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.PackParameters.BinPack == FormatStyle::BPPS_BinPack ||
518 Style.PackParameters.BinPack == FormatStyle::BPPS_UseBreakAfter))) ||
519 (Style.BreakBeforeTernaryOperators && Current.is(TT: TT_ConditionalExpr) &&
520 Previous.isNot(Kind: tok::question)) ||
521 (!Style.BreakBeforeTernaryOperators &&
522 Previous.is(TT: TT_ConditionalExpr))) &&
523 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
524 Current.isNoneOf(Ks: tok::r_paren, Ks: tok::r_brace)) {
525 return true;
526 }
527 if (CurrentState.IsChainedConditional &&
528 ((Style.BreakBeforeTernaryOperators && Current.is(TT: TT_ConditionalExpr) &&
529 Current.is(Kind: tok::colon)) ||
530 (!Style.BreakBeforeTernaryOperators && Previous.is(TT: TT_ConditionalExpr) &&
531 Previous.is(Kind: tok::colon)))) {
532 return true;
533 }
534 if (((Previous.is(TT: TT_DictLiteral) && Previous.is(Kind: tok::l_brace)) ||
535 (Previous.is(TT: TT_ArrayInitializerLSquare) &&
536 Previous.ParameterCount > 1) ||
537 opensProtoMessageField(LessTok: Previous, Style)) &&
538 Style.ColumnLimit > 0 &&
539 getLengthToMatchingParen(Tok: Previous, Stack: State.Stack) + State.Column - 1 >
540 getColumnLimit(State)) {
541 return true;
542 }
543
544 const FormatToken &BreakConstructorInitializersToken =
545 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
546 ? Previous
547 : Current;
548 if (Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterComma &&
549 BreakConstructorInitializersToken.is(TT: TT_CtorInitializerColon) &&
550 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
551 getColumnLimit(State) ||
552 CurrentState.BreakBeforeParameter) &&
553 ((!Current.isTrailingComment() && Style.ColumnLimit > 0) ||
554 Current.NewlinesBefore > 0)) {
555 return true;
556 }
557
558 if (Current.is(TT: TT_ObjCMethodExpr) && Previous.isNot(Kind: TT_SelectorName) &&
559 State.Line->startsWith(Tokens: TT_ObjCMethodSpecifier)) {
560 return true;
561 }
562 if (Current.is(TT: TT_SelectorName) && Previous.isNot(Kind: tok::at) &&
563 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
564 (Style.ObjCBreakBeforeNestedBlockParam ||
565 !Current.startsSequence(K1: TT_SelectorName, Tokens: tok::colon, Tokens: tok::caret))) {
566 return true;
567 }
568
569 unsigned NewLineColumn = getNewLineColumn(State).Total;
570 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
571 State.Column + getLengthToNextOperator(Tok: Current) > Style.ColumnLimit &&
572 (State.Column > NewLineColumn ||
573 Current.NestingLevel < State.StartOfLineLevel)) {
574 return true;
575 }
576
577 if (startsSegmentOfBuilderTypeCall(Tok: Current) &&
578 (CurrentState.CallContinuation != 0 ||
579 CurrentState.BreakBeforeParameter) &&
580 // JavaScript is treated different here as there is a frequent pattern:
581 // SomeFunction(function() {
582 // ...
583 // }.bind(...));
584 // FIXME: We should find a more generic solution to this problem.
585 !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
586 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
587 return true;
588 }
589
590 // If the template declaration spans multiple lines, force wrap before the
591 // function/class declaration.
592 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
593 Current.CanBreakBefore) {
594 return true;
595 }
596
597 if (State.Line->First->isNot(Kind: tok::kw_enum) && State.Column <= NewLineColumn)
598 return false;
599
600 if (Style.AlwaysBreakBeforeMultilineStrings &&
601 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
602 Previous.is(Kind: tok::comma) || Current.NestingLevel < 2) &&
603 Previous.isNoneOf(Ks: tok::kw_return, Ks: tok::lessless, Ks: tok::at,
604 Ks: Keywords.kw_dollar) &&
605 Previous.isNoneOf(Ks: TT_InlineASMColon, Ks: TT_ConditionalExpr) &&
606 nextIsMultilineString(State)) {
607 return true;
608 }
609
610 // Using CanBreakBefore here and below takes care of the decision whether the
611 // current style uses wrapping before or after operators for the given
612 // operator.
613 if (Previous.is(TT: TT_BinaryOperator) && Current.CanBreakBefore) {
614 const auto PreviousPrecedence = Previous.getPrecedence();
615 if (PreviousPrecedence != prec::Assignment &&
616 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
617 const bool LHSIsBinaryExpr =
618 Previous.Previous && Previous.Previous->EndsBinaryExpression;
619 if (LHSIsBinaryExpr)
620 return true;
621 // If we need to break somewhere inside the LHS of a binary expression, we
622 // should also break after the operator. Otherwise, the formatting would
623 // hide the operator precedence, e.g. in:
624 // if (aaaaaaaaaaaaaa ==
625 // bbbbbbbbbbbbbb && c) {..
626 // For comparisons, we only apply this rule, if the LHS is a binary
627 // expression itself as otherwise, the line breaks seem superfluous.
628 // We need special cases for ">>" which we have split into two ">" while
629 // lexing in order to make template parsing easier.
630 const bool IsComparison =
631 (PreviousPrecedence == prec::Relational ||
632 PreviousPrecedence == prec::Equality ||
633 PreviousPrecedence == prec::Spaceship) &&
634 Previous.Previous &&
635 Previous.Previous->isNot(Kind: TT_BinaryOperator); // For >>.
636 if (!IsComparison)
637 return true;
638 }
639 } else if (Current.is(TT: TT_BinaryOperator) && Current.CanBreakBefore &&
640 Current.getPrecedence() != prec::Assignment &&
641 CurrentState.BreakBeforeParameter) {
642 return true;
643 }
644
645 // Same as above, but for the first "<<" operator.
646 if (Current.is(Kind: tok::lessless) && Current.isNot(Kind: TT_OverloadedOperator) &&
647 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
648 return true;
649 }
650
651 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
652 // Always break after "template <...>"(*) and leading annotations. This is
653 // only for cases where the entire line does not fit on a single line as a
654 // different LineFormatter would be used otherwise.
655 // *: Except when another option interferes with that, like concepts.
656 if (Previous.ClosesTemplateDeclaration) {
657 if (Current.is(Kind: tok::kw_concept)) {
658 switch (Style.BreakBeforeConceptDeclarations) {
659 case FormatStyle::BBCDS_Allowed:
660 break;
661 case FormatStyle::BBCDS_Always:
662 return true;
663 case FormatStyle::BBCDS_Never:
664 return false;
665 }
666 }
667 if (Current.is(TT: TT_RequiresClause)) {
668 switch (Style.RequiresClausePosition) {
669 case FormatStyle::RCPS_SingleLine:
670 case FormatStyle::RCPS_WithPreceding:
671 return false;
672 default:
673 return true;
674 }
675 }
676 return Style.BreakTemplateDeclarations != FormatStyle::BTDS_No &&
677 (Style.BreakTemplateDeclarations != FormatStyle::BTDS_Leave ||
678 Current.NewlinesBefore > 0);
679 }
680 if (Previous.is(TT: TT_FunctionAnnotationRParen) &&
681 State.Line->Type != LT_PreprocessorDirective) {
682 return true;
683 }
684 if (Previous.is(TT: TT_LeadingJavaAnnotation) && Current.isNot(Kind: tok::l_paren) &&
685 Current.isNot(Kind: TT_LeadingJavaAnnotation)) {
686 return true;
687 }
688 }
689
690 if (Style.isJavaScript() && Previous.is(Kind: tok::r_paren) &&
691 Previous.is(TT: TT_JavaAnnotation)) {
692 // Break after the closing parenthesis of TypeScript decorators before
693 // functions, getters and setters.
694 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
695 "function"};
696 if (BreakBeforeDecoratedTokens.contains(key: Current.TokenText))
697 return true;
698 }
699
700 if (Current.is(TT: TT_FunctionDeclarationName) &&
701 !State.Line->ReturnTypeWrapped &&
702 // Don't break before a C# function when no break after return type.
703 (!Style.isCSharp() ||
704 Style.BreakAfterReturnType > FormatStyle::RTBS_ExceptShortType) &&
705 // Don't always break between a JavaScript `function` and the function
706 // name.
707 !Style.isJavaScript() && Previous.isNot(Kind: tok::kw_template) &&
708 CurrentState.BreakBeforeParameter) {
709 for (const auto *Tok = &Previous; Tok; Tok = Tok->Previous) {
710 if (Tok->is(TT: TT_LineComment))
711 return false;
712 if (Tok->is(TT: TT_TemplateCloser)) {
713 Tok = Tok->MatchingParen;
714 if (!Tok)
715 return false;
716 }
717 if (Tok->FirstAfterPPLine)
718 return false;
719 }
720
721 return true;
722 }
723
724 // The following could be precomputed as they do not depend on the state.
725 // However, as they should take effect only if the UnwrappedLine does not fit
726 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
727 if (Style.ColumnLimit != 0 && Previous.is(BBK: BK_Block) &&
728 Previous.is(Kind: tok::l_brace) &&
729 Current.isNoneOf(Ks: tok::r_brace, Ks: tok::comment)) {
730 return true;
731 }
732
733 if (Current.is(Kind: tok::lessless) &&
734 ((Previous.is(Kind: tok::identifier) && Previous.TokenText == "endl") ||
735 (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with(Suffix: "\\n\"") ||
736 Previous.TokenText == "\'\\n\'")))) {
737 return true;
738 }
739
740 if (Previous.is(TT: TT_BlockComment) && Previous.IsMultiline)
741 return true;
742
743 if (State.NoContinuation)
744 return true;
745
746 return false;
747}
748
749unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
750 bool DryRun,
751 unsigned ExtraSpaces) {
752 const FormatToken &Current = *State.NextToken;
753 assert(State.NextToken->Previous);
754 const FormatToken &Previous = *State.NextToken->Previous;
755
756 assert(!State.Stack.empty());
757 State.NoContinuation = false;
758
759 if (Current.is(TT: TT_ImplicitStringLiteral) &&
760 (!Previous.Tok.getIdentifierInfo() ||
761 Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
762 tok::pp_not_keyword)) {
763 unsigned EndColumn =
764 SourceMgr.getSpellingColumnNumber(Loc: Current.WhitespaceRange.getEnd());
765 if (Current.LastNewlineOffset != 0) {
766 // If there is a newline within this token, the final column will solely
767 // determined by the current end column.
768 State.Column = EndColumn;
769 } else {
770 unsigned StartColumn =
771 SourceMgr.getSpellingColumnNumber(Loc: Current.WhitespaceRange.getBegin());
772 assert(EndColumn >= StartColumn);
773 State.Column += EndColumn - StartColumn;
774 }
775 moveStateToNextToken(State, DryRun, /*Newline=*/false);
776 return 0;
777 }
778
779 unsigned Penalty = 0;
780 if (Newline)
781 Penalty = addTokenOnNewLine(State, DryRun);
782 else
783 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
784
785 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
786}
787
788void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
789 unsigned ExtraSpaces) {
790 FormatToken &Current = *State.NextToken;
791 assert(State.NextToken->Previous);
792 const FormatToken &Previous = *State.NextToken->Previous;
793 auto &CurrentState = State.Stack.back();
794
795 // Deal with lambda arguments in C++. The aim here is to ensure that we don't
796 // over-indent lambda function bodies when lambdas are passed as arguments to
797 // function calls. We do this by ensuring that either all arguments (including
798 // any lambdas) go on the same line as the function call, or we break before
799 // the first argument.
800 auto DisallowLineBreaks = [&] {
801 if (!Style.isCpp() ||
802 Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope) {
803 return false;
804 }
805
806 // For example, `/*Newline=*/false`.
807 if (Previous.is(TT: TT_BlockComment) && Current.SpacesRequiredBefore == 0)
808 return false;
809
810 if (Current.isOneOf(K1: tok::comment, K2: tok::l_paren, Ks: TT_LambdaLSquare))
811 return false;
812
813 const auto *Prev = Current.getPreviousNonComment();
814 if (!Prev || Prev->isNot(Kind: tok::l_paren))
815 return false;
816
817 if (Prev->BlockParameterCount == 0)
818 return false;
819
820 // Multiple lambdas in the same function call.
821 if (Prev->BlockParameterCount > 1)
822 return true;
823
824 // A lambda followed by another arg.
825 if (!Prev->Role)
826 return false;
827
828 const auto *Comma = Prev->Role->lastComma();
829 if (!Comma)
830 return false;
831
832 const auto *Next = Comma->getNextNonComment();
833 return Next && Next->isNoneOf(Ks: TT_LambdaLSquare, Ks: tok::l_brace, Ks: tok::caret);
834 };
835
836 if (DisallowLineBreaks())
837 State.NoLineBreak = true;
838
839 if (Current.is(Kind: tok::equal) &&
840 (State.Line->First->is(Kind: tok::kw_for) || Current.NestingLevel == 0) &&
841 CurrentState.VariablePos == 0 &&
842 (!Previous.Previous ||
843 Previous.Previous->isNot(Kind: TT_DesignatedInitializerPeriod))) {
844 CurrentState.VariablePos = State.Column;
845 // Move over * and & if they are bound to the variable name.
846 const FormatToken *Tok = &Previous;
847 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
848 CurrentState.VariablePos -= Tok->ColumnWidth;
849 if (Tok->SpacesRequiredBefore != 0)
850 break;
851 Tok = Tok->Previous;
852 }
853 if (Previous.PartOfMultiVariableDeclStmt)
854 CurrentState.LastSpace = CurrentState.VariablePos;
855 }
856
857 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
858
859 // Indent preprocessor directives after the hash if required.
860 int PPColumnCorrection = 0;
861 if (&Previous == State.Line->First && Previous.is(Kind: tok::hash) &&
862 (State.Line->Type == LT_PreprocessorDirective ||
863 State.Line->Type == LT_ImportStatement)) {
864 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash) {
865 Spaces += State.FirstIndent;
866
867 // For preprocessor indent with tabs, State.Column will be 1 because of
868 // the hash. This causes second-level indents onward to have an extra
869 // space after the tabs. We avoid this misalignment by subtracting 1 from
870 // the column value passed to replaceWhitespace().
871 if (Style.UseTab != FormatStyle::UT_Never)
872 PPColumnCorrection = -1;
873 } else if (Style.IndentPPDirectives == FormatStyle::PPDIS_Leave) {
874 Spaces += Current.OriginalColumn - Previous.OriginalColumn - 1;
875 }
876 }
877
878 if (!DryRun) {
879 const bool ContinuePPDirective =
880 State.Line->InMacroBody && Current.isNot(Kind: TT_LineComment);
881 Whitespaces.replaceWhitespace(Tok&: Current, /*Newlines=*/0, Spaces,
882 StartOfTokenColumn: State.Column + Spaces + PPColumnCorrection,
883 /*AlignTo=*/AlignedTo: nullptr, InPPDirective: ContinuePPDirective);
884 }
885
886 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
887 // declaration unless there is multiple inheritance.
