clang 19.0.0git
ContinuationIndenter.cpp
Go to the documentation of this file.
1//===--- ContinuationIndenter.cpp - Format C++ code -----------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8///
9/// \file
10/// This file implements the continuation indenter.
11///
12//===----------------------------------------------------------------------===//
13
15#include "BreakableToken.h"
16#include "FormatInternal.h"
17#include "FormatToken.h"
18#include "WhitespaceManager.h"
22#include "clang/Format/Format.h"
23#include "llvm/ADT/StringSet.h"
24#include "llvm/Support/Debug.h"
25#include <optional>
26
27#define DEBUG_TYPE "format-indenter"
28
29namespace clang {
30namespace format {
31
32// Returns true if a TT_SelectorName should be indented when wrapped,
33// false otherwise.
37}
38
39// Returns true if a binary operator following \p Tok should be unindented when
40// the style permits it.
41static bool shouldUnindentNextOperator(const FormatToken &Tok) {
43 return Previous && (Previous->getPrecedence() == prec::Assignment ||
44 Previous->isOneOf(tok::kw_return, TT_RequiresClause));
45}
46
47// Returns the length of everything up to the first possible line break after
48// the ), ], } or > matching \c Tok.
49static unsigned getLengthToMatchingParen(const FormatToken &Tok,
51 // Normally whether or not a break before T is possible is calculated and
52 // stored in T.CanBreakBefore. Braces, array initializers and text proto
53 // messages like `key: < ... >` are an exception: a break is possible
54 // before a closing brace R if a break was inserted after the corresponding
55 // opening brace. The information about whether or not a break is needed
56 // before a closing brace R is stored in the ParenState field
57 // S.BreakBeforeClosingBrace where S is the state that R closes.
58 //
59 // In order to decide whether there can be a break before encountered right
60 // braces, this implementation iterates over the sequence of tokens and over
61 // the paren stack in lockstep, keeping track of the stack level which visited
62 // right braces correspond to in MatchingStackIndex.
63 //
64 // For example, consider:
65 // L. <- line number
66 // 1. {
67 // 2. {1},
68 // 3. {2},
69 // 4. {{3}}}
70 // ^ where we call this method with this token.
71 // The paren stack at this point contains 3 brace levels:
72 // 0. { at line 1, BreakBeforeClosingBrace: true
73 // 1. first { at line 4, BreakBeforeClosingBrace: false
74 // 2. second { at line 4, BreakBeforeClosingBrace: false,
75 // where there might be fake parens levels in-between these levels.
76 // The algorithm will start at the first } on line 4, which is the matching
77 // brace of the initial left brace and at level 2 of the stack. Then,
78 // examining BreakBeforeClosingBrace: false at level 2, it will continue to
79 // the second } on line 4, and will traverse the stack downwards until it
80 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
81 // false at level 1, it will continue to the third } on line 4 and will
82 // traverse the stack downwards until it finds the matching { on level 0.
83 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
84 // will stop and will use the second } on line 4 to determine the length to
85 // return, as in this example the range will include the tokens: {3}}
86 //
87 // The algorithm will only traverse the stack if it encounters braces, array
88 // initializer squares or text proto angle brackets.
89 if (!Tok.MatchingParen)
90 return 0;
91 FormatToken *End = Tok.MatchingParen;
92 // Maintains a stack level corresponding to the current End token.
93 int MatchingStackIndex = Stack.size() - 1;
94 // Traverses the stack downwards, looking for the level to which LBrace
95 // corresponds. Returns either a pointer to the matching level or nullptr if
96 // LParen is not found in the initial portion of the stack up to
97 // MatchingStackIndex.
98 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
99 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
100 --MatchingStackIndex;
101 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
102 };
103 for (; End->Next; End = End->Next) {
104 if (End->Next->CanBreakBefore)
105 break;
106 if (!End->Next->closesScope())
107 continue;
108 if (End->Next->MatchingParen &&
109 End->Next->MatchingParen->isOneOf(
110 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
111 const ParenState *State = FindParenState(End->Next->MatchingParen);
112 if (State && State->BreakBeforeClosingBrace)
113 break;
114 }
115 }
116 return End->TotalLength - Tok.TotalLength + 1;
117}
118
119static unsigned getLengthToNextOperator(const FormatToken &Tok) {
120 if (!Tok.NextOperator)
121 return 0;
122 return Tok.NextOperator->TotalLength - Tok.TotalLength;
123}
124
125// Returns \c true if \c Tok is the "." or "->" of a call and starts the next
126// segment of a builder type call.
128 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
129}
130
131// Returns \c true if \c Current starts a new parameter.
132static bool startsNextParameter(const FormatToken &Current,
133 const FormatStyle &Style) {
134 const FormatToken &Previous = *Current.Previous;
135 if (Current.is(TT_CtorInitializerComma) &&
137 return true;
138 }
139 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
140 return true;
141 return Previous.is(tok::comma) && !Current.isTrailingComment() &&
142 ((Previous.isNot(TT_CtorInitializerComma) ||
145 (Previous.isNot(TT_InheritanceComma) ||
147}
148
149static bool opensProtoMessageField(const FormatToken &LessTok,
150 const FormatStyle &Style) {
151 if (LessTok.isNot(tok::less))
152 return false;
153 return Style.Language == FormatStyle::LK_TextProto ||
155 (LessTok.NestingLevel > 0 ||
156 (LessTok.Previous && LessTok.Previous->is(tok::equal))));
157}
158
159// Returns the delimiter of a raw string literal, or std::nullopt if TokenText
160// is not the text of a raw string literal. The delimiter could be the empty
161// string. For example, the delimiter of R"deli(cont)deli" is deli.
162static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
163 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
164 || !TokenText.starts_with("R\"") || !TokenText.ends_with("\"")) {
165 return std::nullopt;
166 }
167
168 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
169 // size at most 16 by the standard, so the first '(' must be among the first
170 // 19 bytes.
171 size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
172 if (LParenPos == StringRef::npos)
173 return std::nullopt;
174 StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
175
176 // Check that the string ends in ')Delimiter"'.
177 size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
178 if (TokenText[RParenPos] != ')')
179 return std::nullopt;
180 if (!TokenText.substr(RParenPos + 1).starts_with(Delimiter))
181 return std::nullopt;
182 return Delimiter;
183}
184
185// Returns the canonical delimiter for \p Language, or the empty string if no
186// canonical delimiter is specified.
187static StringRef
190 for (const auto &Format : Style.RawStringFormats)
191 if (Format.Language == Language)
192 return StringRef(Format.CanonicalDelimiter);
193 return "";
194}
195
197 const FormatStyle &CodeStyle) {
198 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
199 std::optional<FormatStyle> LanguageStyle =
200 CodeStyle.GetLanguageStyle(RawStringFormat.Language);
201 if (!LanguageStyle) {
202 FormatStyle PredefinedStyle;
203 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
204 RawStringFormat.Language, &PredefinedStyle)) {
205 PredefinedStyle = getLLVMStyle();
206 PredefinedStyle.Language = RawStringFormat.Language;
207 }
208 LanguageStyle = PredefinedStyle;
209 }
210 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
211 for (StringRef Delimiter : RawStringFormat.Delimiters)
212 DelimiterStyle.insert({Delimiter, *LanguageStyle});
213 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions)
214 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
215 }
216}
217
218std::optional<FormatStyle>
220 auto It = DelimiterStyle.find(Delimiter);
221 if (It == DelimiterStyle.end())
222 return std::nullopt;
223 return It->second;
224}
225
226std::optional<FormatStyle>
228 StringRef EnclosingFunction) const {
229 auto It = EnclosingFunctionStyle.find(EnclosingFunction);
230 if (It == EnclosingFunctionStyle.end())
231 return std::nullopt;
232 return It->second;
233}
234
236 const AdditionalKeywords &Keywords,
237 const SourceManager &SourceMgr,
238 WhitespaceManager &Whitespaces,
239 encoding::Encoding Encoding,
240 bool BinPackInconclusiveFunctions)
241 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
242 Whitespaces(Whitespaces), Encoding(Encoding),
243 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
244 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
245
247 unsigned FirstStartColumn,
248 const AnnotatedLine *Line,
249 bool DryRun) {
250 LineState State;
251 State.FirstIndent = FirstIndent;
252 if (FirstStartColumn && Line->First->NewlinesBefore == 0)
253 State.Column = FirstStartColumn;
254 else
255 State.Column = FirstIndent;
256 // With preprocessor directive indentation, the line starts on column 0
257 // since it's indented after the hash, but FirstIndent is set to the
258 // preprocessor indent.
260 (Line->Type == LT_PreprocessorDirective ||
261 Line->Type == LT_ImportStatement)) {
262 State.Column = 0;
263 }
264 State.Line = Line;
265 State.NextToken = Line->First;
266 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
267 /*AvoidBinPacking=*/false,
268 /*NoLineBreak=*/false));
269 State.NoContinuation = false;
270 State.StartOfStringLiteral = 0;
271 State.NoLineBreak = false;
272 State.StartOfLineLevel = 0;
273 State.LowestLevelOnLine = 0;
274 State.IgnoreStackForComparison = false;
275
276 if (Style.Language == FormatStyle::LK_TextProto) {
277 // We need this in order to deal with the bin packing of text fields at
278 // global scope.
279 auto &CurrentState = State.Stack.back();
280 CurrentState.AvoidBinPacking = true;
281 CurrentState.BreakBeforeParameter = true;
282 CurrentState.AlignColons = false;
283 }
284
285 // The first token has already been indented and thus consumed.
286 moveStateToNextToken(State, DryRun, /*Newline=*/false);
287 return State;
288}
289
291 const FormatToken &Current = *State.NextToken;
292 const FormatToken &Previous = *Current.Previous;
293 const auto &CurrentState = State.Stack.back();
294 assert(&Previous == Current.Previous);
295 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace &&
296 Current.closesBlockOrBlockTypeList(Style))) {
297 return false;
298 }
299 // The opening "{" of a braced list has to be on the same line as the first
300 // element if it is nested in another braced init list or function call.
301 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
302 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) &&
303 Previous.Previous &&
304 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) {
305 return false;
306 }
307 // This prevents breaks like:
308 // ...
309 // SomeParameter, OtherParameter).DoSomething(
310 // ...
311 // As they hide "DoSomething" and are generally bad for readability.
312 if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
313 State.LowestLevelOnLine < State.StartOfLineLevel &&
314 State.LowestLevelOnLine < Current.NestingLevel) {
315 return false;
316 }
317 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder)
318 return false;
319
320 // Don't create a 'hanging' indent if there are multiple blocks in a single
321 // statement and we are aligning lambda blocks to their signatures.
322 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
323 State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
324 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks &&
326 return false;
327 }
328
329 // Don't break after very short return types (e.g. "void") as that is often
330 // unexpected.
331 if (Current.is(TT_FunctionDeclarationName)) {
333 State.Column < 6) {
334 return false;
335 }
336
338 assert(State.Column >= State.FirstIndent);
339 if (State.Column - State.FirstIndent < 6)
340 return false;
341 }
342 }
343
344 // If binary operators are moved to the next line (including commas for some
345 // styles of constructor initializers), that's always ok.
346 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
347 // Allow breaking opening brace of lambdas (when passed as function
348 // arguments) to a new line when BeforeLambdaBody brace wrapping is
349 // enabled.
351 Current.isNot(TT_LambdaLBrace)) &&
352 CurrentState.NoLineBreakInOperand) {
353 return false;
354 }
355
356 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr))
357 return false;
358
359 if (Current.is(TT_ConditionalExpr) && Previous.is(tok::r_paren) &&
360 Previous.MatchingParen && Previous.MatchingParen->Previous &&
361 Previous.MatchingParen->Previous->MatchingParen &&
362 Previous.MatchingParen->Previous->MatchingParen->is(TT_LambdaLBrace)) {
363 // We have a lambda within a conditional expression, allow breaking here.
364 assert(Previous.MatchingParen->Previous->is(tok::r_brace));
365 return true;
366 }
367
368 return !State.NoLineBreak && !CurrentState.NoLineBreak;
369}
370
372 const FormatToken &Current = *State.NextToken;
373 const FormatToken &Previous = *Current.Previous;
374 const auto &CurrentState = State.Stack.back();
375 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore &&
376 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) {
377 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack);
378 return LambdaBodyLength > getColumnLimit(State);
379 }
380 if (Current.MustBreakBefore ||
381 (Current.is(TT_InlineASMColon) &&
384 Style.ColumnLimit > 0)))) {
385 return true;
386 }
387 if (CurrentState.BreakBeforeClosingBrace &&
388 (Current.closesBlockOrBlockTypeList(Style) ||
389 (Current.is(tok::r_brace) &&
390 Current.isBlockIndentedInitRBrace(Style)))) {
391 return true;
392 }
393 if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren))
394 return true;
395 if (Style.Language == FormatStyle::LK_ObjC &&
397 Current.ObjCSelectorNameParts > 1 &&
398 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
399 return true;
400 }
401 // Avoid producing inconsistent states by requiring breaks where they are not
402 // permitted for C# generic type constraints.
403 if (CurrentState.IsCSharpGenericTypeConstraint &&
404 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) {
405 return false;
406 }
407 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
408 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
409 State.Line->First->isNot(TT_AttributeSquare) && Style.isCpp() &&
410 // FIXME: This is a temporary workaround for the case where clang-format
411 // sets BreakBeforeParameter to avoid bin packing and this creates a
412 // completely unnecessary line break after a template type that isn't
413 // line-wrapped.
414 (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
415 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
416 Previous.isNot(tok::question)) ||
418 Previous.is(TT_ConditionalExpr))) &&
419 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
420 !Current.isOneOf(tok::r_paren, tok::r_brace)) {
421 return true;
422 }
423 if (CurrentState.IsChainedConditional &&
424 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
425 Current.is(tok::colon)) ||
426 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) &&
427 Previous.is(tok::colon)))) {
428 return true;
429 }
430 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
431 (Previous.is(TT_ArrayInitializerLSquare) &&
432 Previous.ParameterCount > 1) ||
434 Style.ColumnLimit > 0 &&
435 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
436 getColumnLimit(State)) {
437 return true;
438 }
439
440 const FormatToken &BreakConstructorInitializersToken =
442 ? Previous
443 : Current;
444 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
445 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
446 getColumnLimit(State) ||
447 CurrentState.BreakBeforeParameter) &&
448 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
451 Style.ColumnLimit != 0)) {
452 return true;
453 }
454
455 if (Current.is(TT_ObjCMethodExpr) && Previous.isNot(TT_SelectorName) &&
456 State.Line->startsWith(TT_ObjCMethodSpecifier)) {
457 return true;
458 }
459 if (Current.is(TT_SelectorName) && Previous.isNot(tok::at) &&
460 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
462 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) {
463 return true;
464 }
465
466 unsigned NewLineColumn = getNewLineColumn(State);
467 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
468 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
469 (State.Column > NewLineColumn ||
470 Current.NestingLevel < State.StartOfLineLevel)) {
471 return true;
472 }
473
474 if (startsSegmentOfBuilderTypeCall(Current) &&
475 (CurrentState.CallContinuation != 0 ||
476 CurrentState.BreakBeforeParameter) &&
477 // JavaScript is treated different here as there is a frequent pattern:
478 // SomeFunction(function() {
479 // ...
