[ELF] ObjFile::parse: check EM_AARCH64 for SHT_AARCH64_MEMTAG_GLOBAL_STATIC
[llvm-project.git] / clang / lib / Format / ContinuationIndenter.cpp
blobaed86c1fb99551ca9ada659a8fe9683c511f9064
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 //===----------------------------------------------------------------------===//
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>
27 #define DEBUG_TYPE "format-indenter"
29 namespace clang {
30 namespace format {
32 // Returns true if a TT_SelectorName should be indented when wrapped,
33 // false otherwise.
34 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
35 LineType LineType) {
36 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
39 // Returns true if a binary operator following \p Tok should be unindented when
40 // the style permits it.
41 static bool shouldUnindentNextOperator(const FormatToken &Tok) {
42 const FormatToken *Previous = Tok.getPreviousNonComment();
43 return Previous && (Previous->getPrecedence() == prec::Assignment ||
44 Previous->isOneOf(tok::kw_return, TT_RequiresClause));
47 // Returns the length of everything up to the first possible line break after
48 // the ), ], } or > matching \c Tok.
49 static 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.
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.
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}}
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;
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 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
111 const ParenState *State = FindParenState(End->Next->MatchingParen);
112 if (State && State->BreakBeforeClosingBrace)
113 break;
116 return End->TotalLength - Tok.TotalLength + 1;
119 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
120 if (!Tok.NextOperator)
121 return 0;
122 return Tok.NextOperator->TotalLength - Tok.TotalLength;
125 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
126 // segment of a builder type call.
127 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
128 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
131 // Returns \c true if \c Token in an alignable binary operator
132 static 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_BinaryOperator) && !HasTwoOperands &&
136 Token.getPrecedence() > prec::Conditional &&
137 Token.getPrecedence() < prec::PointerToMember;
140 // Returns \c true if \c Current starts the next operand in a binary operation.
141 static bool startsNextOperand(const FormatToken &Current) {
142 assert(Current.Previous);
143 const auto &Previous = *Current.Previous;
144 return isAlignableBinaryOperator(Previous) && !Current.isTrailingComment();
147 // Returns \c true if \c Current is a binary operation that must break.
148 static bool mustBreakBinaryOperation(const FormatToken &Current,
149 const FormatStyle &Style) {
150 return Style.BreakBinaryOperations != FormatStyle::BBO_Never &&
151 (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None
152 ? startsNextOperand
153 : isAlignableBinaryOperator)(Current);
156 static bool opensProtoMessageField(const FormatToken &LessTok,
157 const FormatStyle &Style) {
158 if (LessTok.isNot(tok::less))
159 return false;
160 return Style.Language == FormatStyle::LK_TextProto ||
161 (Style.Language == FormatStyle::LK_Proto &&
162 (LessTok.NestingLevel > 0 ||
163 (LessTok.Previous && LessTok.Previous->is(tok::equal))));
166 // Returns the delimiter of a raw string literal, or std::nullopt if TokenText
167 // is not the text of a raw string literal. The delimiter could be the empty
168 // string. For example, the delimiter of R"deli(cont)deli" is deli.
169 static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
170 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
171 || !TokenText.starts_with("R\"") || !TokenText.ends_with("\"")) {
172 return std::nullopt;
175 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
176 // size at most 16 by the standard, so the first '(' must be among the first
177 // 19 bytes.
178 size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
179 if (LParenPos == StringRef::npos)
180 return std::nullopt;
181 StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
183 // Check that the string ends in ')Delimiter"'.
184 size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
185 if (TokenText[RParenPos] != ')')
186 return std::nullopt;
187 if (!TokenText.substr(RParenPos + 1).starts_with(Delimiter))
188 return std::nullopt;
189 return Delimiter;
192 // Returns the canonical delimiter for \p Language, or the empty string if no
193 // canonical delimiter is specified.
194 static StringRef
195 getCanonicalRawStringDelimiter(const FormatStyle &Style,
196 FormatStyle::LanguageKind Language) {
197 for (const auto &Format : Style.RawStringFormats)
198 if (Format.Language == Language)
199 return StringRef(Format.CanonicalDelimiter);
200 return "";
203 RawStringFormatStyleManager::RawStringFormatStyleManager(
204 const FormatStyle &CodeStyle) {
205 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
206 std::optional<FormatStyle> LanguageStyle =
207 CodeStyle.GetLanguageStyle(RawStringFormat.Language);
208 if (!LanguageStyle) {
209 FormatStyle PredefinedStyle;
210 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
211 RawStringFormat.Language, &PredefinedStyle)) {
212 PredefinedStyle = getLLVMStyle();
213 PredefinedStyle.Language = RawStringFormat.Language;
215 LanguageStyle = PredefinedStyle;
217 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
218 for (StringRef Delimiter : RawStringFormat.Delimiters)
219 DelimiterStyle.insert({Delimiter, *LanguageStyle});
220 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions)
221 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
225 std::optional<FormatStyle>
226 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
227 auto It = DelimiterStyle.find(Delimiter);
228 if (It == DelimiterStyle.end())
229 return std::nullopt;
230 return It->second;
233 std::optional<FormatStyle>
234 RawStringFormatStyleManager::getEnclosingFunctionStyle(
235 StringRef EnclosingFunction) const {
236 auto It = EnclosingFunctionStyle.find(EnclosingFunction);
237 if (It == EnclosingFunctionStyle.end())
238 return std::nullopt;
239 return It->second;
242 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
243 const AdditionalKeywords &Keywords,
244 const SourceManager &SourceMgr,
245 WhitespaceManager &Whitespaces,
246 encoding::Encoding Encoding,
247 bool BinPackInconclusiveFunctions)
248 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
249 Whitespaces(Whitespaces), Encoding(Encoding),
250 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
251 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
253 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
254 unsigned FirstStartColumn,
255 const AnnotatedLine *Line,
256 bool DryRun) {
257 LineState State;
258 State.FirstIndent = FirstIndent;
259 if (FirstStartColumn && Line->First->NewlinesBefore == 0)
260 State.Column = FirstStartColumn;
261 else
262 State.Column = FirstIndent;
263 // With preprocessor directive indentation, the line starts on column 0
264 // since it's indented after the hash, but FirstIndent is set to the
265 // preprocessor indent.
266 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
267 (Line->Type == LT_PreprocessorDirective ||
268 Line->Type == LT_ImportStatement)) {
269 State.Column = 0;
271 State.Line = Line;
272 State.NextToken = Line->First;
273 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
274 /*AvoidBinPacking=*/false,
275 /*NoLineBreak=*/false));
276 State.NoContinuation = false;
277 State.StartOfStringLiteral = 0;
278 State.NoLineBreak = false;
279 State.StartOfLineLevel = 0;
280 State.LowestLevelOnLine = 0;
281 State.IgnoreStackForComparison = false;
283 if (Style.Language == FormatStyle::LK_TextProto) {
284 // We need this in order to deal with the bin packing of text fields at
285 // global scope.
286 auto &CurrentState = State.Stack.back();
287 CurrentState.AvoidBinPacking = true;
288 CurrentState.BreakBeforeParameter = true;
289 CurrentState.AlignColons = false;
292 // The first token has already been indented and thus consumed.
293 moveStateToNextToken(State, DryRun, /*Newline=*/false);
294 return State;
297 bool ContinuationIndenter::canBreak(const LineState &State) {
298 const FormatToken &Current = *State.NextToken;
299 const FormatToken &Previous = *Current.Previous;
300 const auto &CurrentState = State.Stack.back();
301 assert(&Previous == Current.Previous);
302 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace &&
303 Current.closesBlockOrBlockTypeList(Style))) {
304 return false;
306 // The opening "{" of a braced list has to be on the same line as the first
307 // element if it is nested in another braced init list or function call.
308 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
309 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) &&
310 Previous.Previous &&
311 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) {
312 return false;
314 // This prevents breaks like:
315 // ...
316 // SomeParameter, OtherParameter).DoSomething(
317 // ...
318 // As they hide "DoSomething" and are generally bad for readability.
319 if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
320 State.LowestLevelOnLine < State.StartOfLineLevel &&
321 State.LowestLevelOnLine < Current.NestingLevel) {
322 return false;
324 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder)
325 return false;
327 // Don't create a 'hanging' indent if there are multiple blocks in a single
328 // statement and we are aligning lambda blocks to their signatures.
329 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
330 State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
331 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks &&
332 Style.LambdaBodyIndentation == FormatStyle::LBI_Signature) {
333 return false;
336 // Don't break after very short return types (e.g. "void") as that is often
337 // unexpected.
338 if (Current.is(TT_FunctionDeclarationName)) {
339 if (Style.BreakAfterReturnType == FormatStyle::RTBS_None &&
340 State.Column < 6) {
341 return false;
344 if (Style.BreakAfterReturnType == FormatStyle::RTBS_ExceptShortType) {
345 assert(State.Column >= State.FirstIndent);
346 if (State.Column - State.FirstIndent < 6)
347 return false;
351 // Don't allow breaking before a closing brace of a block-indented braced list
352 // initializer if there isn't already a break.
353 if (Current.is(tok::r_brace) && Current.MatchingParen &&
354 Current.isBlockIndentedInitRBrace(Style)) {
355 return CurrentState.BreakBeforeClosingBrace;
358 // If binary operators are moved to the next line (including commas for some
359 // styles of constructor initializers), that's always ok.
360 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
361 // Allow breaking opening brace of lambdas (when passed as function
362 // arguments) to a new line when BeforeLambdaBody brace wrapping is
363 // enabled.
364 (!Style.BraceWrapping.BeforeLambdaBody ||
365 Current.isNot(TT_LambdaLBrace)) &&
366 CurrentState.NoLineBreakInOperand) {
367 return false;
370 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr))
371 return false;
373 if (Current.is(TT_ConditionalExpr) && Previous.is(tok::r_paren) &&
374 Previous.MatchingParen && Previous.MatchingParen->Previous &&
375 Previous.MatchingParen->Previous->MatchingParen &&
376 Previous.MatchingParen->Previous->MatchingParen->is(TT_LambdaLBrace)) {
377 // We have a lambda within a conditional expression, allow breaking here.
378 assert(Previous.MatchingParen->Previous->is(tok::r_brace));
379 return true;
382 return !State.NoLineBreak && !CurrentState.NoLineBreak;
385 bool ContinuationIndenter::mustBreak(const LineState &State) {
386 const FormatToken &Current = *State.NextToken;
387 const FormatToken &Previous = *Current.Previous;
388 const auto &CurrentState = State.Stack.back();
389 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore &&
390 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) {
391 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack);
392 return LambdaBodyLength > getColumnLimit(State);
394 if (Current.MustBreakBefore ||
395 (Current.is(TT_InlineASMColon) &&
396 (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always ||
397 (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_OnlyMultiline &&
398 Style.ColumnLimit > 0)))) {
399 return true;
401 if (CurrentState.BreakBeforeClosingBrace &&
402 (Current.closesBlockOrBlockTypeList(Style) ||
403 (Current.is(tok::r_brace) &&
404 Current.isBlockIndentedInitRBrace(Style)))) {
405 return true;
407 if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren))
408 return true;
409 if (Style.Language == FormatStyle::LK_ObjC &&
410 Style.ObjCBreakBeforeNestedBlockParam &&
411 Current.ObjCSelectorNameParts > 1 &&
412 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
413 return true;
415 // Avoid producing inconsistent states by requiring breaks where they are not
416 // permitted for C# generic type constraints.
417 if (CurrentState.IsCSharpGenericTypeConstraint &&
418 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) {
419 return false;
421 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
422 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
423 State.Line->First->isNot(TT_AttributeSquare) && Style.isCpp() &&
424 // FIXME: This is a temporary workaround for the case where clang-format
425 // sets BreakBeforeParameter to avoid bin packing and this creates a
426 // completely unnecessary line break after a template type that isn't
427 // line-wrapped.
