State.Stack.size() - 2].LastSpace;
1560 }
1561 if (NextNonComment->is(TT: TT_TemplateString) && NextNonComment->closesScope())
1562 return State.Stack[State.Stack.size() - 2].LastSpace;
1563 // Field labels in a nested type should be aligned to the brace. For example
1564 // in ProtoBuf:
1565 // optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123,
1566 // bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}];
1567 // For Verilog, a quote following a brace is treated as an identifier. And
1568 // Both braces and colons get annotated as TT_DictLiteral. So we have to
1569 // check.
1570 if (Current.is(Kind: tok::identifier) && Current.Next &&
1571 (!Style.isVerilog() || Current.Next->is(Kind: tok::colon)) &&
1572 (Current.Next->is(TT: TT_DictLiteral) ||
1573 (Style.isProto() && Current.Next->isOneOf(K1: tok::less, K2: tok::l_brace)))) {
1574 return CurrentState.Indent;
1575 }
1576 if (NextNonComment->is(TT: TT_ObjCStringLiteral) &&
1577 State.StartOfStringLiteral != 0) {
1578 return State.StartOfStringLiteral - 1;
1579 }
1580 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1581 return State.StartOfStringLiteral;
1582 if (NextNonComment->is(Kind: tok::lessless) && CurrentState.FirstLessLess != 0)
1583 return CurrentState.FirstLessLess;
1584 if (NextNonComment->isMemberAccess()) {
1585 if (CurrentState.CallContinuation == 0)
1586 return ContinuationIndent;
1587 return CurrentState.CallContinuation;
1588 }
1589 if (CurrentState.QuestionColumn != 0 &&
1590 ((NextNonComment->is(Kind: tok::colon) &&
1591 NextNonComment->is(TT: TT_ConditionalExpr)) ||
1592 Previous.is(TT: TT_ConditionalExpr))) {
1593 if (((NextNonComment->is(Kind: tok::colon) && NextNonComment->Next &&
1594 !NextNonComment->Next->FakeLParens.empty() &&
1595 NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1596 (Previous.is(Kind: tok::colon) && !Current.FakeLParens.empty() &&
1597 Current.FakeLParens.back() == prec::Conditional)) &&
1598 !CurrentState.IsWrappedConditional) {
1599 // NOTE: we may tweak this slightly:
1600 // * not remove the 'lead' ContinuationIndentWidth
1601 // * always un-indent by the operator when
1602 // BreakBeforeTernaryOperators=true
1603 unsigned Indent = CurrentState.Indent.Total;
1604 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1605 Indent -= Style.ContinuationIndentWidth;
1606 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1607 Indent -= 2;
1608 return Indent;
1609 }
1610 return CurrentState.QuestionColumn;
1611 }
1612 if (Previous.is(Kind: tok::comma) && CurrentState.VariablePos != 0)
1613 return CurrentState.VariablePos;
1614 if (Current.is(TT: TT_RequiresClause)) {
1615 if (Style.IndentRequiresClause)
1616 return CurrentState.Indent + Style.IndentWidth;
1617 switch (Style.RequiresClausePosition) {
1618 case FormatStyle::RCPS_OwnLine:
1619 case FormatStyle::RCPS_WithFollowing:
1620 case FormatStyle::RCPS_OwnLineWithBrace:
1621 return CurrentState.Indent;
1622 default:
1623 break;
1624 }
1625 }
1626 if (NextNonComment->isOneOf(K1: TT_CtorInitializerColon, K2: TT_InheritanceColon,
1627 Ks: TT_InheritanceComma)) {
1628 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1629 }
1630 if ((PreviousNonComment &&
1631 (PreviousNonComment->ClosesTemplateDeclaration ||
1632 PreviousNonComment->ClosesRequiresClause ||
1633 (PreviousNonComment->is(TT: TT_AttributeMacro) &&
1634 Current.isNot(Kind: tok::l_paren) &&
1635 !Current.endsSequence(K1: TT_StartOfName, Tokens: TT_AttributeMacro,
1636 Tokens: TT_PointerOrReference)) ||
1637 PreviousNonComment->isOneOf(K1: TT_AttributeRParen, K2: TT_AttributeRSquare,
1638 Ks: TT_FunctionAnnotationRParen,
1639 Ks: TT_JavaAnnotation,
1640 Ks: TT_LeadingJavaAnnotation))) ||
1641 (!Style.IndentWrappedFunctionNames &&
1642 NextNonComment->isOneOf(K1: tok::kw_operator, K2: TT_FunctionDeclarationName))) {
1643 return std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1644 b: CurrentState.Indent);
1645 }
1646 if (NextNonComment->is(TT: TT_SelectorName)) {
1647 if (!CurrentState.ObjCSelectorNameFound) {
1648 auto MinIndent = CurrentState.Indent;
1649 if (shouldIndentWrappedSelectorName(Style, LineType: State.Line->Type)) {
1650 MinIndent =
1651 std::max(a: MinIndent, b: IndentationAndAlignment(State.FirstIndent) +
1652 Style.ContinuationIndentWidth);
1653 }
1654 // If LongestObjCSelectorName is 0, we are indenting the first
1655 // part of an ObjC selector (or a selector component which is
1656 // not colon-aligned due to block formatting).
1657 //
1658 // Otherwise, we are indenting a subsequent part of an ObjC
1659 // selector which should be colon-aligned to the longest
1660 // component of the ObjC selector.
1661 //
1662 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1663 return MinIndent.addPadding(
1664 Spaces: std::max(a: NextNonComment->LongestObjCSelectorName,
1665 b: NextNonComment->ColumnWidth) -
1666 NextNonComment->ColumnWidth);
1667 }
1668 if (!CurrentState.AlignColons)
1669 return CurrentState.Indent;
1670 if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1671 return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1672 return CurrentState.Indent;
1673 }
1674 if (NextNonComment->is(Kind: tok::colon) && NextNonComment->is(TT: TT_ObjCMethodExpr))
1675 return CurrentState.ColonPos;
1676 if (NextNonComment->is(TT: TT_ArraySubscriptLSquare)) {
1677 if (CurrentState.StartOfArraySubscripts != 0) {
1678 return CurrentState.StartOfArraySubscripts;
1679 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1680 // initializers.
1681 return CurrentState.Indent;
1682 }
1683 return ContinuationIndent;
1684 }
1685
1686 // OpenMP clauses want to get additional indentation when they are pushed onto
1687 // the next line.
1688 if (State.Line->InPragmaDirective) {
1689 FormatToken *PragmaType = State.Line->First->Next->Next;
1690 if (PragmaType && PragmaType->TokenText == "omp")
1691 return CurrentState.Indent + Style.ContinuationIndentWidth;
1692 }
1693
1694 // This ensure that we correctly format ObjC methods calls without inputs,
1695 // i.e. where the last element isn't selector like: [callee method];
1696 if (NextNonComment->is(Kind: tok::identifier) && NextNonComment->FakeRParens == 0 &&
1697 NextNonComment->Next && NextNonComment->Next->is(TT: TT_ObjCMethodExpr)) {
1698 return CurrentState.Indent;
1699 }
1700
1701 if (NextNonComment->isOneOf(K1: TT_StartOfName, K2: TT_PointerOrReference) ||
1702 Previous.isOneOf(K1: tok::coloncolon, K2: tok::equal, Ks: TT_JsTypeColon)) {
1703 return ContinuationIndent;
1704 }
1705 if (PreviousNonComment && PreviousNonComment->is(Kind: tok::colon) &&
1706 PreviousNonComment->isOneOf(K1: TT_ObjCMethodExpr, K2: TT_DictLiteral)) {
1707 return ContinuationIndent;
1708 }
1709 if (NextNonComment->is(TT: TT_CtorInitializerComma))
1710 return CurrentState.Indent;
1711 if (PreviousNonComment && PreviousNonComment->is(TT: TT_CtorInitializerColon) &&
1712 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1713 return CurrentState.Indent;
1714 }
1715 if (PreviousNonComment && PreviousNonComment->is(TT: TT_InheritanceColon) &&
1716 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
1717 return CurrentState.Indent;
1718 }
1719 if (Previous.is(Kind: tok::r_paren) &&
1720 Previous.isNot(Kind: TT_TableGenDAGArgOperatorToBreak) &&
1721 !Current.isBinaryOperator() &&
1722 Current.isNoneOf(Ks: tok::colon, Ks: tok::comment)) {
1723 return ContinuationIndent;
1724 }
1725 if (Current.is(TT: TT_ProtoExtensionLSquare))
1726 return CurrentState.Indent;
1727 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1728 return CurrentState.Indent - Current.Tok.getLength() -
1729 Current.SpacesRequiredBefore;
1730 }
1731 if (Current.is(Kind: tok::comment) && NextNonComment->isBinaryOperator() &&
1732 CurrentState.UnindentOperator) {
1733 return CurrentState.Indent - NextNonComment->Tok.getLength() -
1734 NextNonComment->SpacesRequiredBefore;
1735 }
1736 if (CurrentState.Indent.Total == State.FirstIndent && PreviousNonComment &&
1737 PreviousNonComment->isNoneOf(Ks: tok::r_brace, Ks: TT_CtorInitializerComma)) {
1738 // Ensure that we fall back to the continuation indent width instead of
1739 // just flushing continuations left.