888 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
889 Current.is(TT: TT_InheritanceColon)) {
890 CurrentState.NoLineBreak = true;
891 }
892 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
893 Previous.is(TT: TT_InheritanceColon)) {
894 CurrentState.NoLineBreak = true;
895 }
896
897 if (Current.is(TT: TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
898 unsigned MinIndent =
899 std::max(a: State.FirstIndent + Style.ContinuationIndentWidth,
900 b: CurrentState.Indent.Total);
901 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
902 if (Current.LongestObjCSelectorName == 0)
903 CurrentState.AlignColons = false;
904 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
905 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
906 else
907 CurrentState.ColonPos = FirstColonPos;
908 }
909
910 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
911 // parenthesis by disallowing any further line breaks if there is no line
912 // break after the opening parenthesis. Don't break if it doesn't conserve
913 // columns.
914 auto IsOpeningBracket = [&](const FormatToken &Tok) {
915 auto IsStartOfBracedList = [&]() {
916 return Tok.is(Kind: tok::l_brace) && Tok.isNot(Kind: BK_Block) &&
917 Style.Cpp11BracedListStyle != FormatStyle::BLS_Block;
918 };
919 if (IsStartOfBracedList())
920 return Style.BreakAfterOpenBracketBracedList;
921 if (Tok.isNoneOf(Ks: tok::l_paren, Ks: TT_TemplateOpener, Ks: tok::l_square))
922 return false;
923 if (!Tok.Previous)
924 return true;
925 if (Tok.Previous->isIf())
926 return Style.BreakAfterOpenBracketIf;
927 if (Tok.Previous->isLoop(Style))
928 return Style.BreakAfterOpenBracketLoop;
929 if (Tok.Previous->is(Kind: tok::kw_switch))
930 return Style.BreakAfterOpenBracketSwitch;
931 if (Style.BreakAfterOpenBracketFunction) {
932 return !Tok.Previous->is(TT: TT_CastRParen) &&
933 !(Style.isJavaScript() && Tok.is(II: Keywords.kw_await));
934 }
935 return false;
936 };
937 auto IsFunctionCallParen = [](const FormatToken &Tok) {
938 return Tok.is(Kind: tok::l_paren) && Tok.ParameterCount > 0 && Tok.Previous &&
939 Tok.Previous->is(Kind: tok::identifier);
940 };
941 auto IsInTemplateString = [this](const FormatToken &Tok, bool NestBlocks) {
942 if (!Style.isJavaScript())
943 return false;
944 for (const auto *Prev = &Tok; Prev; Prev = Prev->Previous) {
945 if (Prev->is(TT: TT_TemplateString) && Prev->opensScope())
946 return true;
947 if (Prev->opensScope() && !NestBlocks)
948 return false;
949 if (Prev->is(TT: TT_TemplateString) && Prev->closesScope())
950 return false;
951 }
952 return false;
953 };
954 // Identifies simple (no expression) one-argument function calls.
955 auto StartsSimpleOneArgList = [&](const FormatToken &TokAfterLParen) {
956 assert(TokAfterLParen.isNot(tok::comment) || TokAfterLParen.Next);
957 const auto &Tok =
958 TokAfterLParen.is(Kind: tok::comment) ? *TokAfterLParen.Next : TokAfterLParen;
959 if (!Tok.FakeLParens.empty() && Tok.FakeLParens.back() > prec::Unknown)
960 return false;
961 // Nested calls that involve `new` expressions also look like simple
962 // function calls, eg:
963 // - foo(new Bar())
964 // - foo(::new Bar())
965 if (Tok.is(Kind: tok::kw_new) || Tok.startsSequence(K1: tok::coloncolon, Tokens: tok::kw_new))
966 return true;
967 if (Tok.is(TT: TT_UnaryOperator) ||
968 (Style.isJavaScript() &&
969 Tok.isOneOf(K1: tok::ellipsis, K2: Keywords.kw_await))) {
970 return true;
971 }
972 const auto *Previous = TokAfterLParen.Previous;
973 assert(Previous); // IsOpeningBracket(Previous)
974 if (Previous->Previous &&
975 (Previous->Previous->isIf() || Previous->Previous->isLoop(Style) ||
976 Previous->Previous->is(Kind: tok::kw_switch))) {
977 return false;
978 }
979 if (Previous->isNoneOf(Ks: TT_FunctionDeclarationLParen,
980 Ks: TT_LambdaDefinitionLParen) &&
981 !IsFunctionCallParen(*Previous)) {
982 return true;
983 }
984 if (IsOpeningBracket(Tok) || IsInTemplateString(Tok, true))
985 return true;
986 const auto *Next = Tok.Next;
987 return !Next || Next->isMemberAccess() ||
988 Next->is(TT: TT_FunctionDeclarationLParen) || IsFunctionCallParen(*Next);
989 };
990 if (IsOpeningBracket(Previous) &&
991 State.Column > getNewLineColumn(State).Total &&
992 // Don't do this for simple (no expressions) one-argument function calls
993 // as that feels like needlessly wasting whitespace, e.g.:
994 //
995 // caaaaaaaaaaaall(
996 // caaaaaaaaaaaall(
997 // caaaaaaaaaaaall(
998 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
999 // or
1000 // caaaaaaaaaaaaaaaaaaaaal(
1001 // new SomethingElseeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee());
1002 !StartsSimpleOneArgList(Current)) {
1003 CurrentState.NoLineBreak = true;
1004 }
1005
1006 if (Previous.is(TT: TT_TemplateString) && Previous.opensScope())
1007 CurrentState.NoLineBreak = true;
1008
1009 // Align following lines within parentheses / brackets if configured.
1010 // Note: This doesn't apply to macro expansion lines, which are MACRO( , , )
1011 // with args as children of the '(' and ',' tokens. It does not make sense to
1012 // align the commas with the opening paren.
1013 if (Style.AlignAfterOpenBracket &&
1014 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
1015 Previous.isNoneOf(Ks: TT_ObjCMethodExpr, Ks: TT_RequiresClause,
1016 Ks: TT_TableGenDAGArgOpener,
1017 Ks: TT_TableGenDAGArgOpenerToBreak) &&
1018 !(Current.MacroParent && Previous.MacroParent) &&
1019 (Current.isNot(Kind: TT_LineComment) ||
1020 (Previous.is(BBK: BK_BracedInit) &&
1021 Style.Cpp11BracedListStyle != FormatStyle::BLS_FunctionCall) ||
1022 Previous.is(TT: TT_VerilogMultiLineListLParen)) &&
1023 !IsInTemplateString(Current, false)) {
1024 CurrentState.Indent = State.Column + Spaces;
1025 CurrentState.AlignedTo = &Previous;
1026 }
1027 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
1028 CurrentState.NoLineBreak = true;
1029 if (mustBreakBinaryOperation(Current, Style))
1030 CurrentState.NoLineBreak = true;
1031
1032 if (startsSegmentOfBuilderTypeCall(Tok: Current) &&
1033 State.Column > getNewLineColumn(State).Total) {
1034 CurrentState.ContainsUnwrappedBuilder = true;
1035 }
1036
1037 if (Current.is(TT: TT_LambdaArrow) && Style.isJava())
1038 CurrentState.NoLineBreak = true;
1039 if (Current.isMemberAccess() && Previous.is(Kind: tok::r_paren) &&
1040 (Previous.MatchingParen &&
1041 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
1042 // If there is a function call with long parameters, break before trailing
1043 // calls. This prevents things like:
1044 // EXPECT_CALL(SomeLongParameter).Times(
1045 // 2);
1046 // We don't want to do this for short parameters as they can just be
1047 // indexes.
1048 CurrentState.NoLineBreak = true;
1049 }
1050
1051 // Don't allow the RHS of an operator to be split over multiple lines unless
1052 // there is a line-break right after the operator.
1053 // Exclude relational operators, as there, it is always more desirable to
1054 // have the LHS 'left' of the RHS.
1055 const FormatToken *P = Current.getPreviousNonComment();
1056 if (Current.isNot(Kind: tok::comment) && P &&
1057 (P->isOneOf(K1: TT_BinaryOperator, K2: tok::comma) ||
1058 (P->is(TT: TT_ConditionalExpr) && P->is(Kind: tok::colon))) &&
1059 P->isNoneOf(Ks: TT_OverloadedOperator, Ks: TT_CtorInitializerComma) &&
1060 P->getPrecedence() != prec::Assignment &&
1061 P->getPrecedence() != prec::Relational &&
1062 P->getPrecedence() != prec::Spaceship) {
1063 bool BreakBeforeOperator =
1064 P->MustBreakBefore || P->is(Kind: tok::lessless) ||
1065 (P->is(TT: TT_BinaryOperator) &&
1066 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
1067 (P->is(TT: TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
1068 // Don't do this if there are only two operands. In these cases, there is
1069 // always a nice vertical separation between them and the extra line break
1070 // does not help.
1071 bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator &&
1072 P->isNot(Kind: TT_ConditionalExpr);
1073 if ((!BreakBeforeOperator &&
1074 !(HasTwoOperands &&
1075 Style.AlignOperands != FormatStyle::OAS_DontAlign)) ||
1076 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
1077 CurrentState.NoLineBreakInOperand = true;
1078 }
1079 }
1080
1081 State.Column += Spaces;
1082 if (Current.isNot(Kind: tok::comment) && Previous.is(Kind: tok::l_paren) &&
1083 Previous.Previous &&
1084 (Previous.Previous->is(Kind: tok::kw_for) || Previous.Previous->isIf())) {
1085 // Treat the condition inside an if as if it was a second function
1086 // parameter, i.e. let nested calls have a continuation indent.
1087 CurrentState.LastSpace = State.Column;
1088 CurrentState.NestedBlockIndent = State.Column;
1089 } else if (Current.isNoneOf(Ks: tok::comment, Ks: tok::caret) &&
1090 ((Previous.is(Kind: tok::comma) &&
1091 Previous.isNot(Kind: TT_OverloadedOperator)) ||
1092 (Previous.is(Kind: tok::colon) && Previous.is(TT: TT_ObjCMethodExpr)))) {
1093 CurrentState.LastSpace = State.Column;
1094 } else if (Previous.is(TT: TT_CtorInitializerColon) &&
1095 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
1096 Style.BreakConstructorInitializers ==
1097 FormatStyle::BCIS_AfterColon) {
1098 CurrentState.Indent = State.Column;
1099 CurrentState.LastSpace = State.Column;
1100 } else if (Previous.isOneOf(K1: TT_ConditionalExpr, K2: TT_CtorInitializerColon)) {
1101 CurrentState.LastSpace = State.Column;
1102 } else if (Previous.is(TT: TT_BinaryOperator) &&
1103 ((Previous.getPrecedence() != prec::Assignment &&
1104 (Previous.isNot(Kind: tok::lessless) || Previous.OperatorIndex != 0 ||
1105 Previous.NextOperator)) ||
1106 Current.StartsBinaryExpression)) {
1107 // Indent relative to the RHS of the expression unless this is a simple
1108 // assignment without binary expression on the RHS.
1109 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
1110 CurrentState.LastSpace = State.Column;
1111 } else if (Previous.is(TT: TT_InheritanceColon)) {
1112 CurrentState.Indent = State.Column;
1113 CurrentState.LastSpace = State.Column;
1114 } else if (Current.is(TT: TT_CSharpGenericTypeConstraintColon)) {
1115 CurrentState.ColonPos = State.Column;
1116 } else if (Previous.opensScope()) {
1117 // If a function has a trailing call, indent all parameters from the
1118 // opening parenthesis. This avoids confusing indents like:
1119 // OuterFunction(InnerFunctionCall( // break
1120 // ParameterToInnerFunction)) // break
1121 // .SecondInnerFunctionCall();
1122 if (Previous.MatchingParen) {
1123 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
1124 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
1125 State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
1126 CurrentState.LastSpace = State.Column;
1127 }
1128 }
1129 }
1130}
1131
1132unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
1133 bool DryRun) {
1134 FormatToken &Current = *State.NextToken;
1135 assert(State.NextToken->Previous);
1136 const FormatToken &Previous = *State.NextToken->Previous;
1137 auto &CurrentState = State.Stack.back();
1138
1139 // Extra penalty that needs to be added because of the way certain line
1140 // breaks are chosen.
1141 unsigned Penalty = 0;
1142
1143 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1144 const FormatToken *NextNonComment = Previous.getNextNonComment();
1145 if (!NextNonComment)
1146 NextNonComment = &Current;
1147 // The first line break on any NestingLevel causes an extra penalty in order
1148 // prefer similar line breaks.
1149 if (!CurrentState.ContainsLineBreak)
1150 Penalty += 15;
1151 CurrentState.ContainsLineBreak = true;
1152
1153 Penalty += State.NextToken->SplitPenalty;
1154
1155 // Breaking before the first "<<" is generally not desirable if the LHS is
1156 // short. Also always add the penalty if the LHS is split over multiple lines
1157 // to avoid unnecessary line breaks that just work around this penalty.
1158 if (NextNonComment->is(Kind: tok::lessless) && CurrentState.FirstLessLess == 0 &&
1159 (State.Column <= Style.ColumnLimit / 3 ||
1160 CurrentState.BreakBeforeParameter)) {
1161 Penalty += Style.PenaltyBreakFirstLessLess;
1162 }
1163
1164 const auto [TotalColumn, IndentedFromColumn] = getNewLineColumn(State);
1165 State.Column = TotalColumn;
1166
1167 // Add Penalty proportional to amount of whitespace away from FirstColumn
1168 // This tends to penalize several lines that are far-right indented,
1169 // and prefers a line-break prior to such a block, e.g:
1170 //
1171 // Constructor() :
1172 // member(value), looooooooooooooooong_member(
1173 // looooooooooong_call(param_1, param_2, param_3))
1174 // would then become
1175 // Constructor() :
1176 // member(value),
1177 // looooooooooooooooong_member(
1178 // looooooooooong_call(param_1, param_2, param_3))
1179 if (State.Column > State.FirstIndent) {
1180 Penalty +=
1181 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
1182 }
1183
1184 // Indent nested blocks relative to this column, unless in a very specific
1185 // JavaScript special case where:
1186 //
1187 // var loooooong_name =
1188 // function() {
1189 // // code
1190 // }
1191 //
1192 // is common and should be formatted like a free-standing function. The same
1193 // goes for wrapping before the lambda return type arrow.