428 (Previous.NestingLevel == 1 ||
429 Style.BinPackParameters == FormatStyle::BPPS_BinPack)) ||
430 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
431 Previous.isNot(tok::question)) ||
432 (!Style.BreakBeforeTernaryOperators &&
433 Previous.is(TT_ConditionalExpr))) &&
434 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
435 !Current.isOneOf(tok::r_paren, tok::r_brace)) {
436 return true;
438 if (CurrentState.IsChainedConditional &&
439 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
440 Current.is(tok::colon)) ||
441 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) &&
442 Previous.is(tok::colon)))) {
443 return true;
445 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
446 (Previous.is(TT_ArrayInitializerLSquare) &&
447 Previous.ParameterCount > 1) ||
448 opensProtoMessageField(Previous, Style)) &&
449 Style.ColumnLimit > 0 &&
450 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
451 getColumnLimit(State)) {
452 return true;
455 const FormatToken &BreakConstructorInitializersToken =
456 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
457 ? Previous
458 : Current;
459 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
460 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
461 getColumnLimit(State) ||
462 CurrentState.BreakBeforeParameter) &&
463 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
464 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
465 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
466 Style.ColumnLimit != 0)) {
467 return true;
470 if (Current.is(TT_ObjCMethodExpr) && Previous.isNot(TT_SelectorName) &&
471 State.Line->startsWith(TT_ObjCMethodSpecifier)) {
472 return true;
474 if (Current.is(TT_SelectorName) && Previous.isNot(tok::at) &&
475 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
476 (Style.ObjCBreakBeforeNestedBlockParam ||
477 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) {
478 return true;
481 unsigned NewLineColumn = getNewLineColumn(State);
482 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
483 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
484 (State.Column > NewLineColumn ||
485 Current.NestingLevel < State.StartOfLineLevel)) {
486 return true;
489 if (startsSegmentOfBuilderTypeCall(Current) &&
490 (CurrentState.CallContinuation != 0 ||
491 CurrentState.BreakBeforeParameter) &&
492 // JavaScript is treated different here as there is a frequent pattern:
493 // SomeFunction(function() {
494 // ...
495 // }.bind(...));
496 // FIXME: We should find a more generic solution to this problem.
497 !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
498 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
499 return true;
502 // If the template declaration spans multiple lines, force wrap before the
503 // function/class declaration.
504 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
505 Current.CanBreakBefore) {
506 return true;
509 if (State.Line->First->isNot(tok::kw_enum) && State.Column <= NewLineColumn)
510 return false;
512 if (Style.AlwaysBreakBeforeMultilineStrings &&
513 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
514 Previous.is(tok::comma) || Current.NestingLevel < 2) &&
515 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at,
516 Keywords.kw_dollar) &&
517 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
518 nextIsMultilineString(State)) {
519 return true;
522 // Using CanBreakBefore here and below takes care of the decision whether the
523 // current style uses wrapping before or after operators for the given
524 // operator.
525 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
526 const auto PreviousPrecedence = Previous.getPrecedence();
527 if (PreviousPrecedence != prec::Assignment &&
528 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
529 const bool LHSIsBinaryExpr =
530 Previous.Previous && Previous.Previous->EndsBinaryExpression;
531 if (LHSIsBinaryExpr)
532 return true;
533 // If we need to break somewhere inside the LHS of a binary expression, we
534 // should also break after the operator. Otherwise, the formatting would
535 // hide the operator precedence, e.g. in:
536 // if (aaaaaaaaaaaaaa ==
537 // bbbbbbbbbbbbbb && c) {..
538 // For comparisons, we only apply this rule, if the LHS is a binary
539 // expression itself as otherwise, the line breaks seem superfluous.
540 // We need special cases for ">>" which we have split into two ">" while
541 // lexing in order to make template parsing easier.
542 const bool IsComparison =
543 (PreviousPrecedence == prec::Relational ||
544 PreviousPrecedence == prec::Equality ||
545 PreviousPrecedence == prec::Spaceship) &&
546 Previous.Previous &&
547 Previous.Previous->isNot(TT_BinaryOperator); // For >>.
548 if (!IsComparison)
549 return true;
551 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
552 CurrentState.BreakBeforeParameter) {
553 return true;
556 // Same as above, but for the first "<<" operator.
557 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
558 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
559 return true;
562 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
563 // Always break after "template <...>"(*) and leading annotations. This is
564 // only for cases where the entire line does not fit on a single line as a
565 // different LineFormatter would be used otherwise.
566 // *: Except when another option interferes with that, like concepts.
567 if (Previous.ClosesTemplateDeclaration) {
568 if (Current.is(tok::kw_concept)) {
569 switch (Style.BreakBeforeConceptDeclarations) {
570 case FormatStyle::BBCDS_Allowed:
571 break;
572 case FormatStyle::BBCDS_Always:
573 return true;
574 case FormatStyle::BBCDS_Never:
575 return false;
578 if (Current.is(TT_RequiresClause)) {
579 switch (Style.RequiresClausePosition) {
580 case FormatStyle::RCPS_SingleLine:
581 case FormatStyle::RCPS_WithPreceding:
582 return false;
583 default:
584 return true;
587 return Style.BreakTemplateDeclarations != FormatStyle::BTDS_No &&
588 (Style.BreakTemplateDeclarations != FormatStyle::BTDS_Leave ||
589 Current.NewlinesBefore > 0);
591 if (Previous.is(TT_FunctionAnnotationRParen) &&
592 State.Line->Type != LT_PreprocessorDirective) {
593 return true;
595 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
596 Current.isNot(TT_LeadingJavaAnnotation)) {
597 return true;
601 if (Style.isJavaScript() && Previous.is(tok::r_paren) &&
602 Previous.is(TT_JavaAnnotation)) {
603 // Break after the closing parenthesis of TypeScript decorators before
604 // functions, getters and setters.
605 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
606 "function"};
607 if (BreakBeforeDecoratedTokens.contains(Current.TokenText))
608 return true;
611 if (Current.is(TT_FunctionDeclarationName) &&
612 !State.Line->ReturnTypeWrapped &&
613 // Don't break before a C# function when no break after return type.
614 (!Style.isCSharp() ||
615 Style.BreakAfterReturnType > FormatStyle::RTBS_ExceptShortType) &&
616 // Don't always break between a JavaScript `function` and the function
617 // name.
618 !Style.isJavaScript() && Previous.isNot(tok::kw_template) &&
619 CurrentState.BreakBeforeParameter) {
620 return true;
623 // The following could be precomputed as they do not depend on the state.
624 // However, as they should take effect only if the UnwrappedLine does not fit
625 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
626 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) &&
627 Previous.is(tok::l_brace) &&
628 !Current.isOneOf(tok::r_brace, tok::comment)) {
629 return true;
632 if (Current.is(tok::lessless) &&
633 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
634 (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with("\\n\"") ||
635 Previous.TokenText == "\'\\n\'")))) {
636 return true;
639 if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
640 return true;
642 if (State.NoContinuation)
643 return true;
645 return false;
648 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
649 bool DryRun,
650 unsigned ExtraSpaces) {
651 const FormatToken &Current = *State.NextToken;
652 assert(State.NextToken->Previous);
653 const FormatToken &Previous = *State.NextToken->Previous;
655 assert(!State.Stack.empty());
656 State.NoContinuation = false;
658 if (Current.is(TT_ImplicitStringLiteral) &&
659 (!Previous.Tok.getIdentifierInfo() ||
660 Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
661 tok::pp_not_keyword)) {
662 unsigned EndColumn =
663 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
664 if (Current.LastNewlineOffset != 0) {
665 // If there is a newline within this token, the final column will solely
666 // determined by the current end column.
667 State.Column = EndColumn;
668 } else {
669 unsigned StartColumn =
670 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
671 assert(EndColumn >= StartColumn);
672 State.Column += EndColumn - StartColumn;
674 moveStateToNextToken(State, DryRun, /*Newline=*/false);
675 return 0;
678 unsigned Penalty = 0;
679 if (Newline)
680 Penalty = addTokenOnNewLine(State, DryRun);
681 else
682 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
684 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
687 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
688 unsigned ExtraSpaces) {
689 FormatToken &Current = *State.NextToken;
690 assert(State.NextToken->Previous);
691 const FormatToken &Previous = *State.NextToken->Previous;
692 auto &CurrentState = State.Stack.back();
694 bool DisallowLineBreaksOnThisLine =
695 Style.LambdaBodyIndentation == FormatStyle::LBI_Signature &&
696 // Deal with lambda arguments in C++. The aim here is to ensure that we
697 // don't over-indent lambda function bodies when lambdas are passed as
698 // arguments to function calls. We do this by ensuring that either all
699 // arguments (including any lambdas) go on the same line as the function
700 // call, or we break before the first argument.
701 Style.isCpp() && [&] {
702 // For example, `/*Newline=*/false`.
703 if (Previous.is(TT_BlockComment) && Current.SpacesRequiredBefore == 0)
704 return false;
705 const auto *PrevNonComment = Current.getPreviousNonComment();
706 if (!PrevNonComment || PrevNonComment->isNot(tok::l_paren))
707 return false;
708 if (Current.isOneOf(tok::comment, tok::l_paren, TT_LambdaLSquare))
709 return false;
710 auto BlockParameterCount = PrevNonComment->BlockParameterCount;
711 if (BlockParameterCount == 0)
712 return false;
714 // Multiple lambdas in the same function call.
715 if (BlockParameterCount > 1)
716 return true;
718 // A lambda followed by another arg.
719 if (!PrevNonComment->Role)
720 return false;
721 auto Comma = PrevNonComment->Role->lastComma();
722 if (!Comma)
723 return false;
724 auto Next = Comma->getNextNonComment();
725 return Next &&
726 !Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret);
727 }();
729 if (DisallowLineBreaksOnThisLine)
730 State.NoLineBreak = true;
732 if (Current.is(tok::equal) &&
733 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
734 CurrentState.VariablePos == 0 &&
735 (!Previous.Previous ||
736 Previous.Previous->isNot(TT_DesignatedInitializerPeriod))) {
737 CurrentState.VariablePos = State.Column;
738 // Move over * and & if they are bound to the variable name.
739 const FormatToken *Tok = &Previous;
740 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
741 CurrentState.VariablePos -= Tok->ColumnWidth;
742 if (Tok->SpacesRequiredBefore != 0)
743 break;
744 Tok = Tok->Previous;
746 if (Previous.PartOfMultiVariableDeclStmt)
747 CurrentState.LastSpace = CurrentState.VariablePos;
750 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
752 // Indent preprocessor directives after the hash if required.
753 int PPColumnCorrection = 0;
754 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
755 Previous.is(tok::hash) && State.FirstIndent > 0 &&
756 &Previous == State.Line->First &&
757 (State.Line->Type == LT_PreprocessorDirective ||
758 State.Line->Type == LT_ImportStatement)) {
759 Spaces += State.FirstIndent;
761 // For preprocessor indent with tabs, State.Column will be 1 because of the
762 // hash. This causes second-level indents onward to have an extra space
763 // after the tabs. We avoid this misalignment by subtracting 1 from the
764 // column value passed to replaceWhitespace().
765 if (Style.UseTab != FormatStyle::UT_Never)
766 PPColumnCorrection = -1;
769 if (!DryRun) {
770 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
771 State.Column + Spaces + PPColumnCorrection,
772 /*IsAligned=*/false, State.Line->InMacroBody);
775 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
776 // declaration unless there is multiple inheritance.
777 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
778 Current.is(TT_InheritanceColon)) {
779 CurrentState.NoLineBreak = true;
781 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
782 Previous.is(TT_InheritanceColon)) {
783 CurrentState.NoLineBreak = true;
786 if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
787 unsigned MinIndent = std::max(
788 State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent);
789 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
790 if (Current.LongestObjCSelectorName == 0)
791 CurrentState.AlignColons = false;
792 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
793 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
794 else
795 CurrentState.ColonPos = FirstColonPos;
798 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
799 // parenthesis by disallowing any further line breaks if there is no line
800 // break after the opening parenthesis. Don't break if it doesn't conserve
801 // columns.
802 auto IsOpeningBracket = [&](const FormatToken &Tok) {
803 auto IsStartOfBracedList = [&]() {
804 return Tok.is(tok::l_brace) && Tok.isNot(BK_Block) &&
805 Style.Cpp11BracedListStyle;
807 if (!Tok.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) &&
808 !IsStartOfBracedList()) {
809 return false;
811 if (!Tok.Previous)
812 return true;
813 if (Tok.Previous->isIf())
814 return Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak;
815 return !Tok.Previous->isOneOf(TT_CastRParen, tok::kw_for, tok::kw_while,
816 tok::kw_switch) &&
817 !(Style.isJavaScript() && Tok.Previous->is(Keywords.kw_await));
819 auto IsFunctionCallParen = [](const FormatToken &Tok) {
820 return Tok.is(tok::l_paren) && Tok.ParameterCount > 0 && Tok.Previous &&
821 Tok.Previous->is(tok::identifier);
823 auto IsInTemplateString = [this](const FormatToken &Tok) {
824 if (!Style.isJavaScript())
825 return false;
826 for (const auto *Prev = &Tok; Prev; Prev = Prev->Previous) {
827 if (Prev->is(TT_TemplateString) && Prev->opensScope())
828 return true;
829 if (Prev->is(TT_TemplateString) && Prev->closesScope())
830 break;
832 return false;
834 // Identifies simple (no expression) one-argument function calls.