1740 return CurrentState.Indent + Style.ContinuationIndentWidth;
1741 }
1742 return CurrentState.Indent;
1743}
1744
1745static bool hasNestedBlockInlined(const FormatToken *Previous,
1746 const FormatToken &Current,
1747 const FormatStyle &Style) {
1748 if (Previous->isNot(Kind: tok::l_paren))
1749 return true;
1750 if (Previous->ParameterCount > 1)
1751 return true;
1752
1753 // Also a nested block if contains a lambda inside function with 1 parameter.
1754 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT: TT_LambdaLSquare);
1755}
1756
1757unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1758 bool DryRun, bool Newline) {
1759 assert(State.Stack.size());
1760 const FormatToken &Current = *State.NextToken;
1761 auto &CurrentState = State.Stack.back();
1762
1763 if (Current.is(TT: TT_CSharpGenericTypeConstraint))
1764 CurrentState.IsCSharpGenericTypeConstraint = true;
1765 if (Current.isOneOf(K1: tok::comma, K2: TT_BinaryOperator))
1766 CurrentState.NoLineBreakInOperand = false;
1767 if (Current.isOneOf(K1: TT_InheritanceColon, K2: TT_CSharpGenericTypeConstraintColon))
1768 CurrentState.AvoidBinPacking = true;
1769 if (Current.is(Kind: tok::lessless) && Current.isNot(Kind: TT_OverloadedOperator)) {
1770 if (CurrentState.FirstLessLess == 0)
1771 CurrentState.FirstLessLess = State.Column;
1772 else
1773 CurrentState.LastOperatorWrapped = Newline;
1774 }
1775 if (Current.is(TT: TT_BinaryOperator) && Current.isNot(Kind: tok::lessless))
1776 CurrentState.LastOperatorWrapped = Newline;
1777 if (Current.is(TT: TT_ConditionalExpr) && Current.Previous &&
1778 Current.Previous->isNot(Kind: TT_ConditionalExpr)) {
1779 CurrentState.LastOperatorWrapped = Newline;
1780 }
1781 if (Current.is(TT: TT_ArraySubscriptLSquare) &&
1782 CurrentState.StartOfArraySubscripts == 0) {
1783 CurrentState.StartOfArraySubscripts = State.Column;
1784 }
1785
1786 auto IsWrappedConditional = [](const FormatToken &Tok) {
1787 if (!(Tok.is(TT: TT_ConditionalExpr) && Tok.is(Kind: tok::question)))
1788 return false;
1789 if (Tok.MustBreakBefore)
1790 return true;
1791
1792 const FormatToken *Next = Tok.getNextNonComment();
1793 return Next && Next->MustBreakBefore;
1794 };
1795 if (IsWrappedConditional(Current))
1796 CurrentState.IsWrappedConditional = true;
1797 if (Style.BreakBeforeTernaryOperators && Current.is(Kind: tok::question))
1798 CurrentState.QuestionColumn = State.Column;
1799 if (!Style.BreakBeforeTernaryOperators && Current.isNot(Kind: tok::colon)) {
1800 const FormatToken *Previous = Current.Previous;
1801 while (Previous && Previous->isTrailingComment())
1802 Previous = Previous->Previous;
1803 if (Previous && Previous->is(Kind: tok::question))
1804 CurrentState.QuestionColumn = State.Column;
1805 }
1806 if (!Current.opensScope() && !Current.closesScope() &&
1807 Current.isNot(Kind: TT_PointerOrReference)) {
1808 State.LowestLevelOnLine =
1809 std::min(a: State.LowestLevelOnLine, b: Current.NestingLevel);
1810 }
1811 if (Current.isMemberAccess())
1812 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1813 if (Current.is(TT: TT_SelectorName))
1814 CurrentState.ObjCSelectorNameFound = true;
1815 if (Current.is(TT: TT_CtorInitializerColon) &&
1816 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1817 // Indent 2 from the column, so:
1818 // SomeClass::SomeClass()
1819 // : First(...), ...
1820 // Next(...)
1821 // ^ line up here.
1822 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1823 FormatStyle::BCIS_BeforeComma
1824 ? 0
1825 : 2);
1826 CurrentState.NestedBlockIndent = CurrentState.Indent.Total;
1827 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) {
1828 CurrentState.AvoidBinPacking = true;
1829 CurrentState.BreakBeforeParameter =
1830 Style.ColumnLimit > 0 &&
1831 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine &&
1832 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLineOnly;
1833 } else {
1834 CurrentState.BreakBeforeParameter = false;
1835 }
1836 }
1837 if (Current.is(TT: TT_CtorInitializerColon) &&
1838 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1839 CurrentState.Indent =
1840 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1841 CurrentState.NestedBlockIndent = CurrentState.Indent.Total;
1842 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack)
1843 CurrentState.AvoidBinPacking = true;
1844 else
1845 CurrentState.BreakBeforeParameter = false;
1846 }
1847 if (Current.is(TT: TT_InheritanceColon)) {
1848 CurrentState.Indent =
1849 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1850 }
1851 if (Current.isOneOf(K1: TT_BinaryOperator, K2: TT_ConditionalExpr) && Newline)
1852 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1853 if (Current.isOneOf(K1: TT_LambdaLSquare, K2: TT_LambdaArrow))
1854 CurrentState.LastSpace = State.Column;
1855 if (Current.is(TT: TT_RequiresExpression) &&
1856 Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) {
1857 CurrentState.NestedBlockIndent = State.Column;
1858 }
1859
1860 // Insert scopes created by fake parenthesis.
1861 const FormatToken *Previous = Current.getPreviousNonComment();
1862
1863 // Add special behavior to support a format commonly used for JavaScript
1864 // closures:
1865 // SomeFunction(function() {
1866 // foo();
1867 // bar();
1868 // }, a, b, c);
1869 if (Current.isNot(Kind: tok::comment) && !Current.ClosesRequiresClause &&
1870 Previous && Previous->isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) &&
1871 Previous->isNot(Kind: TT_DictLiteral) && State.Stack.size() > 1 &&
1872 !CurrentState.HasMultipleNestedBlocks) {
1873 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1874 for (ParenState &PState : llvm::drop_end(RangeOrContainer&: State.Stack))
1875 PState.NoLineBreak = true;
1876 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1877 }
1878 if (Previous && (Previous->isOneOf(K1: TT_BinaryOperator, K2: TT_ConditionalExpr) ||
1879 (Previous->isOneOf(K1: tok::l_paren, K2: tok::comma, Ks: tok::colon) &&
1880 Previous->isNoneOf(Ks: TT_DictLiteral, Ks: TT_ObjCMethodExpr,
1881 Ks: TT_CtorInitializerColon)))) {
1882 CurrentState.NestedBlockInlined =
1883 !Newline && hasNestedBlockInlined(Previous, Current, Style);
1884 }
1885
1886 moveStatePastFakeLParens(State, Newline);
1887 moveStatePastScopeCloser(State);
1888 // Do not use CurrentState here, since the two functions before may change the
1889 // Stack.
1890 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1891 !State.Stack.back().NoLineBreakInOperand;
1892 moveStatePastScopeOpener(State, Newline);
1893 moveStatePastFakeRParens(State);
1894
1895 if (Current.is(TT: TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1896 State.StartOfStringLiteral = State.Column + 1;
1897 if (Current.is(TT: TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1898 State.StartOfStringLiteral = State.Column + 1;
1899 } else if (Current.is(TT: TT_TableGenMultiLineString) &&
1900 State.StartOfStringLiteral == 0) {
1901 State.StartOfStringLiteral = State.Column + 1;
1902 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1903 State.StartOfStringLiteral = State.Column;
1904 } else if (Current.isNoneOf(Ks: tok::comment, Ks: tok::identifier, Ks: tok::hash) &&
1905 !Current.isStringLiteral()) {
1906 State.StartOfStringLiteral = 0;
1907 }
1908
1909 State.Column += Current.ColumnWidth;
1910 State.NextToken = State.NextToken->Next;
1911 // Verilog case labels are on the same unwrapped lines as the statements that
1912 // follow. TokenAnnotator identifies them and sets MustBreakBefore.
1913 // Indentation is taken care of here. A case label can only have 1 statement
1914 // in Verilog, so we don't have to worry about lines that follow.