1194 if (Current.isNot(Kind: TT_LambdaArrow) &&
1195 (!Style.isJavaScript() || Current.NestingLevel != 0 ||
1196 !PreviousNonComment || PreviousNonComment->isNot(Kind: tok::equal) ||
1197 Current.isNoneOf(Ks: Keywords.kw_async, Ks: Keywords.kw_function))) {
1198 CurrentState.NestedBlockIndent = State.Column;
1199 }
1200
1201 if (NextNonComment->isMemberAccess()) {
1202 if (CurrentState.CallContinuation == 0)
1203 CurrentState.CallContinuation = State.Column;
1204 } else if (NextNonComment->is(TT: TT_SelectorName)) {
1205 if (!CurrentState.ObjCSelectorNameFound) {
1206 if (NextNonComment->LongestObjCSelectorName == 0) {
1207 CurrentState.AlignColons = false;
1208 } else {
1209 CurrentState.ColonPos =
1210 (shouldIndentWrappedSelectorName(Style, LineType: State.Line->Type)
1211 ? std::max(a: CurrentState.Indent.Total,
1212 b: State.FirstIndent + Style.ContinuationIndentWidth)
1213 : CurrentState.Indent.Total) +
1214 std::max(a: NextNonComment->LongestObjCSelectorName,
1215 b: NextNonComment->ColumnWidth);
1216 }
1217 } else if (CurrentState.AlignColons &&
1218 CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
1219 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
1220 }
1221 } else if (PreviousNonComment && PreviousNonComment->is(Kind: tok::colon) &&
1222 PreviousNonComment->isOneOf(K1: TT_ObjCMethodExpr, K2: TT_DictLiteral)) {
1223 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
1224 // method expression, the block should be aligned to the line starting it,
1225 // e.g.:
1226 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
1227 // ^(int *i) {
1228 // // ...
1229 // }];
1230 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
1231 // when we consume all of the "}"'s FakeRParens at the "{".
1232 if (State.Stack.size() > 1) {
1233 State.Stack[State.Stack.size() - 2].LastSpace =
1234 std::max(a: CurrentState.LastSpace, b: CurrentState.Indent.Total) +
1235 Style.ContinuationIndentWidth;
1236 }
1237 }
1238
1239 switch (Style.BreakInheritanceList) {
1240 case FormatStyle::BILS_BeforeColon:
1241 case FormatStyle::BILS_AfterComma:
1242 if (Current.is(TT: TT_InheritanceColon) || Previous.is(TT: TT_InheritanceComma)) {
1243 CurrentState.AlignedTo = Previous.getPreviousOneOf(
1244 Ks: tok::kw_class, Ks: tok::kw_struct, Ks: tok::kw_union);
1245 }
1246 break;
1247 case FormatStyle::BILS_BeforeComma:
1248 if (Current.isOneOf(K1: TT_InheritanceColon, K2: TT_InheritanceComma)) {
1249 CurrentState.AlignedTo = Previous.getPreviousOneOf(
1250 Ks: tok::kw_class, Ks: tok::kw_struct, Ks: tok::kw_union);
1251 }
1252 break;
1253 case FormatStyle::BILS_AfterColon:
1254 if (Previous.isOneOf(K1: TT_InheritanceColon, K2: TT_InheritanceComma))
1255 CurrentState.AlignedTo = &Previous;
1256 break;
1257 }
1258
1259 if ((PreviousNonComment &&
1260 PreviousNonComment->isOneOf(K1: tok::comma, K2: tok::semi) &&
1261 !CurrentState.AvoidBinPacking) ||
1262 Previous.is(TT: TT_BinaryOperator)) {
1263 CurrentState.BreakBeforeParameter = false;
1264 }
1265 if (PreviousNonComment &&
1266 (PreviousNonComment->isOneOf(K1: TT_TemplateCloser, K2: TT_JavaAnnotation) ||
1267 PreviousNonComment->ClosesRequiresClause) &&
1268 Current.NestingLevel == 0) {
1269 CurrentState.BreakBeforeParameter = false;
1270 }
1271 if (NextNonComment->is(Kind: tok::question) ||
1272 (PreviousNonComment && PreviousNonComment->is(Kind: tok::question))) {
1273 CurrentState.BreakBeforeParameter = true;
1274 }
1275 if (Current.is(TT: TT_BinaryOperator) && Current.CanBreakBefore) {
1276 CurrentState.BreakBeforeParameter = false;
1277 CurrentState.AlignedTo = &Current;
1278 }
1279 if (Style.AlignOperands != FormatStyle::OAS_DontAlign &&
1280 Current.is(TT: TT_ConditionalExpr)) {
1281 switch (Style.AlignOperands) {
1282 case FormatStyle::OAS_Align:
1283 CurrentState.AlignedTo = Current.is(Kind: tok::question)
1284 ? Current.getPrevious(A1: tok::equal)
1285 : Current.getPrevious(A1: tok::question);
1286 break;
1287 case FormatStyle::OAS_AlignAfterOperator:
1288 if (Current.is(Kind: tok::colon))
1289 CurrentState.AlignedTo = Current.getPrevious(A1: tok::question);
1290 break;
1291 case FormatStyle::OAS_DontAlign:
1292 break;
1293 }
1294 }
1295
1296 if (!DryRun) {
1297 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
1298 if (Current.is(Kind: tok::r_brace) && Current.MatchingParen &&
1299 // Only strip trailing empty lines for l_braces that have children, i.e.
1300 // for function expressions (lambdas, arrows, etc).
1301 !Current.MatchingParen->Children.empty()) {
1302 // lambdas and arrow functions are expressions, thus their r_brace is not
1303 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
1304 // about removing empty lines on closing blocks. Special case them here.
1305 MaxEmptyLinesToKeep = 1;
1306 }
1307 const unsigned Newlines =
1308 std::max(a: 1u, b: std::min(a: Current.NewlinesBefore, b: MaxEmptyLinesToKeep));
1309 const bool ContinuePPDirective = State.Line->InPPDirective &&
1310 State.Line->Type != LT_ImportStatement &&
1311 Current.isNot(Kind: TT_LineComment);
1312 Whitespaces.replaceWhitespace(Tok&: Current, Newlines, Spaces: State.Column, StartOfTokenColumn: State.Column,
1313 AlignedTo: CurrentState.AlignedTo, InPPDirective: ContinuePPDirective,
1314 IndentedFromColumn);
1315 }
1316
1317 if (!Current.isTrailingComment())
1318 CurrentState.LastSpace = State.Column;
1319 if (Current.is(Kind: tok::lessless)) {
1320 // If we are breaking before a "<<", we always want to indent relative to
1321 // RHS. This is necessary only for "<<", as we special-case it and don't
1322 // always indent relative to the RHS.
1323 CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1324 }
1325
1326 State.StartOfLineLevel = Current.NestingLevel;
1327 State.LowestLevelOnLine = Current.NestingLevel;
1328
1329 // Any break on this level means that the parent level has been broken
1330 // and we need to avoid bin packing there.
1331 bool NestedBlockSpecialCase =
1332 (!Style.isCpp() && Current.is(Kind: tok::r_brace) && State.Stack.size() > 1 &&
1333 State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1334 (Style.Language == FormatStyle::LK_ObjC && Current.is(Kind: tok::r_brace) &&
1335 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1336 // Do not force parameter break for statements with requires expressions.
1337 NestedBlockSpecialCase =
1338 NestedBlockSpecialCase ||
1339 (Current.MatchingParen &&
1340 Current.MatchingParen->is(TT: TT_RequiresExpressionLBrace));
1341 if (!NestedBlockSpecialCase) {
1342 auto ParentLevelIt = std::next(x: State.Stack.rbegin());
1343 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1344 Current.MatchingParen && Current.MatchingParen->is(TT: TT_LambdaLBrace)) {
1345 // If the first character on the new line is a lambda's closing brace, the
1346 // stack still contains that lambda's parenthesis. As such, we need to
1347 // recurse further down the stack than usual to find the parenthesis level
1348 // containing the lambda, which is where we want to set
1349 // BreakBeforeParameter.
1350 //
1351 // We specifically special case "OuterScope"-formatted lambdas here
1352 // because, when using that setting, breaking before the parameter
1353 // directly following the lambda is particularly unsightly. However, when
1354 // "OuterScope" is not set, the logic to find the parent parenthesis level
1355 // still appears to be sometimes incorrect. It has not been fixed yet
1356 // because it would lead to significant changes in existing behaviour.
1357 //
1358 // TODO: fix the non-"OuterScope" case too.
1359 auto FindCurrentLevel = [&](const auto &It) {
1360 return std::find_if(It, State.Stack.rend(), [](const auto &PState) {
1361 return PState.Tok != nullptr; // Ignore fake parens.
1362 });
1363 };
1364 auto MaybeIncrement = [&](const auto &It) {
1365 return It != State.Stack.rend() ? std::next(It) : It;
1366 };
1367 auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin());
1368 auto LevelContainingLambdaIt =
1369 FindCurrentLevel(MaybeIncrement(LambdaLevelIt));
1370 ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt);
1371 }
1372 for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I)
1373 I->BreakBeforeParameter = true;
1374 }
1375
1376 if (PreviousNonComment &&
1377 PreviousNonComment->isNoneOf(Ks: tok::comma, Ks: tok::colon, Ks: tok::semi) &&
1378 ((PreviousNonComment->isNot(Kind: TT_TemplateCloser) &&
1379 !PreviousNonComment->ClosesRequiresClause) ||
1380 Current.NestingLevel != 0) &&
1381 PreviousNonComment->isNoneOf(
1382 Ks: TT_BinaryOperator, Ks: TT_FunctionAnnotationRParen, Ks: TT_JavaAnnotation,
1383 Ks: TT_LeadingJavaAnnotation) &&
1384 Current.isNot(Kind: TT_BinaryOperator) && !PreviousNonComment->opensScope() &&
1385 // We don't want to enforce line breaks for subsequent arguments just
1386 // because we have been forced to break before a lambda body.
1387 (!Style.BraceWrapping.BeforeLambdaBody ||
1388 Current.isNot(Kind: TT_LambdaLBrace))) {
1389 CurrentState.BreakBeforeParameter = true;
1390 }
1391
1392 // If we break after { or the [ of an array initializer, we should also break
1393 // before the corresponding } or ].
1394 if (PreviousNonComment &&
1395 (PreviousNonComment->isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) ||
1396 opensProtoMessageField(LessTok: *PreviousNonComment, Style))) {
1397 CurrentState.BreakBeforeClosingBrace = true;
1398 }
1399
1400 if (PreviousNonComment && PreviousNonComment->is(Kind: tok::l_paren)) {
1401 if (auto Previous = PreviousNonComment->Previous) {
1402 if (Previous->isIf()) {
1403 CurrentState.BreakBeforeClosingParen = Style.BreakBeforeCloseBracketIf;
1404 } else if (Previous->isLoop(Style)) {
1405 CurrentState.BreakBeforeClosingParen =
1406 Style.BreakBeforeCloseBracketLoop;
1407 } else if (Previous->is(Kind: tok::kw_switch)) {
1408 CurrentState.BreakBeforeClosingParen =
1409 Style.BreakBeforeCloseBracketSwitch;
1410 } else {
1411 CurrentState.BreakBeforeClosingParen =
1412 Style.BreakBeforeCloseBracketFunction;
1413 }
1414 }
1415 }
1416
1417 if (PreviousNonComment && PreviousNonComment->is(TT: TT_TemplateOpener))
1418 CurrentState.BreakBeforeClosingAngle = Style.BreakBeforeTemplateCloser;
1419
1420 if (CurrentState.AvoidBinPacking) {
1421 // If we are breaking after '(', '{', '<', or this is the break after a ':'
1422 // to start a member initializer list in a constructor, this should not
1423 // be considered bin packing unless the relevant AllowAll option is false or
1424 // this is a dict/object literal.
1425 bool PreviousIsBreakingCtorInitializerColon =
1426 PreviousNonComment && PreviousNonComment->is(TT: TT_CtorInitializerColon) &&
1427 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
1428 bool AllowAllConstructorInitializersOnNextLine =
1429 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine ||
1430 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLineOnly;
1431 if ((Previous.isNoneOf(Ks: tok::l_paren, Ks: tok::l_brace, Ks: TT_BinaryOperator) &&
1432 !PreviousIsBreakingCtorInitializerColon) ||
1433 (!Style.AllowAllParametersOfDeclarationOnNextLine &&
1434 State.Line->MustBeDeclaration) ||
1435 (!Style.AllowAllArgumentsOnNextLine &&
1436 !State.Line->MustBeDeclaration) ||
1437 (!AllowAllConstructorInitializersOnNextLine &&
1438 PreviousIsBreakingCtorInitializerColon) ||
1439 Previous.is(TT: TT_DictLiteral)) {
1440 CurrentState.BreakBeforeParameter = true;
1441 }
1442
1443 // If we are breaking after a ':' to start a member initializer list,
1444 // and we allow all arguments on the next line, we should not break
1445 // before the next parameter.
1446 if (PreviousIsBreakingCtorInitializerColon &&
1447 AllowAllConstructorInitializersOnNextLine) {
1448 CurrentState.BreakBeforeParameter = false;
1449 }
1450 }
1451
1452 if (mustBreakBinaryOperation(Current, Style))
1453 CurrentState.BreakBeforeParameter = true;
1454
1455 return Penalty;
1456}
1457
1458IndentationAndAlignment
1459ContinuationIndenter::getNewLineColumn(const LineState &State) {
1460 if (!State.NextToken || !State.NextToken->Previous)
1461 return 0;
1462
1463 FormatToken &Current = *State.NextToken;
1464 const auto &CurrentState = State.Stack.back();
1465
1466 if (CurrentState.IsCSharpGenericTypeConstraint &&
1467 Current.isNot(Kind: TT_CSharpGenericTypeConstraint)) {
1468 return CurrentState.ColonPos + 2;
1469 }
1470
1471 const FormatToken &Previous = *Current.Previous;
1472 // If we are continuing an expression, we want to use the continuation indent.
1473 const auto ContinuationIndent =
1474 std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1475 b: CurrentState.Indent) +
1476 Style.ContinuationIndentWidth;
1477 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1478 const FormatToken *NextNonComment = Previous.getNextNonComment();
1479 if (!NextNonComment)
1480 NextNonComment = &Current;
1481
1482 // Java specific bits.
1483 if (Style.isJava() &&
1484 Current.isOneOf(K1: Keywords.kw_implements, K2: Keywords.kw_extends)) {
1485 return std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1486 b: CurrentState.Indent + Style.ContinuationIndentWidth);
1487 }
1488
1489 // Indentation of the statement following a Verilog case label is taken care
1490 // of in moveStateToNextToken.