480 // }.bind(...));
481 // FIXME: We should find a more generic solution to this problem.
482 !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
483 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
484 return true;
485 }
486
487 // If the template declaration spans multiple lines, force wrap before the
488 // function/class declaration.
489 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
490 Current.CanBreakBefore) {
491 return true;
492 }
493
494 if (State.Line->First->isNot(tok::kw_enum) && State.Column <= NewLineColumn)
495 return false;
496
498 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
499 Previous.is(tok::comma) || Current.NestingLevel < 2) &&
500 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at,
501 Keywords.kw_dollar) &&
502 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
503 nextIsMultilineString(State)) {
504 return true;
505 }
506
507 // Using CanBreakBefore here and below takes care of the decision whether the
508 // current style uses wrapping before or after operators for the given
509 // operator.
510 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
511 const auto PreviousPrecedence = Previous.getPrecedence();
512 if (PreviousPrecedence != prec::Assignment &&
513 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
514 const bool LHSIsBinaryExpr =
515 Previous.Previous && Previous.Previous->EndsBinaryExpression;
516 if (LHSIsBinaryExpr)
517 return true;
518 // If we need to break somewhere inside the LHS of a binary expression, we
519 // should also break after the operator. Otherwise, the formatting would
520 // hide the operator precedence, e.g. in:
521 // if (aaaaaaaaaaaaaa ==
522 // bbbbbbbbbbbbbb && c) {..
523 // For comparisons, we only apply this rule, if the LHS is a binary
524 // expression itself as otherwise, the line breaks seem superfluous.
525 // We need special cases for ">>" which we have split into two ">" while
526 // lexing in order to make template parsing easier.
527 const bool IsComparison =
528 (PreviousPrecedence == prec::Relational ||
529 PreviousPrecedence == prec::Equality ||
530 PreviousPrecedence == prec::Spaceship) &&
531 Previous.Previous &&
532 Previous.Previous->isNot(TT_BinaryOperator); // For >>.
533 if (!IsComparison)
534 return true;
535 }
536 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
537 CurrentState.BreakBeforeParameter) {
538 return true;
539 }
540
541 // Same as above, but for the first "<<" operator.
542 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
543 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
544 return true;
545 }
546
547 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
548 // Always break after "template <...>"(*) and leading annotations. This is
549 // only for cases where the entire line does not fit on a single line as a
550 // different LineFormatter would be used otherwise.
551 // *: Except when another option interferes with that, like concepts.
552 if (Previous.ClosesTemplateDeclaration) {
553 if (Current.is(tok::kw_concept)) {
554 switch (Style.BreakBeforeConceptDeclarations) {
556 break;
558 return true;
560 return false;
561 }
562 }
563 if (Current.is(TT_RequiresClause)) {
564 switch (Style.RequiresClausePosition) {
567 return false;
568 default:
569 return true;
570 }
571 }
574 Current.NewlinesBefore > 0);
575 }
576 if (Previous.is(TT_FunctionAnnotationRParen) &&
577 State.Line->Type != LT_PreprocessorDirective) {
578 return true;
579 }
580 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
581 Current.isNot(TT_LeadingJavaAnnotation)) {
582 return true;
583 }
584 }
585
586 if (Style.isJavaScript() && Previous.is(tok::r_paren) &&
587 Previous.is(TT_JavaAnnotation)) {
588 // Break after the closing parenthesis of TypeScript decorators before
589 // functions, getters and setters.
590 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
591 "function"};
592 if (BreakBeforeDecoratedTokens.contains(Current.TokenText))
593 return true;
594 }
595
596 if (Current.is(TT_FunctionDeclarationName) &&
597 !State.Line->ReturnTypeWrapped &&
598 // Don't break before a C# function when no break after return type.
599 (!Style.isCSharp() ||
601 // Don't always break between a JavaScript `function` and the function
602 // name.
603 !Style.isJavaScript() && Previous.isNot(tok::kw_template) &&
604 CurrentState.BreakBeforeParameter) {
605 return true;
606 }
607
608 // The following could be precomputed as they do not depend on the state.
609 // However, as they should take effect only if the UnwrappedLine does not fit
610 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
611 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) &&
612 Previous.is(tok::l_brace) &&
613 !Current.isOneOf(tok::r_brace, tok::comment)) {
614 return true;
615 }
616
617 if (Current.is(tok::lessless) &&
618 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
619 (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with("\\n\"") ||
620 Previous.TokenText == "\'\\n\'")))) {
621 return true;
622 }
623
624 if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
625 return true;
626
627 if (State.NoContinuation)
628 return true;
629
630 return false;
631}
632
634 bool DryRun,
635 unsigned ExtraSpaces) {
636 const FormatToken &Current = *State.NextToken;
637 assert(State.NextToken->Previous);
638 const FormatToken &Previous = *State.NextToken->Previous;
639
640 assert(!State.Stack.empty());
641 State.NoContinuation = false;
642
643 if (Current.is(TT_ImplicitStringLiteral) &&
644 (!Previous.Tok.getIdentifierInfo() ||
645 Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
646 tok::pp_not_keyword)) {
647 unsigned EndColumn =
648 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
649 if (Current.LastNewlineOffset != 0) {
650 // If there is a newline within this token, the final column will solely
651 // determined by the current end column.
652 State.Column = EndColumn;
653 } else {
654 unsigned StartColumn =
655 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
656 assert(EndColumn >= StartColumn);
657 State.Column += EndColumn - StartColumn;
658 }
659 moveStateToNextToken(State, DryRun, /*Newline=*/false);
660 return 0;
661 }
662
663 unsigned Penalty = 0;
664 if (Newline)
665 Penalty = addTokenOnNewLine(State, DryRun);
666 else
667 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
668
669 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
670}
671
672void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
673 unsigned ExtraSpaces) {
674 FormatToken &Current = *State.NextToken;
675 assert(State.NextToken->Previous);
676 const FormatToken &Previous = *State.NextToken->Previous;
677 auto &CurrentState = State.Stack.back();
678
679 bool DisallowLineBreaksOnThisLine =
681 Style.isCpp() && [&Current] {
682 // Deal with lambda arguments in C++. The aim here is to ensure that we
683 // don't over-indent lambda function bodies when lambdas are passed as
684 // arguments to function calls. We do this by ensuring that either all
685 // arguments (including any lambdas) go on the same line as the function
686 // call, or we break before the first argument.
687 const auto *Prev = Current.Previous;
688 if (!Prev)
689 return false;
690 // For example, `/*Newline=*/false`.
691 if (Prev->is(TT_BlockComment) && Current.SpacesRequiredBefore == 0)
692 return false;
693 const auto *PrevNonComment = Current.getPreviousNonComment();
694 if (!PrevNonComment || PrevNonComment->isNot(tok::l_paren))
695 return false;
696 if (Current.isOneOf(tok::comment, tok::l_paren, TT_LambdaLSquare))
697 return false;
698 auto BlockParameterCount = PrevNonComment->BlockParameterCount;
699 if (BlockParameterCount == 0)
700 return false;
701
702 // Multiple lambdas in the same function call.
703 if (BlockParameterCount > 1)
704 return true;
705
706 // A lambda followed by another arg.
707 if (!PrevNonComment->Role)
708 return false;
709 auto Comma = PrevNonComment->Role->lastComma();
710 if (!Comma)
711 return false;
712 auto Next = Comma->getNextNonComment();
713 return Next &&
714 !Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret);
715 }();
716
717 if (DisallowLineBreaksOnThisLine)
718 State.NoLineBreak = true;
719
720 if (Current.is(tok::equal) &&
721 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
722 CurrentState.VariablePos == 0 &&
723 (!Previous.Previous ||
724 Previous.Previous->isNot(TT_DesignatedInitializerPeriod))) {
725 CurrentState.VariablePos = State.Column;
726 // Move over * and & if they are bound to the variable name.
727 const FormatToken *Tok = &Previous;
728 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
729 CurrentState.VariablePos -= Tok->ColumnWidth;
730 if (Tok->SpacesRequiredBefore != 0)
731 break;
732 Tok = Tok->Previous;
733 }
734 if (Previous.PartOfMultiVariableDeclStmt)
735 CurrentState.LastSpace = CurrentState.VariablePos;
736 }
737
738 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
739
740 // Indent preprocessor directives after the hash if required.
741 int PPColumnCorrection = 0;
743 Previous.is(tok::hash) && State.FirstIndent > 0 &&
744 &Previous == State.Line->First &&
745 (State.Line->Type == LT_PreprocessorDirective ||
746 State.Line->Type == LT_ImportStatement)) {
747 Spaces += State.FirstIndent;
748
749 // For preprocessor indent with tabs, State.Column will be 1 because of the
750 // hash. This causes second-level indents onward to have an extra space
751 // after the tabs. We avoid this misalignment by subtracting 1 from the
752 // column value passed to replaceWhitespace().
753 if (Style.UseTab != FormatStyle::UT_Never)
754 PPColumnCorrection = -1;
755 }
756
757 if (!DryRun) {
758 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
759 State.Column + Spaces + PPColumnCorrection,
760 /*IsAligned=*/false, State.Line->InMacroBody);
761 }
762
763 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
764 // declaration unless there is multiple inheritance.
766 Current.is(TT_InheritanceColon)) {
767 CurrentState.NoLineBreak = true;
768 }
770 Previous.is(TT_InheritanceColon)) {
771 CurrentState.NoLineBreak = true;
772 }
773
774 if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
775 unsigned MinIndent = std::max(
776 State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent);
777 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
778 if (Current.LongestObjCSelectorName == 0)
779 CurrentState.AlignColons = false;
780 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
781 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
782 else
783 CurrentState.ColonPos = FirstColonPos;
784 }
785
786 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
787 // parenthesis by disallowing any further line breaks if there is no line
788 // break after the opening parenthesis. Don't break if it doesn't conserve
789 // columns.
790 auto IsOpeningBracket = [&](const FormatToken &Tok) {
791 auto IsStartOfBracedList = [&]() {
792 return Tok.is(tok::l_brace) && Tok.isNot(BK_Block) &&
794 };
795 if (!Tok.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) &&
796 !IsStartOfBracedList()) {
797 return false;
798 }
799 if (!Tok.Previous)
800 return true;
801 if (Tok.Previous->isIf())
803 return !Tok.Previous->isOneOf(TT_CastRParen, tok::kw_for, tok::kw_while,
804 tok::kw_switch);
805 };
808 IsOpeningBracket(Previous) && State.Column > getNewLineColumn(State) &&
809 // Don't do this for simple (no expressions) one-argument function calls
810 // as that feels like needlessly wasting whitespace, e.g.:
811 //
812 // caaaaaaaaaaaall(
813 // caaaaaaaaaaaall(
814 // caaaaaaaaaaaall(
815 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
816 Current.FakeLParens.size() > 0 &&
817 Current.FakeLParens.back() > prec::Unknown) {
818 CurrentState.NoLineBreak = true;
819 }
820 if (Previous.is(TT_TemplateString) && Previous.opensScope())
821 CurrentState.NoLineBreak = true;
822
823 // Align following lines within parentheses / brackets if configured.
824 // Note: This doesn't apply to macro expansion lines, which are MACRO( , , )
825 // with args as children of the '(' and ',' tokens. It does not make sense to
826 // align the commas with the opening paren.
828 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
829 Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) &&
830 Previous.isNot(TT_TableGenDAGArgOpener) &&
831 Previous.isNot(TT_TableGenDAGArgOpenerToBreak) &&
832 !(Current.MacroParent && Previous.MacroParent) &&
833 (Current.isNot(TT_LineComment) ||
834 Previous.isOneOf(BK_BracedInit, TT_VerilogMultiLineListLParen))) {
835 CurrentState.Indent = State.Column + Spaces;
836 CurrentState.IsAligned = true;
837 }
838 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
839 CurrentState.NoLineBreak = true;
841 State.Column > getNewLineColumn(State)) {
842 CurrentState.ContainsUnwrappedBuilder = true;
843 }
844
845 if (Current.is(TT_TrailingReturnArrow) &&
847 CurrentState.NoLineBreak = true;
848 }
849 if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
850 (Previous.MatchingParen &&
851 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
852 // If there is a function call with long parameters, break before trailing
853 // calls. This prevents things like:
854 // EXPECT_CALL(SomeLongParameter).Times(
855 // 2);
856 // We don't want to do this for short parameters as they can just be
857 // indexes.
858 CurrentState.NoLineBreak = true;
859 }
860
861 // Don't allow the RHS of an operator to be split over multiple lines unless
862 // there is a line-break right after the operator.
863 // Exclude relational operators, as there, it is always more desirable to
864 // have the LHS 'left' of the RHS.
865 const FormatToken *P = Current.getPreviousNonComment();
866 if (Current.isNot(tok::comment) && P &&
867 (P->isOneOf(TT_BinaryOperator, tok::comma) ||
868 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
869 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
870 P->getPrecedence() != prec::Assignment &&
871 P->getPrecedence() != prec::Relational &&
872 P->getPrecedence() != prec::Spaceship) {
873 bool BreakBeforeOperator =
874 P->MustBreakBefore || P->is(tok::lessless) ||
875 (P->is(TT_BinaryOperator) &&
877 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
878 // Don't do this if there are only two operands. In these cases, there is
879 // always a nice vertical separation between them and the extra line break
880 // does not help.
881 bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator &&
882 P->isNot(TT_ConditionalExpr);
883 if ((!BreakBeforeOperator &&
884 !(HasTwoOperands &&
886 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
887 CurrentState.NoLineBreakInOperand = true;
888 }
889 }
890
891 State.Column += Spaces;
892 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
893 Previous.Previous &&
894 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) {
895 // Treat the condition inside an if as if it was a second function
896 // parameter, i.e. let nested calls have a continuation indent.