1915 if (Style.isVerilog() && State.NextToken &&
1916 State.NextToken->MustBreakBefore &&
1917 Keywords.isVerilogEndOfLabel(Tok: Current)) {
1918 State.FirstIndent += Style.IndentWidth;
1919 CurrentState.Indent = State.FirstIndent;
1920 }
1921
1922 unsigned Penalty =
1923 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1924
1925 if (Current.Role)
1926 Current.Role->formatFromToken(State, Indenter: this, DryRun);
1927 // If the previous has a special role, let it consume tokens as appropriate.
1928 // It is necessary to start at the previous token for the only implemented
1929 // role (comma separated list). That way, the decision whether or not to break
1930 // after the "{" is already done and both options are tried and evaluated.
1931 // FIXME: This is ugly, find a better way.
1932 if (Previous && Previous->Role)
1933 Penalty += Previous->Role->formatAfterToken(State, Indenter: this, DryRun);
1934
1935 return Penalty;
1936}
1937
1938void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1939 bool Newline) {
1940 const FormatToken &Current = *State.NextToken;
1941 if (Current.FakeLParens.empty())
1942 return;
1943
1944 const FormatToken *Previous = Current.getPreviousNonComment();
1945
1946 // Don't add extra indentation for the first fake parenthesis after
1947 // 'return', assignments, opening <({[, or requires clauses. The indentation
1948 // for these cases is special cased.
1949 bool SkipFirstExtraIndent =
1950 Previous &&
1951 (Previous->opensScope() ||
1952 Previous->isOneOf(K1: tok::semi, K2: tok::kw_return, Ks: TT_RequiresClause) ||
1953 (Previous->getPrecedence() == prec::Assignment &&
1954 Style.AlignOperands != FormatStyle::OAS_DontAlign) ||
1955 Previous->is(TT: TT_ObjCMethodExpr));
1956 for (const auto &PrecedenceLevel : llvm::reverse(C: Current.FakeLParens)) {
1957 const auto &CurrentState = State.Stack.back();
1958 ParenState NewParenState = CurrentState;
1959 NewParenState.Tok = nullptr;
1960 NewParenState.ContainsLineBreak = false;
1961 NewParenState.LastOperatorWrapped = true;
1962 NewParenState.IsChainedConditional = false;
1963 NewParenState.IsWrappedConditional = false;
1964 NewParenState.UnindentOperator = false;
1965 NewParenState.NoLineBreak =
1966 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
1967
1968 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1969 if (PrecedenceLevel > prec::Comma)
1970 NewParenState.AvoidBinPacking = false;
1971
1972 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1973 // a builder type call after 'return' or, if the alignment after opening
1974 // brackets is disabled.
1975 if (!Current.isTrailingComment() &&
1976 (Style.AlignOperands != FormatStyle::OAS_DontAlign ||
1977 PrecedenceLevel < prec::Assignment) &&
1978 (!Previous || Previous->isNot(Kind: tok::kw_return) ||
1979 (!Style.isJava() && PrecedenceLevel > 0)) &&
1980 (Style.AlignAfterOpenBracket || PrecedenceLevel > prec::Comma ||
1981 Current.NestingLevel == 0) &&
1982 (!Style.isTableGen() ||
1983 (Previous && Previous->isOneOf(K1: TT_TableGenDAGArgListComma,
1984 K2: TT_TableGenDAGArgListCommaToBreak)))) {
1985 NewParenState.Indent =
1986 std::max(l: {IndentationAndAlignment(State.Column), NewParenState.Indent,
1987 IndentationAndAlignment(CurrentState.LastSpace)});
1988 }
1989
1990 // Special case for generic selection expressions, its comma-separated
1991 // expressions are not aligned to the opening paren like regular calls, but
1992 // rather continuation-indented relative to the _Generic keyword.
1993 if (Previous && Previous->endsSequence(K1: tok::l_paren, Tokens: tok::kw__Generic) &&
1994 State.Stack.size() > 1) {
1995 NewParenState.Indent = State.Stack[State.Stack.size() - 2].Indent +
1996 Style.ContinuationIndentWidth;
1997 }
1998
1999 if ((shouldUnindentNextOperator(Tok: Current) ||
2000 (Previous &&
2001 (PrecedenceLevel == prec::Conditional &&
2002 Previous->is(Kind: tok::question) && Previous->is(TT: TT_ConditionalExpr)))) &&
2003 !Newline) {
2004 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
2005 // the operator and keep the operands aligned.
2006 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator)
2007 NewParenState.UnindentOperator = true;
2008 // Mark indentation as alignment if the expression is aligned.
2009 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
2010 NewParenState.IsAligned = true;
2011 }
2012
2013 // Do not indent relative to the fake parentheses inserted for "." or "->".
2014 // This is a special case to make the following to statements consistent:
2015 // OuterFunction(InnerFunctionCall( // break
2016 // ParameterToInnerFunction));
2017 // OuterFunction(SomeObject.InnerFunctionCall( // break
2018 // ParameterToInnerFunction));
2019 if (PrecedenceLevel > prec::Unknown)
2020 NewParenState.LastSpace = std::max(a: NewParenState.LastSpace, b: State.Column);
2021 if (PrecedenceLevel != prec::Conditional &&
2022 Current.isNot(Kind: TT_UnaryOperator) && Style.AlignAfterOpenBracket) {
2023 NewParenState.StartOfFunctionCall = State.Column;
2024 }
2025
2026 // Indent conditional expressions, unless they are chained "else-if"
2027 // conditionals. Never indent expression where the 'operator' is ',', ';' or
2028 // an assignment (i.e. *I <= prec::Assignment) as those have different
2029 // indentation rules. Indent other expression, unless the indentation needs
2030 // to be skipped.
2031 if (PrecedenceLevel == prec::Conditional && Previous &&
2032 Previous->is(Kind: tok::colon) && Previous->is(TT: TT_ConditionalExpr) &&
2033 &PrecedenceLevel == &Current.FakeLParens.back() &&
2034 !CurrentState.IsWrappedConditional) {
2035 NewParenState.IsChainedConditional = true;
2036 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
2037 } else if (PrecedenceLevel == prec::Conditional ||
2038 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
2039 !Current.isTrailingComment())) {
2040 NewParenState.Indent += Style.ContinuationIndentWidth;
2041 }
2042 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
2043 NewParenState.BreakBeforeParameter = false;
2044 State.Stack.push_back(Elt: NewParenState);
2045 SkipFirstExtraIndent = false;
2046 }
2047}
2048
2049void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
2050 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
2051 unsigned VariablePos = State.Stack.back().VariablePos;
2052 if (State.Stack.size() == 1) {
2053 // Do not pop the last element.
2054 break;
2055 }
2056 State.Stack.pop_back();
2057 State.Stack.back().VariablePos = VariablePos;
2058 }
2059
2060 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
2061 // Remove the indentation of the requires clauses (which is not in Indent,
2062 // but in LastSpace).
2063 State.Stack.back().LastSpace -= Style.IndentWidth;
2064 }
2065}
2066
2067void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
2068 bool Newline) {
2069 const FormatToken &Current = *State.NextToken;
2070 if (!Current.opensScope())
2071 return;
2072
2073 const auto &CurrentState = State.Stack.back();
2074
2075 // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
2076 if (Current.isOneOf(K1: tok::less, K2: tok::l_paren) &&
2077 CurrentState.IsCSharpGenericTypeConstraint) {
2078 return;
2079 }
2080
2081 if (Current.MatchingParen && Current.is(BBK: BK_Block)) {
2082 moveStateToNewBlock(State, NewLine: Newline);
2083 return;
2084 }
2085
2086 const bool EndsInComma = [](const FormatToken *Tok) {
2087 if (!Tok)
2088 return false;
2089 const auto *Prev = Tok->getPreviousNonComment();
2090 if (!Prev)
2091 return false;
2092 return Prev->is(Kind: tok::comma);
2093 }(Current.MatchingParen);
2094
2095 IndentationAndAlignment NewIndent = 0;
2096 unsigned LastSpace = CurrentState.LastSpace;
2097 bool AvoidBinPacking;
2098 bool BreakBeforeParameter = false;
2099 unsigned NestedBlockIndent = std::max(a: CurrentState.StartOfFunctionCall,
2100 b: CurrentState.NestedBlockIndent);
2101 if (Current.isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) ||
2102 opensProtoMessageField(LessTok: Current, Style)) {
2103 if (Current.opensBlockOrBlockTypeList(Style)) {
2104 NewIndent = Style.IndentWidth +
2105 std::min(a: State.Column, b: CurrentState.NestedBlockIndent);
2106 } else if (Current.is(Kind: tok::l_brace)) {
2107 const auto Width = Style.BracedInitializerIndentWidth;
2108 NewIndent = IndentationAndAlignment(CurrentState.LastSpace) +
2109 (Width < 0 ? Style.ContinuationIndentWidth : Width);
2110 } else {
2111 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
2112 }
2113 const FormatToken *NextNonComment = Current.getNextNonComment();
2114 AvoidBinPacking = EndsInComma || Current.is(TT: TT_DictLiteral) ||
2115 Style.isProto() || !Style.BinPackArguments ||
2116 (NextNonComment && NextNonComment->isOneOf(
2117 K1: TT_DesignatedInitializerPeriod,
2118 K2: TT_DesignatedInitializerLSquare));
2119 BreakBeforeParameter = EndsInComma;
2120 if (Current.ParameterCount > 1)
2121 NestedBlockIndent = std::max(a: NestedBlockIndent, b: State.Column + 1);
2122 } else {
2123 NewIndent = IndentationAndAlignment(std::max(
2124 a: CurrentState.LastSpace, b: CurrentState.StartOfFunctionCall)) +
2125 Style.ContinuationIndentWidth;
2126
2127 if (Style.isTableGen() && Current.is(TT: TT_TableGenDAGArgOpenerToBreak) &&
2128 Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakElements) {
2129 // For the case the next token is a TableGen DAGArg operator identifier
2130 // that is not marked to have a line break after it.