1491 if (Style.isVerilog() && PreviousNonComment &&
1492 Keywords.isVerilogEndOfLabel(Tok: *PreviousNonComment)) {
1493 return State.FirstIndent;
1494 }
1495
1496 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths &&
1497 State.Line->First->is(Kind: tok::kw_enum)) {
1498 return IndentationAndAlignment(Style.IndentWidth *
1499 State.Line->First->IndentLevel) +
1500 Style.IndentWidth;
1501 }
1502
1503 if (Style.BraceWrapping.BeforeLambdaBody &&
1504 Style.BraceWrapping.IndentBraces && Current.is(TT: TT_LambdaLBrace)) {
1505 const auto From = Style.LambdaBodyIndentation == FormatStyle::LBI_Signature
1506 ? CurrentState.Indent
1507 : State.FirstIndent;
1508 return From + Style.IndentWidth;
1509 }
1510
1511 if ((NextNonComment->is(Kind: tok::l_brace) && NextNonComment->is(BBK: BK_Block)) ||
1512 (Style.isVerilog() && Keywords.isVerilogBegin(Tok: *NextNonComment))) {
1513 if (Current.NestingLevel == 0 ||
1514 (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1515 State.NextToken->is(TT: TT_LambdaLBrace))) {
1516 return State.FirstIndent;
1517 }
1518 return CurrentState.Indent;
1519 }
1520 if (Current.is(TT: TT_LambdaArrow) &&
1521 Previous.isOneOf(K1: tok::kw_noexcept, K2: tok::kw_mutable, Ks: tok::kw_constexpr,
1522 Ks: tok::kw_consteval, Ks: tok::kw_static,
1523 Ks: TT_AttributeRSquare)) {
1524 return ContinuationIndent;
1525 }
1526 if ((Current.isOneOf(K1: tok::r_brace, K2: tok::r_square) ||
1527 (Current.is(Kind: tok::greater) && (Style.isProto() || Style.isTableGen()))) &&
1528 State.Stack.size() > 1) {
1529 if (Current.closesBlockOrBlockTypeList(Style))
1530 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1531 if (Current.MatchingParen && Current.MatchingParen->is(BBK: BK_BracedInit)) {
1532 // The brace should line up with the start of the line in this case. The
1533 // stack depth is checked to make sure that the brace is at the top
1534 // level. It should contain the levels for the top, the assignment if
1535 // there is an equal sign, and the braces.
1536 //
1537 // SomeStruct //
1538 // s = {
1539 // "xxxxxxxxxxxxx",
1540 // };
1541 if ((State.Stack.size() == 2 &&
1542 Current.MatchingParen->getPreviousNonComment() &&
1543 Current.MatchingParen->getPreviousNonComment()->is(
1544 TT: TT_StartOfName)) ||
1545 (State.Stack.size() == 3 &&
1546 State.Stack[1].Precedence == prec::Assignment)) {
1547 return State.FirstIndent;
1548 }
1549 return State.Stack[State.Stack.size() - 2].LastSpace;
1550 }
1551 return State.FirstIndent;
1552 }
1553 // Indent a closing parenthesis at the previous level if followed by a semi,
1554 // const, or opening brace. This allows indentations such as:
1555 // foo(
1556 // a,
1557 // );
1558 // int Foo::getter(
1559 // //
1560 // ) const {
1561 // return foo;
1562 // }
1563 // function foo(
1564 // a,
1565 // ) {
1566 // code(); //
1567 // }
1568 if (Current.is(Kind: tok::r_paren) && State.Stack.size() > 1 &&
1569 (!Current.Next ||
1570 Current.Next->isOneOf(K1: tok::semi, K2: tok::kw_const, Ks: tok::l_brace))) {
1571 return State.Stack[State.Stack.size() - 2].LastSpace;
1572 }
1573 // When DAGArg closer exists top of line, it should be aligned in the similar
1574 // way as function call above.
1575 if (Style.isTableGen() && Current.is(TT: TT_TableGenDAGArgCloser) &&
1576 State.Stack.size() > 1) {
1577 return State.Stack[State.Stack.size() - 2].LastSpace;
1578 }
1579 if (Style.BreakBeforeCloseBracketBracedList && Current.is(Kind: tok::r_brace) &&
1580 Current.MatchingParen && Current.MatchingParen->is(BBK: BK_BracedInit) &&
1581 State.Stack.size() > 1) {
1582 return State.Stack[State.Stack.size() - 2].LastSpace;
1583 }
1584 if ((Style.BreakBeforeCloseBracketFunction ||
1585 Style.BreakBeforeCloseBracketIf || Style.BreakBeforeCloseBracketLoop ||
1586 Style.BreakBeforeCloseBracketSwitch) &&
1587 Current.is(Kind: tok::r_paren) && State.Stack.size() > 1) {
1588 return State.Stack[State.Stack.size() - 2].LastSpace;
1589 }
1590 if (Style.BreakBeforeTemplateCloser && Current.is(TT: TT_TemplateCloser) &&
1591 State.Stack.size() > 1) {
1592 return State.Stack[State.Stack.size() - 2].LastSpace;
1593 }
1594 if (NextNonComment->is(TT: TT_TemplateString) && NextNonComment->closesScope())
1595 return State.Stack[State.Stack.size() - 2].LastSpace;
1596 // Field labels in a nested type should be aligned to the brace. For example
1597 // in ProtoBuf:
1598 // optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123,
1599 // bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}];
1600 // For Verilog, a quote preceding a brace is treated as an identifier. And
1601 // Both braces and colons get annotated as TT_DictLiteral. So we have to
1602 // check.
1603 if (Current.is(Kind: tok::identifier) && Current.Next &&
1604 (!Style.isVerilog() || Current.Next->is(Kind: tok::colon)) &&
1605 (Current.Next->is(TT: TT_DictLiteral) ||
1606 (Style.isProto() && Current.Next->isOneOf(K1: tok::less, K2: tok::l_brace)))) {
1607 return CurrentState.Indent;
1608 }
1609 if (NextNonComment->is(TT: TT_ObjCStringLiteral) &&
1610 State.StartOfStringLiteral != 0) {
1611 return State.StartOfStringLiteral - 1;
1612 }
1613 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1614 return State.StartOfStringLiteral;
1615 if (NextNonComment->is(Kind: tok::lessless) && CurrentState.FirstLessLess != 0)
1616 return CurrentState.FirstLessLess;
1617 if (NextNonComment->isMemberAccess()) {
1618 if (CurrentState.CallContinuation == 0)
1619 return ContinuationIndent;
1620 return CurrentState.CallContinuation;
1621 }
1622 if (CurrentState.QuestionColumn != 0 &&
1623 ((NextNonComment->is(Kind: tok::colon) &&
1624 NextNonComment->is(TT: TT_ConditionalExpr)) ||
1625 Previous.is(TT: TT_ConditionalExpr))) {
1626 if (((NextNonComment->is(Kind: tok::colon) && NextNonComment->Next &&
1627 !NextNonComment->Next->FakeLParens.empty() &&
1628 NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1629 (Previous.is(Kind: tok::colon) && !Current.FakeLParens.empty() &&
1630 Current.FakeLParens.back() == prec::Conditional)) &&
1631 !CurrentState.IsWrappedConditional) {
1632 // NOTE: we may tweak this slightly:
1633 // * not remove the 'lead' ContinuationIndentWidth
1634 // * always un-indent by the operator when
1635 // BreakBeforeTernaryOperators=true
1636 unsigned Indent = CurrentState.Indent.Total;
1637 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1638 Indent -= Style.ContinuationIndentWidth;
1639 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1640 Indent -= 2;
1641 return Indent;
1642 }
1643 return CurrentState.QuestionColumn;
1644 }
1645 if (Previous.is(Kind: tok::comma) && CurrentState.VariablePos != 0)
1646 return CurrentState.VariablePos;
1647 if (Current.is(TT: TT_RequiresClause)) {
1648 if (Style.IndentRequiresClause)
1649 return CurrentState.Indent + Style.IndentWidth;
1650 switch (Style.RequiresClausePosition) {
1651 case FormatStyle::RCPS_OwnLine:
1652 case FormatStyle::RCPS_WithFollowing:
1653 case FormatStyle::RCPS_OwnLineWithBrace:
1654 return CurrentState.Indent;
1655 default:
1656 break;
1657 }
1658 }
1659 if (NextNonComment->isOneOf(K1: TT_CtorInitializerColon, K2: TT_InheritanceColon,
1660 Ks: TT_InheritanceComma)) {
1661 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1662 }
1663 if ((PreviousNonComment &&
1664 (PreviousNonComment->ClosesTemplateDeclaration ||
1665 PreviousNonComment->ClosesRequiresClause ||
1666 (PreviousNonComment->is(TT: TT_AttributeMacro) &&
1667 Current.isNot(Kind: tok::l_paren) &&
1668 !Current.endsSequence(K1: TT_StartOfName, Tokens: TT_AttributeMacro,
1669 Tokens: TT_PointerOrReference)) ||
1670 PreviousNonComment->isOneOf(K1: TT_AttributeRParen, K2: TT_AttributeRSquare,
1671 Ks: TT_FunctionAnnotationRParen,
1672 Ks: TT_JavaAnnotation,
1673 Ks: TT_LeadingJavaAnnotation))) ||
1674 (!Style.IndentWrappedFunctionNames &&
1675 NextNonComment->isOneOf(K1: tok::kw_operator, K2: TT_FunctionDeclarationName)) ||
1676 (State.Line->ReturnTypeWrapped && PreviousNonComment &&
1677 isReturnTypePrefixSpecifier(Tok: *PreviousNonComment))) {
1678 return std::max(a: IndentationAndAlignment(CurrentState.LastSpace),
1679 b: CurrentState.Indent);
1680 }
1681 if (NextNonComment->is(TT: TT_SelectorName)) {
1682 if (!CurrentState.ObjCSelectorNameFound) {
1683 auto MinIndent = CurrentState.Indent;
1684 if (shouldIndentWrappedSelectorName(Style, LineType: State.Line->Type)) {
1685 MinIndent =
1686 std::max(a: MinIndent, b: IndentationAndAlignment(State.FirstIndent) +
1687 Style.ContinuationIndentWidth);
1688 }
1689 // If LongestObjCSelectorName is 0, we are indenting the first
1690 // part of an ObjC selector (or a selector component which is
1691 // not colon-aligned due to block formatting).
1692 //
1693 // Otherwise, we are indenting a subsequent part of an ObjC
1694 // selector which should be colon-aligned to the longest
1695 // component of the ObjC selector.
1696 //
1697 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1698 return MinIndent.addPadding(
1699 Spaces: std::max(a: NextNonComment->LongestObjCSelectorName,
1700 b: NextNonComment->ColumnWidth) -
1701 NextNonComment->ColumnWidth);
1702 }
1703 if (!CurrentState.AlignColons)
1704 return CurrentState.Indent;
1705 if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1706 return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1707 return CurrentState.Indent;
1708 }
1709 if (NextNonComment->is(Kind: tok::colon) && NextNonComment->is(TT: TT_ObjCMethodExpr))
1710 return CurrentState.ColonPos;
1711 if (NextNonComment->is(TT: TT_ArraySubscriptLSquare)) {
1712 if (CurrentState.StartOfArraySubscripts != 0) {
1713 return CurrentState.StartOfArraySubscripts;
1714 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1715 // initializers.
1716 return CurrentState.Indent;
1717 }
1718 return ContinuationIndent;
1719 }
1720
1721 // OpenMP clauses want to get additional indentation when they are pushed onto
1722 // the next line.
1723 if (State.Line->InPragmaDirective) {
1724 FormatToken *PragmaType = State.Line->First->Next->Next;
1725 if (PragmaType && PragmaType->TokenText == "omp")
1726 return CurrentState.Indent + Style.ContinuationIndentWidth;
1727 }
1728
1729 // This ensure that we correctly format ObjC methods calls without inputs,
1730 // i.e. where the last element isn't selector like: [callee method];
1731 if (NextNonComment->is(Kind: tok::identifier) && NextNonComment->FakeRParens == 0 &&
1732 NextNonComment->Next && NextNonComment->Next->is(TT: TT_ObjCMethodExpr)) {
1733 return CurrentState.Indent;
1734 }
1735
1736 if (NextNonComment->isOneOf(K1: TT_StartOfName, K2: TT_PointerOrReference) ||
1737 Previous.isOneOf(K1: tok::coloncolon, K2: tok::equal, Ks: TT_JsTypeColon)) {
1738 return ContinuationIndent;
1739 }
1740 if (PreviousNonComment && PreviousNonComment->is(Kind: tok::colon) &&
1741 PreviousNonComment->isOneOf(K1: TT_ObjCMethodExpr, K2: TT_DictLiteral)) {
1742 return ContinuationIndent;
1743 }
1744 if (NextNonComment->is(TT: TT_CtorInitializerComma))
1745 return CurrentState.Indent;
1746 if (PreviousNonComment && PreviousNonComment->is(TT: TT_CtorInitializerColon) &&
1747 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1748 return CurrentState.Indent;
1749 }
1750 if (PreviousNonComment && PreviousNonComment->is(TT: TT_InheritanceColon) &&
1751 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
1752 return CurrentState.Indent;
1753 }
1754 if (Previous.is(Kind: tok::r_paren) &&
1755 Previous.isNot(Kind: TT_TableGenDAGArgOperatorToBreak) &&
1756 !Current.isBinaryOperator() &&
1757 Current.isNoneOf(Ks: tok::colon, Ks: tok::comment)) {
1758 return ContinuationIndent;
1759 }
1760 if (Current.is(TT: TT_ProtoExtensionLSquare))
1761 return CurrentState.Indent;
1762 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1763 return CurrentState.Indent - Current.Tok.getLength() -
1764 Current.SpacesRequiredBefore;
1765 }
1766 if (Current.is(Kind: tok::comment) && NextNonComment->isBinaryOperator() &&
1767 CurrentState.UnindentOperator) {
1768 return CurrentState.Indent - NextNonComment->Tok.getLength() -
1769 NextNonComment->SpacesRequiredBefore;
1770 }
1771 if (CurrentState.Indent.Total == State.FirstIndent && PreviousNonComment &&
1772 PreviousNonComment->isNoneOf(Ks: tok::r_brace, Ks: TT_CtorInitializerComma)) {
1773 // Ensure that we fall back to the continuation indent width instead of
1774 // just flushing continuations left.
1775 return CurrentState.Indent + Style.ContinuationIndentWidth;
1776 }
1777 return CurrentState.Indent;
1778}
1779
1780static bool hasNestedBlockInlined(const FormatToken *Previous,
1781 const FormatToken &Current,
1782 const FormatStyle &Style) {
1783 if (Previous->isNot(Kind: tok::l_paren))
1784 return true;
1785 if (Previous->ParameterCount > 1)
1786 return true;
1787
1788 // Also a nested block if contains a lambda inside function with 1 parameter.