897 CurrentState.LastSpace = State.Column;
898 CurrentState.NestedBlockIndent = State.Column;
899 } else if (!Current.isOneOf(tok::comment, tok::caret) &&
900 ((Previous.is(tok::comma) &&
901 Previous.isNot(TT_OverloadedOperator)) ||
902 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
903 CurrentState.LastSpace = State.Column;
904 } else if (Previous.is(TT_CtorInitializerColon) &&
905 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
908 CurrentState.Indent = State.Column;
909 CurrentState.LastSpace = State.Column;
910 } else if (Previous.isOneOf(TT_ConditionalExpr, TT_CtorInitializerColon)) {
911 CurrentState.LastSpace = State.Column;
912 } else if (Previous.is(TT_BinaryOperator) &&
913 ((Previous.getPrecedence() != prec::Assignment &&
914 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
915 Previous.NextOperator)) ||
916 Current.StartsBinaryExpression)) {
917 // Indent relative to the RHS of the expression unless this is a simple
918 // assignment without binary expression on the RHS.
920 CurrentState.LastSpace = State.Column;
921 } else if (Previous.is(TT_InheritanceColon)) {
922 CurrentState.Indent = State.Column;
923 CurrentState.LastSpace = State.Column;
924 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) {
925 CurrentState.ColonPos = State.Column;
926 } else if (Previous.opensScope()) {
927 // If a function has a trailing call, indent all parameters from the
928 // opening parenthesis. This avoids confusing indents like:
929 // OuterFunction(InnerFunctionCall( // break
930 // ParameterToInnerFunction)) // break
931 // .SecondInnerFunctionCall();
932 if (Previous.MatchingParen) {
933 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
934 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
935 State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
936 CurrentState.LastSpace = State.Column;
937 }
938 }
939 }
940}
941
942unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
943 bool DryRun) {
944 FormatToken &Current = *State.NextToken;
945 assert(State.NextToken->Previous);
946 const FormatToken &Previous = *State.NextToken->Previous;
947 auto &CurrentState = State.Stack.back();
948
949 // Extra penalty that needs to be added because of the way certain line
950 // breaks are chosen.
951 unsigned Penalty = 0;
952
953 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
954 const FormatToken *NextNonComment = Previous.getNextNonComment();
955 if (!NextNonComment)
956 NextNonComment = &Current;
957 // The first line break on any NestingLevel causes an extra penalty in order
958 // prefer similar line breaks.
959 if (!CurrentState.ContainsLineBreak)
960 Penalty += 15;
961 CurrentState.ContainsLineBreak = true;
962
963 Penalty += State.NextToken->SplitPenalty;
964
965 // Breaking before the first "<<" is generally not desirable if the LHS is
966 // short. Also always add the penalty if the LHS is split over multiple lines
967 // to avoid unnecessary line breaks that just work around this penalty.
968 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 &&
969 (State.Column <= Style.ColumnLimit / 3 ||
970 CurrentState.BreakBeforeParameter)) {
971 Penalty += Style.PenaltyBreakFirstLessLess;
972 }
973
974 State.Column = getNewLineColumn(State);
975
976 // Add Penalty proportional to amount of whitespace away from FirstColumn
977 // This tends to penalize several lines that are far-right indented,
978 // and prefers a line-break prior to such a block, e.g:
979 //
980 // Constructor() :
981 // member(value), looooooooooooooooong_member(
982 // looooooooooong_call(param_1, param_2, param_3))
983 // would then become
984 // Constructor() :
985 // member(value),
986 // looooooooooooooooong_member(
987 // looooooooooong_call(param_1, param_2, param_3))
988 if (State.Column > State.FirstIndent) {
989 Penalty +=
990 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
991 }
992
993 // Indent nested blocks relative to this column, unless in a very specific
994 // JavaScript special case where:
995 //
996 // var loooooong_name =
997 // function() {
998 // // code
999 // }
1000 //
1001 // is common and should be formatted like a free-standing function. The same
1002 // goes for wrapping before the lambda return type arrow.
1003 if (Current.isNot(TT_TrailingReturnArrow) &&
1004 (!Style.isJavaScript() || Current.NestingLevel != 0 ||
1005 !PreviousNonComment || PreviousNonComment->isNot(tok::equal) ||
1006 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) {
1007 CurrentState.NestedBlockIndent = State.Column;
1008 }
1009
1010 if (NextNonComment->isMemberAccess()) {
1011 if (CurrentState.CallContinuation == 0)
1012 CurrentState.CallContinuation = State.Column;
1013 } else if (NextNonComment->is(TT_SelectorName)) {
1014 if (!CurrentState.ObjCSelectorNameFound) {
1015 if (NextNonComment->LongestObjCSelectorName == 0) {
1016 CurrentState.AlignColons = false;
1017 } else {
1018 CurrentState.ColonPos =
1019 (shouldIndentWrappedSelectorName(Style, State.Line->Type)
1020 ? std::max(CurrentState.Indent,
1021 State.FirstIndent + Style.ContinuationIndentWidth)
1022 : CurrentState.Indent) +
1023 std::max(NextNonComment->LongestObjCSelectorName,
1024 NextNonComment->ColumnWidth);
1025 }
1026 } else if (CurrentState.AlignColons &&
1027 CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
1028 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
1029 }
1030 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1031 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1032 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
1033 // method expression, the block should be aligned to the line starting it,
1034 // e.g.:
1035 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
1036 // ^(int *i) {
1037 // // ...
1038 // }];
1039 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
1040 // when we consume all of the "}"'s FakeRParens at the "{".
1041 if (State.Stack.size() > 1) {
1042 State.Stack[State.Stack.size() - 2].LastSpace =
1043 std::max(CurrentState.LastSpace, CurrentState.Indent) +
1045 }
1046 }
1047
1048 if ((PreviousNonComment &&
1049 PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
1050 !CurrentState.AvoidBinPacking) ||
1051 Previous.is(TT_BinaryOperator)) {
1052 CurrentState.BreakBeforeParameter = false;
1053 }
1054 if (PreviousNonComment &&
1055 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) ||
1056 PreviousNonComment->ClosesRequiresClause) &&
1057 Current.NestingLevel == 0) {
1058 CurrentState.BreakBeforeParameter = false;
1059 }
1060 if (NextNonComment->is(tok::question) ||
1061 (PreviousNonComment && PreviousNonComment->is(tok::question))) {
1062 CurrentState.BreakBeforeParameter = true;
1063 }
1064 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
1065 CurrentState.BreakBeforeParameter = false;
1066
1067 if (!DryRun) {
1068 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
1069 if (Current.is(tok::r_brace) && Current.MatchingParen &&
1070 // Only strip trailing empty lines for l_braces that have children, i.e.
1071 // for function expressions (lambdas, arrows, etc).
1072 !Current.MatchingParen->Children.empty()) {
1073 // lambdas and arrow functions are expressions, thus their r_brace is not
1074 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
1075 // about removing empty lines on closing blocks. Special case them here.
1076 MaxEmptyLinesToKeep = 1;
1077 }
1078 unsigned Newlines =
1079 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
1080 bool ContinuePPDirective =
1081 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
1082 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
1083 CurrentState.IsAligned, ContinuePPDirective);
1084 }
1085
1086 if (!Current.isTrailingComment())
1087 CurrentState.LastSpace = State.Column;
1088 if (Current.is(tok::lessless)) {
1089 // If we are breaking before a "<<", we always want to indent relative to
1090 // RHS. This is necessary only for "<<", as we special-case it and don't
1091 // always indent relative to the RHS.
1092 CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1093 }
1094
1095 State.StartOfLineLevel = Current.NestingLevel;
1096 State.LowestLevelOnLine = Current.NestingLevel;
1097
1098 // Any break on this level means that the parent level has been broken
1099 // and we need to avoid bin packing there.
1100 bool NestedBlockSpecialCase =
1101 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
1102 State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1103 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) &&
1104 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1105 // Do not force parameter break for statements with requires expressions.
1106 NestedBlockSpecialCase =
1107 NestedBlockSpecialCase ||
1108 (Current.MatchingParen &&
1109 Current.MatchingParen->is(TT_RequiresExpressionLBrace));
1110 if (!NestedBlockSpecialCase) {
1111 auto ParentLevelIt = std::next(State.Stack.rbegin());
1113 Current.MatchingParen && Current.MatchingParen->is(TT_LambdaLBrace)) {
1114 // If the first character on the new line is a lambda's closing brace, the
1115 // stack still contains that lambda's parenthesis. As such, we need to
1116 // recurse further down the stack than usual to find the parenthesis level
1117 // containing the lambda, which is where we want to set
1118 // BreakBeforeParameter.
1119 //
1120 // We specifically special case "OuterScope"-formatted lambdas here
1121 // because, when using that setting, breaking before the parameter
1122 // directly following the lambda is particularly unsightly. However, when
1123 // "OuterScope" is not set, the logic to find the parent parenthesis level
1124 // still appears to be sometimes incorrect. It has not been fixed yet
1125 // because it would lead to significant changes in existing behaviour.
1126 //
1127 // TODO: fix the non-"OuterScope" case too.
1128 auto FindCurrentLevel = [&](const auto &It) {
1129 return std::find_if(It, State.Stack.rend(), [](const auto &PState) {
1130 return PState.Tok != nullptr; // Ignore fake parens.
1131 });
1132 };
1133 auto MaybeIncrement = [&](const auto &It) {
1134 return It != State.Stack.rend() ? std::next(It) : It;
1135 };
1136 auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin());
1137 auto LevelContainingLambdaIt =
1138 FindCurrentLevel(MaybeIncrement(LambdaLevelIt));
1139 ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt);
1140 }
1141 for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I)
1142 I->BreakBeforeParameter = true;
1143 }
1144
1145 if (PreviousNonComment &&
1146 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
1147 ((PreviousNonComment->isNot(TT_TemplateCloser) &&
1148 !PreviousNonComment->ClosesRequiresClause) ||
1149 Current.NestingLevel != 0) &&
1150 !PreviousNonComment->isOneOf(
1151 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
1152 TT_LeadingJavaAnnotation) &&
1153 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope() &&
1154 // We don't want to enforce line breaks for subsequent arguments just
1155 // because we have been forced to break before a lambda body.
1157 Current.isNot(TT_LambdaLBrace))) {
1158 CurrentState.BreakBeforeParameter = true;
1159 }
1160
1161 // If we break after { or the [ of an array initializer, we should also break
1162 // before the corresponding } or ].
1163 if (PreviousNonComment &&
1164 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1165 opensProtoMessageField(*PreviousNonComment, Style))) {
1166 CurrentState.BreakBeforeClosingBrace = true;
1167 }
1168
1169 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) {
1170 CurrentState.BreakBeforeClosingParen =
1172 }
1173
1174 if (CurrentState.AvoidBinPacking) {
1175 // If we are breaking after '(', '{', '<', or this is the break after a ':'
1176 // to start a member initializer list in a constructor, this should not
1177 // be considered bin packing unless the relevant AllowAll option is false or
1178 // this is a dict/object literal.
1179 bool PreviousIsBreakingCtorInitializerColon =
1180 PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1182 bool AllowAllConstructorInitializersOnNextLine =
1185 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
1186 PreviousIsBreakingCtorInitializerColon) ||
1188 State.Line->MustBeDeclaration) ||
1190 !State.Line->MustBeDeclaration) ||
1191 (!AllowAllConstructorInitializersOnNextLine &&
1192 PreviousIsBreakingCtorInitializerColon) ||
1193 Previous.is(TT_DictLiteral)) {
1194 CurrentState.BreakBeforeParameter = true;
1195 }
1196
1197 // If we are breaking after a ':' to start a member initializer list,
1198 // and we allow all arguments on the next line, we should not break
1199 // before the next parameter.
1200 if (PreviousIsBreakingCtorInitializerColon &&
1201 AllowAllConstructorInitializersOnNextLine) {
1202 CurrentState.BreakBeforeParameter = false;
1203 }
1204 }
1205
1206 return Penalty;
1207}
1208
1209unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
1210 if (!State.NextToken || !State.NextToken->Previous)
1211 return 0;
1212
1213 FormatToken &Current = *State.NextToken;
1214 const auto &CurrentState = State.Stack.back();
1215
1216 if (CurrentState.IsCSharpGenericTypeConstraint &&
1217 Current.isNot(TT_CSharpGenericTypeConstraint)) {
1218 return CurrentState.ColonPos + 2;
1219 }
1220
1221 const FormatToken &Previous = *Current.Previous;
1222 // If we are continuing an expression, we want to use the continuation indent.
1223 unsigned ContinuationIndent =
1224 std::max(CurrentState.LastSpace, CurrentState.Indent) +
1226 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1227 const FormatToken *NextNonComment = Previous.getNextNonComment();
1228 if (!NextNonComment)
1229 NextNonComment = &Current;
1230
1231 // Java specific bits.
1232 if (Style.Language == FormatStyle::LK_Java &&
1233 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) {
1234 return std::max(CurrentState.LastSpace,
1235 CurrentState.Indent + Style.ContinuationIndentWidth);
1236 }
1237
1238 // Indentation of the statement following a Verilog case label is taken care
1239 // of in moveStateToNextToken.
1240 if (Style.isVerilog() && PreviousNonComment &&
1241 Keywords.isVerilogEndOfLabel(*PreviousNonComment)) {
1242 return State.FirstIndent;
1243 }
1244
1246 State.Line->First->is(tok::kw_enum)) {
1247 return (Style.IndentWidth * State.Line->First->IndentLevel) +
1248 Style.IndentWidth;
1249 }
1250
1251 if ((NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) ||
1252 (Style.isVerilog() && Keywords.isVerilogBegin(*NextNonComment))) {
1253 if (Current.NestingLevel == 0 ||
1255 State.NextToken->is(TT_LambdaLBrace))) {
1256 return State.FirstIndent;
1257 }
1258 return CurrentState.Indent;
1259 }
1260 if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
1261 (Current.is(tok::greater) && (Style.isProto() || Style.isTableGen()))) &&
1262 State.Stack.size() > 1) {
1263 if (Current.closesBlockOrBlockTypeList(Style))
1264 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1265 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit))
1266 return State.Stack[State.Stack.size() - 2].LastSpace;
1267 return State.FirstIndent;
1268 }
1269 // Indent a closing parenthesis at the previous level if followed by a semi,
1270 // const, or opening brace. This allows indentations such as:
1271 // foo(
1272 // a,
1273 // );
1274 // int Foo::getter(
1275 // //
1276 // ) const {
1277 // return foo;
1278 // }
1279 // function foo(
1280 // a,
1281 // ) {
1282 // code(); //
1283 // }
1284 if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
1285 (!Current.Next ||
1286 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) {
1287 return State.Stack[State.Stack.size() - 2].LastSpace;
1288 }
1289 // When DAGArg closer exists top of line, it should be aligned in the similar
1290 // way as function call above.