2131 // In this case the option DAS_BreakElements requires to align the
2132 // DAGArg elements to the operator.
2133 const FormatToken *Next = Current.Next;
2134 if (Next && Next->is(TT: TT_TableGenDAGArgOperatorID))
2135 NewIndent = State.Column + Next->TokenText.size() + 2;
2136 }
2137
2138 // Ensure that different different brackets force relative alignment, e.g.:
2139 // void SomeFunction(vector< // break
2140 // int> v);
2141 // FIXME: We likely want to do this for more combinations of brackets.
2142 if (Current.is(Kind: tok::less) && Current.ParentBracket == tok::l_paren) {
2143 NewIndent = std::max(a: NewIndent, b: CurrentState.Indent);
2144 LastSpace = std::max(a: LastSpace, b: CurrentState.Indent.Total);
2145 }
2146
2147 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
2148 // for backwards compatibility.
2149 bool ObjCBinPackProtocolList =
2150 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
2151 Style.BinPackParameters == FormatStyle::BPPS_BinPack) ||
2152 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
2153
2154 bool BinPackDeclaration =
2155 (State.Line->Type != LT_ObjCDecl &&
2156 Style.BinPackParameters == FormatStyle::BPPS_BinPack) ||
2157 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
2158
2159 bool GenericSelection =
2160 Current.getPreviousNonComment() &&
2161 Current.getPreviousNonComment()->is(Kind: tok::kw__Generic);
2162
2163 AvoidBinPacking =
2164 (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
2165 (Style.isJavaScript() && EndsInComma) ||
2166 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
2167 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
2168 (Style.ExperimentalAutoDetectBinPacking &&
2169 (Current.is(PPK: PPK_OnePerLine) ||
2170 (!BinPackInconclusiveFunctions && Current.is(PPK: PPK_Inconclusive))));
2171
2172 if (Current.is(TT: TT_ObjCMethodExpr) && Current.MatchingParen &&
2173 Style.ObjCBreakBeforeNestedBlockParam) {
2174 if (Style.ColumnLimit) {
2175 // If this '[' opens an ObjC call, determine whether all parameters fit
2176 // into one line and put one per line if they don't.
2177 if (getLengthToMatchingParen(Tok: Current, Stack: State.Stack) + State.Column >
2178 getColumnLimit(State)) {
2179 BreakBeforeParameter = true;
2180 }
2181 } else {
2182 // For ColumnLimit = 0, we have to figure out whether there is or has to
2183 // be a line break within this call.
2184 for (const FormatToken *Tok = &Current;
2185 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
2186 if (Tok->MustBreakBefore ||
2187 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
2188 BreakBeforeParameter = true;
2189 break;
2190 }
2191 }
2192 }
2193 }
2194
2195 if (Style.isJavaScript() && EndsInComma)
2196 BreakBeforeParameter = true;
2197 }
2198 // Generally inherit NoLineBreak from the current scope to nested scope.
2199 // However, don't do this for non-empty nested blocks, dict literals and
2200 // array literals as these follow different indentation rules.
2201 bool NoLineBreak =
2202 Current.Children.empty() &&
2203 Current.isNoneOf(Ks: TT_DictLiteral, Ks: TT_ArrayInitializerLSquare) &&
2204 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
2205 (Current.is(TT: TT_TemplateOpener) &&
2206 CurrentState.ContainsUnwrappedBuilder));
2207 State.Stack.push_back(
2208 Elt: ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
2209 auto &NewState = State.Stack.back();
2210 NewState.NestedBlockIndent = NestedBlockIndent;
2211 NewState.BreakBeforeParameter = BreakBeforeParameter;
2212 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
2213
2214 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next &&
2215 Current.is(Kind: tok::l_paren)) {
2216 // Search for any parameter that is a lambda.
2217 FormatToken const *next = Current.Next;
2218 while (next) {
2219 if (next->is(TT: TT_LambdaLSquare)) {
2220 NewState.HasMultipleNestedBlocks = true;
2221 break;
2222 }
2223 next = next->Next;
2224 }
2225 }
2226
2227 NewState.IsInsideObjCArrayLiteral = Current.is(TT: TT_ArrayInitializerLSquare) &&
2228 Current.Previous &&
2229 Current.Previous->is(Kind: tok::at);
2230}
2231
2232void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
2233 const FormatToken &Current = *State.NextToken;
2234 if (!Current.closesScope())
2235 return;
2236
2237 // If we encounter a closing ), ], } or >, we can remove a level from our
2238 // stacks.
2239 if (State.Stack.size() > 1 &&
2240 (Current.isOneOf(K1: tok::r_paren, K2: tok::r_square, Ks: TT_TemplateString) ||
2241 (Current.is(Kind: tok::r_brace) && State.NextToken != State.Line->First) ||
2242 State.NextToken->is(TT: TT_TemplateCloser) ||
2243 State.NextToken->is(TT: TT_TableGenListCloser) ||
2244 (Current.is(Kind: tok::greater) && Current.is(TT: TT_DictLiteral)))) {
2245 State.Stack.pop_back();
2246 }
2247
2248 auto &CurrentState = State.Stack.back();
2249
2250 // Reevaluate whether ObjC message arguments fit into one line.
2251 // If a receiver spans multiple lines, e.g.:
2252 // [[object block:^{
2253 // return 42;
2254 // }] a:42 b:42];
2255 // BreakBeforeParameter is calculated based on an incorrect assumption
2256 // (it is checked whether the whole expression fits into one line without
2257 // considering a line break inside a message receiver).
2258 // We check whether arguments fit after receiver scope closer (into the same
2259 // line).
2260 if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
2261 Current.MatchingParen->Previous) {
2262 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
2263 if (CurrentScopeOpener.is(TT: TT_ObjCMethodExpr) &&
2264 CurrentScopeOpener.MatchingParen) {
2265 int NecessarySpaceInLine =
2266 getLengthToMatchingParen(Tok: CurrentScopeOpener, Stack: State.Stack) +
2267 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
2268 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
2269 Style.ColumnLimit) {
2270 CurrentState.BreakBeforeParameter = false;
2271 }
2272 }
2273 }
2274
2275 if (Current.is(Kind: tok::r_square)) {
2276 // If this ends the array subscript expr, reset the corresponding value.
2277 const FormatToken *NextNonComment = Current.getNextNonComment();
2278 if (NextNonComment && NextNonComment->isNot(Kind: tok::l_square))
2279 CurrentState.StartOfArraySubscripts = 0;
2280 }
2281}
2282
2283void ContinuationIndenter::moveStateToNewBlock(LineState &State, bool NewLine) {
2284 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
2285 State.NextToken->is(TT: TT_LambdaLBrace) &&
2286 !State.Line->MightBeFunctionDecl) {
2287 const auto Indent = Style.IndentWidth * Style.BraceWrapping.IndentBraces;
2288 State.Stack.back().NestedBlockIndent = State.FirstIndent + Indent;
2289 }
2290 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
2291 // ObjC block sometimes follow special indentation rules.
2292 unsigned NewIndent =
2293 NestedBlockIndent + (State.NextToken->is(TT: TT_ObjCBlockLBrace)
2294 ? Style.ObjCBlockIndentWidth
2295 : Style.IndentWidth);
2296
2297 // Even when wrapping before lambda body, the left brace can still be added to
2298 // the same line. This occurs when checking whether the whole lambda body can
2299 // go on a single line. In this case we have to make sure there are no line
2300 // breaks in the body, otherwise we could just end up with a regular lambda
2301 // body without the brace wrapped.