1789 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT: TT_LambdaLSquare);
1790}
1791
1792unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1793 bool DryRun, bool Newline) {
1794 assert(State.Stack.size());
1795 const FormatToken &Current = *State.NextToken;
1796 auto &CurrentState = State.Stack.back();
1797
1798 if (Current.is(TT: TT_CSharpGenericTypeConstraint))
1799 CurrentState.IsCSharpGenericTypeConstraint = true;
1800 if (Current.isOneOf(K1: tok::comma, K2: TT_BinaryOperator))
1801 CurrentState.NoLineBreakInOperand = false;
1802 if (Current.isOneOf(K1: TT_InheritanceColon, K2: TT_CSharpGenericTypeConstraintColon))
1803 CurrentState.AvoidBinPacking = true;
1804 if (Current.is(Kind: tok::lessless) && Current.isNot(Kind: TT_OverloadedOperator)) {
1805 if (CurrentState.FirstLessLess == 0)
1806 CurrentState.FirstLessLess = State.Column;
1807 else
1808 CurrentState.LastOperatorWrapped = Newline;
1809 }
1810 if (Current.is(TT: TT_BinaryOperator) && Current.isNot(Kind: tok::lessless))
1811 CurrentState.LastOperatorWrapped = Newline;
1812 if (Current.is(TT: TT_ConditionalExpr) && Current.Previous &&
1813 Current.Previous->isNot(Kind: TT_ConditionalExpr)) {
1814 CurrentState.LastOperatorWrapped = Newline;
1815 }
1816 if (Current.is(TT: TT_ArraySubscriptLSquare) &&
1817 CurrentState.StartOfArraySubscripts == 0) {
1818 CurrentState.StartOfArraySubscripts = State.Column;
1819 }
1820
1821 auto IsWrappedConditional = [](const FormatToken &Tok) {
1822 if (!(Tok.is(TT: TT_ConditionalExpr) && Tok.is(Kind: tok::question)))
1823 return false;
1824 if (Tok.MustBreakBefore)
1825 return true;
1826
1827 const FormatToken *Next = Tok.getNextNonComment();
1828 return Next && Next->MustBreakBefore;
1829 };
1830 if (IsWrappedConditional(Current))
1831 CurrentState.IsWrappedConditional = true;
1832 if (Style.BreakBeforeTernaryOperators && Current.is(Kind: tok::question))
1833 CurrentState.QuestionColumn = State.Column;
1834 if (!Style.BreakBeforeTernaryOperators && Current.isNot(Kind: tok::colon)) {
1835 const FormatToken *Previous = Current.Previous;
1836 while (Previous && Previous->isTrailingComment())
1837 Previous = Previous->Previous;
1838 if (Previous && Previous->is(Kind: tok::question))
1839 CurrentState.QuestionColumn = State.Column;
1840 }
1841 if (!Current.opensScope() && !Current.closesScope() &&
1842 Current.isNot(Kind: TT_PointerOrReference)) {
1843 State.LowestLevelOnLine =
1844 std::min(a: State.LowestLevelOnLine, b: Current.NestingLevel);
1845 }
1846 if (Current.isMemberAccess())
1847 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1848 if (Current.is(TT: TT_SelectorName))
1849 CurrentState.ObjCSelectorNameFound = true;
1850 if (Current.is(TT: TT_CtorInitializerColon) &&
1851 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1852 // Indent 2 from the column, so:
1853 // SomeClass::SomeClass()
1854 // : First(...), ...
1855 // Next(...)
1856 // ^ line up here.
1857 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1858 FormatStyle::BCIS_BeforeComma
1859 ? 0
1860 : 2);
1861 CurrentState.NestedBlockIndent = CurrentState.Indent.Total;
1862 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) {
1863 CurrentState.AvoidBinPacking = true;
1864 CurrentState.BreakBeforeParameter =
1865 Style.ColumnLimit > 0 &&
1866 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine &&
1867 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLineOnly;
1868 } else {
1869 CurrentState.BreakBeforeParameter = false;
1870 }
1871 }
1872 if (Current.is(TT: TT_CtorInitializerColon) &&
1873 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1874 CurrentState.Indent =
1875 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1876 CurrentState.NestedBlockIndent = CurrentState.Indent.Total;
1877 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack)
1878 CurrentState.AvoidBinPacking = true;
1879 else
1880 CurrentState.BreakBeforeParameter = false;
1881 }
1882 if (Current.is(TT: TT_InheritanceColon)) {
1883 CurrentState.Indent =
1884 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1885 }
1886 if (Current.isOneOf(K1: TT_BinaryOperator, K2: TT_ConditionalExpr) && Newline)
1887 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1888 if (Current.isOneOf(K1: TT_LambdaLSquare, K2: TT_LambdaArrow))
1889 CurrentState.LastSpace = State.Column;
1890 if (Current.is(TT: TT_RequiresExpression) &&
1891 Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) {
1892 CurrentState.NestedBlockIndent = State.Column;
1893 }
1894
1895 // Insert scopes created by fake parenthesis.
1896 const FormatToken *Previous = Current.getPreviousNonComment();
1897
1898 // Add special behavior to support a format commonly used for JavaScript
1899 // closures:
1900 // SomeFunction(function() {
1901 // foo();
1902 // bar();
1903 // }, a, b, c);
1904 if (Current.isNot(Kind: tok::comment) && !Current.ClosesRequiresClause &&
1905 Previous && Previous->isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) &&
1906 Previous->isNot(Kind: TT_DictLiteral) && State.Stack.size() > 1 &&
1907 !CurrentState.HasMultipleNestedBlocks) {
1908 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1909 for (ParenState &PState : llvm::drop_end(RangeOrContainer&: State.Stack))
1910 PState.NoLineBreak = true;
1911 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1912 }
1913 if (Previous && (Previous->isOneOf(K1: TT_BinaryOperator, K2: TT_ConditionalExpr) ||
1914 (Previous->isOneOf(K1: tok::l_paren, K2: tok::comma, Ks: tok::colon) &&
1915 Previous->isNoneOf(Ks: TT_DictLiteral, Ks: TT_ObjCMethodExpr,
1916 Ks: TT_CtorInitializerColon)))) {
1917 CurrentState.NestedBlockInlined =
1918 !Newline && hasNestedBlockInlined(Previous, Current, Style);
1919 }
1920
1921 moveStatePastFakeLParens(State, Newline);
1922 moveStatePastScopeCloser(State);
1923 // Do not use CurrentState here, since the two functions before may change the
1924 // Stack.
1925 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1926 !State.Stack.back().NoLineBreakInOperand;
1927 moveStatePastScopeOpener(State, Newline);
1928 moveStatePastFakeRParens(State);
1929
1930 if (Current.is(TT: TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1931 State.StartOfStringLiteral = State.Column + 1;
1932 if (Current.is(TT: TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1933 State.StartOfStringLiteral = State.Column + 1;
1934 } else if (Current.is(TT: TT_TableGenMultiLineString) &&
1935 State.StartOfStringLiteral == 0) {
1936 State.StartOfStringLiteral = State.Column + 1;
1937 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1938 State.StartOfStringLiteral = State.Column;
1939 } else if (Current.isNoneOf(Ks: tok::comment, Ks: tok::identifier, Ks: tok::hash) &&
1940 !Current.isStringLiteral()) {
1941 State.StartOfStringLiteral = 0;
1942 }
1943
1944 State.Column += Current.ColumnWidth;
1945 State.NextToken = State.NextToken->Next;
1946 // Verilog case labels are on the same unwrapped lines as the statements that
1947 // follow. TokenAnnotator identifies them and sets MustBreakBefore.
1948 // Indentation is taken care of here. A case label can only have 1 statement
1949 // in Verilog, so we don't have to worry about lines that follow.
1950 if (Style.isVerilog() && State.NextToken &&
1951 State.NextToken->MustBreakBefore &&
1952 Keywords.isVerilogEndOfLabel(Tok: Current)) {
1953 State.FirstIndent += Style.IndentWidth;
1954 CurrentState.Indent = State.FirstIndent;
1955 }
1956
1957 unsigned Penalty =
1958 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1959
1960 if (Current.Role)
1961 Current.Role->formatFromToken(State, Indenter: this, DryRun);
1962 // If the previous has a special role, let it consume tokens as appropriate.
1963 // It is necessary to start at the previous token for the only implemented
1964 // role (comma separated list). That way, the decision whether or not to break
1965 // after the "{" is already done and both options are tried and evaluated.
1966 // FIXME: This is ugly, find a better way.
1967 if (Previous && Previous->Role)
1968 Penalty += Previous->Role->formatAfterToken(State, Indenter: this, DryRun);
1969
1970 return Penalty;
1971}
1972
1973void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1974 bool Newline) {
1975 const FormatToken &Current = *State.NextToken;
1976 if (Current.FakeLParens.empty())
1977 return;
1978
1979 const FormatToken *Previous = Current.getPreviousNonComment();
1980
1981 // Don't add extra indentation for the first fake parenthesis after
1982 // 'return', assignments, opening <({[, or requires clauses. The indentation
1983 // for these cases is special cased.
1984 bool SkipFirstExtraIndent =
1985 Previous &&
1986 (Previous->opensScope() ||
1987 Previous->isOneOf(K1: tok::semi, K2: tok::kw_return, Ks: TT_RequiresClause) ||
1988 (Previous->getPrecedence() == prec::Assignment &&
1989 Style.AlignOperands != FormatStyle::OAS_DontAlign) ||
1990 Previous->is(TT: TT_ObjCMethodExpr));
1991 for (const auto &PrecedenceLevel : llvm::reverse(C: Current.FakeLParens)) {
1992 const auto &CurrentState = State.Stack.back();
1993 ParenState NewParenState = CurrentState;
1994 NewParenState.Tok = nullptr;
1995 NewParenState.ContainsLineBreak = false;
1996 NewParenState.LastOperatorWrapped = true;
1997 NewParenState.IsChainedConditional = false;
1998 NewParenState.IsWrappedConditional = false;
1999 NewParenState.UnindentOperator = false;
2000 NewParenState.NoLineBreak =
2001 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
2002 NewParenState.Precedence = PrecedenceLevel;
2003
2004 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
2005 if (PrecedenceLevel > prec::Comma)
2006 NewParenState.AvoidBinPacking = false;
2007
2008 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
2009 // a builder type call after 'return' or, if the alignment after opening
2010 // brackets is disabled.
2011 if (!Current.isTrailingComment() &&
2012 (Style.AlignOperands != FormatStyle::OAS_DontAlign ||
2013 PrecedenceLevel < prec::Assignment) &&
2014 (!Previous || Previous->isNot(Kind: tok::kw_return) ||
2015 (!Style.isJava() && PrecedenceLevel > 0)) &&
2016 (Style.AlignAfterOpenBracket || PrecedenceLevel > prec::Comma ||
2017 Current.NestingLevel == 0) &&
2018 (!Style.isTableGen() ||
2019 (Previous && Previous->isOneOf(K1: TT_TableGenDAGArgListComma,
2020 K2: TT_TableGenDAGArgListCommaToBreak)))) {
2021 NewParenState.Indent =
2022 std::max(l: {IndentationAndAlignment(State.Column), NewParenState.Indent,
2023 IndentationAndAlignment(CurrentState.LastSpace)});
2024 }
2025
2026 // Special case for generic selection expressions, its comma-separated
2027 // expressions are not aligned to the opening paren like regular calls, but
2028 // rather continuation-indented relative to the _Generic keyword.
2029 if (Previous && Previous->endsSequence(K1: tok::l_paren, Tokens: tok::kw__Generic) &&
2030 State.Stack.size() > 1) {
2031 NewParenState.Indent = State.Stack[State.Stack.size() - 2].Indent +
2032 Style.ContinuationIndentWidth;
2033 }
2034
2035 if ((shouldUnindentNextOperator(Tok: Current) ||
2036 (Previous &&
2037 (PrecedenceLevel == prec::Conditional &&
2038 Previous->is(Kind: tok::question) && Previous->is(TT: TT_ConditionalExpr)))) &&
2039 !Newline) {
2040 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
2041 // the operator and keep the operands aligned.
2042 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator)
2043 NewParenState.UnindentOperator = true;
2044 // Mark indentation as alignment if the expression is aligned.
2045 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
2046 NewParenState.AlignedTo = Previous;
2047 }
2048
2049 // Do not indent relative to the fake parentheses inserted for "." or "->".
2050 // This is a special case to make the following to statements consistent:
2051 // OuterFunction(InnerFunctionCall( // break
2052 // ParameterToInnerFunction));
2053 // OuterFunction(SomeObject.InnerFunctionCall( // break
2054 // ParameterToInnerFunction));
2055 if (PrecedenceLevel > prec::Unknown)
2056 NewParenState.LastSpace = std::max(a: NewParenState.LastSpace, b: State.Column);
2057 if (PrecedenceLevel != prec::Conditional &&
2058 Current.isNot(Kind: TT_UnaryOperator) && Style.AlignAfterOpenBracket) {
2059 NewParenState.StartOfFunctionCall = State.Column;
2060 }
2061
2062 // Indent conditional expressions, unless they are chained "else-if"
2063 // conditionals. Never indent expression where the 'operator' is ',', ';' or
2064 // an assignment (i.e. *I <= prec::Assignment) as those have different
2065 // indentation rules. Indent other expression, unless the indentation needs
2066 // to be skipped.
2067 if (PrecedenceLevel == prec::Conditional && Previous &&
2068 Previous->is(Kind: tok::colon) && Previous->is(TT: TT_ConditionalExpr) &&
2069 &PrecedenceLevel == &Current.FakeLParens.back() &&
2070 !CurrentState.IsWrappedConditional) {
2071 NewParenState.IsChainedConditional = true;
2072 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
2073 } else if (PrecedenceLevel == prec::Conditional ||
2074 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
2075 !Current.isTrailingComment())) {
2076 NewParenState.Indent += Style.ContinuationIndentWidth;
2077 }
2078 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
2079 NewParenState.BreakBeforeParameter = false;
2080 State.Stack.push_back(Elt: NewParenState);
2081 SkipFirstExtraIndent = false;
2082 }
2083}
2084
2085void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
2086 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
2087 unsigned VariablePos = State.Stack.back().VariablePos;
2088 if (State.Stack.size() == 1) {
2089 // Do not pop the last element.
2090 break;
2091 }
2092 State.Stack.pop_back();
2093 State.Stack.back().VariablePos = VariablePos;
2094 }
2095
2096 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
2097 // Remove the indentation of the requires clauses (which is not in Indent,
2098 // but in LastSpace).