1291 if (Style.isTableGen() && Current.is(TT_TableGenDAGArgCloser) &&
1292 State.Stack.size() > 1) {
1293 return State.Stack[State.Stack.size() - 2].LastSpace;
1294 }
1296 (Current.is(tok::r_paren) ||
1297 (Current.is(tok::r_brace) && Current.MatchingParen &&
1298 Current.MatchingParen->is(BK_BracedInit))) &&
1299 State.Stack.size() > 1) {
1300 return State.Stack[State.Stack.size() - 2].LastSpace;
1301 }
1302 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
1303 return State.Stack[State.Stack.size() - 2].LastSpace;
1304 // Field labels in a nested type should be aligned to the brace. For example
1305 // in ProtoBuf:
1306 // optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123,
1307 // bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}];
1308 // For Verilog, a quote following a brace is treated as an identifier. And
1309 // Both braces and colons get annotated as TT_DictLiteral. So we have to
1310 // check.
1311 if (Current.is(tok::identifier) && Current.Next &&
1312 (!Style.isVerilog() || Current.Next->is(tok::colon)) &&
1313 (Current.Next->is(TT_DictLiteral) ||
1314 (Style.isProto() && Current.Next->isOneOf(tok::less, tok::l_brace)))) {
1315 return CurrentState.Indent;
1316 }
1317 if (NextNonComment->is(TT_ObjCStringLiteral) &&
1318 State.StartOfStringLiteral != 0) {
1319 return State.StartOfStringLiteral - 1;
1320 }
1321 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1322 return State.StartOfStringLiteral;
1323 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0)
1324 return CurrentState.FirstLessLess;
1325 if (NextNonComment->isMemberAccess()) {
1326 if (CurrentState.CallContinuation == 0)
1327 return ContinuationIndent;
1328 return CurrentState.CallContinuation;
1329 }
1330 if (CurrentState.QuestionColumn != 0 &&
1331 ((NextNonComment->is(tok::colon) &&
1332 NextNonComment->is(TT_ConditionalExpr)) ||
1333 Previous.is(TT_ConditionalExpr))) {
1334 if (((NextNonComment->is(tok::colon) && NextNonComment->Next &&
1335 !NextNonComment->Next->FakeLParens.empty() &&
1336 NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1337 (Previous.is(tok::colon) && !Current.FakeLParens.empty() &&
1338 Current.FakeLParens.back() == prec::Conditional)) &&
1339 !CurrentState.IsWrappedConditional) {
1340 // NOTE: we may tweak this slightly:
1341 // * not remove the 'lead' ContinuationIndentWidth
1342 // * always un-indent by the operator when
1343 // BreakBeforeTernaryOperators=true
1344 unsigned Indent = CurrentState.Indent;
1346 Indent -= Style.ContinuationIndentWidth;
1347 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1348 Indent -= 2;
1349 return Indent;
1350 }
1351 return CurrentState.QuestionColumn;
1352 }
1353 if (Previous.is(tok::comma) && CurrentState.VariablePos != 0)
1354 return CurrentState.VariablePos;
1355 if (Current.is(TT_RequiresClause)) {
1356 if (Style.IndentRequiresClause)
1357 return CurrentState.Indent + Style.IndentWidth;
1358 switch (Style.RequiresClausePosition) {
1361 return CurrentState.Indent;
1362 default:
1363 break;
1364 }
1365 }
1366 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1367 TT_InheritanceComma)) {
1368 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1369 }
1370 if ((PreviousNonComment &&
1371 (PreviousNonComment->ClosesTemplateDeclaration ||
1372 PreviousNonComment->ClosesRequiresClause ||
1373 (PreviousNonComment->is(TT_AttributeMacro) &&
1374 Current.isNot(tok::l_paren)) ||
1375 PreviousNonComment->isOneOf(
1376 TT_AttributeRParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
1377 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
1379 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) {
1380 return std::max(CurrentState.LastSpace, CurrentState.Indent);
1381 }
1382 if (NextNonComment->is(TT_SelectorName)) {
1383 if (!CurrentState.ObjCSelectorNameFound) {
1384 unsigned MinIndent = CurrentState.Indent;
1385 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) {
1386 MinIndent = std::max(MinIndent,
1387 State.FirstIndent + Style.ContinuationIndentWidth);
1388 }
1389 // If LongestObjCSelectorName is 0, we are indenting the first
1390 // part of an ObjC selector (or a selector component which is
1391 // not colon-aligned due to block formatting).
1392 //
1393 // Otherwise, we are indenting a subsequent part of an ObjC
1394 // selector which should be colon-aligned to the longest
1395 // component of the ObjC selector.
1396 //
1397 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1398 return MinIndent +
1399 std::max(NextNonComment->LongestObjCSelectorName,
1400 NextNonComment->ColumnWidth) -
1401 NextNonComment->ColumnWidth;
1402 }
1403 if (!CurrentState.AlignColons)
1404 return CurrentState.Indent;
1405 if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1406 return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1407 return CurrentState.Indent;
1408 }
1409 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1410 return CurrentState.ColonPos;
1411 if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1412 if (CurrentState.StartOfArraySubscripts != 0) {
1413 return CurrentState.StartOfArraySubscripts;
1414 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1415 // initializers.
1416 return CurrentState.Indent;
1417 }
1418 return ContinuationIndent;
1419 }
1420
1421 // OpenMP clauses want to get additional indentation when they are pushed onto
1422 // the next line.
1423 if (State.Line->InPragmaDirective) {
1424 FormatToken *PragmaType = State.Line->First->Next->Next;
1425 if (PragmaType && PragmaType->TokenText.equals("omp"))
1426 return CurrentState.Indent + Style.ContinuationIndentWidth;
1427 }
1428
1429 // This ensure that we correctly format ObjC methods calls without inputs,
1430 // i.e. where the last element isn't selector like: [callee method];
1431 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1432 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) {
1433 return CurrentState.Indent;
1434 }
1435
1436 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1437 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) {
1438 return ContinuationIndent;
1439 }
1440 if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1441 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1442 return ContinuationIndent;
1443 }
1444 if (NextNonComment->is(TT_CtorInitializerComma))
1445 return CurrentState.Indent;
1446 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1448 return CurrentState.Indent;
1449 }
1450 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1452 return CurrentState.Indent;
1453 }
1454 if (Previous.is(tok::r_paren) &&
1455 Previous.isNot(TT_TableGenDAGArgOperatorToBreak) &&
1456 !Current.isBinaryOperator() &&
1457 !Current.isOneOf(tok::colon, tok::comment)) {
1458 return ContinuationIndent;
1459 }
1460 if (Current.is(TT_ProtoExtensionLSquare))
1461 return CurrentState.Indent;
1462 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1463 return CurrentState.Indent - Current.Tok.getLength() -
1464 Current.SpacesRequiredBefore;
1465 }
1466 if (Current.is(tok::comment) && NextNonComment->isBinaryOperator() &&
1467 CurrentState.UnindentOperator) {
1468 return CurrentState.Indent - NextNonComment->Tok.getLength() -
1469 NextNonComment->SpacesRequiredBefore;
1470 }
1471 if (CurrentState.Indent == State.FirstIndent && PreviousNonComment &&
1472 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) {
1473 // Ensure that we fall back to the continuation indent width instead of
1474 // just flushing continuations left.
1475 return CurrentState.Indent + Style.ContinuationIndentWidth;
1476 }
1477 return CurrentState.Indent;
1478}
1479
1481 const FormatToken &Current,
1482 const FormatStyle &Style) {
1483 if (Previous->isNot(tok::l_paren))
1484 return true;
1485 if (Previous->ParameterCount > 1)
1486 return true;
1487
1488 // Also a nested block if contains a lambda inside function with 1 parameter.
1489 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare);
1490}
1491
1492unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1493 bool DryRun, bool Newline) {
1494 assert(State.Stack.size());
1495 const FormatToken &Current = *State.NextToken;
1496 auto &CurrentState = State.Stack.back();
1497
1498 if (Current.is(TT_CSharpGenericTypeConstraint))
1499 CurrentState.IsCSharpGenericTypeConstraint = true;
1500 if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1501 CurrentState.NoLineBreakInOperand = false;
1502 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon))
1503 CurrentState.AvoidBinPacking = true;
1504 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1505 if (CurrentState.FirstLessLess == 0)
1506 CurrentState.FirstLessLess = State.Column;
1507 else
1508 CurrentState.LastOperatorWrapped = Newline;
1509 }
1510 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1511 CurrentState.LastOperatorWrapped = Newline;
1512 if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1513 Current.Previous->isNot(TT_ConditionalExpr)) {
1514 CurrentState.LastOperatorWrapped = Newline;
1515 }
1516 if (Current.is(TT_ArraySubscriptLSquare) &&
1517 CurrentState.StartOfArraySubscripts == 0) {
1518 CurrentState.StartOfArraySubscripts = State.Column;
1519 }
1520
1521 auto IsWrappedConditional = [](const FormatToken &Tok) {
1522 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question)))
1523 return false;
1524 if (Tok.MustBreakBefore)
1525 return true;
1526
1527 const FormatToken *Next = Tok.getNextNonComment();
1528 return Next && Next->MustBreakBefore;
1529 };
1530 if (IsWrappedConditional(Current))
1531 CurrentState.IsWrappedConditional = true;
1532 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1533 CurrentState.QuestionColumn = State.Column;
1534 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1535 const FormatToken *Previous = Current.Previous;
1536 while (Previous && Previous->isTrailingComment())
1537 Previous = Previous->Previous;
1538 if (Previous && Previous->is(tok::question))
1539 CurrentState.QuestionColumn = State.Column;
1540 }
1541 if (!Current.opensScope() && !Current.closesScope() &&
1542 Current.isNot(TT_PointerOrReference)) {
1543 State.LowestLevelOnLine =
1544 std::min(State.LowestLevelOnLine, Current.NestingLevel);
1545 }
1546 if (Current.isMemberAccess())
1547 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1548 if (Current.is(TT_SelectorName))
1549 CurrentState.ObjCSelectorNameFound = true;
1550 if (Current.is(TT_CtorInitializerColon) &&
1552 // Indent 2 from the column, so:
1553 // SomeClass::SomeClass()
1554 // : First(...), ...
1555 // Next(...)
1556 // ^ line up here.
1557 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1559 ? 0
1560 : 2);
1561 CurrentState.NestedBlockIndent = CurrentState.Indent;
1563 CurrentState.AvoidBinPacking = true;
1564 CurrentState.BreakBeforeParameter =
1565 Style.ColumnLimit > 0 &&
1568 } else {
1569 CurrentState.BreakBeforeParameter = false;
1570 }
1571 }
1572 if (Current.is(TT_CtorInitializerColon) &&
1574 CurrentState.Indent =
1575 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1576 CurrentState.NestedBlockIndent = CurrentState.Indent;
1578 CurrentState.AvoidBinPacking = true;
1579 else
1580 CurrentState.BreakBeforeParameter = false;
1581 }
1582 if (Current.is(TT_InheritanceColon)) {
1583 CurrentState.Indent =
1584 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1585 }
1586 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1587 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1588 if (Current.isOneOf(TT_LambdaLSquare, TT_TrailingReturnArrow))
1589 CurrentState.LastSpace = State.Column;
1590 if (Current.is(TT_RequiresExpression) &&
1592 CurrentState.NestedBlockIndent = State.Column;
1593 }
1594
1595 // Insert scopes created by fake parenthesis.
1596 const FormatToken *Previous = Current.getPreviousNonComment();
1597
1598 // Add special behavior to support a format commonly used for JavaScript
1599 // closures:
1600 // SomeFunction(function() {
1601 // foo();
1602 // bar();
1603 // }, a, b, c);
1604 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause &&
1605 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1606 Previous->isNot(TT_DictLiteral) && State.Stack.size() > 1 &&
1607 !CurrentState.HasMultipleNestedBlocks) {
1608 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1609 for (ParenState &PState : llvm::drop_end(State.Stack))
1610 PState.NoLineBreak = true;
1611 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1612 }
1613 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) ||
1614 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) &&
1615 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) {
1616 CurrentState.NestedBlockInlined =
1617 !Newline && hasNestedBlockInlined(Previous, Current, Style);
1618 }
1619
1620 moveStatePastFakeLParens(State, Newline);
1621 moveStatePastScopeCloser(State);
1622 // Do not use CurrentState here, since the two functions before may change the
1623 // Stack.
1624 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1625 !State.Stack.back().NoLineBreakInOperand;
1626 moveStatePastScopeOpener(State, Newline);
1627 moveStatePastFakeRParens(State);
1628
1629 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1630 State.StartOfStringLiteral = State.Column + 1;
1631 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1632 State.StartOfStringLiteral = State.Column + 1;
1633 } else if (Current.is(TT_TableGenMultiLineString) &&
1634 State.StartOfStringLiteral == 0) {
1635 State.StartOfStringLiteral = State.Column + 1;
1636 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1637 State.StartOfStringLiteral = State.Column;
1638 } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1639 !Current.isStringLiteral()) {
1640 State.StartOfStringLiteral = 0;
1641 }
1642
1643 State.Column += Current.ColumnWidth;
1644 State.NextToken = State.NextToken->Next;
1645 // Verilog case labels are on the same unwrapped lines as the statements that
1646 // follow. TokenAnnotator identifies them and sets MustBreakBefore.
1647 // Indentation is taken care of here. A case label can only have 1 statement
1648 // in Verilog, so we don't have to worry about lines that follow.
1649 if (Style.isVerilog() && State.NextToken &&
1650 State.NextToken->MustBreakBefore &&
1651 Keywords.isVerilogEndOfLabel(Current)) {
1652 State.FirstIndent += Style.IndentWidth;
1653 CurrentState.Indent = State.FirstIndent;
1654 }
1655
1656 unsigned Penalty =
1657 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1658
1659 if (Current.Role)
1660 Current.Role->formatFromToken(State, this, DryRun);
1661 // If the previous has a special role, let it consume tokens as appropriate.
1662 // It is necessary to start at the previous token for the only implemented
1663 // role (comma separated list). That way, the decision whether or not to break
1664 // after the "{" is already done and both options are tried and evaluated.
1665 // FIXME: This is ugly, find a better way.
1666 if (Previous && Previous->Role)
1667 Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1668
1669 return Penalty;
1670}
1671
1672void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1673 bool Newline) {
1674 const FormatToken &Current = *State.NextToken;
1675 if (Current.FakeLParens.empty())
1676 return;
1677
1678 const FormatToken *Previous = Current.getPreviousNonComment();
1679
1680 // Don't add extra indentation for the first fake parenthesis after
1681 // 'return', assignments, opening <({[, or requires clauses. The indentation
1682 // for these cases is special cased.