2302 bool NoLineBreak = Style.BraceWrapping.BeforeLambdaBody && !NewLine &&
2303 State.NextToken->is(TT: TT_LambdaLBrace);
2304
2305 State.Stack.push_back(Elt: ParenState(State.NextToken, NewIndent,
2306 State.Stack.back().LastSpace,
2307 /*AvoidBinPacking=*/true, NoLineBreak));
2308 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
2309 State.Stack.back().BreakBeforeParameter = true;
2310}
2311
2312static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
2313 unsigned TabWidth,
2314 encoding::Encoding Encoding) {
2315 size_t LastNewlinePos = Text.find_last_of(Chars: "\n");
2316 if (LastNewlinePos == StringRef::npos) {
2317 return StartColumn +
2318 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
2319 } else {
2320 return encoding::columnWidthWithTabs(Text: Text.substr(Start: LastNewlinePos),
2321 /*StartColumn=*/0, TabWidth, Encoding);
2322 }
2323}
2324
2325unsigned ContinuationIndenter::reformatRawStringLiteral(
2326 const FormatToken &Current, LineState &State,
2327 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
2328 unsigned StartColumn = State.Column - Current.ColumnWidth;
2329 StringRef OldDelimiter = *getRawStringDelimiter(TokenText: Current.TokenText);
2330 StringRef NewDelimiter =
2331 getCanonicalRawStringDelimiter(Style, Language: RawStringStyle.Language);
2332 if (NewDelimiter.empty())
2333 NewDelimiter = OldDelimiter;
2334 // The text of a raw string is between the leading 'R"delimiter(' and the
2335 // trailing 'delimiter)"'.
2336 unsigned OldPrefixSize = 3 + OldDelimiter.size();
2337 unsigned OldSuffixSize = 2 + OldDelimiter.size();
2338 // We create a virtual text environment which expects a null-terminated
2339 // string, so we cannot use StringRef.
2340 std::string RawText = std::string(
2341 Current.TokenText.substr(Start: OldPrefixSize).drop_back(N: OldSuffixSize));
2342 if (NewDelimiter != OldDelimiter) {
2343 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
2344 // raw string.
2345 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
2346 if (StringRef(RawText).contains(Other: CanonicalDelimiterSuffix))
2347 NewDelimiter = OldDelimiter;
2348 }
2349
2350 unsigned NewPrefixSize = 3 + NewDelimiter.size();
2351 unsigned NewSuffixSize = 2 + NewDelimiter.size();
2352
2353 // The first start column is the column the raw text starts after formatting.
2354 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
2355
2356 // The next start column is the intended indentation a line break inside
2357 // the raw string at level 0. It is determined by the following rules:
2358 // - if the content starts on newline, it is one level more than the current
2359 // indent, and
2360 // - if the content does not start on a newline, it is the first start
2361 // column.
2362 // These rules have the advantage that the formatted content both does not
2363 // violate the rectangle rule and visually flows within the surrounding
2364 // source.
2365 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
2366 // If this token is the last parameter (checked by looking if it's followed by
2367 // `)` and is not on a newline, the base the indent off the line's nested
2368 // block indent. Otherwise, base the indent off the arguments indent, so we
2369 // can achieve:
2370 //
2371 // fffffffffff(1, 2, 3, R"pb(
2372 // key1: 1 #
2373 // key2: 2)pb");
2374 //
2375 // fffffffffff(1, 2, 3,
2376 // R"pb(
2377 // key1: 1 #
2378 // key2: 2
2379 // )pb");
2380 //
2381 // fffffffffff(1, 2, 3,
2382 // R"pb(
2383 // key1: 1 #
2384 // key2: 2
2385 // )pb",
2386 // 5);
2387 unsigned CurrentIndent =
2388 (!Newline && Current.Next && Current.Next->is(Kind: tok::r_paren))
2389 ? State.Stack.back().NestedBlockIndent
2390 : State.Stack.back().Indent.Total;
2391 unsigned NextStartColumn = ContentStartsOnNewline
2392 ? CurrentIndent + Style.IndentWidth
2393 : FirstStartColumn;
2394
2395 // The last start column is the column the raw string suffix starts if it is
2396 // put on a newline.
2397 // The last start column is the intended indentation of the raw string postfix
2398 // if it is put on a newline. It is determined by the following rules:
2399 // - if the raw string prefix starts on a newline, it is the column where
2400 // that raw string prefix starts, and
2401 // - if the raw string prefix does not start on a newline, it is the current
2402 // indent.
2403 unsigned LastStartColumn =
2404 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
2405
2406 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
2407 Style: RawStringStyle, Code: RawText, Ranges: {tooling::Range(0, RawText.size())},
2408 FirstStartColumn, NextStartColumn, LastStartColumn, FileName: "<stdin>",
2409 /*Status=*/nullptr);
2410
2411 auto NewCode = applyAllReplacements(Code: RawText, Replaces: Fixes.first);
2412 if (!NewCode)
2413 return addMultilineToken(Current, State);
2414 if (!DryRun) {
2415 if (NewDelimiter != OldDelimiter) {
2416 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
2417 // of the token.
2418 SourceLocation PrefixDelimiterStart =
2419 Current.Tok.getLocation().getLocWithOffset(Offset: 2);
2420 auto PrefixErr = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2421 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2422 if (PrefixErr) {
2423 llvm::errs()
2424 << "Failed to update the prefix delimiter of a raw string: "
2425 << llvm::toString(E: std::move(PrefixErr)) << "\n";
2426 }
2427 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
2428 // position length - 1 - |delimiter|.
2429 SourceLocation SuffixDelimiterStart =
2430 Current.Tok.getLocation().getLocWithOffset(Offset: Current.TokenText.size() -
2431 1 - OldDelimiter.size());
2432 auto SuffixErr = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2433 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2434 if (SuffixErr) {
2435 llvm::errs()
2436 << "Failed to update the suffix delimiter of a raw string: "
2437 << llvm::toString(E: std::move(SuffixErr)) << "\n";
2438 }
2439 }
2440 SourceLocation OriginLoc =
2441 Current.Tok.getLocation().getLocWithOffset(Offset: OldPrefixSize);
2442 for (const tooling::Replacement &Fix : Fixes.first) {
2443 auto Err = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2444 SourceMgr, OriginLoc.getLocWithOffset(Offset: Fix.getOffset()),
2445 Fix.getLength(), Fix.getReplacementText()));
2446 if (Err) {
2447 llvm::errs() << "Failed to reformat raw string: "
2448 << llvm::toString(E: std::move(Err)) << "\n";
2449 }
2450 }
2451 }
2452 unsigned RawLastLineEndColumn = getLastLineEndColumn(
2453 Text: *NewCode, StartColumn: FirstStartColumn, TabWidth: Style.TabWidth, Encoding);
2454 State.Column = RawLastLineEndColumn + NewSuffixSize;
2455 // Since we're updating the column to after the raw string literal here, we
2456 // have to manually add the penalty for the prefix R"delim( over the column
2457 // limit.
2458 unsigned PrefixExcessCharacters =
2459 StartColumn + NewPrefixSize > Style.ColumnLimit
2460 ? StartColumn + NewPrefixSize - Style.ColumnLimit
2461 : 0;
2462 bool IsMultiline =
2463 ContentStartsOnNewline || (NewCode->find(c: '\n') != std::string::npos);
2464 if (IsMultiline) {
2465 // Break before further function parameters on all levels.
2466 for (ParenState &Paren : State.Stack)
2467 Paren.BreakBeforeParameter = true;
2468 }
2469 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
2470}
2471
2472unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
2473 LineState &State) {
2474 // Break before further function parameters on all levels.
2475 for (ParenState &Paren : State.Stack)
2476 Paren.BreakBeforeParameter = true;
2477
2478 unsigned ColumnsUsed = State.Column;
2479 // We can only affect layout of the first and the last line, so the penalty
2480 // for all other lines is constant, and we ignore it.
2481 State.Column = Current.LastLineColumnWidth;
2482
2483 if (ColumnsUsed > getColumnLimit(State))
2484 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2485 return 0;
2486}
2487
2488unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2489 LineState &State, bool DryRun,
2490 bool AllowBreak, bool Newline) {
2491 unsigned Penalty = 0;
2492 // Compute the raw string style to use in case this is a raw string literal
2493 // that can be reformatted.
2494 auto RawStringStyle = getRawStringStyle(Current, State);
2495 if (RawStringStyle && !Current.Finalized) {
2496 Penalty = reformatRawStringLiteral(Current, State, RawStringStyle: *RawStringStyle, DryRun,
2497 Newline);
2498 } else if (Current.IsMultiline && Current.isNot(Kind: TT_BlockComment)) {
2499 // Don't break multi-line tokens other than block comments and raw string
2500 // literals. Instead, just update the state.