2099 State.Stack.back().LastSpace -= Style.IndentWidth;
2100 }
2101}
2102
2103void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
2104 bool Newline) {
2105 const FormatToken &Current = *State.NextToken;
2106 if (!Current.opensScope())
2107 return;
2108
2109 const auto &CurrentState = State.Stack.back();
2110
2111 // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
2112 if (Current.isOneOf(K1: tok::less, K2: tok::l_paren) &&
2113 CurrentState.IsCSharpGenericTypeConstraint) {
2114 return;
2115 }
2116
2117 if (Current.MatchingParen && Current.is(BBK: BK_Block)) {
2118 moveStateToNewBlock(State, NewLine: Newline);
2119 return;
2120 }
2121
2122 const bool EndsInComma = [](const FormatToken *Tok) {
2123 if (!Tok)
2124 return false;
2125 const auto *Prev = Tok->getPreviousNonComment();
2126 if (!Prev)
2127 return false;
2128 return Prev->is(Kind: tok::comma);
2129 }(Current.MatchingParen);
2130
2131 IndentationAndAlignment NewIndent = 0;
2132 unsigned LastSpace = CurrentState.LastSpace;
2133 bool AvoidBinPacking;
2134 bool BreakBeforeParameter = false;
2135 unsigned NestedBlockIndent = std::max(a: CurrentState.StartOfFunctionCall,
2136 b: CurrentState.NestedBlockIndent);
2137 if (Current.isOneOf(K1: tok::l_brace, K2: TT_ArrayInitializerLSquare) ||
2138 opensProtoMessageField(LessTok: Current, Style)) {
2139 if (Current.opensBlockOrBlockTypeList(Style)) {
2140 NewIndent = Style.IndentWidth +
2141 std::min(a: State.Column, b: CurrentState.NestedBlockIndent);
2142 } else if (Current.is(Kind: tok::l_brace)) {
2143 const auto Width = Style.BracedInitializerIndentWidth;
2144 NewIndent = IndentationAndAlignment(CurrentState.LastSpace) +
2145 (Width < 0 ? Style.ContinuationIndentWidth : Width);
2146 } else {
2147 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
2148 }
2149 const FormatToken *NextNonComment = Current.getNextNonComment();
2150 AvoidBinPacking =
2151 EndsInComma || Current.is(TT: TT_DictLiteral) || Style.isProto() ||
2152 Style.PackArguments.BinPack == FormatStyle::BPAS_OnePerLine ||
2153 (NextNonComment &&
2154 NextNonComment->isOneOf(K1: TT_DesignatedInitializerPeriod,
2155 K2: TT_DesignatedInitializerLSquare));
2156 BreakBeforeParameter = EndsInComma;
2157 if (Current.ParameterCount > 1)
2158 NestedBlockIndent = std::max(a: NestedBlockIndent, b: State.Column + 1);
2159 } else {
2160 NewIndent = IndentationAndAlignment(std::max(
2161 a: CurrentState.LastSpace, b: CurrentState.StartOfFunctionCall)) +
2162 Style.ContinuationIndentWidth;
2163
2164 if (Style.isTableGen() && Current.is(TT: TT_TableGenDAGArgOpenerToBreak) &&
2165 Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakElements) {
2166 // For the case the next token is a TableGen DAGArg operator identifier
2167 // that is not marked to have a line break after it.
2168 // In this case the option DAS_BreakElements requires to align the
2169 // DAGArg elements to the operator.
2170 const FormatToken *Next = Current.Next;
2171 if (Next && Next->is(TT: TT_TableGenDAGArgOperatorID))
2172 NewIndent = State.Column + Next->TokenText.size() + 2;
2173 }
2174
2175 // Ensure that different different brackets force relative alignment, e.g.:
2176 // void SomeFunction(vector< // break
2177 // int> v);
2178 // FIXME: We likely want to do this for more combinations of brackets.
2179 if (Current.is(Kind: tok::less) && Current.ParentBracket == tok::l_paren) {
2180 NewIndent = std::max(a: NewIndent, b: CurrentState.Indent);
2181 LastSpace = std::max(a: LastSpace, b: CurrentState.Indent.Total);
2182 }
2183
2184 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
2185 // for backwards compatibility.
2186 bool ObjCBinPackProtocolList =
2187 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
2188 (Style.PackParameters.BinPack == FormatStyle::BPPS_BinPack ||
2189 Style.PackParameters.BinPack == FormatStyle::BPPS_UseBreakAfter)) ||
2190 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
2191
2192 bool BinPackDeclaration =
2193 (State.Line->Type != LT_ObjCDecl &&
2194 (Style.PackParameters.BinPack == FormatStyle::BPPS_BinPack ||
2195 Style.PackParameters.BinPack == FormatStyle::BPPS_UseBreakAfter)) ||
2196 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
2197
2198 bool GenericSelection =
2199 Current.getPreviousNonComment() &&
2200 Current.getPreviousNonComment()->is(Kind: tok::kw__Generic);
2201
2202 AvoidBinPacking =
2203 (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
2204 (Style.isJavaScript() && EndsInComma) ||
2205 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
2206 (!State.Line->MustBeDeclaration &&
2207 Style.PackArguments.BinPack == FormatStyle::BPAS_OnePerLine) ||
2208 (Style.ExperimentalAutoDetectBinPacking &&
2209 (Current.is(PPK: PPK_OnePerLine) ||
2210 (!BinPackInconclusiveFunctions && Current.is(PPK: PPK_Inconclusive))));
2211
2212 if (Current.is(TT: TT_ObjCMethodExpr) && Current.MatchingParen &&
2213 Style.ObjCBreakBeforeNestedBlockParam) {
2214 if (Style.ColumnLimit) {
2215 // If this '[' opens an ObjC call, determine whether all parameters fit
2216 // into one line and put one per line if they don't.
2217 if (getLengthToMatchingParen(Tok: Current, Stack: State.Stack) + State.Column >
2218 getColumnLimit(State)) {
2219 BreakBeforeParameter = true;
2220 }
2221 } else {
2222 // For ColumnLimit = 0, we have to figure out whether there is or has to
2223 // be a line break within this call.
2224 for (const FormatToken *Tok = &Current;
2225 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
2226 if (Tok->MustBreakBefore ||
2227 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
2228 BreakBeforeParameter = true;
2229 break;
2230 }
2231 }
2232 }
2233 }
2234
2235 if (Style.isJavaScript() && EndsInComma)
2236 BreakBeforeParameter = true;
2237 }
2238 // Generally inherit NoLineBreak from the current scope to nested scope.
2239 // However, don't do this for non-empty nested blocks, dict literals and
2240 // array literals as these follow different indentation rules.
2241 bool NoLineBreak =
2242 Current.Children.empty() &&
2243 Current.isNoneOf(Ks: TT_DictLiteral, Ks: TT_ArrayInitializerLSquare) &&
2244 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
2245 (Current.is(TT: TT_TemplateOpener) &&
2246 CurrentState.ContainsUnwrappedBuilder));
2247 State.Stack.push_back(
2248 Elt: ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
2249 auto &NewState = State.Stack.back();
2250 NewState.NestedBlockIndent = NestedBlockIndent;
2251 NewState.BreakBeforeParameter = BreakBeforeParameter;
2252 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
2253
2254 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next &&
2255 Current.is(Kind: tok::l_paren)) {
2256 // Search for any parameter that is a lambda.
2257 FormatToken const *next = Current.Next;
2258 while (next) {
2259 if (next->is(TT: TT_LambdaLSquare)) {
2260 NewState.HasMultipleNestedBlocks = true;
2261 break;
2262 }
2263 next = next->Next;
2264 }
2265 }
2266
2267 NewState.IsInsideObjCArrayLiteral = Current.is(TT: TT_ArrayInitializerLSquare) &&
2268 Current.Previous &&
2269 Current.Previous->is(Kind: tok::at);
2270}
2271
2272void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
2273 const FormatToken &Current = *State.NextToken;
2274 if (!Current.closesScope())
2275 return;
2276
2277 // If we encounter a closing ), ], } or >, we can remove a level from our
2278 // stacks.
2279 if (State.Stack.size() > 1 &&
2280 (Current.isOneOf(K1: tok::r_paren, K2: tok::r_square, Ks: TT_TemplateString) ||
2281 (Current.is(Kind: tok::r_brace) && State.NextToken != State.Line->First) ||
2282 State.NextToken->is(TT: TT_TemplateCloser) ||
2283 State.NextToken->is(TT: TT_TableGenListCloser) ||
2284 (Current.is(Kind: tok::greater) && Current.is(TT: TT_DictLiteral)))) {
2285 State.Stack.pop_back();
2286 }
2287
2288 auto &CurrentState = State.Stack.back();
2289
2290 // Reevaluate whether ObjC message arguments fit into one line.
2291 // If a receiver spans multiple lines, e.g.:
2292 // [[object block:^{
2293 // return 42;
2294 // }] a:42 b:42];
2295 // BreakBeforeParameter is calculated based on an incorrect assumption
2296 // (it is checked whether the whole expression fits into one line without
2297 // considering a line break inside a message receiver).
2298 // We check whether arguments fit after receiver scope closer (into the same
2299 // line).
2300 if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
2301 Current.MatchingParen->Previous) {
2302 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
2303 if (CurrentScopeOpener.is(TT: TT_ObjCMethodExpr) &&
2304 CurrentScopeOpener.MatchingParen) {
2305 int NecessarySpaceInLine =
2306 getLengthToMatchingParen(Tok: CurrentScopeOpener, Stack: State.Stack) +
2307 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
2308 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
2309 Style.ColumnLimit) {
2310 CurrentState.BreakBeforeParameter = false;
2311 }
2312 }
2313 }
2314
2315 if (Current.is(Kind: tok::r_square)) {
2316 // If this ends the array subscript expr, reset the corresponding value.
2317 const FormatToken *NextNonComment = Current.getNextNonComment();
2318 if (NextNonComment && NextNonComment->isNot(Kind: tok::l_square))
2319 CurrentState.StartOfArraySubscripts = 0;
2320 }
2321}
2322
2323void ContinuationIndenter::moveStateToNewBlock(LineState &State, bool NewLine) {
2324 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
2325 State.NextToken->is(TT: TT_LambdaLBrace) &&
2326 !State.Line->MightBeFunctionDecl) {
2327 const auto Indent = Style.IndentWidth * Style.BraceWrapping.IndentBraces;
2328 State.Stack.back().NestedBlockIndent = State.FirstIndent + Indent;
2329 }
2330 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
2331 // ObjC block sometimes follow special indentation rules.
2332 unsigned NewIndent =
2333 NestedBlockIndent + (State.NextToken->is(TT: TT_ObjCBlockLBrace)
2334 ? Style.ObjCBlockIndentWidth
2335 : Style.IndentWidth);
2336
2337 // Even when wrapping before lambda body, the left brace can still be added to
2338 // the same line. This occurs when checking whether the whole lambda body can
2339 // go on a single line. In this case we have to make sure there are no line
2340 // breaks in the body, otherwise we could just end up with a regular lambda
2341 // body without the brace wrapped.
2342 bool NoLineBreak = Style.BraceWrapping.BeforeLambdaBody && !NewLine &&
2343 State.NextToken->is(TT: TT_LambdaLBrace);
2344
2345 State.Stack.push_back(Elt: ParenState(State.NextToken, NewIndent,
2346 State.Stack.back().LastSpace,
2347 /*AvoidBinPacking=*/true, NoLineBreak));
2348 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
2349 State.Stack.back().BreakBeforeParameter = true;
2350}
2351
2352static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
2353 unsigned TabWidth,
2354 encoding::Encoding Encoding) {
2355 size_t LastNewlinePos = Text.find_last_of(Chars: "\n");
2356 if (LastNewlinePos == StringRef::npos) {
2357 return StartColumn +
2358 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
2359 } else {
2360 return encoding::columnWidthWithTabs(Text: Text.substr(Start: LastNewlinePos),
2361 /*StartColumn=*/0, TabWidth, Encoding);
2362 }
2363}
2364
2365unsigned ContinuationIndenter::reformatRawStringLiteral(
2366 const FormatToken &Current, LineState &State,
2367 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
2368 unsigned StartColumn = State.Column - Current.ColumnWidth;
2369 StringRef OldDelimiter = *getRawStringDelimiter(TokenText: Current.TokenText);
2370 StringRef NewDelimiter =
2371 getCanonicalRawStringDelimiter(Style, Language: RawStringStyle.Language);
2372 if (NewDelimiter.empty())
2373 NewDelimiter = OldDelimiter;
2374 // The text of a raw string is between the leading 'R"delimiter(' and the
2375 // trailing 'delimiter)"'.
2376 unsigned OldPrefixSize = 3 + OldDelimiter.size();
2377 unsigned OldSuffixSize = 2 + OldDelimiter.size();
2378 // We create a virtual text environment which expects a null-terminated
2379 // string, so we cannot use StringRef.
2380 std::string RawText = std::string(
2381 Current.TokenText.substr(Start: OldPrefixSize).drop_back(N: OldSuffixSize));
2382 if (NewDelimiter != OldDelimiter) {
2383 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
2384 // raw string.
2385 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
2386 if (StringRef(RawText).contains(Other: CanonicalDelimiterSuffix))
2387 NewDelimiter = OldDelimiter;
2388 }
2389
2390 unsigned NewPrefixSize = 3 + NewDelimiter.size();
2391 unsigned NewSuffixSize = 2 + NewDelimiter.size();
2392
2393 // The first start column is the column the raw text starts after formatting.
2394 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
2395
2396 // The next start column is the intended indentation a line break inside
2397 // the raw string at level 0. It is determined by the following rules:
2398 // - if the content starts on newline, it is one level more than the current
2399 // indent, and
2400 // - if the content does not start on a newline, it is the first start
2401 // column.
2402 // These rules have the advantage that the formatted content both does not
2403 // violate the rectangle rule and visually flows within the surrounding
2404 // source.
2405 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
2406 // If this token is the last parameter (checked by looking if it's followed by
2407 // `)` and is not on a newline, the base the indent off the line's nested
2408 // block indent. Otherwise, base the indent off the arguments indent, so we
2409 // can achieve:
2410 //
2411 // fffffffffff(1, 2, 3, R"pb(
2412 // key1: 1 #
2413 // key2: 2)pb");
2414 //
2415 // fffffffffff(1, 2, 3,
2416 // R"pb(
2417 // key1: 1 #
2418 // key2: 2
2419 // )pb");
2420 //
2421 // fffffffffff(1, 2, 3,
2422 // R"pb(
2423 // key1: 1 #
2424 // key2: 2
2425 // )pb",
2426 // 5);
2427 unsigned CurrentIndent =
2428 (!Newline && Current.Next && Current.Next->is(Kind: tok::r_paren))
2429 ? State.Stack.back().NestedBlockIndent
2430 : State.Stack.back().Indent.Total;
2431 unsigned NextStartColumn = ContentStartsOnNewline
2432 ? CurrentIndent + Style.IndentWidth
2433 : FirstStartColumn;
2434
2435 // The last start column is the column the raw string suffix starts if it is
2436 // put on a newline.
2437 // The last start column is the intended indentation of the raw string postfix
2438 // if it is put on a newline. It is determined by the following rules:
2439 // - if the raw string prefix starts on a newline, it is the column where
2440 // that raw string prefix starts, and
2441 // - if the raw string prefix does not start on a newline, it is the current
2442 // indent.
2443 unsigned LastStartColumn =
2444 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
2445
2446 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
2447 Style: RawStringStyle, Code: RawText, Ranges: {tooling::Range(0, RawText.size())},
2448 FirstStartColumn, NextStartColumn, LastStartColumn, FileName: "<stdin>",
2449 /*Status=*/nullptr);
2450
2451 auto NewCode = applyAllReplacements(Code: RawText, Replaces: Fixes.first);
2452 if (!NewCode)
2453 return addMultilineToken(Current, State);
2454 if (!DryRun) {
2455 if (NewDelimiter != OldDelimiter) {
2456 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
2457 // of the token.