1683 bool SkipFirstExtraIndent =
1684 Previous &&
1685 (Previous->opensScope() ||
1686 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) ||
1687 (Previous->getPrecedence() == prec::Assignment &&
1689 Previous->is(TT_ObjCMethodExpr));
1690 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) {
1691 const auto &CurrentState = State.Stack.back();
1692 ParenState NewParenState = CurrentState;
1693 NewParenState.Tok = nullptr;
1694 NewParenState.ContainsLineBreak = false;
1695 NewParenState.LastOperatorWrapped = true;
1696 NewParenState.IsChainedConditional = false;
1697 NewParenState.IsWrappedConditional = false;
1698 NewParenState.UnindentOperator = false;
1699 NewParenState.NoLineBreak =
1700 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
1701
1702 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1703 if (PrecedenceLevel > prec::Comma)
1704 NewParenState.AvoidBinPacking = false;
1705
1706 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1707 // a builder type call after 'return' or, if the alignment after opening
1708 // brackets is disabled.
1709 if (!Current.isTrailingComment() &&
1711 PrecedenceLevel < prec::Assignment) &&
1712 (!Previous || Previous->isNot(tok::kw_return) ||
1713 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) &&
1715 PrecedenceLevel != prec::Comma || Current.NestingLevel == 0) &&
1716 (!Style.isTableGen() ||
1717 (Previous && Previous->isOneOf(TT_TableGenDAGArgListComma,
1718 TT_TableGenDAGArgListCommaToBreak)))) {
1719 NewParenState.Indent = std::max(
1720 std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace);
1721 }
1722
1723 // Special case for generic selection expressions, its comma-separated
1724 // expressions are not aligned to the opening paren like regular calls, but
1725 // rather continuation-indented relative to the _Generic keyword.
1726 if (Previous && Previous->endsSequence(tok::l_paren, tok::kw__Generic) &&
1727 State.Stack.size() > 1) {
1728 NewParenState.Indent = State.Stack[State.Stack.size() - 2].Indent +
1730 }
1731
1732 if ((shouldUnindentNextOperator(Current) ||
1733 (Previous &&
1734 (PrecedenceLevel == prec::Conditional &&
1735 Previous->is(tok::question) && Previous->is(TT_ConditionalExpr)))) &&
1736 !Newline) {
1737 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
1738 // the operator and keep the operands aligned.
1740 NewParenState.UnindentOperator = true;
1741 // Mark indentation as alignment if the expression is aligned.
1743 NewParenState.IsAligned = true;
1744 }
1745
1746 // Do not indent relative to the fake parentheses inserted for "." or "->".
1747 // This is a special case to make the following to statements consistent:
1748 // OuterFunction(InnerFunctionCall( // break
1749 // ParameterToInnerFunction));
1750 // OuterFunction(SomeObject.InnerFunctionCall( // break
1751 // ParameterToInnerFunction));
1752 if (PrecedenceLevel > prec::Unknown)
1753 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1754 if (PrecedenceLevel != prec::Conditional &&
1755 Current.isNot(TT_UnaryOperator) &&
1757 NewParenState.StartOfFunctionCall = State.Column;
1758 }
1759
1760 // Indent conditional expressions, unless they are chained "else-if"
1761 // conditionals. Never indent expression where the 'operator' is ',', ';' or
1762 // an assignment (i.e. *I <= prec::Assignment) as those have different
1763 // indentation rules. Indent other expression, unless the indentation needs
1764 // to be skipped.
1765 if (PrecedenceLevel == prec::Conditional && Previous &&
1766 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) &&
1767 &PrecedenceLevel == &Current.FakeLParens.back() &&
1768 !CurrentState.IsWrappedConditional) {
1769 NewParenState.IsChainedConditional = true;
1770 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
1771 } else if (PrecedenceLevel == prec::Conditional ||
1772 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
1773 !Current.isTrailingComment())) {
1774 NewParenState.Indent += Style.ContinuationIndentWidth;
1775 }
1776 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
1777 NewParenState.BreakBeforeParameter = false;
1778 State.Stack.push_back(NewParenState);
1779 SkipFirstExtraIndent = false;
1780 }
1781}
1782
1783void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1784 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1785 unsigned VariablePos = State.Stack.back().VariablePos;
1786 if (State.Stack.size() == 1) {
1787 // Do not pop the last element.
1788 break;
1789 }
1790 State.Stack.pop_back();
1791 State.Stack.back().VariablePos = VariablePos;
1792 }
1793
1794 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
1795 // Remove the indentation of the requires clauses (which is not in Indent,
1796 // but in LastSpace).
1797 State.Stack.back().LastSpace -= Style.IndentWidth;
1798 }
1799}
1800
1801void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1802 bool Newline) {
1803 const FormatToken &Current = *State.NextToken;
1804 if (!Current.opensScope())
1805 return;
1806
1807 const auto &CurrentState = State.Stack.back();
1808
1809 // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
1810 if (Current.isOneOf(tok::less, tok::l_paren) &&
1811 CurrentState.IsCSharpGenericTypeConstraint) {
1812 return;
1813 }
1814
1815 if (Current.MatchingParen && Current.is(BK_Block)) {
1816 moveStateToNewBlock(State, Newline);
1817 return;
1818 }
1819
1820 unsigned NewIndent;
1821 unsigned LastSpace = CurrentState.LastSpace;
1822 bool AvoidBinPacking;
1823 bool BreakBeforeParameter = false;
1824 unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall,
1825 CurrentState.NestedBlockIndent);
1826 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1827 opensProtoMessageField(Current, Style)) {
1828 if (Current.opensBlockOrBlockTypeList(Style)) {
1829 NewIndent = Style.IndentWidth +
1830 std::min(State.Column, CurrentState.NestedBlockIndent);
1831 } else if (Current.is(tok::l_brace)) {
1832 NewIndent =
1833 CurrentState.LastSpace + Style.BracedInitializerIndentWidth.value_or(
1835 } else {
1836 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
1837 }
1838 const FormatToken *NextNonComment = Current.getNextNonComment();
1839 bool EndsInComma = Current.MatchingParen &&
1840 Current.MatchingParen->Previous &&
1841 Current.MatchingParen->Previous->is(tok::comma);
1842 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1843 Style.isProto() || !Style.BinPackArguments ||
1844 (NextNonComment && NextNonComment->isOneOf(
1845 TT_DesignatedInitializerPeriod,
1846 TT_DesignatedInitializerLSquare));
1847 BreakBeforeParameter = EndsInComma;
1848 if (Current.ParameterCount > 1)
1849 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1850 } else {
1851 NewIndent =
1853 std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall);
1854
1855 if (Style.isTableGen() && Current.is(TT_TableGenDAGArgOpenerToBreak) &&
1857 // For the case the next token is a TableGen DAGArg operator identifier
1858 // that is not marked to have a line break after it.
1859 // In this case the option DAS_BreakElements requires to align the
1860 // DAGArg elements to the operator.
1861 const FormatToken *Next = Current.Next;
1862 if (Next && Next->is(TT_TableGenDAGArgOperatorID))
1863 NewIndent = State.Column + Next->TokenText.size() + 2;
1864 }
1865
1866 // Ensure that different different brackets force relative alignment, e.g.:
1867 // void SomeFunction(vector< // break
1868 // int> v);
1869 // FIXME: We likely want to do this for more combinations of brackets.
1870 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1871 NewIndent = std::max(NewIndent, CurrentState.Indent);
1872 LastSpace = std::max(LastSpace, CurrentState.Indent);
1873 }
1874
1875 bool EndsInComma =
1876 Current.MatchingParen &&
1877 Current.MatchingParen->getPreviousNonComment() &&
1878 Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1879
1880 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1881 // for backwards compatibility.
1882 bool ObjCBinPackProtocolList =
1884 Style.BinPackParameters) ||
1886
1887 bool BinPackDeclaration =
1888 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1889 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1890
1891 bool GenericSelection =
1892 Current.getPreviousNonComment() &&
1893 Current.getPreviousNonComment()->is(tok::kw__Generic);
1894
1895 AvoidBinPacking =
1896 (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
1897 (Style.isJavaScript() && EndsInComma) ||
1898 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1899 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1901 (Current.is(PPK_OnePerLine) ||
1902 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive))));
1903
1904 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen &&
1906 if (Style.ColumnLimit) {
1907 // If this '[' opens an ObjC call, determine whether all parameters fit
1908 // into one line and put one per line if they don't.
1909 if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1910 getColumnLimit(State)) {
1911 BreakBeforeParameter = true;
1912 }
1913 } else {
1914 // For ColumnLimit = 0, we have to figure out whether there is or has to
1915 // be a line break within this call.
1916 for (const FormatToken *Tok = &Current;
1917 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1918 if (Tok->MustBreakBefore ||
1919 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1920 BreakBeforeParameter = true;
1921 break;
1922 }
1923 }
1924 }
1925 }
1926
1927 if (Style.isJavaScript() && EndsInComma)
1928 BreakBeforeParameter = true;
1929 }
1930 // Generally inherit NoLineBreak from the current scope to nested scope.
1931 // However, don't do this for non-empty nested blocks, dict literals and
1932 // array literals as these follow different indentation rules.
1933 bool NoLineBreak =
1934 Current.Children.empty() &&
1935 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1936 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
1937 (Current.is(TT_TemplateOpener) &&
1938 CurrentState.ContainsUnwrappedBuilder));
1939 State.Stack.push_back(
1940 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1941 auto &NewState = State.Stack.back();
1942 NewState.NestedBlockIndent = NestedBlockIndent;
1943 NewState.BreakBeforeParameter = BreakBeforeParameter;
1944 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
1945
1946 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next &&
1947 Current.is(tok::l_paren)) {
1948 // Search for any parameter that is a lambda.
1949 FormatToken const *next = Current.Next;
1950 while (next) {
1951 if (next->is(TT_LambdaLSquare)) {
1952 NewState.HasMultipleNestedBlocks = true;
1953 break;
1954 }
1955 next = next->Next;
1956 }
1957 }
1958
1959 NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) &&
1960 Current.Previous &&
1961 Current.Previous->is(tok::at);
1962}
1963
1964void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1965 const FormatToken &Current = *State.NextToken;
1966 if (!Current.closesScope())
1967 return;
1968
1969 // If we encounter a closing ), ], } or >, we can remove a level from our
1970 // stacks.
1971 if (State.Stack.size() > 1 &&
1972 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1973 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1974 State.NextToken->is(TT_TemplateCloser) ||
1975 State.NextToken->is(TT_TableGenListCloser) ||
1976 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) {
1977 State.Stack.pop_back();
1978 }
1979
1980 auto &CurrentState = State.Stack.back();
1981
1982 // Reevaluate whether ObjC message arguments fit into one line.
1983 // If a receiver spans multiple lines, e.g.:
1984 // [[object block:^{
1985 // return 42;
1986 // }] a:42 b:42];
1987 // BreakBeforeParameter is calculated based on an incorrect assumption
1988 // (it is checked whether the whole expression fits into one line without
1989 // considering a line break inside a message receiver).
1990 // We check whether arguments fit after receiver scope closer (into the same
1991 // line).
1992 if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
1993 Current.MatchingParen->Previous) {
1994 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
1995 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) &&
1996 CurrentScopeOpener.MatchingParen) {
1997 int NecessarySpaceInLine =
1998 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) +
1999 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
2000 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
2001 Style.ColumnLimit) {
2002 CurrentState.BreakBeforeParameter = false;
2003 }
2004 }
2005 }
2006
2007 if (Current.is(tok::r_square)) {
2008 // If this ends the array subscript expr, reset the corresponding value.
2009 const FormatToken *NextNonComment = Current.getNextNonComment();
2010 if (NextNonComment && NextNonComment->isNot(tok::l_square))
2011 CurrentState.StartOfArraySubscripts = 0;
2012 }
2013}
2014
2015void ContinuationIndenter::moveStateToNewBlock(LineState &State, bool NewLine) {
2017 State.NextToken->is(TT_LambdaLBrace) &&
2018 !State.Line->MightBeFunctionDecl) {
2019 State.Stack.back().NestedBlockIndent = State.FirstIndent;
2020 }
2021 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
2022 // ObjC block sometimes follow special indentation rules.
2023 unsigned NewIndent =
2024 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
2025 ? Style.ObjCBlockIndentWidth
2026 : Style.IndentWidth);
2027
2028 // Even when wrapping before lambda body, the left brace can still be added to
2029 // the same line. This occurs when checking whether the whole lambda body can
2030 // go on a single line. In this case we have to make sure there are no line
2031 // breaks in the body, otherwise we could just end up with a regular lambda
2032 // body without the brace wrapped.
2033 bool NoLineBreak = Style.BraceWrapping.BeforeLambdaBody && !NewLine &&
2034 State.NextToken->is(TT_LambdaLBrace);
2035
2036 State.Stack.push_back(ParenState(State.NextToken, NewIndent,
2037 State.Stack.back().LastSpace,
2038 /*AvoidBinPacking=*/true, NoLineBreak));
2039 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
2040 State.Stack.back().BreakBeforeParameter = true;
2041}
2042
2043static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
2044 unsigned TabWidth,
2045 encoding::Encoding Encoding) {
2046 size_t LastNewlinePos = Text.find_last_of("\n");
2047 if (LastNewlinePos == StringRef::npos) {
2048 return StartColumn +
2049 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
2050 } else {
2051 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
2052 /*StartColumn=*/0, TabWidth, Encoding);
2053 }
2054}
2055
2056unsigned ContinuationIndenter::reformatRawStringLiteral(
2057 const FormatToken &Current, LineState &State,
2058 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
2059 unsigned StartColumn = State.Column - Current.ColumnWidth;
2060 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
2061 StringRef NewDelimiter =
2062 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
2063 if (NewDelimiter.empty())
2064 NewDelimiter = OldDelimiter;
2065 // The text of a raw string is between the leading 'R"delimiter(' and the
2066 // trailing 'delimiter)"'.
2067 unsigned OldPrefixSize = 3 + OldDelimiter.size();
2068 unsigned OldSuffixSize = 2 + OldDelimiter.size();
2069 // We create a virtual text environment which expects a null-terminated
2070 // string, so we cannot use StringRef.
2071 std::string RawText = std::string(
2072 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize));
2073 if (NewDelimiter != OldDelimiter) {
2074 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
2075 // raw string.