835 auto StartsSimpleOneArgList = [&](const FormatToken &TokAfterLParen) {
836 assert(TokAfterLParen.isNot(tok::comment) || TokAfterLParen.Next);
837 const auto &Tok =
838 TokAfterLParen.is(tok::comment) ? *TokAfterLParen.Next : TokAfterLParen;
839 if (!Tok.FakeLParens.empty() && Tok.FakeLParens.back() > prec::Unknown)
840 return false;
841 // Nested calls that involve `new` expressions also look like simple
842 // function calls, eg:
843 // - foo(new Bar())
844 // - foo(::new Bar())
845 if (Tok.is(tok::kw_new) || Tok.startsSequence(tok::coloncolon, tok::kw_new))
846 return true;
847 if (Tok.is(TT_UnaryOperator) ||
848 (Style.isJavaScript() &&
849 Tok.isOneOf(tok::ellipsis, Keywords.kw_await))) {
850 return true;
852 const auto *Previous = Tok.Previous;
853 if (!Previous || (!Previous->isOneOf(TT_FunctionDeclarationLParen,
854 TT_LambdaDefinitionLParen) &&
855 !IsFunctionCallParen(*Previous))) {
856 return true;
858 if (IsOpeningBracket(Tok) || IsInTemplateString(Tok))
859 return true;
860 const auto *Next = Tok.Next;
861 return !Next || Next->isMemberAccess() ||
862 Next->is(TT_FunctionDeclarationLParen) || IsFunctionCallParen(*Next);
864 if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak ||
865 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) &&
866 IsOpeningBracket(Previous) && State.Column > getNewLineColumn(State) &&
867 // Don't do this for simple (no expressions) one-argument function calls
868 // as that feels like needlessly wasting whitespace, e.g.:
870 // caaaaaaaaaaaall(
871 // caaaaaaaaaaaall(
872 // caaaaaaaaaaaall(
873 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
874 // or
875 // caaaaaaaaaaaaaaaaaaaaal(
876 // new SomethingElseeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee());
877 !StartsSimpleOneArgList(Current)) {
878 CurrentState.NoLineBreak = true;
881 if (Previous.is(TT_TemplateString) && Previous.opensScope())
882 CurrentState.NoLineBreak = true;
884 // Align following lines within parentheses / brackets if configured.
885 // Note: This doesn't apply to macro expansion lines, which are MACRO( , , )
886 // with args as children of the '(' and ',' tokens. It does not make sense to
887 // align the commas with the opening paren.
888 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
889 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
890 Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) &&
891 Previous.isNot(TT_TableGenDAGArgOpener) &&
892 Previous.isNot(TT_TableGenDAGArgOpenerToBreak) &&
893 !(Current.MacroParent && Previous.MacroParent) &&
894 (Current.isNot(TT_LineComment) ||
895 Previous.isOneOf(BK_BracedInit, TT_VerilogMultiLineListLParen)) &&
896 !IsInTemplateString(Current)) {
897 CurrentState.Indent = State.Column + Spaces;
898 CurrentState.IsAligned = true;
900 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
901 CurrentState.NoLineBreak = true;
902 if (mustBreakBinaryOperation(Current, Style))
903 CurrentState.NoLineBreak = true;
905 if (startsSegmentOfBuilderTypeCall(Current) &&
906 State.Column > getNewLineColumn(State)) {
907 CurrentState.ContainsUnwrappedBuilder = true;
910 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java)
911 CurrentState.NoLineBreak = true;
912 if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
913 (Previous.MatchingParen &&
914 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
915 // If there is a function call with long parameters, break before trailing
916 // calls. This prevents things like:
917 // EXPECT_CALL(SomeLongParameter).Times(
918 // 2);
919 // We don't want to do this for short parameters as they can just be
920 // indexes.
921 CurrentState.NoLineBreak = true;
924 // Don't allow the RHS of an operator to be split over multiple lines unless
925 // there is a line-break right after the operator.
926 // Exclude relational operators, as there, it is always more desirable to
927 // have the LHS 'left' of the RHS.
928 const FormatToken *P = Current.getPreviousNonComment();
929 if (Current.isNot(tok::comment) && P &&
930 (P->isOneOf(TT_BinaryOperator, tok::comma) ||
931 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
932 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
933 P->getPrecedence() != prec::Assignment &&
934 P->getPrecedence() != prec::Relational &&
935 P->getPrecedence() != prec::Spaceship) {
936 bool BreakBeforeOperator =
937 P->MustBreakBefore || P->is(tok::lessless) ||
938 (P->is(TT_BinaryOperator) &&
939 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
940 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
941 // Don't do this if there are only two operands. In these cases, there is
942 // always a nice vertical separation between them and the extra line break
943 // does not help.
944 bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator &&
945 P->isNot(TT_ConditionalExpr);
946 if ((!BreakBeforeOperator &&
947 !(HasTwoOperands &&
948 Style.AlignOperands != FormatStyle::OAS_DontAlign)) ||
949 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
950 CurrentState.NoLineBreakInOperand = true;
954 State.Column += Spaces;
955 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
956 Previous.Previous &&
957 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) {
958 // Treat the condition inside an if as if it was a second function
959 // parameter, i.e. let nested calls have a continuation indent.
960 CurrentState.LastSpace = State.Column;
961 CurrentState.NestedBlockIndent = State.Column;
962 } else if (!Current.isOneOf(tok::comment, tok::caret) &&
963 ((Previous.is(tok::comma) &&
964 Previous.isNot(TT_OverloadedOperator)) ||
965 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
966 CurrentState.LastSpace = State.Column;
967 } else if (Previous.is(TT_CtorInitializerColon) &&
968 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
969 Style.BreakConstructorInitializers ==
970 FormatStyle::BCIS_AfterColon) {
971 CurrentState.Indent = State.Column;
972 CurrentState.LastSpace = State.Column;
973 } else if (Previous.isOneOf(TT_ConditionalExpr, TT_CtorInitializerColon)) {
974 CurrentState.LastSpace = State.Column;
975 } else if (Previous.is(TT_BinaryOperator) &&
976 ((Previous.getPrecedence() != prec::Assignment &&
977 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
978 Previous.NextOperator)) ||
979 Current.StartsBinaryExpression)) {
980 // Indent relative to the RHS of the expression unless this is a simple
981 // assignment without binary expression on the RHS.
982 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
983 CurrentState.LastSpace = State.Column;
984 } else if (Previous.is(TT_InheritanceColon)) {
985 CurrentState.Indent = State.Column;
986 CurrentState.LastSpace = State.Column;
987 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) {
988 CurrentState.ColonPos = State.Column;
989 } else if (Previous.opensScope()) {
990 // If a function has a trailing call, indent all parameters from the
991 // opening parenthesis. This avoids confusing indents like:
992 // OuterFunction(InnerFunctionCall( // break
993 // ParameterToInnerFunction)) // break
994 // .SecondInnerFunctionCall();
995 if (Previous.MatchingParen) {
996 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
997 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
998 State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
999 CurrentState.LastSpace = State.Column;
1005 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
1006 bool DryRun) {
1007 FormatToken &Current = *State.NextToken;
1008 assert(State.NextToken->Previous);
1009 const FormatToken &Previous = *State.NextToken->Previous;
1010 auto &CurrentState = State.Stack.back();
1012 // Extra penalty that needs to be added because of the way certain line
1013 // breaks are chosen.
1014 unsigned Penalty = 0;
1016 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1017 const FormatToken *NextNonComment = Previous.getNextNonComment();
1018 if (!NextNonComment)
1019 NextNonComment = &Current;
1020 // The first line break on any NestingLevel causes an extra penalty in order
1021 // prefer similar line breaks.
1022 if (!CurrentState.ContainsLineBreak)
1023 Penalty += 15;
1024 CurrentState.ContainsLineBreak = true;
1026 Penalty += State.NextToken->SplitPenalty;
1028 // Breaking before the first "<<" is generally not desirable if the LHS is
1029 // short. Also always add the penalty if the LHS is split over multiple lines
1030 // to avoid unnecessary line breaks that just work around this penalty.
1031 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 &&
1032 (State.Column <= Style.ColumnLimit / 3 ||
1033 CurrentState.BreakBeforeParameter)) {
1034 Penalty += Style.PenaltyBreakFirstLessLess;
1037 State.Column = getNewLineColumn(State);
1039 // Add Penalty proportional to amount of whitespace away from FirstColumn
1040 // This tends to penalize several lines that are far-right indented,
1041 // and prefers a line-break prior to such a block, e.g:
1043 // Constructor() :
1044 // member(value), looooooooooooooooong_member(
1045 // looooooooooong_call(param_1, param_2, param_3))
1046 // would then become
1047 // Constructor() :
1048 // member(value),
1049 // looooooooooooooooong_member(
1050 // looooooooooong_call(param_1, param_2, param_3))
1051 if (State.Column > State.FirstIndent) {
1052 Penalty +=
1053 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
1056 // Indent nested blocks relative to this column, unless in a very specific
1057 // JavaScript special case where:
1059 // var loooooong_name =
1060 // function() {
1061 // // code
1062 // }
1064 // is common and should be formatted like a free-standing function. The same
1065 // goes for wrapping before the lambda return type arrow.
1066 if (Current.isNot(TT_LambdaArrow) &&
1067 (!Style.isJavaScript() || Current.NestingLevel != 0 ||
1068 !PreviousNonComment || PreviousNonComment->isNot(tok::equal) ||
1069 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) {
1070 CurrentState.NestedBlockIndent = State.Column;
1073 if (NextNonComment->isMemberAccess()) {
1074 if (CurrentState.CallContinuation == 0)
1075 CurrentState.CallContinuation = State.Column;
1076 } else if (NextNonComment->is(TT_SelectorName)) {
1077 if (!CurrentState.ObjCSelectorNameFound) {
1078 if (NextNonComment->LongestObjCSelectorName == 0) {
1079 CurrentState.AlignColons = false;
1080 } else {
1081 CurrentState.ColonPos =
1082 (shouldIndentWrappedSelectorName(Style, State.Line->Type)
1083 ? std::max(CurrentState.Indent,
1084 State.FirstIndent + Style.ContinuationIndentWidth)
1085 : CurrentState.Indent) +
1086 std::max(NextNonComment->LongestObjCSelectorName,
1087 NextNonComment->ColumnWidth);
1089 } else if (CurrentState.AlignColons &&
1090 CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
1091 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
1093 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1094 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1095 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
1096 // method expression, the block should be aligned to the line starting it,
1097 // e.g.:
1098 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
1099 // ^(int *i) {
1100 // // ...
1101 // }];
1102 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
1103 // when we consume all of the "}"'s FakeRParens at the "{".
1104 if (State.Stack.size() > 1) {
1105 State.Stack[State.Stack.size() - 2].LastSpace =
1106 std::max(CurrentState.LastSpace, CurrentState.Indent) +
1107 Style.ContinuationIndentWidth;
1111 if ((PreviousNonComment &&
1112 PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
1113 !CurrentState.AvoidBinPacking) ||
1114 Previous.is(TT_BinaryOperator)) {
1115 CurrentState.BreakBeforeParameter = false;
1117 if (PreviousNonComment &&
1118 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) ||
1119 PreviousNonComment->ClosesRequiresClause) &&
1120 Current.NestingLevel == 0) {
1121 CurrentState.BreakBeforeParameter = false;
1123 if (NextNonComment->is(tok::question) ||
1124 (PreviousNonComment && PreviousNonComment->is(tok::question))) {
1125 CurrentState.BreakBeforeParameter = true;
1127 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
1128 CurrentState.BreakBeforeParameter = false;
1130 if (!DryRun) {
1131 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
1132 if (Current.is(tok::r_brace) && Current.MatchingParen &&
1133 // Only strip trailing empty lines for l_braces that have children, i.e.
1134 // for function expressions (lambdas, arrows, etc).
1135 !Current.MatchingParen->Children.empty()) {
1136 // lambdas and arrow functions are expressions, thus their r_brace is not
1137 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
1138 // about removing empty lines on closing blocks. Special case them here.
1139 MaxEmptyLinesToKeep = 1;
1141 unsigned Newlines =
1142 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
1143 bool ContinuePPDirective =
1144 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
1145 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
1146 CurrentState.IsAligned, ContinuePPDirective);
1149 if (!Current.isTrailingComment())
1150 CurrentState.LastSpace = State.Column;
1151 if (Current.is(tok::lessless)) {
1152 // If we are breaking before a "<<", we always want to indent relative to
1153 // RHS. This is necessary only for "<<", as we special-case it and don't
1154 // always indent relative to the RHS.
1155 CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1158 State.StartOfLineLevel = Current.NestingLevel;
1159 State.LowestLevelOnLine = Current.NestingLevel;
1161 // Any break on this level means that the parent level has been broken
1162 // and we need to avoid bin packing there.
1163 bool NestedBlockSpecialCase =
1164 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
1165 State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1166 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) &&
1167 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1168 // Do not force parameter break for statements with requires expressions.
1169 NestedBlockSpecialCase =
1170 NestedBlockSpecialCase ||
1171 (Current.MatchingParen &&
1172 Current.MatchingParen->is(TT_RequiresExpressionLBrace));
1173 if (!NestedBlockSpecialCase) {
1174 auto ParentLevelIt = std::next(State.Stack.rbegin());
1175 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1176 Current.MatchingParen && Current.MatchingParen->is(TT_LambdaLBrace)) {
1177 // If the first character on the new line is a lambda's closing brace, the
1178 // stack still contains that lambda's parenthesis. As such, we need to
1179 // recurse further down the stack than usual to find the parenthesis level
1180 // containing the lambda, which is where we want to set
1181 // BreakBeforeParameter.