2501 Penalty = addMultilineToken(Current, State);
2502 } else if (State.Line->Type != LT_ImportStatement) {
2503 // We generally don't break import statements.
2504 LineState OriginalState = State;
2505
2506 // Whether we force the reflowing algorithm to stay strictly within the
2507 // column limit.
2508 bool Strict = false;
2509 // Whether the first non-strict attempt at reflowing did intentionally
2510 // exceed the column limit.
2511 bool Exceeded = false;
2512 std::tie(args&: Penalty, args&: Exceeded) = breakProtrudingToken(
2513 Current, State, AllowBreak, /*DryRun=*/true, Strict);
2514 if (Exceeded) {
2515 // If non-strict reflowing exceeds the column limit, try whether strict
2516 // reflowing leads to an overall lower penalty.
2517 LineState StrictState = OriginalState;
2518 unsigned StrictPenalty =
2519 breakProtrudingToken(Current, State&: StrictState, AllowBreak,
2520 /*DryRun=*/true, /*Strict=*/true)
2521 .first;
2522 Strict = StrictPenalty <= Penalty;
2523 if (Strict) {
2524 Penalty = StrictPenalty;
2525 State = std::move(StrictState);
2526 }
2527 }
2528 if (!DryRun) {
2529 // If we're not in dry-run mode, apply the changes with the decision on
2530 // strictness made above.
2531 breakProtrudingToken(Current, State&: OriginalState, AllowBreak, /*DryRun=*/false,
2532 Strict);
2533 }
2534 }
2535 if (State.Column > getColumnLimit(State)) {
2536 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2537 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2538 }
2539 return Penalty;
2540}
2541
2542// Returns the enclosing function name of a token, or the empty string if not
2543// found.
2544static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2545 // Look for: 'function(' or 'function<templates>(' before Current.
2546 auto Tok = Current.getPreviousNonComment();
2547 if (!Tok || Tok->isNot(Kind: tok::l_paren))
2548 return "";
2549 Tok = Tok->getPreviousNonComment();
2550 if (!Tok)
2551 return "";
2552 if (Tok->is(TT: TT_TemplateCloser)) {
2553 Tok = Tok->MatchingParen;
2554 if (Tok)
2555 Tok = Tok->getPreviousNonComment();
2556 }
2557 if (!Tok || Tok->isNot(Kind: tok::identifier))
2558 return "";
2559 return Tok->TokenText;
2560}
2561
2562std::optional<FormatStyle>
2563ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2564 const LineState &State) {
2565 if (!Current.isStringLiteral())
2566 return std::nullopt;
2567 auto Delimiter = getRawStringDelimiter(TokenText: Current.TokenText);
2568 if (!Delimiter)
2569 return std::nullopt;
2570 auto RawStringStyle = RawStringFormats.getDelimiterStyle(Delimiter: *Delimiter);
2571 if (!RawStringStyle && Delimiter->empty()) {
2572 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2573 EnclosingFunction: getEnclosingFunctionName(Current));
2574 }
2575 if (!RawStringStyle)
2576 return std::nullopt;
2577 RawStringStyle->ColumnLimit = getColumnLimit(State);
2578 return RawStringStyle;
2579}
2580
2581std::unique_ptr<BreakableToken>
2582ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2583 LineState &State, bool AllowBreak) {
2584 unsigned StartColumn = State.Column - Current.ColumnWidth;
2585 if (Current.isStringLiteral()) {
2586 // Strings in JSON cannot be broken. Breaking strings in JavaScript is
2587 // disabled for now.
2588 if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals ||
2589 !AllowBreak) {
2590 return nullptr;
2591 }
2592
2593 // Don't break string literals inside preprocessor directives (except for
2594 // #define directives, as their contents are stored in separate lines and
2595 // are not affected by this check).
2596 // This way we avoid breaking code with line directives and unknown
2597 // preprocessor directives that contain long string literals.
2598 if (State.Line->Type == LT_PreprocessorDirective)
2599 return nullptr;
2600 // Exempts unterminated string literals from line breaking. The user will
2601 // likely want to terminate the string before any line breaking is done.
2602 if (Current.IsUnterminatedLiteral)
2603 return nullptr;
2604 // Don't break string literals inside Objective-C array literals (doing so
2605 // raises the warning -Wobjc-string-concatenation).
2606 if (State.Stack.back().IsInsideObjCArrayLiteral)
2607 return nullptr;
2608
2609 // The "DPI"/"DPI-C" in SystemVerilog direct programming interface
2610 // imports/exports cannot be split, e.g.
2611 // `import "DPI" function foo();`
2612 // FIXME: make this use same infra as C++ import checks
2613 if (Style.isVerilog() && Current.Previous &&
2614 Current.Previous->isOneOf(K1: tok::kw_export, K2: Keywords.kw_import)) {
2615 return nullptr;
2616 }
2617 StringRef Text = Current.TokenText;
2618
2619 // We need this to address the case where there is an unbreakable tail only
2620 // if certain other formatting decisions have been taken. The
2621 // UnbreakableTailLength of Current is an overapproximation in that case and
2622 // we need to be correct here.
2623 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2624 ? 0
2625 : Current.UnbreakableTailLength;
2626
2627 if (Style.isVerilog() || Style.isJava() || Style.isJavaScript() ||
2628 Style.isCSharp()) {
2629 BreakableStringLiteralUsingOperators::QuoteStyleType QuoteStyle;
2630 if (Style.isJavaScript() && Text.starts_with(Prefix: "'") &&
2631 Text.ends_with(Suffix: "'")) {
2632 QuoteStyle = BreakableStringLiteralUsingOperators::SingleQuotes;
2633 } else if (Style.isCSharp() && Text.starts_with(Prefix: "@\"") &&
2634 Text.ends_with(Suffix: "\"")) {
2635 QuoteStyle = BreakableStringLiteralUsingOperators::AtDoubleQuotes;
2636 } else if (Text.starts_with(Prefix: "\"") && Text.ends_with(Suffix: "\"")) {
2637 QuoteStyle = BreakableStringLiteralUsingOperators::DoubleQuotes;
2638 } else {
2639 return nullptr;
2640 }
2641 return std::make_unique<BreakableStringLiteralUsingOperators>(
2642 args: Current, args&: QuoteStyle,
2643 /*UnindentPlus=*/args: shouldUnindentNextOperator(Tok: Current), args&: StartColumn,
2644 args&: UnbreakableTailLength, args: State.Line->InPPDirective, args&: Encoding, args&: Style);
2645 }
2646
2647 StringRef Prefix;
2648 StringRef Postfix;
2649 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2650 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2651 // reduce the overhead) for each FormatToken, which is a string, so that we
2652 // don't run multiple checks here on the hot path.
2653 if ((Text.ends_with(Suffix: Postfix = "\"") &&
2654 (Text.starts_with(Prefix: Prefix = "@\"") || Text.starts_with(Prefix: Prefix = "\"") ||
2655 Text.starts_with(Prefix: Prefix = "u\"") ||
2656 Text.starts_with(Prefix: Prefix = "U\"") ||
2657 Text.starts_with(Prefix: Prefix = "u8\"") ||
2658 Text.starts_with(Prefix: Prefix = "L\""))) ||
2659 (Text.starts_with(Prefix: Prefix = "_T(\"") &&
2660 Text.ends_with(Suffix: Postfix = "\")"))) {
2661 return std::make_unique<BreakableStringLiteral>(
2662 args: Current, args&: StartColumn, args&: Prefix, args&: Postfix, args&: UnbreakableTailLength,
2663 args: State.Line->InPPDirective, args&: Encoding, args&: Style);
2664 }
2665 } else if (Current.is(TT: TT_BlockComment)) {
2666 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2667 // If a comment token switches formatting, like
2668 // /* clang-format on */, we don't want to break it further,
2669 // but we may still want to adjust its indentation.