2076 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
2077 if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
2078 NewDelimiter = OldDelimiter;
2079 }
2080
2081 unsigned NewPrefixSize = 3 + NewDelimiter.size();
2082 unsigned NewSuffixSize = 2 + NewDelimiter.size();
2083
2084 // The first start column is the column the raw text starts after formatting.
2085 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
2086
2087 // The next start column is the intended indentation a line break inside
2088 // the raw string at level 0. It is determined by the following rules:
2089 // - if the content starts on newline, it is one level more than the current
2090 // indent, and
2091 // - if the content does not start on a newline, it is the first start
2092 // column.
2093 // These rules have the advantage that the formatted content both does not
2094 // violate the rectangle rule and visually flows within the surrounding
2095 // source.
2096 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
2097 // If this token is the last parameter (checked by looking if it's followed by
2098 // `)` and is not on a newline, the base the indent off the line's nested
2099 // block indent. Otherwise, base the indent off the arguments indent, so we
2100 // can achieve:
2101 //
2102 // fffffffffff(1, 2, 3, R"pb(
2103 // key1: 1 #
2104 // key2: 2)pb");
2105 //
2106 // fffffffffff(1, 2, 3,
2107 // R"pb(
2108 // key1: 1 #
2109 // key2: 2
2110 // )pb");
2111 //
2112 // fffffffffff(1, 2, 3,
2113 // R"pb(
2114 // key1: 1 #
2115 // key2: 2
2116 // )pb",
2117 // 5);
2118 unsigned CurrentIndent =
2119 (!Newline && Current.Next && Current.Next->is(tok::r_paren))
2120 ? State.Stack.back().NestedBlockIndent
2121 : State.Stack.back().Indent;
2122 unsigned NextStartColumn = ContentStartsOnNewline
2123 ? CurrentIndent + Style.IndentWidth
2124 : FirstStartColumn;
2125
2126 // The last start column is the column the raw string suffix starts if it is
2127 // put on a newline.
2128 // The last start column is the intended indentation of the raw string postfix
2129 // if it is put on a newline. It is determined by the following rules:
2130 // - if the raw string prefix starts on a newline, it is the column where
2131 // that raw string prefix starts, and
2132 // - if the raw string prefix does not start on a newline, it is the current
2133 // indent.
2134 unsigned LastStartColumn =
2135 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
2136
2137 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
2138 RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
2139 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
2140 /*Status=*/nullptr);
2141
2142 auto NewCode = applyAllReplacements(RawText, Fixes.first);
2143 tooling::Replacements NoFixes;
2144 if (!NewCode)
2145 return addMultilineToken(Current, State);
2146 if (!DryRun) {
2147 if (NewDelimiter != OldDelimiter) {
2148 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
2149 // of the token.
2150 SourceLocation PrefixDelimiterStart =
2151 Current.Tok.getLocation().getLocWithOffset(2);
2152 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
2153 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2154 if (PrefixErr) {
2155 llvm::errs()
2156 << "Failed to update the prefix delimiter of a raw string: "
2157 << llvm::toString(std::move(PrefixErr)) << "\n";
2158 }
2159 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
2160 // position length - 1 - |delimiter|.
2161 SourceLocation SuffixDelimiterStart =
2162 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
2163 1 - OldDelimiter.size());
2164 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
2165 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
2166 if (SuffixErr) {
2167 llvm::errs()
2168 << "Failed to update the suffix delimiter of a raw string: "
2169 << llvm::toString(std::move(SuffixErr)) << "\n";
2170 }
2171 }
2172 SourceLocation OriginLoc =
2173 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
2174 for (const tooling::Replacement &Fix : Fixes.first) {
2175 auto Err = Whitespaces.addReplacement(tooling::Replacement(
2176 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
2177 Fix.getLength(), Fix.getReplacementText()));
2178 if (Err) {
2179 llvm::errs() << "Failed to reformat raw string: "
2180 << llvm::toString(std::move(Err)) << "\n";
2181 }
2182 }
2183 }
2184 unsigned RawLastLineEndColumn = getLastLineEndColumn(
2185 *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
2186 State.Column = RawLastLineEndColumn + NewSuffixSize;
2187 // Since we're updating the column to after the raw string literal here, we
2188 // have to manually add the penalty for the prefix R"delim( over the column
2189 // limit.
2190 unsigned PrefixExcessCharacters =
2191 StartColumn + NewPrefixSize > Style.ColumnLimit
2192 ? StartColumn + NewPrefixSize - Style.ColumnLimit
2193 : 0;
2194 bool IsMultiline =
2195 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos);
2196 if (IsMultiline) {
2197 // Break before further function parameters on all levels.
2198 for (ParenState &Paren : State.Stack)
2199 Paren.BreakBeforeParameter = true;
2200 }
2201 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
2202}
2203
2204unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
2205 LineState &State) {
2206 // Break before further function parameters on all levels.
2207 for (ParenState &Paren : State.Stack)
2208 Paren.BreakBeforeParameter = true;
2209
2210 unsigned ColumnsUsed = State.Column;
2211 // We can only affect layout of the first and the last line, so the penalty
2212 // for all other lines is constant, and we ignore it.
2213 State.Column = Current.LastLineColumnWidth;
2214
2215 if (ColumnsUsed > getColumnLimit(State))
2216 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2217 return 0;
2218}
2219
2220unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2221 LineState &State, bool DryRun,
2222 bool AllowBreak, bool Newline) {
2223 unsigned Penalty = 0;
2224 // Compute the raw string style to use in case this is a raw string literal
2225 // that can be reformatted.
2226 auto RawStringStyle = getRawStringStyle(Current, State);
2227 if (RawStringStyle && !Current.Finalized) {
2228 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun,
2229 Newline);
2230 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
2231 // Don't break multi-line tokens other than block comments and raw string
2232 // literals. Instead, just update the state.
2233 Penalty = addMultilineToken(Current, State);
2234 } else if (State.Line->Type != LT_ImportStatement) {
2235 // We generally don't break import statements.
2236 LineState OriginalState = State;
2237
2238 // Whether we force the reflowing algorithm to stay strictly within the
2239 // column limit.
2240 bool Strict = false;
2241 // Whether the first non-strict attempt at reflowing did intentionally
2242 // exceed the column limit.
2243 bool Exceeded = false;
2244 std::tie(Penalty, Exceeded) = breakProtrudingToken(
2245 Current, State, AllowBreak, /*DryRun=*/true, Strict);
2246 if (Exceeded) {
2247 // If non-strict reflowing exceeds the column limit, try whether strict
2248 // reflowing leads to an overall lower penalty.
2249 LineState StrictState = OriginalState;
2250 unsigned StrictPenalty =
2251 breakProtrudingToken(Current, StrictState, AllowBreak,
2252 /*DryRun=*/true, /*Strict=*/true)
2253 .first;
2254 Strict = StrictPenalty <= Penalty;
2255 if (Strict) {
2256 Penalty = StrictPenalty;
2257 State = StrictState;
2258 }
2259 }
2260 if (!DryRun) {
2261 // If we're not in dry-run mode, apply the changes with the decision on
2262 // strictness made above.
2263 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
2264 Strict);
2265 }
2266 }
2267 if (State.Column > getColumnLimit(State)) {
2268 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2269 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2270 }
2271 return Penalty;
2272}
2273
2274// Returns the enclosing function name of a token, or the empty string if not
2275// found.
2276static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2277 // Look for: 'function(' or 'function<templates>(' before Current.
2278 auto Tok = Current.getPreviousNonComment();
2279 if (!Tok || Tok->isNot(tok::l_paren))
2280 return "";
2281 Tok = Tok->getPreviousNonComment();
2282 if (!Tok)
2283 return "";
2284 if (Tok->is(TT_TemplateCloser)) {
2285 Tok = Tok->MatchingParen;
2286 if (Tok)
2287 Tok = Tok->getPreviousNonComment();
2288 }
2289 if (!Tok || Tok->isNot(tok::identifier))
2290 return "";
2291 return Tok->TokenText;
2292}
2293
2294std::optional<FormatStyle>
2295ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2296 const LineState &State) {
2297 if (!Current.isStringLiteral())
2298 return std::nullopt;
2299 auto Delimiter = getRawStringDelimiter(Current.TokenText);
2300 if (!Delimiter)
2301 return std::nullopt;
2302 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
2303 if (!RawStringStyle && Delimiter->empty()) {
2304 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2305 getEnclosingFunctionName(Current));
2306 }
2307 if (!RawStringStyle)
2308 return std::nullopt;
2309 RawStringStyle->ColumnLimit = getColumnLimit(State);
2310 return RawStringStyle;
2311}
2312
2313std::unique_ptr<BreakableToken>
2314ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2315 LineState &State, bool AllowBreak) {
2316 unsigned StartColumn = State.Column - Current.ColumnWidth;
2317 if (Current.isStringLiteral()) {
2318 // Strings in JSON cannot be broken. Breaking strings in JavaScript is
2319 // disabled for now.
2320 if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals ||
2321 !AllowBreak) {
2322 return nullptr;
2323 }
2324
2325 // Don't break string literals inside preprocessor directives (except for
2326 // #define directives, as their contents are stored in separate lines and
2327 // are not affected by this check).
2328 // This way we avoid breaking code with line directives and unknown
2329 // preprocessor directives that contain long string literals.
2330 if (State.Line->Type == LT_PreprocessorDirective)
2331 return nullptr;
2332 // Exempts unterminated string literals from line breaking. The user will
2333 // likely want to terminate the string before any line breaking is done.
2334 if (Current.IsUnterminatedLiteral)
2335 return nullptr;
2336 // Don't break string literals inside Objective-C array literals (doing so
2337 // raises the warning -Wobjc-string-concatenation).
2338 if (State.Stack.back().IsInsideObjCArrayLiteral)
2339 return nullptr;
2340
2341 // The "DPI"/"DPI-C" in SystemVerilog direct programming interface
2342 // imports/exports cannot be split, e.g.
2343 // `import "DPI" function foo();`
2344 // FIXME: make this use same infra as C++ import checks
2345 if (Style.isVerilog() && Current.Previous &&
2346 Current.Previous->isOneOf(tok::kw_export, Keywords.kw_import)) {
2347 return nullptr;
2348 }
2349 StringRef Text = Current.TokenText;
2350
2351 // We need this to address the case where there is an unbreakable tail only
2352 // if certain other formatting decisions have been taken. The
2353 // UnbreakableTailLength of Current is an overapproximation in that case and
2354 // we need to be correct here.
2355 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2356 ? 0
2357 : Current.UnbreakableTailLength;
2358
2359 if (Style.isVerilog() || Style.Language == FormatStyle::LK_Java ||
2360 Style.isJavaScript() || Style.isCSharp()) {
2362 if (Style.isJavaScript() && Text.starts_with("'") &&
2363 Text.ends_with("'")) {
2365 } else if (Style.isCSharp() && Text.starts_with("@\"") &&
2366 Text.ends_with("\"")) {
2368 } else if (Text.starts_with("\"") && Text.ends_with("\"")) {
2370 } else {
2371 return nullptr;
2372 }
2373 return std::make_unique<BreakableStringLiteralUsingOperators>(
2374 Current, QuoteStyle,
2375 /*UnindentPlus=*/shouldUnindentNextOperator(Current), StartColumn,
2376 UnbreakableTailLength, State.Line->InPPDirective, Encoding, Style);
2377 }
2378
2379 StringRef Prefix;
2380 StringRef Postfix;
2381 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2382 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2383 // reduce the overhead) for each FormatToken, which is a string, so that we
2384 // don't run multiple checks here on the hot path.
2385 if ((Text.ends_with(Postfix = "\"") &&
2386 (Text.starts_with(Prefix = "@\"") || Text.starts_with(Prefix = "\"") ||
2387 Text.starts_with(Prefix = "u\"") ||
2388 Text.starts_with(Prefix = "U\"") ||
2389 Text.starts_with(Prefix = "u8\"") ||
2390 Text.starts_with(Prefix = "L\""))) ||
2391 (Text.starts_with(Prefix = "_T(\"") &&
2392 Text.ends_with(Postfix = "\")"))) {
2393 return std::make_unique<BreakableStringLiteral>(
2394 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
2395 State.Line->InPPDirective, Encoding, Style);
2396 }
2397 } else if (Current.is(TT_BlockComment)) {
2398 if (!Style.ReflowComments ||
2399 // If a comment token switches formatting, like
2400 // /* clang-format on */, we don't want to break it further,
2401 // but we may still want to adjust its indentation.
2402 switchesFormatting(Current)) {
2403 return nullptr;
2404 }
2405 return std::make_unique<BreakableBlockComment>(
2406 Current, StartColumn, Current.OriginalColumn, !Current.Previous,
2407 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF());
2408 } else if (Current.is(TT_LineComment) &&
2409 (!Current.Previous ||
2410 Current.Previous->isNot(TT_ImplicitStringLiteral))) {
2411 bool RegularComments = [&]() {
2412 for (const FormatToken *T = &Current; T && T->is(TT_LineComment);
2413 T = T->Next) {
2414 if (!(T->TokenText.starts_with("//") || T->TokenText.starts_with("#")))
2415 return false;
2416 }
2417 return true;
2418 }();
2419 if (!Style.ReflowComments ||
2420 CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
2421 switchesFormatting(Current) || !RegularComments) {
2422 return nullptr;
2423 }
2424 return std::make_unique<BreakableLineCommentSection>(
2425 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style);
2426 }
2427 return nullptr;
2428}
2429
2430std::pair<unsigned, bool>
2431ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2432 LineState &State, bool AllowBreak,
2433 bool DryRun, bool Strict) {
2434 std::unique_ptr<const BreakableToken> Token =
2435 createBreakableToken(Current, State, AllowBreak);
2436 if (!Token)
2437 return {0, false};
2438 assert(Token->getLineCount() > 0);
2439 unsigned ColumnLimit = getColumnLimit(State);
2440 if (Current.is(TT_LineComment)) {
2441 // We don't insert backslashes when breaking line comments.
2442 ColumnLimit = Style.ColumnLimit;
2443 }
2444 if (ColumnLimit == 0) {
2445 // To make the rest of the function easier set the column limit to the
2446 // maximum, if there should be no limit.
2447 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2448 }
2449 if (Current.UnbreakableTailLength >= ColumnLimit)
2450 return {0, false};
2451 // ColumnWidth was already accounted into State.Column before calling
2452 // breakProtrudingToken.
2453 unsigned StartColumn = State.Column - Current.ColumnWidth;
2454 unsigned NewBreakPenalty = Current.isStringLiteral()
2455 ? Style.PenaltyBreakString
2456 : Style.PenaltyBreakComment;
2457 // Stores whether we intentionally decide to let a line exceed the column
2458 // limit.