2458 SourceLocation PrefixDelimiterStart =
2459 Current.Tok.getLocation().getLocWithOffset(Offset: 2);
2460 auto PrefixErr = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2461 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2462 if (PrefixErr) {
2463 llvm::errs()
2464 << "Failed to update the prefix delimiter of a raw string: "
2465 << llvm::toString(E: std::move(PrefixErr)) << "\n";
2466 }
2467 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
2468 // position length - 1 - |delimiter|.
2469 SourceLocation SuffixDelimiterStart =
2470 Current.Tok.getLocation().getLocWithOffset(Offset: Current.TokenText.size() -
2471 1 - OldDelimiter.size());
2472 auto SuffixErr = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2473 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2474 if (SuffixErr) {
2475 llvm::errs()
2476 << "Failed to update the suffix delimiter of a raw string: "
2477 << llvm::toString(E: std::move(SuffixErr)) << "\n";
2478 }
2479 }
2480 SourceLocation OriginLoc =
2481 Current.Tok.getLocation().getLocWithOffset(Offset: OldPrefixSize);
2482 for (const tooling::Replacement &Fix : Fixes.first) {
2483 auto Err = Whitespaces.addReplacement(Replacement: tooling::Replacement(
2484 SourceMgr, OriginLoc.getLocWithOffset(Offset: Fix.getOffset()),
2485 Fix.getLength(), Fix.getReplacementText()));
2486 if (Err) {
2487 llvm::errs() << "Failed to reformat raw string: "
2488 << llvm::toString(E: std::move(Err)) << "\n";
2489 }
2490 }
2491 }
2492 unsigned RawLastLineEndColumn = getLastLineEndColumn(
2493 Text: *NewCode, StartColumn: FirstStartColumn, TabWidth: Style.TabWidth, Encoding);
2494 State.Column = RawLastLineEndColumn + NewSuffixSize;
2495 // Since we're updating the column to after the raw string literal here, we
2496 // have to manually add the penalty for the prefix R"delim( over the column
2497 // limit.
2498 unsigned PrefixExcessCharacters =
2499 StartColumn + NewPrefixSize > Style.ColumnLimit
2500 ? StartColumn + NewPrefixSize - Style.ColumnLimit
2501 : 0;
2502 bool IsMultiline =
2503 ContentStartsOnNewline || (NewCode->find(c: '\n') != std::string::npos);
2504 if (IsMultiline) {
2505 // Break before further function parameters on all levels.
2506 for (ParenState &Paren : State.Stack)
2507 Paren.BreakBeforeParameter = true;
2508 }
2509 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
2510}
2511
2512unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
2513 LineState &State) {
2514 // Break before further function parameters on all levels.
2515 for (ParenState &Paren : State.Stack)
2516 Paren.BreakBeforeParameter = true;
2517
2518 unsigned ColumnsUsed = State.Column;
2519 // We can only affect layout of the first and the last line, so the penalty
2520 // for all other lines is constant, and we ignore it.
2521 State.Column = Current.LastLineColumnWidth;
2522
2523 if (ColumnsUsed > getColumnLimit(State))
2524 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2525 return 0;
2526}
2527
2528unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2529 LineState &State, bool DryRun,
2530 bool AllowBreak, bool Newline) {
2531 unsigned Penalty = 0;
2532 // Compute the raw string style to use in case this is a raw string literal
2533 // that can be reformatted.
2534 auto RawStringStyle = getRawStringStyle(Current, State);
2535 if (RawStringStyle && !Current.Finalized) {
2536 Penalty = reformatRawStringLiteral(Current, State, RawStringStyle: *RawStringStyle, DryRun,
2537 Newline);
2538 } else if (Current.IsMultiline && Current.isNot(Kind: TT_BlockComment)) {
2539 // Don't break multi-line tokens other than block comments and raw string
2540 // literals. Instead, just update the state.
2541 Penalty = addMultilineToken(Current, State);
2542 } else if (State.Line->Type != LT_ImportStatement) {
2543 // We generally don't break import statements.
2544 LineState OriginalState = State;
2545
2546 // Whether we force the reflowing algorithm to stay strictly within the
2547 // column limit.
2548 bool Strict = false;
2549 // Whether the first non-strict attempt at reflowing did intentionally
2550 // exceed the column limit.
2551 bool Exceeded = false;
2552 std::tie(args&: Penalty, args&: Exceeded) = breakProtrudingToken(
2553 Current, State, AllowBreak, /*DryRun=*/true, Strict);
2554 if (Exceeded) {
2555 // If non-strict reflowing exceeds the column limit, try whether strict
2556 // reflowing leads to an overall lower penalty.
2557 LineState StrictState = OriginalState;
2558 unsigned StrictPenalty =
2559 breakProtrudingToken(Current, State&: StrictState, AllowBreak,
2560 /*DryRun=*/true, /*Strict=*/true)
2561 .first;
2562 Strict = StrictPenalty <= Penalty;
2563 if (Strict) {
2564 Penalty = StrictPenalty;
2565 State = std::move(StrictState);
2566 }
2567 }
2568 if (!DryRun) {
2569 // If we're not in dry-run mode, apply the changes with the decision on
2570 // strictness made above.
2571 breakProtrudingToken(Current, State&: OriginalState, AllowBreak, /*DryRun=*/false,
2572 Strict);
2573 }
2574 }
2575 if (State.Column > getColumnLimit(State)) {
2576 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2577 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2578 }
2579 return Penalty;
2580}
2581
2582// Returns the enclosing function name of a token, or the empty string if not
2583// found.
2584static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2585 // Look for: 'function(' or 'function<templates>(' before Current.
2586 auto Tok = Current.getPreviousNonComment();
2587 if (!Tok || Tok->isNot(Kind: tok::l_paren))
2588 return "";
2589 Tok = Tok->getPreviousNonComment();
2590 if (!Tok)
2591 return "";
2592 if (Tok->is(TT: TT_TemplateCloser)) {
2593 Tok = Tok->MatchingParen;
2594 if (Tok)
2595 Tok = Tok->getPreviousNonComment();
2596 }
2597 if (!Tok || Tok->isNot(Kind: tok::identifier))
2598 return "";
2599 return Tok->TokenText;
2600}
2601
2602std::optional<FormatStyle>
2603ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2604 const LineState &State) {
2605 if (!Current.isStringLiteral())
2606 return std::nullopt;
2607 auto Delimiter = getRawStringDelimiter(TokenText: Current.TokenText);
2608 if (!Delimiter)
2609 return std::nullopt;
2610 auto RawStringStyle = RawStringFormats.getDelimiterStyle(Delimiter: *Delimiter);
2611 if (!RawStringStyle && Delimiter->empty()) {
2612 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2613 EnclosingFunction: getEnclosingFunctionName(Current));
2614 }
2615 if (!RawStringStyle)
2616 return std::nullopt;
2617 RawStringStyle->ColumnLimit = getColumnLimit(State);
2618 return RawStringStyle;
2619}
2620
2621std::unique_ptr<BreakableToken>
2622ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2623 LineState &State, bool AllowBreak) {
2624 unsigned StartColumn = State.Column - Current.ColumnWidth;
2625 if (Current.isStringLiteral()) {
2626 // Strings in JSON cannot be broken. Breaking strings in JavaScript is
2627 // disabled for now.
2628 if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals ||
2629 !AllowBreak) {
2630 return nullptr;
2631 }
2632
2633 // Don't break string literals inside preprocessor directives (except for
2634 // #define directives, as their contents are stored in separate lines and
2635 // are not affected by this check).
2636 // This way we avoid breaking code with line directives and unknown
2637 // preprocessor directives that contain long string literals.
2638 if (State.Line->Type == LT_PreprocessorDirective)
2639 return nullptr;
2640 // Exempts unterminated string literals from line breaking. The user will
2641 // likely want to terminate the string before any line breaking is done.
2642 if (Current.IsUnterminatedLiteral)
2643 return nullptr;
2644 // Don't break string literals inside Objective-C array literals (doing so
2645 // raises the warning -Wobjc-string-concatenation).
2646 if (State.Stack.back().IsInsideObjCArrayLiteral)
2647 return nullptr;
2648
2649 // The "DPI"/"DPI-C" in SystemVerilog direct programming interface
2650 // imports/exports cannot be split, e.g.
2651 // `import "DPI" function foo();`
2652 // FIXME: make this use same infra as C++ import checks
2653 if (Style.isVerilog() && Current.Previous &&
2654 Current.Previous->isOneOf(K1: tok::kw_export, K2: Keywords.kw_import)) {
2655 return nullptr;
2656 }
2657 StringRef Text = Current.TokenText;
2658
2659 // We need this to address the case where there is an unbreakable tail only
2660 // if certain other formatting decisions have been taken. The
2661 // UnbreakableTailLength of Current is an overapproximation in that case and
2662 // we need to be correct here.
2663 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2664 ? 0
2665 : Current.UnbreakableTailLength;
2666
2667 if (Style.isVerilog() || Style.isJava() || Style.isJavaScript() ||
2668 Style.isCSharp()) {
2669 BreakableStringLiteralUsingOperators::QuoteStyleType QuoteStyle;
2670 if (Style.isJavaScript() && Text.starts_with(Prefix: "'") &&
2671 Text.ends_with(Suffix: "'")) {
2672 QuoteStyle = BreakableStringLiteralUsingOperators::SingleQuotes;
2673 } else if (Style.isCSharp() && Text.starts_with(Prefix: "@\"") &&
2674 Text.ends_with(Suffix: "\"")) {
2675 QuoteStyle = BreakableStringLiteralUsingOperators::AtDoubleQuotes;
2676 } else if (Text.starts_with(Prefix: "\"") && Text.ends_with(Suffix: "\"")) {
2677 QuoteStyle = BreakableStringLiteralUsingOperators::DoubleQuotes;
2678 } else {
2679 return nullptr;
2680 }
2681 return std::make_unique<BreakableStringLiteralUsingOperators>(
2682 args: Current, args&: QuoteStyle,
2683 /*UnindentPlus=*/args: shouldUnindentNextOperator(Tok: Current), args&: StartColumn,
2684 args&: UnbreakableTailLength, args: State.Line->InPPDirective, args&: Encoding, args&: Style);
2685 }
2686
2687 StringRef Prefix;
2688 StringRef Postfix;
2689 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2690 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2691 // reduce the overhead) for each FormatToken, which is a string, so that we
2692 // don't run multiple checks here on the hot path.
2693 if ((Text.ends_with(Suffix: Postfix = "\"") &&
2694 (Text.starts_with(Prefix: Prefix = "@\"") || Text.starts_with(Prefix: Prefix = "\"") ||
2695 Text.starts_with(Prefix: Prefix = "u\"") ||
2696 Text.starts_with(Prefix: Prefix = "U\"") ||
2697 Text.starts_with(Prefix: Prefix = "u8\"") ||
2698 Text.starts_with(Prefix: Prefix = "L\""))) ||
2699 (Text.starts_with(Prefix: Prefix = "_T(\"") &&
2700 Text.ends_with(Suffix: Postfix = "\")"))) {
2701 return std::make_unique<BreakableStringLiteral>(
2702 args: Current, args&: StartColumn, args&: Prefix, args&: Postfix, args&: UnbreakableTailLength,
2703 args: State.Line->InPPDirective, args&: Encoding, args&: Style);
2704 }
2705 } else if (Current.is(TT: TT_BlockComment)) {
2706 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2707 // If a comment token switches formatting, like
2708 // /* clang-format on */, we don't want to break it further,
2709 // but we may still want to adjust its indentation.
2710 switchesFormatting(Token: Current)) {
2711 return nullptr;
2712 }
2713 return std::make_unique<BreakableBlockComment>(
2714 args: Current, args&: StartColumn, args: Current.OriginalColumn, args: !Current.Previous,
2715 args: State.Line->InPPDirective, args&: Encoding, args&: Style, args: Whitespaces.useCRLF());
2716 } else if (Current.is(TT: TT_LineComment) &&
2717 (!Current.Previous ||
2718 Current.Previous->isNot(Kind: TT_ImplicitStringLiteral))) {
2719 bool RegularComments = [&]() {
2720 for (const FormatToken *T = &Current; T && T->is(TT: TT_LineComment);
2721 T = T->Next) {
2722 if (!(T->TokenText.starts_with(Prefix: "//") || T->TokenText.starts_with(Prefix: "#")))
2723 return false;
2724 }
2725 return true;
2726 }();
2727 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2728 CommentPragmasRegex.match(String: Current.TokenText.substr(Start: 2)) ||
2729 switchesFormatting(Token: Current) || !RegularComments) {
2730 return nullptr;
2731 }
2732 return std::make_unique<BreakableLineCommentSection>(
2733 args: Current, args&: StartColumn, /*InPPDirective=*/args: false, args&: Encoding, args&: Style);
2734 }
2735 return nullptr;
2736}
2737
2738std::pair<unsigned, bool>
2739ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2740 LineState &State, bool AllowBreak,
2741 bool DryRun, bool Strict) {
2742 std::unique_ptr<const BreakableToken> Token =
2743 createBreakableToken(Current, State, AllowBreak);
2744 if (!Token)
2745 return {0, false};
2746 assert(Token->getLineCount() > 0);
2747 unsigned ColumnLimit = getColumnLimit(State);
2748 if (Current.is(TT: TT_LineComment)) {
2749 // We don't insert backslashes when breaking line comments.
2750 ColumnLimit = Style.ColumnLimit;
2751 }
2752 if (ColumnLimit == 0) {
2753 // To make the rest of the function easier set the column limit to the
2754 // maximum, if there should be no limit.
2755 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2756 }
2757 if (Current.UnbreakableTailLength >= ColumnLimit)
2758 return {0, false};
2759 // ColumnWidth was already accounted into State.Column before calling
2760 // breakProtrudingToken.
2761 unsigned StartColumn = State.Column - Current.ColumnWidth;
2762 unsigned NewBreakPenalty = Current.isStringLiteral()
2763 ? Style.PenaltyBreakString
2764 : Style.PenaltyBreakComment;
2765 // Stores whether we intentionally decide to let a line exceed the column
2766 // limit.
2767 bool Exceeded = false;
2768 // Stores whether we introduce a break anywhere in the token.
2769 bool BreakInserted = Token->introducesBreakBeforeToken();
2770 // Store whether we inserted a new line break at the end of the previous
2771 // logical line.
2772 bool NewBreakBefore = false;
2773 // We use a conservative reflowing strategy. Reflow starts after a line is
2774 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2775 // line that doesn't get reflown with the previous line is reached.
2776 bool Reflow = false;
2777 // Keep track of where we are in the token:
2778 // Where we are in the content of the current logical line.
2779 unsigned TailOffset = 0;
2780 // The column number we're currently at.