2670 switchesFormatting(Token: Current)) {
2671 return nullptr;
2672 }
2673 return std::make_unique<BreakableBlockComment>(
2674 args: Current, args&: StartColumn, args: Current.OriginalColumn, args: !Current.Previous,
2675 args: State.Line->InPPDirective, args&: Encoding, args&: Style, args: Whitespaces.useCRLF());
2676 } else if (Current.is(TT: TT_LineComment) &&
2677 (!Current.Previous ||
2678 Current.Previous->isNot(Kind: TT_ImplicitStringLiteral))) {
2679 bool RegularComments = [&]() {
2680 for (const FormatToken *T = &Current; T && T->is(TT: TT_LineComment);
2681 T = T->Next) {
2682 if (!(T->TokenText.starts_with(Prefix: "//") || T->TokenText.starts_with(Prefix: "#")))
2683 return false;
2684 }
2685 return true;
2686 }();
2687 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2688 CommentPragmasRegex.match(String: Current.TokenText.substr(Start: 2)) ||
2689 switchesFormatting(Token: Current) || !RegularComments) {
2690 return nullptr;
2691 }
2692 return std::make_unique<BreakableLineCommentSection>(
2693 args: Current, args&: StartColumn, /*InPPDirective=*/args: false, args&: Encoding, args&: Style);
2694 }
2695 return nullptr;
2696}
2697
2698std::pair<unsigned, bool>
2699ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2700 LineState &State, bool AllowBreak,
2701 bool DryRun, bool Strict) {
2702 std::unique_ptr<const BreakableToken> Token =
2703 createBreakableToken(Current, State, AllowBreak);
2704 if (!Token)
2705 return {0, false};
2706 assert(Token->getLineCount() > 0);
2707 unsigned ColumnLimit = getColumnLimit(State);
2708 if (Current.is(TT: TT_LineComment)) {
2709 // We don't insert backslashes when breaking line comments.
2710 ColumnLimit = Style.ColumnLimit;
2711 }
2712 if (ColumnLimit == 0) {
2713 // To make the rest of the function easier set the column limit to the
2714 // maximum, if there should be no limit.
2715 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2716 }
2717 if (Current.UnbreakableTailLength >= ColumnLimit)
2718 return {0, false};
2719 // ColumnWidth was already accounted into State.Column before calling
2720 // breakProtrudingToken.
2721 unsigned StartColumn = State.Column - Current.ColumnWidth;
2722 unsigned NewBreakPenalty = Current.isStringLiteral()
2723 ? Style.PenaltyBreakString
2724 : Style.PenaltyBreakComment;
2725 // Stores whether we intentionally decide to let a line exceed the column
2726 // limit.
2727 bool Exceeded = false;
2728 // Stores whether we introduce a break anywhere in the token.
2729 bool BreakInserted = Token->introducesBreakBeforeToken();
2730 // Store whether we inserted a new line break at the end of the previous
2731 // logical line.
2732 bool NewBreakBefore = false;
2733 // We use a conservative reflowing strategy. Reflow starts after a line is
2734 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2735 // line that doesn't get reflown with the previous line is reached.
2736 bool Reflow = false;
2737 // Keep track of where we are in the token:
2738 // Where we are in the content of the current logical line.
2739 unsigned TailOffset = 0;
2740 // The column number we're currently at.
2741 unsigned ContentStartColumn =
2742 Token->getContentStartColumn(LineIndex: 0, /*Break=*/false);
2743 // The number of columns left in the current logical line after TailOffset.
2744 unsigned RemainingTokenColumns =
2745 Token->getRemainingLength(LineIndex: 0, Offset: TailOffset, StartColumn: ContentStartColumn);
2746 // Adapt the start of the token, for example indent.
2747 if (!DryRun)
2748 Token->adaptStartOfLine(LineIndex: 0, Whitespaces);
2749
2750 unsigned ContentIndent = 0;
2751 unsigned Penalty = 0;
2752 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2753 << StartColumn << ".\n");
2754 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2755 LineIndex != EndIndex; ++LineIndex) {
2756 LLVM_DEBUG(llvm::dbgs()
2757 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2758 NewBreakBefore = false;
2759 // If we did reflow the previous line, we'll try reflowing again. Otherwise
2760 // we'll start reflowing if the current line is broken or whitespace is
2761 // compressed.
2762 bool TryReflow = Reflow;
2763 // Break the current token until we can fit the rest of the line.
2764 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2765 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
2766 << (ContentStartColumn + RemainingTokenColumns)
2767 << ", space: " << ColumnLimit
2768 << ", reflown prefix: " << ContentStartColumn
2769 << ", offset in line: " << TailOffset << "\n");
2770 // If the current token doesn't fit, find the latest possible split in the
2771 // current line so that breaking at it will be under the column limit.
2772 // FIXME: Use the earliest possible split while reflowing to correctly
2773 // compress whitespace within a line.
2774 BreakableToken::Split Split =
2775 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2776 ContentStartColumn, CommentPragmasRegex);
2777 if (Split.first == StringRef::npos) {
2778 // No break opportunity - update the penalty and continue with the next
2779 // logical line.
2780 if (LineIndex < EndIndex - 1) {
2781 // The last line's penalty is handled in addNextStateToQueue() or when
2782 // calling replaceWhitespaceAfterLastLine below.
2783 Penalty += Style.PenaltyExcessCharacter *
2784 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2785 }
2786 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
2787 break;
2788 }
2789 assert(Split.first != 0);
2790
2791 if (Token->supportsReflow()) {
2792 // Check whether the next natural split point after the current one can
2793 // still fit the line, either because we can compress away whitespace,
2794 // or because the penalty the excess characters introduce is lower than
2795 // the break penalty.
2796 // We only do this for tokens that support reflowing, and thus allow us
2797 // to change the whitespace arbitrarily (e.g. comments).
2798 // Other tokens, like string literals, can be broken on arbitrary
2799 // positions.
2800
2801 // First, compute the columns from TailOffset to the next possible split
2802 // position.
2803 // For example:
2804 // ColumnLimit: |
2805 // // Some text that breaks
2806 // ^ tail offset
2807 // ^-- split
2808 // ^-------- to split columns
2809 // ^--- next split
2810 // ^--------------- to next split columns
2811 unsigned ToSplitColumns = Token->getRangeLength(
2812 LineIndex, Offset: TailOffset, Length: Split.first, StartColumn: ContentStartColumn);
2813 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
2814
2815 BreakableToken::Split NextSplit = Token->getSplit(
2816 LineIndex, TailOffset: TailOffset + Split.first + Split.second, ColumnLimit,
2817 ContentStartColumn: ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2818 // Compute the columns necessary to fit the next non-breakable sequence
2819 // into the current line.
2820 unsigned ToNextSplitColumns = 0;
2821 if (NextSplit.first == StringRef::npos) {
2822 ToNextSplitColumns = Token->getRemainingLength(LineIndex, Offset: TailOffset,
2823 StartColumn: ContentStartColumn);
2824 } else {
2825 ToNextSplitColumns = Token->getRangeLength(
2826 LineIndex, Offset: TailOffset,
2827 Length: Split.first + Split.second + NextSplit.first, StartColumn: ContentStartColumn);
2828 }
2829 // Compress the whitespace between the break and the start of the next
2830 // unbreakable sequence.
2831 ToNextSplitColumns =
2832 Token->getLengthAfterCompression(RemainingTokenColumns: ToNextSplitColumns, Split);
2833 LLVM_DEBUG(llvm::dbgs()
2834 << " ContentStartColumn: " << ContentStartColumn << "\n");
2835 LLVM_DEBUG(llvm::dbgs()
2836 << " ToNextSplit: " << ToNextSplitColumns << "\n");
2837 // If the whitespace compression makes us fit, continue on the current
2838 // line.
2839 bool ContinueOnLine =
2840 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2841 unsigned ExcessCharactersPenalty = 0;
2842 if (!ContinueOnLine && !Strict) {
2843 // Similarly, if the excess characters' penalty is lower than the
2844 // penalty of introducing a new break, continue on the current line.
2845 ExcessCharactersPenalty =
2846 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2847 Style.PenaltyExcessCharacter;
2848 LLVM_DEBUG(llvm::dbgs()
2849 << " Penalty excess: " << ExcessCharactersPenalty
2850 << "\n break : " << NewBreakPenalty << "\n");
2851 if (ExcessCharactersPenalty < NewBreakPenalty) {
2852 Exceeded = true;
2853 ContinueOnLine = true;
2854 }
2855 }
2856 if (ContinueOnLine) {
2857 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
2858 // The current line fits after compressing the whitespace - reflow
2859 // the next line into it if possible.
2860 TryReflow = true;
2861 if (!DryRun) {
2862 Token->compressWhitespace(LineIndex, TailOffset, Split,
2863 Whitespaces);
2864 }
2865 // When we continue on the same line, leave one space between content.
2866 ContentStartColumn += ToSplitColumns + 1;
2867 Penalty += ExcessCharactersPenalty;
2868 TailOffset += Split.first + Split.second;
2869 RemainingTokenColumns = Token->getRemainingLength(
2870 LineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
2871 continue;
2872 }
2873 }
2874 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
2875 // Update the ContentIndent only if the current line was not reflown with
2876 // the previous line, since in that case the previous line should still
2877 // determine the ContentIndent. Also never intent the last line.