1183 // We specifically special case "OuterScope"-formatted lambdas here
1184 // because, when using that setting, breaking before the parameter
1185 // directly following the lambda is particularly unsightly. However, when
1186 // "OuterScope" is not set, the logic to find the parent parenthesis level
1187 // still appears to be sometimes incorrect. It has not been fixed yet
1188 // because it would lead to significant changes in existing behaviour.
1190 // TODO: fix the non-"OuterScope" case too.
1191 auto FindCurrentLevel = [&](const auto &It) {
1192 return std::find_if(It, State.Stack.rend(), [](const auto &PState) {
1193 return PState.Tok != nullptr; // Ignore fake parens.
1196 auto MaybeIncrement = [&](const auto &It) {
1197 return It != State.Stack.rend() ? std::next(It) : It;
1199 auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin());
1200 auto LevelContainingLambdaIt =
1201 FindCurrentLevel(MaybeIncrement(LambdaLevelIt));
1202 ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt);
1204 for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I)
1205 I->BreakBeforeParameter = true;
1208 if (PreviousNonComment &&
1209 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
1210 ((PreviousNonComment->isNot(TT_TemplateCloser) &&
1211 !PreviousNonComment->ClosesRequiresClause) ||
1212 Current.NestingLevel != 0) &&
1213 !PreviousNonComment->isOneOf(
1214 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
1215 TT_LeadingJavaAnnotation) &&
1216 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope() &&
1217 // We don't want to enforce line breaks for subsequent arguments just
1218 // because we have been forced to break before a lambda body.
1219 (!Style.BraceWrapping.BeforeLambdaBody ||
1220 Current.isNot(TT_LambdaLBrace))) {
1221 CurrentState.BreakBeforeParameter = true;
1224 // If we break after { or the [ of an array initializer, we should also break
1225 // before the corresponding } or ].
1226 if (PreviousNonComment &&
1227 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1228 opensProtoMessageField(*PreviousNonComment, Style))) {
1229 CurrentState.BreakBeforeClosingBrace = true;
1232 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) {
1233 CurrentState.BreakBeforeClosingParen =
1234 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent;
1237 if (CurrentState.AvoidBinPacking) {
1238 // If we are breaking after '(', '{', '<', or this is the break after a ':'
1239 // to start a member initializer list in a constructor, this should not
1240 // be considered bin packing unless the relevant AllowAll option is false or
1241 // this is a dict/object literal.
1242 bool PreviousIsBreakingCtorInitializerColon =
1243 PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1244 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
1245 bool AllowAllConstructorInitializersOnNextLine =
1246 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine ||
1247 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLineOnly;
1248 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
1249 PreviousIsBreakingCtorInitializerColon) ||
1250 (!Style.AllowAllParametersOfDeclarationOnNextLine &&
1251 State.Line->MustBeDeclaration) ||
1252 (!Style.AllowAllArgumentsOnNextLine &&
1253 !State.Line->MustBeDeclaration) ||
1254 (!AllowAllConstructorInitializersOnNextLine &&
1255 PreviousIsBreakingCtorInitializerColon) ||
1256 Previous.is(TT_DictLiteral)) {
1257 CurrentState.BreakBeforeParameter = true;
1260 // If we are breaking after a ':' to start a member initializer list,
1261 // and we allow all arguments on the next line, we should not break
1262 // before the next parameter.
1263 if (PreviousIsBreakingCtorInitializerColon &&
1264 AllowAllConstructorInitializersOnNextLine) {
1265 CurrentState.BreakBeforeParameter = false;
1269 if (mustBreakBinaryOperation(Current, Style))
1270 CurrentState.BreakBeforeParameter = true;
1272 return Penalty;
1275 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
1276 if (!State.NextToken || !State.NextToken->Previous)
1277 return 0;
1279 FormatToken &Current = *State.NextToken;
1280 const auto &CurrentState = State.Stack.back();
1282 if (CurrentState.IsCSharpGenericTypeConstraint &&
1283 Current.isNot(TT_CSharpGenericTypeConstraint)) {
1284 return CurrentState.ColonPos + 2;
1287 const FormatToken &Previous = *Current.Previous;
1288 // If we are continuing an expression, we want to use the continuation indent.
1289 unsigned ContinuationIndent =
1290 std::max(CurrentState.LastSpace, CurrentState.Indent) +
1291 Style.ContinuationIndentWidth;
1292 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1293 const FormatToken *NextNonComment = Previous.getNextNonComment();
1294 if (!NextNonComment)
1295 NextNonComment = &Current;
1297 // Java specific bits.
1298 if (Style.Language == FormatStyle::LK_Java &&
1299 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) {
1300 return std::max(CurrentState.LastSpace,
1301 CurrentState.Indent + Style.ContinuationIndentWidth);
1304 // Indentation of the statement following a Verilog case label is taken care
1305 // of in moveStateToNextToken.
1306 if (Style.isVerilog() && PreviousNonComment &&
1307 Keywords.isVerilogEndOfLabel(*PreviousNonComment)) {
1308 return State.FirstIndent;
1311 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths &&
1312 State.Line->First->is(tok::kw_enum)) {
1313 return (Style.IndentWidth * State.Line->First->IndentLevel) +
1314 Style.IndentWidth;
1317 if ((NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) ||
1318 (Style.isVerilog() && Keywords.isVerilogBegin(*NextNonComment))) {
1319 if (Current.NestingLevel == 0 ||
1320 (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
1321 State.NextToken->is(TT_LambdaLBrace))) {
1322 return State.FirstIndent;
1324 return CurrentState.Indent;
1326 if (Current.is(TT_LambdaArrow) &&
1327 Previous.isOneOf(tok::kw_noexcept, tok::kw_mutable, tok::kw_constexpr,
1328 tok::kw_consteval, tok::kw_static, TT_AttributeSquare)) {
1329 return ContinuationIndent;
1331 if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
1332 (Current.is(tok::greater) && (Style.isProto() || Style.isTableGen()))) &&
1333 State.Stack.size() > 1) {
1334 if (Current.closesBlockOrBlockTypeList(Style))
1335 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1336 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit))
1337 return State.Stack[State.Stack.size() - 2].LastSpace;
1338 return State.FirstIndent;
1340 // Indent a closing parenthesis at the previous level if followed by a semi,
1341 // const, or opening brace. This allows indentations such as:
1342 // foo(
1343 // a,
1344 // );
1345 // int Foo::getter(
1346 // //
1347 // ) const {
1348 // return foo;
1349 // }
1350 // function foo(
1351 // a,
1352 // ) {
1353 // code(); //
1354 // }
1355 if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
1356 (!Current.Next ||
1357 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) {
1358 return State.Stack[State.Stack.size() - 2].LastSpace;
1360 // When DAGArg closer exists top of line, it should be aligned in the similar
1361 // way as function call above.
1362 if (Style.isTableGen() && Current.is(TT_TableGenDAGArgCloser) &&
1363 State.Stack.size() > 1) {
1364 return State.Stack[State.Stack.size() - 2].LastSpace;
1366 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent &&
1367 (Current.is(tok::r_paren) ||
1368 (Current.is(tok::r_brace) && Current.MatchingParen &&
1369 Current.MatchingParen->is(BK_BracedInit))) &&
1370 State.Stack.size() > 1) {
1371 return State.Stack[State.Stack.size() - 2].LastSpace;
1373 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
1374 return State.Stack[State.Stack.size() - 2].LastSpace;
1375 // Field labels in a nested type should be aligned to the brace. For example
1376 // in ProtoBuf:
1377 // optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123,
1378 // bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}];
1379 // For Verilog, a quote following a brace is treated as an identifier. And
1380 // Both braces and colons get annotated as TT_DictLiteral. So we have to
1381 // check.
1382 if (Current.is(tok::identifier) && Current.Next &&
1383 (!Style.isVerilog() || Current.Next->is(tok::colon)) &&
1384 (Current.Next->is(TT_DictLiteral) ||
1385 (Style.isProto() && Current.Next->isOneOf(tok::less, tok::l_brace)))) {
1386 return CurrentState.Indent;
1388 if (NextNonComment->is(TT_ObjCStringLiteral) &&
1389 State.StartOfStringLiteral != 0) {
1390 return State.StartOfStringLiteral - 1;
1392 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1393 return State.StartOfStringLiteral;
1394 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0)
1395 return CurrentState.FirstLessLess;
1396 if (NextNonComment->isMemberAccess()) {
1397 if (CurrentState.CallContinuation == 0)
1398 return ContinuationIndent;
1399 return CurrentState.CallContinuation;
1401 if (CurrentState.QuestionColumn != 0 &&
1402 ((NextNonComment->is(tok::colon) &&
1403 NextNonComment->is(TT_ConditionalExpr)) ||
1404 Previous.is(TT_ConditionalExpr))) {
1405 if (((NextNonComment->is(tok::colon) && NextNonComment->Next &&
1406 !NextNonComment->Next->FakeLParens.empty() &&
1407 NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1408 (Previous.is(tok::colon) && !Current.FakeLParens.empty() &&
1409 Current.FakeLParens.back() == prec::Conditional)) &&
1410 !CurrentState.IsWrappedConditional) {
1411 // NOTE: we may tweak this slightly:
1412 // * not remove the 'lead' ContinuationIndentWidth
1413 // * always un-indent by the operator when
1414 // BreakBeforeTernaryOperators=true
1415 unsigned Indent = CurrentState.Indent;
1416 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1417 Indent -= Style.ContinuationIndentWidth;
1418 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1419 Indent -= 2;
1420 return Indent;
1422 return CurrentState.QuestionColumn;
1424 if (Previous.is(tok::comma) && CurrentState.VariablePos != 0)
1425 return CurrentState.VariablePos;
1426 if (Current.is(TT_RequiresClause)) {
1427 if (Style.IndentRequiresClause)
1428 return CurrentState.Indent + Style.IndentWidth;
1429 switch (Style.RequiresClausePosition) {
1430 case FormatStyle::RCPS_OwnLine:
1431 case FormatStyle::RCPS_WithFollowing:
1432 case FormatStyle::RCPS_OwnLineWithBrace:
1433 return CurrentState.Indent;
1434 default:
1435 break;
1438 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1439 TT_InheritanceComma)) {
1440 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1442 if ((PreviousNonComment &&
1443 (PreviousNonComment->ClosesTemplateDeclaration ||
1444 PreviousNonComment->ClosesRequiresClause ||
1445 (PreviousNonComment->is(TT_AttributeMacro) &&
1446 Current.isNot(tok::l_paren)) ||
1447 PreviousNonComment->isOneOf(
1448 TT_AttributeRParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
1449 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
1450 (!Style.IndentWrappedFunctionNames &&
1451 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) {
1452 return std::max(CurrentState.LastSpace, CurrentState.Indent);
1454 if (NextNonComment->is(TT_SelectorName)) {
1455 if (!CurrentState.ObjCSelectorNameFound) {
1456 unsigned MinIndent = CurrentState.Indent;
1457 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) {
1458 MinIndent = std::max(MinIndent,
1459 State.FirstIndent + Style.ContinuationIndentWidth);
1461 // If LongestObjCSelectorName is 0, we are indenting the first
1462 // part of an ObjC selector (or a selector component which is
1463 // not colon-aligned due to block formatting).
1465 // Otherwise, we are indenting a subsequent part of an ObjC
1466 // selector which should be colon-aligned to the longest
1467 // component of the ObjC selector.
1469 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1470 return MinIndent +
1471 std::max(NextNonComment->LongestObjCSelectorName,
1472 NextNonComment->ColumnWidth) -
1473 NextNonComment->ColumnWidth;
1475 if (!CurrentState.AlignColons)
1476 return CurrentState.Indent;
1477 if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1478 return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1479 return CurrentState.Indent;
1481 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1482 return CurrentState.ColonPos;
1483 if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1484 if (CurrentState.StartOfArraySubscripts != 0) {
1485 return CurrentState.StartOfArraySubscripts;
1486 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1487 // initializers.
1488 return CurrentState.Indent;
1490 return ContinuationIndent;
1493 // OpenMP clauses want to get additional indentation when they are pushed onto
1494 // the next line.