2459 bool Exceeded = false;
2460 // Stores whether we introduce a break anywhere in the token.
2461 bool BreakInserted = Token->introducesBreakBeforeToken();
2462 // Store whether we inserted a new line break at the end of the previous
2463 // logical line.
2464 bool NewBreakBefore = false;
2465 // We use a conservative reflowing strategy. Reflow starts after a line is
2466 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2467 // line that doesn't get reflown with the previous line is reached.
2468 bool Reflow = false;
2469 // Keep track of where we are in the token:
2470 // Where we are in the content of the current logical line.
2471 unsigned TailOffset = 0;
2472 // The column number we're currently at.
2473 unsigned ContentStartColumn =
2474 Token->getContentStartColumn(0, /*Break=*/false);
2475 // The number of columns left in the current logical line after TailOffset.
2476 unsigned RemainingTokenColumns =
2477 Token->getRemainingLength(0, TailOffset, ContentStartColumn);
2478 // Adapt the start of the token, for example indent.
2479 if (!DryRun)
2480 Token->adaptStartOfLine(0, Whitespaces);
2481
2482 unsigned ContentIndent = 0;
2483 unsigned Penalty = 0;
2484 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2485 << StartColumn << ".\n");
2486 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2487 LineIndex != EndIndex; ++LineIndex) {
2488 LLVM_DEBUG(llvm::dbgs()
2489 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2490 NewBreakBefore = false;
2491 // If we did reflow the previous line, we'll try reflowing again. Otherwise
2492 // we'll start reflowing if the current line is broken or whitespace is
2493 // compressed.
2494 bool TryReflow = Reflow;
2495 // Break the current token until we can fit the rest of the line.
2496 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2497 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
2498 << (ContentStartColumn + RemainingTokenColumns)
2499 << ", space: " << ColumnLimit
2500 << ", reflown prefix: " << ContentStartColumn
2501 << ", offset in line: " << TailOffset << "\n");
2502 // If the current token doesn't fit, find the latest possible split in the
2503 // current line so that breaking at it will be under the column limit.
2504 // FIXME: Use the earliest possible split while reflowing to correctly
2505 // compress whitespace within a line.
2507 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2508 ContentStartColumn, CommentPragmasRegex);
2509 if (Split.first == StringRef::npos) {
2510 // No break opportunity - update the penalty and continue with the next
2511 // logical line.
2512 if (LineIndex < EndIndex - 1) {
2513 // The last line's penalty is handled in addNextStateToQueue() or when
2514 // calling replaceWhitespaceAfterLastLine below.
2515 Penalty += Style.PenaltyExcessCharacter *
2516 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2517 }
2518 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
2519 break;
2520 }
2521 assert(Split.first != 0);
2522
2523 if (Token->supportsReflow()) {
2524 // Check whether the next natural split point after the current one can
2525 // still fit the line, either because we can compress away whitespace,
2526 // or because the penalty the excess characters introduce is lower than
2527 // the break penalty.
2528 // We only do this for tokens that support reflowing, and thus allow us
2529 // to change the whitespace arbitrarily (e.g. comments).
2530 // Other tokens, like string literals, can be broken on arbitrary
2531 // positions.
2532
2533 // First, compute the columns from TailOffset to the next possible split
2534 // position.
2535 // For example:
2536 // ColumnLimit: |
2537 // // Some text that breaks
2538 // ^ tail offset
2539 // ^-- split
2540 // ^-------- to split columns
2541 // ^--- next split
2542 // ^--------------- to next split columns
2543 unsigned ToSplitColumns = Token->getRangeLength(
2544 LineIndex, TailOffset, Split.first, ContentStartColumn);
2545 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
2546
2547 BreakableToken::Split NextSplit = Token->getSplit(
2548 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
2549 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2550 // Compute the columns necessary to fit the next non-breakable sequence
2551 // into the current line.
2552 unsigned ToNextSplitColumns = 0;
2553 if (NextSplit.first == StringRef::npos) {
2554 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
2555 ContentStartColumn);
2556 } else {
2557 ToNextSplitColumns = Token->getRangeLength(
2558 LineIndex, TailOffset,
2559 Split.first + Split.second + NextSplit.first, ContentStartColumn);
2560 }
2561 // Compress the whitespace between the break and the start of the next
2562 // unbreakable sequence.
2563 ToNextSplitColumns =
2564 Token->getLengthAfterCompression(ToNextSplitColumns, Split);
2565 LLVM_DEBUG(llvm::dbgs()
2566 << " ContentStartColumn: " << ContentStartColumn << "\n");
2567 LLVM_DEBUG(llvm::dbgs()
2568 << " ToNextSplit: " << ToNextSplitColumns << "\n");
2569 // If the whitespace compression makes us fit, continue on the current
2570 // line.
2571 bool ContinueOnLine =
2572 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2573 unsigned ExcessCharactersPenalty = 0;
2574 if (!ContinueOnLine && !Strict) {
2575 // Similarly, if the excess characters' penalty is lower than the
2576 // penalty of introducing a new break, continue on the current line.
2577 ExcessCharactersPenalty =
2578 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2580 LLVM_DEBUG(llvm::dbgs()
2581 << " Penalty excess: " << ExcessCharactersPenalty
2582 << "\n break : " << NewBreakPenalty << "\n");
2583 if (ExcessCharactersPenalty < NewBreakPenalty) {
2584 Exceeded = true;
2585 ContinueOnLine = true;
2586 }
2587 }
2588 if (ContinueOnLine) {
2589 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
2590 // The current line fits after compressing the whitespace - reflow
2591 // the next line into it if possible.
2592 TryReflow = true;
2593 if (!DryRun) {
2594 Token->compressWhitespace(LineIndex, TailOffset, Split,
2595 Whitespaces);
2596 }
2597 // When we continue on the same line, leave one space between content.
2598 ContentStartColumn += ToSplitColumns + 1;
2599 Penalty += ExcessCharactersPenalty;
2600 TailOffset += Split.first + Split.second;
2601 RemainingTokenColumns = Token->getRemainingLength(
2602 LineIndex, TailOffset, ContentStartColumn);
2603 continue;
2604 }
2605 }
2606 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
2607 // Update the ContentIndent only if the current line was not reflown with
2608 // the previous line, since in that case the previous line should still
2609 // determine the ContentIndent. Also never intent the last line.
2610 if (!Reflow)
2611 ContentIndent = Token->getContentIndent(LineIndex);
2612 LLVM_DEBUG(llvm::dbgs()
2613 << " ContentIndent: " << ContentIndent << "\n");
2614 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2615 LineIndex, /*Break=*/true);
2616
2617 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2618 LineIndex, TailOffset + Split.first + Split.second,
2619 ContentStartColumn);
2620 if (NewRemainingTokenColumns == 0) {
2621 // No content to indent.
2622 ContentIndent = 0;
2623 ContentStartColumn =
2624 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2625 NewRemainingTokenColumns = Token->getRemainingLength(
2626 LineIndex, TailOffset + Split.first + Split.second,
2627 ContentStartColumn);
2628 }
2629
2630 // When breaking before a tab character, it may be moved by a few columns,
2631 // but will still be expanded to the next tab stop, so we don't save any
2632 // columns.
2633 if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2634 // FIXME: Do we need to adjust the penalty?
2635 break;
2636 }
2637
2638 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2639 << ", " << Split.second << "\n");
2640 if (!DryRun) {
2641 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2642 Whitespaces);
2643 }
2644
2645 Penalty += NewBreakPenalty;
2646 TailOffset += Split.first + Split.second;
2647 RemainingTokenColumns = NewRemainingTokenColumns;
2648 BreakInserted = true;
2649 NewBreakBefore = true;
2650 }
2651 // In case there's another line, prepare the state for the start of the next
2652 // line.
2653 if (LineIndex + 1 != EndIndex) {
2654 unsigned NextLineIndex = LineIndex + 1;
2655 if (NewBreakBefore) {
2656 // After breaking a line, try to reflow the next line into the current
2657 // one once RemainingTokenColumns fits.
2658 TryReflow = true;
2659 }
2660 if (TryReflow) {
2661 // We decided that we want to try reflowing the next line into the
2662 // current one.
2663 // We will now adjust the state as if the reflow is successful (in
2664 // preparation for the next line), and see whether that works. If we
2665 // decide that we cannot reflow, we will later reset the state to the
2666 // start of the next line.
2667 Reflow = false;
2668 // As we did not continue breaking the line, RemainingTokenColumns is
2669 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2670 // the position at which we want to format the next line if we do
2671 // actually reflow.
2672 // When we reflow, we need to add a space between the end of the current
2673 // line and the next line's start column.
2674 ContentStartColumn += RemainingTokenColumns + 1;
2675 // Get the split that we need to reflow next logical line into the end
2676 // of the current one; the split will include any leading whitespace of
2677 // the next logical line.
2678 BreakableToken::Split SplitBeforeNext =
2679 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
2680 LLVM_DEBUG(llvm::dbgs()
2681 << " Size of reflown text: " << ContentStartColumn
2682 << "\n Potential reflow split: ");
2683 if (SplitBeforeNext.first != StringRef::npos) {
2684 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2685 << SplitBeforeNext.second << "\n");
2686 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2687 // If the rest of the next line fits into the current line below the
2688 // column limit, we can safely reflow.
2689 RemainingTokenColumns = Token->getRemainingLength(
2690 NextLineIndex, TailOffset, ContentStartColumn);
2691 Reflow = true;
2692 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2693 LLVM_DEBUG(llvm::dbgs()
2694 << " Over limit after reflow, need: "
2695 << (ContentStartColumn + RemainingTokenColumns)
2696 << ", space: " << ColumnLimit
2697 << ", reflown prefix: " << ContentStartColumn
2698 << ", offset in line: " << TailOffset << "\n");
2699 // If the whole next line does not fit, try to find a point in
2700 // the next line at which we can break so that attaching the part
2701 // of the next line to that break point onto the current line is
2702 // below the column limit.
2704 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
2705 ContentStartColumn, CommentPragmasRegex);
2706 if (Split.first == StringRef::npos) {
2707 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
2708 Reflow = false;
2709 } else {
2710 // Check whether the first split point gets us below the column
2711 // limit. Note that we will execute this split below as part of
2712 // the normal token breaking and reflow logic within the line.
2713 unsigned ToSplitColumns = Token->getRangeLength(
2714 NextLineIndex, TailOffset, Split.first, ContentStartColumn);
2715 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2716 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
2717 << (ContentStartColumn + ToSplitColumns)
2718 << ", space: " << ColumnLimit);
2719 unsigned ExcessCharactersPenalty =
2720 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2722 if (NewBreakPenalty < ExcessCharactersPenalty)
2723 Reflow = false;
2724 }
2725 }
2726 }
2727 } else {
2728 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2729 }
2730 }
2731 if (!Reflow) {
2732 // If we didn't reflow into the next line, the only space to consider is
2733 // the next logical line. Reset our state to match the start of the next
2734 // line.
2735 TailOffset = 0;
2736 ContentStartColumn =
2737 Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2738 RemainingTokenColumns = Token->getRemainingLength(
2739 NextLineIndex, TailOffset, ContentStartColumn);
2740 // Adapt the start of the token, for example indent.
2741 if (!DryRun)
2742 Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2743 } else {
2744 // If we found a reflow split and have added a new break before the next
2745 // line, we are going to remove the line break at the start of the next
2746 // logical line. For example, here we'll add a new line break after
2747 // 'text', and subsequently delete the line break between 'that' and
2748 // 'reflows'.
2749 // // some text that
2750 // // reflows
2751 // ->
2752 // // some text
2753 // // that reflows
2754 // When adding the line break, we also added the penalty for it, so we
2755 // need to subtract that penalty again when we remove the line break due
2756 // to reflowing.
2757 if (NewBreakBefore) {
2758 assert(Penalty >= NewBreakPenalty);
2759 Penalty -= NewBreakPenalty;
2760 }
2761 if (!DryRun)
2762 Token->reflow(NextLineIndex, Whitespaces);
2763 }
2764 }
2765 }
2766
2767 BreakableToken::Split SplitAfterLastLine =
2768 Token->getSplitAfterLastLine(TailOffset);
2769 if (SplitAfterLastLine.first != StringRef::npos) {
2770 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2771
2772 // We add the last line's penalty here, since that line is going to be split
2773 // now.
2774 Penalty += Style.PenaltyExcessCharacter *
2775 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2776
2777 if (!DryRun) {
2778 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2779 Whitespaces);
2780 }
2781 ContentStartColumn =
2782 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2783 RemainingTokenColumns = Token->getRemainingLength(
2784 Token->getLineCount() - 1,
2785 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2786 ContentStartColumn);
2787 }
2788
2789 State.Column = ContentStartColumn + RemainingTokenColumns -
2790 Current.UnbreakableTailLength;
2791
2792 if (BreakInserted) {
2793 if (!DryRun)
2794 Token->updateAfterBroken(Whitespaces);
2795
2796 // If we break the token inside a parameter list, we need to break before
2797 // the next parameter on all levels, so that the next parameter is clearly
2798 // visible. Line comments already introduce a break.
2799 if (Current.isNot(TT_LineComment))
2800 for (ParenState &Paren : State.Stack)
2801 Paren.BreakBeforeParameter = true;
2802
2803 if (Current.is(TT_BlockComment))
2804 State.NoContinuation = true;
2805
2806 State.Stack.back().LastSpace = StartColumn;
2807 }
2808
2809 Token->updateNextToken(State);
2810
2811 return {Penalty, Exceeded};
2812}
2813
2815 // In preprocessor directives reserve two chars for trailing " \".
2816 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2817}
2818
2819bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2820 const FormatToken &Current = *State.NextToken;
2821 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2822 return false;
2823 // We never consider raw string literals "multiline" for the purpose of
2824 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2825 // (see TokenAnnotator::mustBreakBefore().
2826 if (Current.TokenText.starts_with("R\""))
2827 return false;
2828 if (Current.IsMultiline)
2829 return true;
2830 if (Current.getNextNonComment() &&
2831 Current.getNextNonComment()->isStringLiteral()) {
2832 return true; // Implicit concatenation.
2833 }
2834 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2835 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2836 Style.ColumnLimit) {
2837 return true; // String will be split.
2838 }
2839 return false;
2840}
2841
2842} // namespace format
2843} // namespace clang
StringRef P
Declares BreakableToken, BreakableStringLiteral, BreakableComment, BreakableBlockComment and Breakabl...
This file implements an indenter that manages the indentation of continuations.
This file declares Format APIs to be used internally by the formatting library implementation.
This file contains the declaration of the FormatToken, a wrapper around Token with additional informa...
StringRef Text
Definition: Format.cpp:2972
Various functions to configurably format source code.