2878 if (!Reflow)
2879 ContentIndent = Token->getContentIndent(LineIndex);
2880 LLVM_DEBUG(llvm::dbgs()
2881 << " ContentIndent: " << ContentIndent << "\n");
2882 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2883 LineIndex, /*Break=*/true);
2884
2885 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2886 LineIndex, Offset: TailOffset + Split.first + Split.second,
2887 StartColumn: ContentStartColumn);
2888 if (NewRemainingTokenColumns == 0) {
2889 // No content to indent.
2890 ContentIndent = 0;
2891 ContentStartColumn =
2892 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2893 NewRemainingTokenColumns = Token->getRemainingLength(
2894 LineIndex, Offset: TailOffset + Split.first + Split.second,
2895 StartColumn: ContentStartColumn);
2896 }
2897
2898 // When breaking before a tab character, it may be moved by a few columns,
2899 // but will still be expanded to the next tab stop, so we don't save any
2900 // columns.
2901 if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2902 // FIXME: Do we need to adjust the penalty?
2903 break;
2904 }
2905
2906 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2907 << ", " << Split.second << "\n");
2908 if (!DryRun) {
2909 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2910 Whitespaces);
2911 }
2912
2913 Penalty += NewBreakPenalty;
2914 TailOffset += Split.first + Split.second;
2915 RemainingTokenColumns = NewRemainingTokenColumns;
2916 BreakInserted = true;
2917 NewBreakBefore = true;
2918 }
2919 // In case there's another line, prepare the state for the start of the next
2920 // line.
2921 if (LineIndex + 1 != EndIndex) {
2922 unsigned NextLineIndex = LineIndex + 1;
2923 if (NewBreakBefore) {
2924 // After breaking a line, try to reflow the next line into the current
2925 // one once RemainingTokenColumns fits.
2926 TryReflow = true;
2927 }
2928 if (TryReflow) {
2929 // We decided that we want to try reflowing the next line into the
2930 // current one.
2931 // We will now adjust the state as if the reflow is successful (in
2932 // preparation for the next line), and see whether that works. If we
2933 // decide that we cannot reflow, we will later reset the state to the
2934 // start of the next line.
2935 Reflow = false;
2936 // As we did not continue breaking the line, RemainingTokenColumns is
2937 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2938 // the position at which we want to format the next line if we do
2939 // actually reflow.
2940 // When we reflow, we need to add a space between the end of the current
2941 // line and the next line's start column.
2942 ContentStartColumn += RemainingTokenColumns + 1;
2943 // Get the split that we need to reflow next logical line into the end
2944 // of the current one; the split will include any leading whitespace of
2945 // the next logical line.
2946 BreakableToken::Split SplitBeforeNext =
2947 Token->getReflowSplit(LineIndex: NextLineIndex, CommentPragmasRegex);
2948 LLVM_DEBUG(llvm::dbgs()
2949 << " Size of reflown text: " << ContentStartColumn
2950 << "\n Potential reflow split: ");
2951 if (SplitBeforeNext.first != StringRef::npos) {
2952 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2953 << SplitBeforeNext.second << "\n");
2954 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2955 // If the rest of the next line fits into the current line below the
2956 // column limit, we can safely reflow.
2957 RemainingTokenColumns = Token->getRemainingLength(
2958 LineIndex: NextLineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
2959 Reflow = true;
2960 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2961 LLVM_DEBUG(llvm::dbgs()
2962 << " Over limit after reflow, need: "
2963 << (ContentStartColumn + RemainingTokenColumns)
2964 << ", space: " << ColumnLimit
2965 << ", reflown prefix: " << ContentStartColumn
2966 << ", offset in line: " << TailOffset << "\n");
2967 // If the whole next line does not fit, try to find a point in
2968 // the next line at which we can break so that attaching the part
2969 // of the next line to that break point onto the current line is
2970 // below the column limit.
2971 BreakableToken::Split Split =
2972 Token->getSplit(LineIndex: NextLineIndex, TailOffset, ColumnLimit,
2973 ContentStartColumn, CommentPragmasRegex);
2974 if (Split.first == StringRef::npos) {
2975 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
2976 Reflow = false;
2977 } else {
2978 // Check whether the first split point gets us below the column
2979 // limit. Note that we will execute this split below as part of
2980 // the normal token breaking and reflow logic within the line.
2981 unsigned ToSplitColumns = Token->getRangeLength(
2982 LineIndex: NextLineIndex, Offset: TailOffset, Length: Split.first, StartColumn: ContentStartColumn);
2983 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2984 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
2985 << (ContentStartColumn + ToSplitColumns)
2986 << ", space: " << ColumnLimit);
2987 unsigned ExcessCharactersPenalty =
2988 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2989 Style.PenaltyExcessCharacter;
2990 if (NewBreakPenalty < ExcessCharactersPenalty)
2991 Reflow = false;
2992 }
2993 }
2994 }
2995 } else {
2996 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2997 }
2998 }
2999 if (!Reflow) {
3000 // If we didn't reflow into the next line, the only space to consider is
3001 // the next logical line. Reset our state to match the start of the next
3002 // line.
3003 TailOffset = 0;
3004 ContentStartColumn =
3005 Token->getContentStartColumn(LineIndex: NextLineIndex, /*Break=*/false);
3006 RemainingTokenColumns = Token->getRemainingLength(
3007 LineIndex: NextLineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
3008 // Adapt the start of the token, for example indent.
3009 if (!DryRun)
3010 Token->adaptStartOfLine(LineIndex: NextLineIndex, Whitespaces);
3011 } else {
3012 // If we found a reflow split and have added a new break before the next
3013 // line, we are going to remove the line break at the start of the next
3014 // logical line. For example, here we'll add a new line break after
3015 // 'text', and subsequently delete the line break between 'that' and
3016 // 'reflows'.
3017 // // some text that
3018 // // reflows
3019 // ->
3020 // // some text
3021 // // that reflows
3022 // When adding the line break, we also added the penalty for it, so we
3023 // need to subtract that penalty again when we remove the line break due
3024 // to reflowing.
3025 if (NewBreakBefore) {
3026 assert(Penalty >= NewBreakPenalty);
3027 Penalty -= NewBreakPenalty;
3028 }
3029 if (!DryRun)
3030 Token->reflow(LineIndex: NextLineIndex, Whitespaces);
3031 }
3032 }
3033 }
3034
3035 BreakableToken::Split SplitAfterLastLine =
3036 Token->getSplitAfterLastLine(TailOffset);
3037 if (SplitAfterLastLine.first != StringRef::npos) {
3038 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
3039
3040 // We add the last line's penalty here, since that line is going to be split
3041 // now.
3042 Penalty += Style.PenaltyExcessCharacter *
3043 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
3044
3045 if (!DryRun) {
3046 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
3047 Whitespaces);
3048 }
3049 ContentStartColumn =
3050 Token->getContentStartColumn(LineIndex: Token->getLineCount() - 1, /*Break=*/true);
3051 RemainingTokenColumns = Token->getRemainingLength(
3052 LineIndex: Token->getLineCount() - 1,
3053 Offset: TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
3054 StartColumn: ContentStartColumn);
3055 }
3056
3057 State.Column = ContentStartColumn + RemainingTokenColumns -
3058 Current.UnbreakableTailLength;
3059
3060 if (BreakInserted) {
3061 if (!DryRun)
3062 Token->updateAfterBroken(Whitespaces);
3063
3064 // If we break the token inside a parameter list, we need to break before
3065 // the next parameter on all levels, so that the next parameter is clearly
3066 // visible. Line comments already introduce a break.
3067 if (Current.isNot(Kind: TT_LineComment))
3068 for (ParenState &Paren : State.Stack)
3069 Paren.BreakBeforeParameter = true;
3070
3071 if (Current.is(TT: TT_BlockComment))
3072 State.NoContinuation = true;
3073
3074 State.Stack.back().LastSpace = StartColumn;
3075 }
3076
3077 Token->updateNextToken(State);
3078
3079 return {Penalty, Exceeded};
3080}
3081
3082unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
3083 // In preprocessor directives reserve two chars for trailing " \".
3084 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
3085}
3086
3087bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
3088 const FormatToken &Current = *State.NextToken;
3089 if (!Current.isStringLiteral() || Current.is(TT: TT_ImplicitStringLiteral))
3090 return false;
3091 // We never consider raw string literals "multiline" for the purpose of
3092 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
3093 // (see TokenAnnotator::mustBreakBefore().
3094 if (Current.TokenText.starts_with(Prefix: "R\""))
3095 return false;
3096 if (Current.IsMultiline)
3097 return true;
3098 if (Current.getNextNonComment() &&
3099 Current.getNextNonComment()->isStringLiteral()) {
3100 return true; // Implicit concatenation.
3101 }
3102 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
3103 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
3104 Style.ColumnLimit) {
3105 return true; // String will be split.
3106 }
3107 return false;
3108}
3109
3110} // namespace format
3111} // namespace clang
3112