1495 if (State.Line->InPragmaDirective) {
1496 FormatToken *PragmaType = State.Line->First->Next->Next;
1497 if (PragmaType && PragmaType->TokenText == "omp")
1498 return CurrentState.Indent + Style.ContinuationIndentWidth;
1501 // This ensure that we correctly format ObjC methods calls without inputs,
1502 // i.e. where the last element isn't selector like: [callee method];
1503 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1504 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) {
1505 return CurrentState.Indent;
1508 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1509 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) {
1510 return ContinuationIndent;
1512 if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1513 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1514 return ContinuationIndent;
1516 if (NextNonComment->is(TT_CtorInitializerComma))
1517 return CurrentState.Indent;
1518 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1519 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1520 return CurrentState.Indent;
1522 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1523 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
1524 return CurrentState.Indent;
1526 if (Previous.is(tok::r_paren) &&
1527 Previous.isNot(TT_TableGenDAGArgOperatorToBreak) &&
1528 !Current.isBinaryOperator() &&
1529 !Current.isOneOf(tok::colon, tok::comment)) {
1530 return ContinuationIndent;
1532 if (Current.is(TT_ProtoExtensionLSquare))
1533 return CurrentState.Indent;
1534 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1535 return CurrentState.Indent - Current.Tok.getLength() -
1536 Current.SpacesRequiredBefore;
1538 if (Current.is(tok::comment) && NextNonComment->isBinaryOperator() &&
1539 CurrentState.UnindentOperator) {
1540 return CurrentState.Indent - NextNonComment->Tok.getLength() -
1541 NextNonComment->SpacesRequiredBefore;
1543 if (CurrentState.Indent == State.FirstIndent && PreviousNonComment &&
1544 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) {
1545 // Ensure that we fall back to the continuation indent width instead of
1546 // just flushing continuations left.
1547 return CurrentState.Indent + Style.ContinuationIndentWidth;
1549 return CurrentState.Indent;
1552 static bool hasNestedBlockInlined(const FormatToken *Previous,
1553 const FormatToken &Current,
1554 const FormatStyle &Style) {
1555 if (Previous->isNot(tok::l_paren))
1556 return true;
1557 if (Previous->ParameterCount > 1)
1558 return true;
1560 // Also a nested block if contains a lambda inside function with 1 parameter.
1561 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare);
1564 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1565 bool DryRun, bool Newline) {
1566 assert(State.Stack.size());
1567 const FormatToken &Current = *State.NextToken;
1568 auto &CurrentState = State.Stack.back();
1570 if (Current.is(TT_CSharpGenericTypeConstraint))
1571 CurrentState.IsCSharpGenericTypeConstraint = true;
1572 if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1573 CurrentState.NoLineBreakInOperand = false;
1574 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon))
1575 CurrentState.AvoidBinPacking = true;
1576 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1577 if (CurrentState.FirstLessLess == 0)
1578 CurrentState.FirstLessLess = State.Column;
1579 else
1580 CurrentState.LastOperatorWrapped = Newline;
1582 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1583 CurrentState.LastOperatorWrapped = Newline;
1584 if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1585 Current.Previous->isNot(TT_ConditionalExpr)) {
1586 CurrentState.LastOperatorWrapped = Newline;
1588 if (Current.is(TT_ArraySubscriptLSquare) &&
1589 CurrentState.StartOfArraySubscripts == 0) {
1590 CurrentState.StartOfArraySubscripts = State.Column;
1593 auto IsWrappedConditional = [](const FormatToken &Tok) {
1594 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question)))
1595 return false;
1596 if (Tok.MustBreakBefore)
1597 return true;
1599 const FormatToken *Next = Tok.getNextNonComment();
1600 return Next && Next->MustBreakBefore;
1602 if (IsWrappedConditional(Current))
1603 CurrentState.IsWrappedConditional = true;
1604 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1605 CurrentState.QuestionColumn = State.Column;
1606 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1607 const FormatToken *Previous = Current.Previous;
1608 while (Previous && Previous->isTrailingComment())
1609 Previous = Previous->Previous;
1610 if (Previous && Previous->is(tok::question))
1611 CurrentState.QuestionColumn = State.Column;
1613 if (!Current.opensScope() && !Current.closesScope() &&
1614 Current.isNot(TT_PointerOrReference)) {
1615 State.LowestLevelOnLine =
1616 std::min(State.LowestLevelOnLine, Current.NestingLevel);
1618 if (Current.isMemberAccess())
1619 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1620 if (Current.is(TT_SelectorName))
1621 CurrentState.ObjCSelectorNameFound = true;
1622 if (Current.is(TT_CtorInitializerColon) &&
1623 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1624 // Indent 2 from the column, so:
1625 // SomeClass::SomeClass()
1626 // : First(...), ...
1627 // Next(...)
1628 // ^ line up here.
1629 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1630 FormatStyle::BCIS_BeforeComma
1632 : 2);
1633 CurrentState.NestedBlockIndent = CurrentState.Indent;
1634 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) {
1635 CurrentState.AvoidBinPacking = true;
1636 CurrentState.BreakBeforeParameter =
1637 Style.ColumnLimit > 0 &&
1638 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine &&
1639 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLineOnly;
1640 } else {
1641 CurrentState.BreakBeforeParameter = false;
1644 if (Current.is(TT_CtorInitializerColon) &&
1645 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1646 CurrentState.Indent =
1647 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1648 CurrentState.NestedBlockIndent = CurrentState.Indent;
1649 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack)
1650 CurrentState.AvoidBinPacking = true;
1651 else
1652 CurrentState.BreakBeforeParameter = false;
1654 if (Current.is(TT_InheritanceColon)) {
1655 CurrentState.Indent =
1656 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1658 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1659 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1660 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow))
1661 CurrentState.LastSpace = State.Column;
1662 if (Current.is(TT_RequiresExpression) &&
1663 Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) {
1664 CurrentState.NestedBlockIndent = State.Column;
1667 // Insert scopes created by fake parenthesis.
1668 const FormatToken *Previous = Current.getPreviousNonComment();
1670 // Add special behavior to support a format commonly used for JavaScript
1671 // closures:
1672 // SomeFunction(function() {
1673 // foo();
1674 // bar();
1675 // }, a, b, c);
1676 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause &&
1677 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1678 Previous->isNot(TT_DictLiteral) && State.Stack.size() > 1 &&
1679 !CurrentState.HasMultipleNestedBlocks) {
1680 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1681 for (ParenState &PState : llvm::drop_end(State.Stack))
1682 PState.NoLineBreak = true;
1683 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1685 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) ||
1686 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) &&
1687 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) {
1688 CurrentState.NestedBlockInlined =
1689 !Newline && hasNestedBlockInlined(Previous, Current, Style);
1692 moveStatePastFakeLParens(State, Newline);
1693 moveStatePastScopeCloser(State);
1694 // Do not use CurrentState here, since the two functions before may change the
1695 // Stack.
1696 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1697 !State.Stack.back().NoLineBreakInOperand;
1698 moveStatePastScopeOpener(State, Newline);
1699 moveStatePastFakeRParens(State);
1701 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1702 State.StartOfStringLiteral = State.Column + 1;
1703 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1704 State.StartOfStringLiteral = State.Column + 1;
1705 } else if (Current.is(TT_TableGenMultiLineString) &&
1706 State.StartOfStringLiteral == 0) {
1707 State.StartOfStringLiteral = State.Column + 1;
1708 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1709 State.StartOfStringLiteral = State.Column;
1710 } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1711 !Current.isStringLiteral()) {
1712 State.StartOfStringLiteral = 0;
1715 State.Column += Current.ColumnWidth;
1716 State.NextToken = State.NextToken->Next;
1717 // Verilog case labels are on the same unwrapped lines as the statements that
1718 // follow. TokenAnnotator identifies them and sets MustBreakBefore.
1719 // Indentation is taken care of here. A case label can only have 1 statement
1720 // in Verilog, so we don't have to worry about lines that follow.
1721 if (Style.isVerilog() && State.NextToken &&
1722 State.NextToken->MustBreakBefore &&
1723 Keywords.isVerilogEndOfLabel(Current)) {
1724 State.FirstIndent += Style.IndentWidth;
1725 CurrentState.Indent = State.FirstIndent;
1728 unsigned Penalty =
1729 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1731 if (Current.Role)
1732 Current.Role->formatFromToken(State, this, DryRun);
1733 // If the previous has a special role, let it consume tokens as appropriate.
1734 // It is necessary to start at the previous token for the only implemented
1735 // role (comma separated list). That way, the decision whether or not to break
1736 // after the "{" is already done and both options are tried and evaluated.
1737 // FIXME: This is ugly, find a better way.
1738 if (Previous && Previous->Role)
1739 Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1741 return Penalty;
1744 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1745 bool Newline) {
1746 const FormatToken &Current = *State.NextToken;
1747 if (Current.FakeLParens.empty())
1748 return;
1750 const FormatToken *Previous = Current.getPreviousNonComment();
1752 // Don't add extra indentation for the first fake parenthesis after
1753 // 'return', assignments, opening <({[, or requires clauses. The indentation
1754 // for these cases is special cased.
1755 bool SkipFirstExtraIndent =
1756 Previous &&
1757 (Previous->opensScope() ||
1758 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) ||
1759 (Previous->getPrecedence() == prec::Assignment &&
1760 Style.AlignOperands != FormatStyle::OAS_DontAlign) ||
1761 Previous->is(TT_ObjCMethodExpr));
1762 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) {
1763 const auto &CurrentState = State.Stack.back();
1764 ParenState NewParenState = CurrentState;
1765 NewParenState.Tok = nullptr;
1766 NewParenState.ContainsLineBreak = false;
1767 NewParenState.LastOperatorWrapped = true;
1768 NewParenState.IsChainedConditional = false;
1769 NewParenState.IsWrappedConditional = false;
1770 NewParenState.UnindentOperator = false;
1771 NewParenState.NoLineBreak =
1772 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
1774 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1775 if (PrecedenceLevel > prec::Comma)
1776 NewParenState.AvoidBinPacking = false;
1778 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1779 // a builder type call after 'return' or, if the alignment after opening
1780 // brackets is disabled.
1781 if (!Current.isTrailingComment() &&
1782 (Style.AlignOperands != FormatStyle::OAS_DontAlign ||
1783 PrecedenceLevel < prec::Assignment) &&
1784 (!Previous || Previous->isNot(tok::kw_return) ||
1785 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) &&
1786 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1787 PrecedenceLevel > prec::Comma || Current.NestingLevel == 0) &&
1788 (!Style.isTableGen() ||
1789 (Previous && Previous->isOneOf(TT_TableGenDAGArgListComma,
1790 TT_TableGenDAGArgListCommaToBreak)))) {
1791 NewParenState.Indent = std::max(
1792 std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace);
1795 // Special case for generic selection expressions, its comma-separated
1796 // expressions are not aligned to the opening paren like regular calls, but
1797 // rather continuation-indented relative to the _Generic keyword.
1798 if (Previous && Previous->endsSequence(tok::l_paren, tok::kw__Generic) &&
1799 State.Stack.size() > 1) {
1800 NewParenState.Indent = State.Stack[State.Stack.size() - 2].Indent +
1801 Style.ContinuationIndentWidth;
1804 if ((shouldUnindentNextOperator(Current) ||
1805 (Previous &&
1806 (PrecedenceLevel == prec::Conditional &&
1807 Previous->is(tok::question) && Previous->is(TT_ConditionalExpr)))) &&
1808 !Newline) {
1809 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
1810 // the operator and keep the operands aligned.
1811 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator)
1812 NewParenState.UnindentOperator = true;
1813 // Mark indentation as alignment if the expression is aligned.
1814 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1815 NewParenState.IsAligned = true;
1818 // Do not indent relative to the fake parentheses inserted for "." or "->".
1819 // This is a special case to make the following to statements consistent:
1820 // OuterFunction(InnerFunctionCall( // break
1821 // ParameterToInnerFunction));
1822 // OuterFunction(SomeObject.InnerFunctionCall( // break
1823 // ParameterToInnerFunction));
1824 if (PrecedenceLevel > prec::Unknown)
1825 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1826 if (PrecedenceLevel != prec::Conditional &&
1827 Current.isNot(TT_UnaryOperator) &&
1828 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) {
1829 NewParenState.StartOfFunctionCall = State.Column;
1832 // Indent conditional expressions, unless they are chained "else-if"
1833 // conditionals. Never indent expression where the 'operator' is ',', ';' or
1834 // an assignment (i.e. *I <= prec::Assignment) as those have different
1835 // indentation rules. Indent other expression, unless the indentation needs
1836 // to be skipped.
1837 if (PrecedenceLevel == prec::Conditional && Previous &&
1838 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) &&
1839 &PrecedenceLevel == &Current.FakeLParens.back() &&
1840 !CurrentState.IsWrappedConditional) {
1841 NewParenState.IsChainedConditional = true;
1842 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
1843 } else if (PrecedenceLevel == prec::Conditional ||
1844 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
1845 !Current.isTrailingComment())) {
1846 NewParenState.Indent += Style.ContinuationIndentWidth;
1848 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
1849 NewParenState.BreakBeforeParameter = false;
1850 State.Stack.push_back(NewParenState);
1851 SkipFirstExtraIndent = false;
1855 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1856 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1857 unsigned VariablePos = State.Stack.back().VariablePos;
1858 if (State.Stack.size() == 1) {
1859 // Do not pop the last element.
