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