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