1860 break;
1862 State.Stack.pop_back();
1863 State.Stack.back().VariablePos = VariablePos;
1866 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
1867 // Remove the indentation of the requires clauses (which is not in Indent,
1868 // but in LastSpace).
1869 State.Stack.back().LastSpace -= Style.IndentWidth;
1873 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1874 bool Newline) {
1875 const FormatToken &Current = *State.NextToken;
1876 if (!Current.opensScope())
1877 return;
1879 const auto &CurrentState = State.Stack.back();
1881 // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
1882 if (Current.isOneOf(tok::less, tok::l_paren) &&
1883 CurrentState.IsCSharpGenericTypeConstraint) {
1884 return;
1887 if (Current.MatchingParen && Current.is(BK_Block)) {
1888 moveStateToNewBlock(State, Newline);
1889 return;
1892 unsigned NewIndent;
1893 unsigned LastSpace = CurrentState.LastSpace;
1894 bool AvoidBinPacking;
1895 bool BreakBeforeParameter = false;
1896 unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall,
1897 CurrentState.NestedBlockIndent);
1898 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1899 opensProtoMessageField(Current, Style)) {
1900 if (Current.opensBlockOrBlockTypeList(Style)) {
1901 NewIndent = Style.IndentWidth +
1902 std::min(State.Column, CurrentState.NestedBlockIndent);
1903 } else if (Current.is(tok::l_brace)) {
1904 NewIndent =
1905 CurrentState.LastSpace + Style.BracedInitializerIndentWidth.value_or(
1906 Style.ContinuationIndentWidth);
1907 } else {
1908 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
1910 const FormatToken *NextNonComment = Current.getNextNonComment();
1911 bool EndsInComma = Current.MatchingParen &&
1912 Current.MatchingParen->Previous &&
1913 Current.MatchingParen->Previous->is(tok::comma);
1914 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1915 Style.isProto() || !Style.BinPackArguments ||
1916 (NextNonComment && NextNonComment->isOneOf(
1917 TT_DesignatedInitializerPeriod,
1918 TT_DesignatedInitializerLSquare));
1919 BreakBeforeParameter = EndsInComma;
1920 if (Current.ParameterCount > 1)
1921 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1922 } else {
1923 NewIndent =
1924 Style.ContinuationIndentWidth +
1925 std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall);
1927 if (Style.isTableGen() && Current.is(TT_TableGenDAGArgOpenerToBreak) &&
1928 Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakElements) {
1929 // For the case the next token is a TableGen DAGArg operator identifier
1930 // that is not marked to have a line break after it.
1931 // In this case the option DAS_BreakElements requires to align the
1932 // DAGArg elements to the operator.
1933 const FormatToken *Next = Current.Next;
1934 if (Next && Next->is(TT_TableGenDAGArgOperatorID))
1935 NewIndent = State.Column + Next->TokenText.size() + 2;
1938 // Ensure that different different brackets force relative alignment, e.g.:
1939 // void SomeFunction(vector< // break
1940 // int> v);
1941 // FIXME: We likely want to do this for more combinations of brackets.
1942 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1943 NewIndent = std::max(NewIndent, CurrentState.Indent);
1944 LastSpace = std::max(LastSpace, CurrentState.Indent);
1947 bool EndsInComma =
1948 Current.MatchingParen &&
1949 Current.MatchingParen->getPreviousNonComment() &&
1950 Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1952 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1953 // for backwards compatibility.
1954 bool ObjCBinPackProtocolList =
1955 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1956 Style.BinPackParameters == FormatStyle::BPPS_BinPack) ||
1957 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1959 bool BinPackDeclaration =
1960 (State.Line->Type != LT_ObjCDecl &&
1961 Style.BinPackParameters == FormatStyle::BPPS_BinPack) ||
1962 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1964 bool GenericSelection =
1965 Current.getPreviousNonComment() &&
1966 Current.getPreviousNonComment()->is(tok::kw__Generic);
1968 AvoidBinPacking =
1969 (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
1970 (Style.isJavaScript() && EndsInComma) ||
1971 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1972 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1973 (Style.ExperimentalAutoDetectBinPacking &&
1974 (Current.is(PPK_OnePerLine) ||
1975 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive))));
1977 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen &&
1978 Style.ObjCBreakBeforeNestedBlockParam) {
1979 if (Style.ColumnLimit) {
1980 // If this '[' opens an ObjC call, determine whether all parameters fit
1981 // into one line and put one per line if they don't.
1982 if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1983 getColumnLimit(State)) {
1984 BreakBeforeParameter = true;
1986 } else {
1987 // For ColumnLimit = 0, we have to figure out whether there is or has to
1988 // be a line break within this call.
1989 for (const FormatToken *Tok = &Current;
1990 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1991 if (Tok->MustBreakBefore ||
1992 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1993 BreakBeforeParameter = true;
1994 break;
2000 if (Style.isJavaScript() && EndsInComma)
2001 BreakBeforeParameter = true;
2003 // Generally inherit NoLineBreak from the current scope to nested scope.
2004 // However, don't do this for non-empty nested blocks, dict literals and
2005 // array literals as these follow different indentation rules.
2006 bool NoLineBreak =
2007 Current.Children.empty() &&
2008 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
2009 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
2010 (Current.is(TT_TemplateOpener) &&
2011 CurrentState.ContainsUnwrappedBuilder));
2012 State.Stack.push_back(
2013 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
2014 auto &NewState = State.Stack.back();
2015 NewState.NestedBlockIndent = NestedBlockIndent;
2016 NewState.BreakBeforeParameter = BreakBeforeParameter;
2017 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
2019 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next &&
2020 Current.is(tok::l_paren)) {
2021 // Search for any parameter that is a lambda.
2022 FormatToken const *next = Current.Next;
2023 while (next) {
2024 if (next->is(TT_LambdaLSquare)) {
2025 NewState.HasMultipleNestedBlocks = true;
2026 break;
2028 next = next->Next;
2032 NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) &&
2033 Current.Previous &&
2034 Current.Previous->is(tok::at);
2037 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
2038 const FormatToken &Current = *State.NextToken;
2039 if (!Current.closesScope())
2040 return;
2042 // If we encounter a closing ), ], } or >, we can remove a level from our
2043 // stacks.
2044 if (State.Stack.size() > 1 &&
2045 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
2046 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
2047 State.NextToken->is(TT_TemplateCloser) ||
2048 State.NextToken->is(TT_TableGenListCloser) ||
2049 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) {
2050 State.Stack.pop_back();
2053 auto &CurrentState = State.Stack.back();
2055 // Reevaluate whether ObjC message arguments fit into one line.
2056 // If a receiver spans multiple lines, e.g.:
2057 // [[object block:^{
2058 // return 42;
2059 // }] a:42 b:42];
2060 // BreakBeforeParameter is calculated based on an incorrect assumption
2061 // (it is checked whether the whole expression fits into one line without
2062 // considering a line break inside a message receiver).
2063 // We check whether arguments fit after receiver scope closer (into the same
2064 // line).
2065 if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
2066 Current.MatchingParen->Previous) {
2067 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
2068 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) &&
2069 CurrentScopeOpener.MatchingParen) {
2070 int NecessarySpaceInLine =
2071 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) +
2072 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
2073 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
2074 Style.ColumnLimit) {
2075 CurrentState.BreakBeforeParameter = false;
2080 if (Current.is(tok::r_square)) {
2081 // If this ends the array subscript expr, reset the corresponding value.
2082 const FormatToken *NextNonComment = Current.getNextNonComment();
2083 if (NextNonComment && NextNonComment->isNot(tok::l_square))
2084 CurrentState.StartOfArraySubscripts = 0;
2088 void ContinuationIndenter::moveStateToNewBlock(LineState &State, bool NewLine) {
2089 if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope &&
2090 State.NextToken->is(TT_LambdaLBrace) &&
2091 !State.Line->MightBeFunctionDecl) {
2092 State.Stack.back().NestedBlockIndent = State.FirstIndent;
2094 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
2095 // ObjC block sometimes follow special indentation rules.
2096 unsigned NewIndent =
2097 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
2098 ? Style.ObjCBlockIndentWidth
2099 : Style.IndentWidth);
2101 // Even when wrapping before lambda body, the left brace can still be added to
2102 // the same line. This occurs when checking whether the whole lambda body can
2103 // go on a single line. In this case we have to make sure there are no line
2104 // breaks in the body, otherwise we could just end up with a regular lambda
2105 // body without the brace wrapped.
2106 bool NoLineBreak = Style.BraceWrapping.BeforeLambdaBody && !NewLine &&
2107 State.NextToken->is(TT_LambdaLBrace);
2109 State.Stack.push_back(ParenState(State.NextToken, NewIndent,
2110 State.Stack.back().LastSpace,
2111 /*AvoidBinPacking=*/true, NoLineBreak));
2112 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
2113 State.Stack.back().BreakBeforeParameter = true;
2116 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
2117 unsigned TabWidth,
2118 encoding::Encoding Encoding) {
2119 size_t LastNewlinePos = Text.find_last_of("\n");
2120 if (LastNewlinePos == StringRef::npos) {
2121 return StartColumn +
2122 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
2123 } else {
2124 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
2125 /*StartColumn=*/0, TabWidth, Encoding);
2129 unsigned ContinuationIndenter::reformatRawStringLiteral(
2130 const FormatToken &Current, LineState &State,
2131 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
2132 unsigned StartColumn = State.Column - Current.ColumnWidth;
2133 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
2134 StringRef NewDelimiter =
2135 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
2136 if (NewDelimiter.empty())
2137 NewDelimiter = OldDelimiter;
2138 // The text of a raw string is between the leading 'R"delimiter(' and the
2139 // trailing 'delimiter)"'.
2140 unsigned OldPrefixSize = 3 + OldDelimiter.size();
2141 unsigned OldSuffixSize = 2 + OldDelimiter.size();
2142 // We create a virtual text environment which expects a null-terminated
2143 // string, so we cannot use StringRef.
2144 std::string RawText = std::string(
2145 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize));
2146 if (NewDelimiter != OldDelimiter) {
2147 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
2148 // raw string.
2149 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
2150 if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
2151 NewDelimiter = OldDelimiter;
2154 unsigned NewPrefixSize = 3 + NewDelimiter.size();
2155 unsigned NewSuffixSize = 2 + NewDelimiter.size();
2157 // The first start column is the column the raw text starts after formatting.
2158 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
2160 // The next start column is the intended indentation a line break inside
2161 // the raw string at level 0. It is determined by the following rules:
2162 // - if the content starts on newline, it is one level more than the current
2163 // indent, and
2164 // - if the content does not start on a newline, it is the first start
2165 // column.
2166 // These rules have the advantage that the formatted content both does not
2167 // violate the rectangle rule and visually flows within the surrounding
2168 // source.
2169 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
2170 // If this token is the last parameter (checked by looking if it's followed by
2171 // `)` and is not on a newline, the base the indent off the line's nested
2172 // block indent. Otherwise, base the indent off the arguments indent, so we
2173 // can achieve:
2175 // fffffffffff(1, 2, 3, R"pb(
2176 // key1: 1 #
2177 // key2: 2)pb");
2179 // fffffffffff(1, 2, 3,
2180 // R"pb(
2181 // key1: 1 #
2182 // key2: 2
2183 // )pb");
2185 // fffffffffff(1, 2, 3,
2186 // R"pb(
2187 // key1: 1 #
2188 // key2: 2
2189 // )pb",
2190 // 5);
2191 unsigned CurrentIndent =
2192 (!Newline && Current.Next && Current.Next->is(tok::r_paren))
2193 ? State.Stack.back().NestedBlockIndent
2194 : State.Stack.back().Indent;
2195 unsigned NextStartColumn = ContentStartsOnNewline
2196 ? CurrentIndent + Style.IndentWidth
2197 : FirstStartColumn;
2199 // The last start column is the column the raw string suffix starts if it is
2200 // put on a newline.
2201 // The last start column is the intended indentation of the raw string postfix
2202 // if it is put on a newline. It is determined by the following rules:
2203 // - if the raw string prefix starts on a newline, it is the column where
2204 // that raw string prefix starts, and
2205 // - if the raw string prefix does not start on a newline, it is the current
2206 // indent.
2207 unsigned LastStartColumn =
2208 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
2210 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
2211 RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
2212 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
2213 /*Status=*/nullptr);
2215 auto NewCode = applyAllReplacements(RawText, Fixes.first);
2216 tooling::Replacements NoFixes;
2217 if (!NewCode)
2218 return addMultilineToken(Current, State);
2219 if (!DryRun) {
2220 if (NewDelimiter != OldDelimiter) {
2221 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
2222 // of the token.