2781 unsigned ContentStartColumn =
2782 Token->getContentStartColumn(LineIndex: 0, /*Break=*/false);
2783 // The number of columns left in the current logical line after TailOffset.
2784 unsigned RemainingTokenColumns =
2785 Token->getRemainingLength(LineIndex: 0, Offset: TailOffset, StartColumn: ContentStartColumn);
2786 // Adapt the start of the token, for example indent.
2787 if (!DryRun)
2788 Token->adaptStartOfLine(LineIndex: 0, Whitespaces);
2789
2790 unsigned ContentIndent = 0;
2791 unsigned Penalty = 0;
2792 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2793 << StartColumn << ".\n");
2794 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2795 LineIndex != EndIndex; ++LineIndex) {
2796 LLVM_DEBUG(llvm::dbgs()
2797 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2798 NewBreakBefore = false;
2799 // If we did reflow the previous line, we'll try reflowing again. Otherwise
2800 // we'll start reflowing if the current line is broken or whitespace is
2801 // compressed.
2802 bool TryReflow = Reflow;
2803 // Break the current token until we can fit the rest of the line.
2804 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2805 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
2806 << (ContentStartColumn + RemainingTokenColumns)
2807 << ", space: " << ColumnLimit
2808 << ", reflown prefix: " << ContentStartColumn
2809 << ", offset in line: " << TailOffset << "\n");
2810 // If the current token doesn't fit, find the latest possible split in the
2811 // current line so that breaking at it will be under the column limit.
2812 // FIXME: Use the earliest possible split while reflowing to correctly
2813 // compress whitespace within a line.
2814 BreakableToken::Split Split =
2815 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2816 ContentStartColumn, CommentPragmasRegex);
2817 if (Split.first == StringRef::npos) {
2818 // No break opportunity - update the penalty and continue with the next
2819 // logical line.
2820 if (LineIndex < EndIndex - 1) {
2821 // The last line's penalty is handled in addNextStateToQueue() or when
2822 // calling replaceWhitespaceAfterLastLine below.
2823 Penalty += Style.PenaltyExcessCharacter *
2824 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2825 }
2826 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
2827 break;
2828 }
2829 assert(Split.first != 0);
2830
2831 if (Token->supportsReflow()) {
2832 // Check whether the next natural split point after the current one can
2833 // still fit the line, either because we can compress away whitespace,
2834 // or because the penalty the excess characters introduce is lower than
2835 // the break penalty.
2836 // We only do this for tokens that support reflowing, and thus allow us
2837 // to change the whitespace arbitrarily (e.g. comments).
2838 // Other tokens, like string literals, can be broken on arbitrary
2839 // positions.
2840
2841 // First, compute the columns from TailOffset to the next possible split
2842 // position.
2843 // For example:
2844 // ColumnLimit: |
2845 // // Some text that breaks
2846 // ^ tail offset
2847 // ^-- split
2848 // ^-------- to split columns
2849 // ^--- next split
2850 // ^--------------- to next split columns
2851 unsigned ToSplitColumns = Token->getRangeLength(
2852 LineIndex, Offset: TailOffset, Length: Split.first, StartColumn: ContentStartColumn);
2853 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
2854
2855 BreakableToken::Split NextSplit = Token->getSplit(
2856 LineIndex, TailOffset: TailOffset + Split.first + Split.second, ColumnLimit,
2857 ContentStartColumn: ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2858 // Compute the columns necessary to fit the next non-breakable sequence
2859 // into the current line.
2860 unsigned ToNextSplitColumns = 0;
2861 if (NextSplit.first == StringRef::npos) {
2862 ToNextSplitColumns = Token->getRemainingLength(LineIndex, Offset: TailOffset,
2863 StartColumn: ContentStartColumn);
2864 } else {
2865 ToNextSplitColumns = Token->getRangeLength(
2866 LineIndex, Offset: TailOffset,
2867 Length: Split.first + Split.second + NextSplit.first, StartColumn: ContentStartColumn);
2868 }
2869 // Compress the whitespace between the break and the start of the next
2870 // unbreakable sequence.
2871 ToNextSplitColumns =
2872 Token->getLengthAfterCompression(RemainingTokenColumns: ToNextSplitColumns, Split);
2873 LLVM_DEBUG(llvm::dbgs()
2874 << " ContentStartColumn: " << ContentStartColumn << "\n");
2875 LLVM_DEBUG(llvm::dbgs()
2876 << " ToNextSplit: " << ToNextSplitColumns << "\n");
2877 // If the whitespace compression makes us fit, continue on the current
2878 // line.
2879 bool ContinueOnLine =
2880 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2881 unsigned ExcessCharactersPenalty = 0;
2882 if (!ContinueOnLine && !Strict) {
2883 // Similarly, if the excess characters' penalty is lower than the
2884 // penalty of introducing a new break, continue on the current line.
2885 ExcessCharactersPenalty =
2886 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2887 Style.PenaltyExcessCharacter;
2888 LLVM_DEBUG(llvm::dbgs()
2889 << " Penalty excess: " << ExcessCharactersPenalty
2890 << "\n break : " << NewBreakPenalty << "\n");
2891 if (ExcessCharactersPenalty < NewBreakPenalty) {
2892 Exceeded = true;
2893 ContinueOnLine = true;
2894 }
2895 }
2896 if (ContinueOnLine) {
2897 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
2898 // The current line fits after compressing the whitespace - reflow
2899 // the next line into it if possible.
2900 TryReflow = true;
2901 if (!DryRun) {
2902 Token->compressWhitespace(LineIndex, TailOffset, Split,
2903 Whitespaces);
2904 }
2905 // When we continue on the same line, leave one space between content.
2906 ContentStartColumn += ToSplitColumns + 1;
2907 Penalty += ExcessCharactersPenalty;
2908 TailOffset += Split.first + Split.second;
2909 RemainingTokenColumns = Token->getRemainingLength(
2910 LineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
2911 continue;
2912 }
2913 }
2914 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
2915 // Update the ContentIndent only if the current line was not reflown with
2916 // the previous line, since in that case the previous line should still
2917 // determine the ContentIndent. Also never intent the last line.
2918 if (!Reflow)
2919 ContentIndent = Token->getContentIndent(LineIndex);
2920 LLVM_DEBUG(llvm::dbgs()
2921 << " ContentIndent: " << ContentIndent << "\n");
2922 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2923 LineIndex, /*Break=*/true);
2924
2925 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2926 LineIndex, Offset: TailOffset + Split.first + Split.second,
2927 StartColumn: ContentStartColumn);
2928 if (NewRemainingTokenColumns == 0) {
2929 // No content to indent.
2930 ContentIndent = 0;
2931 ContentStartColumn =
2932 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2933 NewRemainingTokenColumns = Token->getRemainingLength(
2934 LineIndex, Offset: TailOffset + Split.first + Split.second,
2935 StartColumn: ContentStartColumn);
2936 }
2937
2938 // When breaking before a tab character, it may be moved by a few columns,
2939 // but will still be expanded to the next tab stop, so we don't save any
2940 // columns.
2941 if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2942 // FIXME: Do we need to adjust the penalty?
2943 break;
2944 }
2945
2946 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2947 << ", " << Split.second << "\n");
2948 if (!DryRun) {
2949 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2950 Whitespaces);
2951 }
2952
2953 Penalty += NewBreakPenalty;
2954 TailOffset += Split.first + Split.second;
2955 RemainingTokenColumns = NewRemainingTokenColumns;
2956 BreakInserted = true;
2957 NewBreakBefore = true;
2958 }
2959 // In case there's another line, prepare the state for the start of the next
2960 // line.
2961 if (LineIndex + 1 != EndIndex) {
2962 unsigned NextLineIndex = LineIndex + 1;
2963 if (NewBreakBefore) {
2964 // After breaking a line, try to reflow the next line into the current
2965 // one once RemainingTokenColumns fits.
2966 TryReflow = true;
2967 }
2968 if (TryReflow) {
2969 // We decided that we want to try reflowing the next line into the
2970 // current one.
2971 // We will now adjust the state as if the reflow is successful (in
2972 // preparation for the next line), and see whether that works. If we
2973 // decide that we cannot reflow, we will later reset the state to the
2974 // start of the next line.
2975 Reflow = false;
2976 // As we did not continue breaking the line, RemainingTokenColumns is
2977 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2978 // the position at which we want to format the next line if we do
2979 // actually reflow.
2980 // When we reflow, we need to add a space between the end of the current
2981 // line and the next line's start column.
2982 ContentStartColumn += RemainingTokenColumns + 1;
2983 // Get the split that we need to reflow next logical line into the end
2984 // of the current one; the split will include any leading whitespace of
2985 // the next logical line.
2986 BreakableToken::Split SplitBeforeNext =
2987 Token->getReflowSplit(LineIndex: NextLineIndex, CommentPragmasRegex);
2988 LLVM_DEBUG(llvm::dbgs()
2989 << " Size of reflown text: " << ContentStartColumn
2990 << "\n Potential reflow split: ");
2991 if (SplitBeforeNext.first != StringRef::npos) {
2992 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2993 << SplitBeforeNext.second << "\n");
2994 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2995 // If the rest of the next line fits into the current line below the
2996 // column limit, we can safely reflow.
2997 RemainingTokenColumns = Token->getRemainingLength(
2998 LineIndex: NextLineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
2999 Reflow = true;
3000 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
3001 LLVM_DEBUG(llvm::dbgs()
3002 << " Over limit after reflow, need: "
3003 << (ContentStartColumn + RemainingTokenColumns)
3004 << ", space: " << ColumnLimit
3005 << ", reflown prefix: " << ContentStartColumn
3006 << ", offset in line: " << TailOffset << "\n");
3007 // If the whole next line does not fit, try to find a point in
3008 // the next line at which we can break so that attaching the part
3009 // of the next line to that break point onto the current line is
3010 // below the column limit.
3011 BreakableToken::Split Split =
3012 Token->getSplit(LineIndex: NextLineIndex, TailOffset, ColumnLimit,
3013 ContentStartColumn, CommentPragmasRegex);
3014 if (Split.first == StringRef::npos) {
3015 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
3016 Reflow = false;
3017 } else {
3018 // Check whether the first split point gets us below the column
3019 // limit. Note that we will execute this split below as part of
3020 // the normal token breaking and reflow logic within the line.
3021 unsigned ToSplitColumns = Token->getRangeLength(
3022 LineIndex: NextLineIndex, Offset: TailOffset, Length: Split.first, StartColumn: ContentStartColumn);
3023 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
3024 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
3025 << (ContentStartColumn + ToSplitColumns)
3026 << ", space: " << ColumnLimit);
3027 unsigned ExcessCharactersPenalty =
3028 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
3029 Style.PenaltyExcessCharacter;
3030 if (NewBreakPenalty < ExcessCharactersPenalty)
3031 Reflow = false;
3032 }
3033 }
3034 }
3035 } else {
3036 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
3037 }
3038 }
3039 if (!Reflow) {
3040 // If we didn't reflow into the next line, the only space to consider is
3041 // the next logical line. Reset our state to match the start of the next
3042 // line.
3043 TailOffset = 0;
3044 ContentStartColumn =
3045 Token->getContentStartColumn(LineIndex: NextLineIndex, /*Break=*/false);
3046 RemainingTokenColumns = Token->getRemainingLength(
3047 LineIndex: NextLineIndex, Offset: TailOffset, StartColumn: ContentStartColumn);
3048 // Adapt the start of the token, for example indent.
3049 if (!DryRun)
3050 Token->adaptStartOfLine(LineIndex: NextLineIndex, Whitespaces);
3051 } else {
3052 // If we found a reflow split and have added a new break before the next
3053 // line, we are going to remove the line break at the start of the next
3054 // logical line. For example, here we'll add a new line break after
3055 // 'text', and subsequently delete the line break between 'that' and
3056 // 'reflows'.
3057 // // some text that
3058 // // reflows
3059 // ->
3060 // // some text
3061 // // that reflows
3062 // When adding the line break, we also added the penalty for it, so we
3063 // need to subtract that penalty again when we remove the line break due
3064 // to reflowing.
3065 if (NewBreakBefore) {
3066 assert(Penalty >= NewBreakPenalty);
3067 Penalty -= NewBreakPenalty;
3068 }
3069 if (!DryRun)
3070 Token->reflow(LineIndex: NextLineIndex, Whitespaces);
3071 }
3072 }
3073 }
3074
3075 BreakableToken::Split SplitAfterLastLine =
3076 Token->getSplitAfterLastLine(TailOffset);
3077 if (SplitAfterLastLine.first != StringRef::npos) {
3078 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
3079
3080 // We add the last line's penalty here, since that line is going to be split
3081 // now.
3082 Penalty += Style.PenaltyExcessCharacter *
3083 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
3084
3085 if (!DryRun) {
3086 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
3087 Whitespaces);
3088 }
3089 ContentStartColumn =
3090 Token->getContentStartColumn(LineIndex: Token->getLineCount() - 1, /*Break=*/true);
3091 RemainingTokenColumns = Token->getRemainingLength(
3092 LineIndex: Token->getLineCount() - 1,
3093 Offset: TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
3094 StartColumn: ContentStartColumn);
3095 }
3096
3097 State.Column = ContentStartColumn + RemainingTokenColumns -
3098 Current.UnbreakableTailLength;
3099
3100 if (BreakInserted) {
3101 if (!DryRun)
3102 Token->updateAfterBroken(Whitespaces);
3103
3104 // If we break the token inside a parameter list, we need to break before
3105 // the next parameter on all levels, so that the next parameter is clearly
3106 // visible. Line comments already introduce a break.
3107 if (Current.isNot(Kind: TT_LineComment))
3108 for (ParenState &Paren : State.Stack)
3109 Paren.BreakBeforeParameter = true;
3110
3111 if (Current.is(TT: TT_BlockComment))
3112 State.NoContinuation = true;
3113
3114 State.Stack.back().LastSpace = StartColumn;
3115 }
3116
3117 Token->updateNextToken(State);
3118
3119 return {Penalty, Exceeded};
3120}
3121
3122unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
3123 // In preprocessor directives reserve two chars for trailing " \".
3124 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
3125}
3126
3127bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
3128 const FormatToken &Current = *State.NextToken;
3129 if (!Current.isStringLiteral() || Current.is(TT: TT_ImplicitStringLiteral))
3130 return false;
3131 // We never consider raw string literals "multiline" for the purpose of
3132 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
3133 // (see TokenAnnotator::mustBreakBefore().
3134 if (Current.TokenText.starts_with(Prefix: "R\""))
3135 return false;
3136 if (Current.IsMultiline)
3137 return true;
3138 if (Current.getNextNonComment() &&
3139 Current.getNextNonComment()->isStringLiteral()) {
3140 return true; // Implicit concatenation.
3141 }
3142 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
3143 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
3144 Style.ColumnLimit) {
3145 return true; // String will be split.
3146 }
3147 return false;
3148}
3149
3150} // namespace format
3151} // namespace clang
3152