2223 SourceLocation PrefixDelimiterStart =
2224 Current.Tok.getLocation().getLocWithOffset(2);
2225 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
2226 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2227 if (PrefixErr) {
2228 llvm::errs()
2229 << "Failed to update the prefix delimiter of a raw string: "
2230 << llvm::toString(std::move(PrefixErr)) << "\n";
2232 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
2233 // position length - 1 - |delimiter|.
2234 SourceLocation SuffixDelimiterStart =
2235 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
2236 1 - OldDelimiter.size());
2237 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
2238 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2239 if (SuffixErr) {
2240 llvm::errs()
2241 << "Failed to update the suffix delimiter of a raw string: "
2242 << llvm::toString(std::move(SuffixErr)) << "\n";
2245 SourceLocation OriginLoc =
2246 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
2247 for (const tooling::Replacement &Fix : Fixes.first) {
2248 auto Err = Whitespaces.addReplacement(tooling::Replacement(
2249 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
2250 Fix.getLength(), Fix.getReplacementText()));
2251 if (Err) {
2252 llvm::errs() << "Failed to reformat raw string: "
2253 << llvm::toString(std::move(Err)) << "\n";
2257 unsigned RawLastLineEndColumn = getLastLineEndColumn(
2258 *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
2259 State.Column = RawLastLineEndColumn + NewSuffixSize;
2260 // Since we're updating the column to after the raw string literal here, we
2261 // have to manually add the penalty for the prefix R"delim( over the column
2262 // limit.
2263 unsigned PrefixExcessCharacters =
2264 StartColumn + NewPrefixSize > Style.ColumnLimit
2265 ? StartColumn + NewPrefixSize - Style.ColumnLimit
2266 : 0;
2267 bool IsMultiline =
2268 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos);
2269 if (IsMultiline) {
2270 // Break before further function parameters on all levels.
2271 for (ParenState &Paren : State.Stack)
2272 Paren.BreakBeforeParameter = true;
2274 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
2277 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
2278 LineState &State) {
2279 // Break before further function parameters on all levels.
2280 for (ParenState &Paren : State.Stack)
2281 Paren.BreakBeforeParameter = true;
2283 unsigned ColumnsUsed = State.Column;
2284 // We can only affect layout of the first and the last line, so the penalty
2285 // for all other lines is constant, and we ignore it.
2286 State.Column = Current.LastLineColumnWidth;
2288 if (ColumnsUsed > getColumnLimit(State))
2289 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2290 return 0;
2293 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2294 LineState &State, bool DryRun,
2295 bool AllowBreak, bool Newline) {
2296 unsigned Penalty = 0;
2297 // Compute the raw string style to use in case this is a raw string literal
2298 // that can be reformatted.
2299 auto RawStringStyle = getRawStringStyle(Current, State);
2300 if (RawStringStyle && !Current.Finalized) {
2301 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun,
2302 Newline);
2303 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
2304 // Don't break multi-line tokens other than block comments and raw string
2305 // literals. Instead, just update the state.
2306 Penalty = addMultilineToken(Current, State);
2307 } else if (State.Line->Type != LT_ImportStatement) {
2308 // We generally don't break import statements.
2309 LineState OriginalState = State;
2311 // Whether we force the reflowing algorithm to stay strictly within the
2312 // column limit.
2313 bool Strict = false;
2314 // Whether the first non-strict attempt at reflowing did intentionally
2315 // exceed the column limit.
2316 bool Exceeded = false;
2317 std::tie(Penalty, Exceeded) = breakProtrudingToken(
2318 Current, State, AllowBreak, /*DryRun=*/true, Strict);
2319 if (Exceeded) {
2320 // If non-strict reflowing exceeds the column limit, try whether strict
2321 // reflowing leads to an overall lower penalty.
2322 LineState StrictState = OriginalState;
2323 unsigned StrictPenalty =
2324 breakProtrudingToken(Current, StrictState, AllowBreak,
2325 /*DryRun=*/true, /*Strict=*/true)
2326 .first;
2327 Strict = StrictPenalty <= Penalty;
2328 if (Strict) {
2329 Penalty = StrictPenalty;
2330 State = StrictState;
2333 if (!DryRun) {
2334 // If we're not in dry-run mode, apply the changes with the decision on
2335 // strictness made above.
2336 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
2337 Strict);
2340 if (State.Column > getColumnLimit(State)) {
2341 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2342 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2344 return Penalty;
2347 // Returns the enclosing function name of a token, or the empty string if not
2348 // found.
2349 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2350 // Look for: 'function(' or 'function<templates>(' before Current.
2351 auto Tok = Current.getPreviousNonComment();
2352 if (!Tok || Tok->isNot(tok::l_paren))
2353 return "";
2354 Tok = Tok->getPreviousNonComment();
2355 if (!Tok)
2356 return "";
2357 if (Tok->is(TT_TemplateCloser)) {
2358 Tok = Tok->MatchingParen;
2359 if (Tok)
2360 Tok = Tok->getPreviousNonComment();
2362 if (!Tok || Tok->isNot(tok::identifier))
2363 return "";
2364 return Tok->TokenText;
2367 std::optional<FormatStyle>
2368 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2369 const LineState &State) {
2370 if (!Current.isStringLiteral())
2371 return std::nullopt;
2372 auto Delimiter = getRawStringDelimiter(Current.TokenText);
2373 if (!Delimiter)
2374 return std::nullopt;
2375 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
2376 if (!RawStringStyle && Delimiter->empty()) {
2377 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2378 getEnclosingFunctionName(Current));
2380 if (!RawStringStyle)
2381 return std::nullopt;
2382 RawStringStyle->ColumnLimit = getColumnLimit(State);
2383 return RawStringStyle;
2386 std::unique_ptr<BreakableToken>
2387 ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2388 LineState &State, bool AllowBreak) {
2389 unsigned StartColumn = State.Column - Current.ColumnWidth;
2390 if (Current.isStringLiteral()) {
2391 // Strings in JSON cannot be broken. Breaking strings in JavaScript is
2392 // disabled for now.
2393 if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals ||
2394 !AllowBreak) {
2395 return nullptr;
2398 // Don't break string literals inside preprocessor directives (except for
2399 // #define directives, as their contents are stored in separate lines and
2400 // are not affected by this check).
2401 // This way we avoid breaking code with line directives and unknown
2402 // preprocessor directives that contain long string literals.
2403 if (State.Line->Type == LT_PreprocessorDirective)
2404 return nullptr;
2405 // Exempts unterminated string literals from line breaking. The user will
2406 // likely want to terminate the string before any line breaking is done.
2407 if (Current.IsUnterminatedLiteral)
2408 return nullptr;
2409 // Don't break string literals inside Objective-C array literals (doing so
2410 // raises the warning -Wobjc-string-concatenation).
2411 if (State.Stack.back().IsInsideObjCArrayLiteral)
2412 return nullptr;
2414 // The "DPI"/"DPI-C" in SystemVerilog direct programming interface
2415 // imports/exports cannot be split, e.g.
2416 // `import "DPI" function foo();`
2417 // FIXME: make this use same infra as C++ import checks
2418 if (Style.isVerilog() && Current.Previous &&
2419 Current.Previous->isOneOf(tok::kw_export, Keywords.kw_import)) {
2420 return nullptr;
2422 StringRef Text = Current.TokenText;
2424 // We need this to address the case where there is an unbreakable tail only
2425 // if certain other formatting decisions have been taken. The
2426 // UnbreakableTailLength of Current is an overapproximation in that case and
2427 // we need to be correct here.
2428 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2430 : Current.UnbreakableTailLength;
2432 if (Style.isVerilog() || Style.Language == FormatStyle::LK_Java ||
2433 Style.isJavaScript() || Style.isCSharp()) {
2434 BreakableStringLiteralUsingOperators::QuoteStyleType QuoteStyle;
2435 if (Style.isJavaScript() && Text.starts_with("'") &&
2436 Text.ends_with("'")) {
2437 QuoteStyle = BreakableStringLiteralUsingOperators::SingleQuotes;
2438 } else if (Style.isCSharp() && Text.starts_with("@\"") &&
2439 Text.ends_with("\"")) {
2440 QuoteStyle = BreakableStringLiteralUsingOperators::AtDoubleQuotes;
2441 } else if (Text.starts_with("\"") && Text.ends_with("\"")) {
2442 QuoteStyle = BreakableStringLiteralUsingOperators::DoubleQuotes;
2443 } else {
2444 return nullptr;
2446 return std::make_unique<BreakableStringLiteralUsingOperators>(
2447 Current, QuoteStyle,
2448 /*UnindentPlus=*/shouldUnindentNextOperator(Current), StartColumn,
2449 UnbreakableTailLength, State.Line->InPPDirective, Encoding, Style);
2452 StringRef Prefix;
2453 StringRef Postfix;
2454 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2455 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2456 // reduce the overhead) for each FormatToken, which is a string, so that we
2457 // don't run multiple checks here on the hot path.
2458 if ((Text.ends_with(Postfix = "\"") &&
2459 (Text.starts_with(Prefix = "@\"") || Text.starts_with(Prefix = "\"") ||
2460 Text.starts_with(Prefix = "u\"") ||
2461 Text.starts_with(Prefix = "U\"") ||
2462 Text.starts_with(Prefix = "u8\"") ||
2463 Text.starts_with(Prefix = "L\""))) ||
2464 (Text.starts_with(Prefix = "_T(\"") &&
2465 Text.ends_with(Postfix = "\")"))) {
2466 return std::make_unique<BreakableStringLiteral>(
2467 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
2468 State.Line->InPPDirective, Encoding, Style);
2470 } else if (Current.is(TT_BlockComment)) {
2471 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2472 // If a comment token switches formatting, like
2473 // /* clang-format on */, we don't want to break it further,
2474 // but we may still want to adjust its indentation.
2475 switchesFormatting(Current)) {
2476 return nullptr;
2478 return std::make_unique<BreakableBlockComment>(
2479 Current, StartColumn, Current.OriginalColumn, !Current.Previous,
2480 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF());
2481 } else if (Current.is(TT_LineComment) &&
2482 (!Current.Previous ||
2483 Current.Previous->isNot(TT_ImplicitStringLiteral))) {
2484 bool RegularComments = [&]() {
2485 for (const FormatToken *T = &Current; T && T->is(TT_LineComment);
2486 T = T->Next) {
2487 if (!(T->TokenText.starts_with("//") || T->TokenText.starts_with("#")))
2488 return false;
2490 return true;
2491 }();
2492 if (Style.ReflowComments == FormatStyle::RCS_Never ||
2493 CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
2494 switchesFormatting(Current) || !RegularComments) {
2495 return nullptr;
2497 return std::make_unique<BreakableLineCommentSection>(
2498 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style);
2500 return nullptr;
2503 std::pair<unsigned, bool>
2504 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2505 LineState &State, bool AllowBreak,
2506 bool DryRun, bool Strict) {
2507 std::unique_ptr<const BreakableToken> Token =
2508 createBreakableToken(Current, State, AllowBreak);
2509 if (!Token)
2510 return {0, false};
2511 assert(Token->getLineCount() > 0);
2512 unsigned ColumnLimit = getColumnLimit(State);
2513 if (Current.is(TT_LineComment)) {
2514 // We don't insert backslashes when breaking line comments.
2515 ColumnLimit = Style.ColumnLimit;
2517 if (ColumnLimit == 0) {
2518 // To make the rest of the function easier set the column limit to the
2519 // maximum, if there should be no limit.
2520 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2522 if (Current.UnbreakableTailLength >= ColumnLimit)
2523 return {0, false};
2524 // ColumnWidth was already accounted into State.Column before calling
2525 // breakProtrudingToken.
2526 unsigned StartColumn = State.Column - Current.ColumnWidth;
2527 unsigned NewBreakPenalty = Current.isStringLiteral()
2528 ? Style.PenaltyBreakString
2529 : Style.PenaltyBreakComment;
2530 // Stores whether we intentionally decide to let a line exceed the column
2531 // limit.
2532 bool Exceeded = false;
2533 // Stores whether we introduce a break anywhere in the token.
2534 bool BreakInserted = Token->introducesBreakBeforeToken();
2535 // Store whether we inserted a new line break at the end of the previous
2536 // logical line.
2537 bool NewBreakBefore = false;
2538 // We use a conservative reflowing strategy. Reflow starts after a line is
2539 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2540 // line that doesn't get reflown with the previous line is reached.
2541 bool Reflow = false;
2542 // Keep track of where we are in the token:
2543 // Where we are in the content of the current logical line.
2544 unsigned TailOffset = 0;
2545 // The column number we're currently at.
2546 unsigned ContentStartColumn =
2547 Token->getContentStartColumn(0, /*Break=*/false);
2548 // The number of columns left in the current logical line after TailOffset.
2549 unsigned RemainingTokenColumns =
2550 Token->getRemainingLength(0, TailOffset, ContentStartColumn);
2551 // Adapt the start of the token, for example indent.
2552 if (!DryRun)
2553 Token->adaptStartOfLine(0, Whitespaces);
2555 unsigned ContentIndent = 0;
2556 unsigned Penalty = 0;
2557 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2558 << StartColumn << ".\n");
2559 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2560 LineIndex != EndIndex; ++LineIndex) {
2561 LLVM_DEBUG(llvm::dbgs()
2562 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2563 NewBreakBefore = false;
2564 // If we did reflow the previous line, we'll try reflowing again. Otherwise
2565 // we'll start reflowing if the current line is broken or whitespace is
2566 // compressed.
2567 bool TryReflow = Reflow;
2568 // Break the current token until we can fit the rest of the line.
2569 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2570 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
2571 << (ContentStartColumn + RemainingTokenColumns)
2572 << ", space: " << ColumnLimit
2573 << ", reflown prefix: " << ContentStartColumn
2574 << ", offset in line: " << TailOffset << "\n");
2575 // If the current token doesn't fit, find the latest possible split in the
2576 // current line so that breaking at it will be under the column limit.
2577 // FIXME: Use the earliest possible split while reflowing to correctly
2578 // compress whitespace within a line.
2579 BreakableToken::Split Split =
2580 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2581 ContentStartColumn, CommentPragmasRegex);
2582 if (Split.first == StringRef::npos) {
2583 // No break opportunity - update the penalty and continue with the next
2584 // logical line.
2585 if (LineIndex < EndIndex - 1) {
2586 // The last line's penalty is handled in addNextStateToQueue() or when
2587 // calling replaceWhitespaceAfterLastLine below.
2588 Penalty += Style.PenaltyExcessCharacter *
2589 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2591 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
2592 break;
2594 assert(Split.first != 0);
2596 if (Token->supportsReflow()) {
2597 // Check whether the next natural split point after the current one can
2598 // still fit the line, either because we can compress away whitespace,
2599 // or because the penalty the excess characters introduce is lower than
2600 // the break penalty.
2601 // We only do this for tokens that support reflowing, and thus allow us
2602 // to change the whitespace arbitrarily (e.g. comments).
2603 // Other tokens, like string literals, can be broken on arbitrary
2604 // positions.
2606 // First, compute the columns from TailOffset to the next possible split
2607 // position.
2608 // For example:
2609 // ColumnLimit: |
2610 // // Some text that breaks
2611 // ^ tail offset
2612 // ^-- split
2613 // ^-------- to split columns
2614 // ^--- next split
2615 // ^--------------- to next split columns
2616 unsigned ToSplitColumns = Token->getRangeLength(
2617 LineIndex, TailOffset, Split.first, ContentStartColumn);
2618 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
2620 BreakableToken::Split NextSplit = Token->getSplit(
2621 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
2622 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2623 // Compute the columns necessary to fit the next non-breakable sequence
2624 // into the current line.
2625 unsigned ToNextSplitColumns = 0;
2626 if (NextSplit.first == StringRef::npos) {
2627 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
2628 ContentStartColumn);
2629 } else {
2630 ToNextSplitColumns = Token->getRangeLength(
2631 LineIndex, TailOffset,
2632 Split.first + Split.second + NextSplit.first, ContentStartColumn);
2634 // Compress the whitespace between the break and the start of the next
2635 // unbreakable sequence.
2636 ToNextSplitColumns =
2637 Token->getLengthAfterCompression(ToNextSplitColumns, Split);
2638 LLVM_DEBUG(llvm::dbgs()
2639 << " ContentStartColumn: " << ContentStartColumn << "\n");
2640 LLVM_DEBUG(llvm::dbgs()
2641 << " ToNextSplit: " << ToNextSplitColumns << "\n");
2642 // If the whitespace compression makes us fit, continue on the current
2643 // line.
2644 bool ContinueOnLine =
2645 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2646 unsigned ExcessCharactersPenalty = 0;
2647 if (!ContinueOnLine && !Strict) {
2648 // Similarly, if the excess characters' penalty is lower than the
2649 // penalty of introducing a new break, continue on the current line.
2650 ExcessCharactersPenalty =
2651 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2652 Style.PenaltyExcessCharacter;
2653 LLVM_DEBUG(llvm::dbgs()
2654 << " Penalty excess: " << ExcessCharactersPenalty
2655 << "\n break : " << NewBreakPenalty << "\n");
2656 if (ExcessCharactersPenalty < NewBreakPenalty) {
2657 Exceeded = true;
2658 ContinueOnLine = true;
2661 if (ContinueOnLine) {
2662 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
2663 // The current line fits after compressing the whitespace - reflow
2664 // the next line into it if possible.
2665 TryReflow = true;
2666 if (!DryRun) {
2667 Token->compressWhitespace(LineIndex, TailOffset, Split,
2668 Whitespaces);
2670 // When we continue on the same line, leave one space between content.
2671 ContentStartColumn += ToSplitColumns + 1;
2672 Penalty += ExcessCharactersPenalty;
2673 TailOffset += Split.first + Split.second;
2674 RemainingTokenColumns = Token->getRemainingLength(
2675 LineIndex, TailOffset, ContentStartColumn);
2676 continue;
2679 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
2680 // Update the ContentIndent only if the current line was not reflown with
2681 // the previous line, since in that case the previous line should still
2682 // determine the ContentIndent. Also never intent the last line.
2683 if (!Reflow)
2684 ContentIndent = Token->getContentIndent(LineIndex);
2685 LLVM_DEBUG(llvm::dbgs()
2686 << " ContentIndent: " << ContentIndent << "\n");
2687 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2688 LineIndex, /*Break=*/true);
2690 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2691 LineIndex, TailOffset + Split.first + Split.second,
2692 ContentStartColumn);
2693 if (NewRemainingTokenColumns == 0) {
2694 // No content to indent.
2695 ContentIndent = 0;
2696 ContentStartColumn =
2697 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2698 NewRemainingTokenColumns = Token->getRemainingLength(
2699 LineIndex, TailOffset + Split.first + Split.second,
2700 ContentStartColumn);
2703 // When breaking before a tab character, it may be moved by a few columns,
2704 // but will still be expanded to the next tab stop, so we don't save any
2705 // columns.
2706 if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2707 // FIXME: Do we need to adjust the penalty?
2708 break;
2711 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2712 << ", " << Split.second << "\n");
2713 if (!DryRun) {
2714 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2715 Whitespaces);
2718 Penalty += NewBreakPenalty;
2719 TailOffset += Split.first + Split.second;
2720 RemainingTokenColumns = NewRemainingTokenColumns;
2721 BreakInserted = true;
2722 NewBreakBefore = true;
2724 // In case there's another line, prepare the state for the start of the next
2725 // line.
2726 if (LineIndex + 1 != EndIndex) {
2727 unsigned NextLineIndex = LineIndex + 1;
2728 if (NewBreakBefore) {
2729 // After breaking a line, try to reflow the next line into the current
2730 // one once RemainingTokenColumns fits.
2731 TryReflow = true;
2733 if (TryReflow) {
2734 // We decided that we want to try reflowing the next line into the
2735 // current one.
2736 // We will now adjust the state as if the reflow is successful (in
2737 // preparation for the next line), and see whether that works. If we
2738 // decide that we cannot reflow, we will later reset the state to the
2739 // start of the next line.
2740 Reflow = false;
2741 // As we did not continue breaking the line, RemainingTokenColumns is
2742 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2743 // the position at which we want to format the next line if we do
2744 // actually reflow.
2745 // When we reflow, we need to add a space between the end of the current
2746 // line and the next line's start column.
2747 ContentStartColumn += RemainingTokenColumns + 1;
2748 // Get the split that we need to reflow next logical line into the end
2749 // of the current one; the split will include any leading whitespace of
2750 // the next logical line.
2751 BreakableToken::Split SplitBeforeNext =
2752 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
2753 LLVM_DEBUG(llvm::dbgs()
2754 << " Size of reflown text: " << ContentStartColumn
2755 << "\n Potential reflow split: ");
2756 if (SplitBeforeNext.first != StringRef::npos) {
2757 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2758 << SplitBeforeNext.second << "\n");
2759 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2760 // If the rest of the next line fits into the current line below the
2761 // column limit, we can safely reflow.
2762 RemainingTokenColumns = Token->getRemainingLength(
2763 NextLineIndex, TailOffset, ContentStartColumn);
2764 Reflow = true;
2765 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2766 LLVM_DEBUG(llvm::dbgs()
2767 << " Over limit after reflow, need: "
2768 << (ContentStartColumn + RemainingTokenColumns)
2769 << ", space: " << ColumnLimit
2770 << ", reflown prefix: " << ContentStartColumn
2771 << ", offset in line: " << TailOffset << "\n");
2772 // If the whole next line does not fit, try to find a point in
2773 // the next line at which we can break so that attaching the part
2774 // of the next line to that break point onto the current line is
2775 // below the column limit.
2776 BreakableToken::Split Split =
2777 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
2778 ContentStartColumn, CommentPragmasRegex);
2779 if (Split.first == StringRef::npos) {
2780 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
2781 Reflow = false;
2782 } else {
2783 // Check whether the first split point gets us below the column
2784 // limit. Note that we will execute this split below as part of
2785 // the normal token breaking and reflow logic within the line.
2786 unsigned ToSplitColumns = Token->getRangeLength(
2787 NextLineIndex, TailOffset, Split.first, ContentStartColumn);
2788 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2789 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
2790 << (ContentStartColumn + ToSplitColumns)
2791 << ", space: " << ColumnLimit);
2792 unsigned ExcessCharactersPenalty =
2793 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2794 Style.PenaltyExcessCharacter;
2795 if (NewBreakPenalty < ExcessCharactersPenalty)
2796 Reflow = false;
2800 } else {
2801 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2804 if (!Reflow) {
2805 // If we didn't reflow into the next line, the only space to consider is
2806 // the next logical line. Reset our state to match the start of the next
2807 // line.
2808 TailOffset = 0;
2809 ContentStartColumn =
2810 Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2811 RemainingTokenColumns = Token->getRemainingLength(
2812 NextLineIndex, TailOffset, ContentStartColumn);
2813 // Adapt the start of the token, for example indent.
2814 if (!DryRun)
2815 Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2816 } else {
2817 // If we found a reflow split and have added a new break before the next
2818 // line, we are going to remove the line break at the start of the next
2819 // logical line. For example, here we'll add a new line break after
2820 // 'text', and subsequently delete the line break between 'that' and
2821 // 'reflows'.
2822 // // some text that
2823 // // reflows
2824 // ->
2825 // // some text
2826 // // that reflows
2827 // When adding the line break, we also added the penalty for it, so we
2828 // need to subtract that penalty again when we remove the line break due
2829 // to reflowing.
2830 if (NewBreakBefore) {
2831 assert(Penalty >= NewBreakPenalty);
2832 Penalty -= NewBreakPenalty;
2834 if (!DryRun)
2835 Token->reflow(NextLineIndex, Whitespaces);
2840 BreakableToken::Split SplitAfterLastLine =
2841 Token->getSplitAfterLastLine(TailOffset);
2842 if (SplitAfterLastLine.first != StringRef::npos) {
2843 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2845 // We add the last line's penalty here, since that line is going to be split
2846 // now.
2847 Penalty += Style.PenaltyExcessCharacter *
2848 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2850 if (!DryRun) {
2851 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2852 Whitespaces);
2854 ContentStartColumn =
2855 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2856 RemainingTokenColumns = Token->getRemainingLength(
2857 Token->getLineCount() - 1,
2858 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2859 ContentStartColumn);
2862 State.Column = ContentStartColumn + RemainingTokenColumns -
2863 Current.UnbreakableTailLength;
2865 if (BreakInserted) {
2866 if (!DryRun)
2867 Token->updateAfterBroken(Whitespaces);
2869 // If we break the token inside a parameter list, we need to break before
2870 // the next parameter on all levels, so that the next parameter is clearly
2871 // visible. Line comments already introduce a break.
2872 if (Current.isNot(TT_LineComment))
2873 for (ParenState &Paren : State.Stack)
2874 Paren.BreakBeforeParameter = true;
2876 if (Current.is(TT_BlockComment))
2877 State.NoContinuation = true;
2879 State.Stack.back().LastSpace = StartColumn;
2882 Token->updateNextToken(State);
2884 return {Penalty, Exceeded};
2887 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2888 // In preprocessor directives reserve two chars for trailing " \".
2889 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2892 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2893 const FormatToken &Current = *State.NextToken;
2894 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2895 return false;
2896 // We never consider raw string literals "multiline" for the purpose of
2897 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2898 // (see TokenAnnotator::mustBreakBefore().
2899 if (Current.TokenText.starts_with("R\""))
2900 return false;
2901 if (Current.IsMultiline)
2902 return true;
2903 if (Current.getNextNonComment() &&
2904 Current.getNextNonComment()->isStringLiteral()) {
2905 return true; // Implicit concatenation.
2907 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2908 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2909 Style.ColumnLimit) {
2910 return true; // String will be split.
2912 return false;
2915 } // namespace format
2916 } // namespace clang