Defines and computes precedence levels for binary/ternary operators.
Defines the SourceManager interface.
unsigned LongestObjCSelectorName
Defines the clang::TokenKind enum and support functions.
StateNode * Previous
WhitespaceManager class manages whitespace around tokens and their replacements.
__DEVICE__ int max(int __a, int __b)
This class handles loading and caching of source files into memory.
unsigned getSpellingColumnNumber(SourceLocation Loc, bool *Invalid=nullptr) const
std::pair< StringRef::size_type, unsigned > Split
Contains starting character index and length of split.
bool canBreak(const LineState &State)
Returns true, if a line break after State is allowed.
unsigned addTokenToState(LineState &State, bool Newline, bool DryRun, unsigned ExtraSpaces=0)
Appends the next token to State and updates information necessary for indentation.
unsigned getColumnLimit(const LineState &State) const
Get the column limit for this line.
LineState getInitialState(unsigned FirstIndent, unsigned FirstStartColumn, const AnnotatedLine *Line, bool DryRun)
Get the initial state, i.e.
ContinuationIndenter(const FormatStyle &Style, const AdditionalKeywords &Keywords, const SourceManager &SourceMgr, WhitespaceManager &Whitespaces, encoding::Encoding Encoding, bool BinPackInconclusiveFunctions)
Constructs a ContinuationIndenter to format Line starting in column FirstIndent.
bool mustBreak(const LineState &State)
Returns true, if a line break after State is mandatory.
Manages the whitespaces around tokens and their replacements.
llvm::Error addReplacement(const tooling::Replacement &Replacement)
void replaceWhitespace(FormatToken &Tok, unsigned Newlines, unsigned Spaces, unsigned StartOfTokenColumn, bool IsAligned=false, bool InPPDirective=false)
Replaces the whitespace in front of Tok.
unsigned columnWidthWithTabs(StringRef Text, unsigned StartColumn, unsigned TabWidth, Encoding Encoding)
Returns the number of columns required to display the Text, starting from the StartColumn on a termin...
Definition: Encoding.h:60
std::pair< tooling::Replacements, unsigned > reformat(const FormatStyle &Style, StringRef Code, ArrayRef< tooling::Range > Ranges, unsigned FirstStartColumn, unsigned NextStartColumn, unsigned LastStartColumn, StringRef FileName, FormattingAttemptStatus *Status)
Reformats the given Ranges in the code fragment Code.
Definition: Format.cpp:3598
static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, unsigned TabWidth, encoding::Encoding Encoding)
static bool shouldUnindentNextOperator(const FormatToken &Tok)
FormatStyle getLLVMStyle(FormatStyle::LanguageKind Language=FormatStyle::LanguageKind::LK_Cpp)
Returns a format style complying with the LLVM coding standards: http://llvm.org/docs/CodingStandards...
Definition: Format.cpp:1398
bool switchesFormatting(const FormatToken &Token)
Checks if Token switches formatting, like /* clang-format off *‍/.
static bool hasNestedBlockInlined(const FormatToken *Previous, const FormatToken &Current, const FormatStyle &Style)
static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok)
static unsigned getLengthToNextOperator(const FormatToken &Tok)
static bool startsNextParameter(const FormatToken &Current, const FormatStyle &Style)
static unsigned getLengthToMatchingParen(const FormatToken &Tok, ArrayRef< ParenState > Stack)
static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, LineType LineType)
bool getPredefinedStyle(StringRef Name, FormatStyle::LanguageKind Language, FormatStyle *Style)
Gets a predefined style for the specified language by name.
Definition: Format.cpp:1931
static std::optional< StringRef > getRawStringDelimiter(StringRef TokenText)
static StringRef getCanonicalRawStringDelimiter(const FormatStyle &Style, FormatStyle::LanguageKind Language)
static bool opensProtoMessageField(const FormatToken &LessTok, const FormatStyle &Style)
static StringRef getEnclosingFunctionName(const FormatToken &Current)
bool applyAllReplacements(const Replacements &Replaces, Rewriter &Rewrite)
Apply all replacements in Replaces to the Rewriter Rewrite.
The JSON file list parser is used to communicate input to InstallAPI.
if(T->getSizeExpr()) TRY_TO(TraverseStmt(const_cast< Expr * >(T -> getSizeExpr())))
Language
The language for the input, used to select and validate the language standard and possible actions.
Definition: LangStandard.h:23
const FunctionProtoType * T
Definition: Format.h:5394
Encapsulates keywords that are context sensitive or for languages not properly supported by Clang's l...
Definition: FormatToken.h:996
bool isVerilogBegin(const FormatToken &Tok) const
Returns whether Tok is a Verilog keyword that opens a block.
Definition: FormatToken.h:1820
bool isVerilogEndOfLabel(const FormatToken &Tok) const
Definition: FormatToken.h:1854
The FormatStyle is used to configure the formatting to follow specific guidelines.
Definition: Format.h:55
@ UT_Never
Never use tab.
Definition: Format.h:4852
unsigned ContinuationIndentWidth
Indent width for line continuations.
Definition: Format.h:2398
bool AlwaysBreakBeforeMultilineStrings
This option is renamed to BreakAfterReturnType.
Definition: Format.h:1066
ReturnTypeBreakingStyle BreakAfterReturnType
The function declaration return type breaking style to use.
Definition: Format.h:1617
bool isTableGen() const
Definition: Format.h:3150
LanguageKind
Supported languages.
Definition: Format.h:3114
@ LK_Java
Should be used for Java.
Definition: Format.h:3122
@ LK_ObjC
Should be used for Objective-C, Objective-C++.
Definition: Format.h:3128
@ LK_Proto
Should be used for Protocol Buffers (https://developers.google.com/protocol-buffers/).
Definition: Format.h:3131
@ LK_TextProto
Should be used for Protocol Buffer messages in text format (https://developers.google....
Definition: Format.h:3136
bool Cpp11BracedListStyle
If true, format braced lists as best suited for C++11 braced lists.
Definition: Format.h:2421
BreakInheritanceListStyle BreakInheritanceList
The inheritance list style to use.
Definition: Format.h:2349
unsigned IndentWidth
The number of columns to use for indentation.
Definition: Format.h:2821
bool IndentRequiresClause
Indent the requires clause in a template.
Definition: Format.h:2807
std::vector< RawStringFormat > RawStringFormats
Defines hints for detecting supported languages code blocks in raw strings.
Definition: Format.h:3686
PPDirectiveIndentStyle IndentPPDirectives
The preprocessor directive indenting style to use.
Definition: Format.h:2785
BinaryOperatorStyle BreakBeforeBinaryOperators
The way to wrap binary operators.
Definition: Format.h:1691
@ BPS_Auto
Automatically determine parameter bin-packing behavior.
Definition: Format.h:1642
@ BPS_Always
Always bin-pack parameters.
Definition: Format.h:1644
@ BCIS_AfterColon
Break constructor initializers after the colon and commas.
Definition: Format.h:2219
@ BCIS_BeforeColon
Break constructor initializers before the colon and after the commas.
Definition: Format.h:2204
@ BCIS_BeforeComma
Break constructor initializers before the colon and commas, and align the commas with the colon.
Definition: Format.h:2212
BreakBeforeConceptDeclarationsStyle BreakBeforeConceptDeclarations
The concept declaration style to use.
Definition: Format.h:2150
BreakTemplateDeclarationsStyle BreakTemplateDeclarations
The template declaration breaking style to use.
Definition: Format.h:2353
@ BOS_None
Break after operators.
Definition: Format.h:1662
bool IndentWrappedFunctionNames
Indent if a function definition or declaration is wrapped after the type.
Definition: Format.h:2835
LanguageKind Language
Language, this format style is targeted at.
Definition: Format.h:3154
@ BAS_DontAlign
Don't align, instead use ContinuationIndentWidth, e.g.:
Definition: Format.h:78
@ BAS_AlwaysBreak
Always break after an open bracket, if the parameters don't fit on a single line, e....
Definition: Format.h:85
@ BAS_BlockIndent
Always break after an open bracket, if the parameters don't fit on a single line.
Definition: Format.h:99
@ BBIAS_Always
Always break before inline ASM colon.
Definition: Format.h:2174
@ BBIAS_OnlyMultiline
Break before inline ASM colon if the line length is longer than column limit.
Definition: Format.h:2167
unsigned TabWidth
The number of columns used for tab stops.
Definition: Format.h:4816
@ PPDIS_AfterHash
Indents directives after the hash.
Definition: Format.h:2771
@ LBI_OuterScope
For statements within block scope, align lambda body relative to the indentation level of the outer s...
Definition: Format.h:3098
@ LBI_Signature
Align lambda body relative to the lambda signature.
Definition: Format.h:3084
unsigned PenaltyBreakFirstLessLess
The penalty for breaking before the first <<.
Definition: Format.h:3487
unsigned ObjCBlockIndentWidth
The number of characters to use for indentation of ObjC blocks.
Definition: Format.h:3352
std::optional< FormatStyle > GetLanguageStyle(LanguageKind Language) const
Definition: Format.cpp:2099
std::optional< unsigned > BracedInitializerIndentWidth
The number of columns to use to indent the contents of braced init lists.
Definition: Format.h:1232
bool ExperimentalAutoDetectBinPacking
If true, clang-format detects whether function calls and definitions are formatted with one parameter...
Definition: Format.h:2567
bool ObjCBreakBeforeNestedBlockParam
Break parameters list into lines when there is nested block parameters in a function call.
Definition: Format.h:3376
OperandAlignmentStyle AlignOperands
If true, horizontally align operands of binary and ternary expressions.
Definition: Format.h:524
@ BTDS_No
Do not force break before declaration.
Definition: Format.h:1089
@ BTDS_Leave
Do not change the line breaking before the declaration.
Definition: Format.h:1079
BreakConstructorInitializersStyle BreakConstructorInitializers
The break constructor initializers style to use.
Definition: Format.h:2224
bool BreakStringLiterals
Allow breaking string literals when formatting.
Definition: Format.h:2291
@ PCIS_NextLineOnly
Put all constructor initializers on the next line if they fit.
Definition: Format.h:3466
@ PCIS_BinPack
Bin-pack constructor initializers.
Definition: Format.h:3426
@ PCIS_NextLine
Same as PCIS_CurrentLine except that if all constructor initializers do not fit on the current line,...
Definition: Format.h:3451
BraceBreakingStyle BreakBeforeBraces
The brace breaking style to use.
Definition: Format.h:2126
@ BILS_AfterColon
Break inheritance list after the colon and commas.
Definition: Format.h:2337
@ BILS_BeforeComma
Break inheritance list before the colon and commas, and align the commas with the colon.
Definition: Format.h:2329
bool isCSharp() const
Definition: Format.h:3143
unsigned PenaltyExcessCharacter
The penalty for each character outside of the column limit.
Definition: Format.h:3507
bool BinPackParameters
If false, a function declaration's or function definition's parameters will either all be on the same...
Definition: Format.h:1170
@ DAS_BreakElements
Break inside DAGArg after each list element but for the last.
Definition: Format.h:4799
unsigned ConstructorInitializerIndentWidth
This option is deprecated.
Definition: Format.h:2387
@ RCPS_OwnLine
Always put the requires clause on its own line.
Definition: Format.h:3860
@ RCPS_WithPreceding
Try to put the clause together with the preceding part of a declaration.
Definition: Format.h:3877
@ RCPS_SingleLine
Try to put everything in the same line if possible.
Definition: Format.h:3915
@ RCPS_WithFollowing
Try to put the requires clause together with the class or function declaration.
Definition: Format.h:3891
RequiresClausePositionStyle RequiresClausePosition
The position of the requires clause.
Definition: Format.h:3920
BreakBeforeInlineASMColonStyle BreakBeforeInlineASMColon
The inline ASM colon style to use.
Definition: Format.h:2179
@ BS_Whitesmiths
Like Allman but always indent braces and line up code with braces.
Definition: Format.h:2009
bool BinPackArguments
If false, a function call's arguments will either be all on the same line or will have one line each.
Definition: Format.h:1155
@ SFS_All
Merge all functions fitting on a single line.
Definition: Format.h:819
@ REI_Keyword
Align requires expression body relative to the requires keyword.
Definition: Format.h:3941
PackConstructorInitializersStyle PackConstructorInitializers
The pack constructor initializers style to use.
Definition: Format.h:3471
@ BBCDS_Allowed
Breaking between template declaration and concept is allowed.
Definition: Format.h:2138
@ BBCDS_Never
Keep the template declaration line together with concept.
Definition: Format.h:2134
@ BBCDS_Always
Always break before concept, putting it in the line after the template declaration.
Definition: Format.h:2145
bool AllowAllParametersOfDeclarationOnNextLine
This option is deprecated.
Definition: Format.h:651
BracketAlignmentStyle AlignAfterOpenBracket
If true, horizontally aligns arguments after an open bracket.
Definition: Format.h:107
bool isProto() const
Definition: Format.h:3147
bool ReflowComments
If true, clang-format will attempt to re-flow comments.
Definition: Format.h:3730
unsigned MaxEmptyLinesToKeep
The maximum number of consecutive empty lines to keep.
Definition: Format.h:3257
BinPackStyle ObjCBinPackProtocolList
Controls bin-packing Objective-C protocol conformance list items into as few lines as possible when t...
Definition: Format.h:3341
bool isVerilog() const
Definition: Format.h:3146
bool isJavaScript() const
Definition: Format.h:3145
DAGArgStyle TableGenBreakInsideDAGArg
The styles of the line break inside the DAGArg in TableGen.
Definition: Format.h:4812
LambdaBodyIndentationKind LambdaBodyIndentation
The indentation style of lambda bodies.
Definition: Format.h:3107
BraceWrappingFlags BraceWrapping
Control of individual brace wrapping cases.
Definition: Format.h:1513
unsigned PenaltyBreakString
The penalty for each line break introduced inside a string literal.
Definition: Format.h:3499
RequiresExpressionIndentationKind RequiresExpressionIndentation
The indentation used for requires expression bodies.
Definition: Format.h:3946
unsigned PenaltyIndentedWhitespace
Penalty for each character of whitespace indentation (counted relative to leading non-whitespace colu...
Definition: Format.h:3512
ShortFunctionStyle AllowShortFunctionsOnASingleLine
Dependent on the value, int f() { return 0; } can be put on a single line.
Definition: Format.h:825
bool AllowAllArgumentsOnNextLine
If a function call or braced initializer list doesn't fit on a line, allow putting all arguments onto...
Definition: Format.h:628
unsigned PenaltyBreakComment
The penalty for each line break introduced inside a comment.
Definition: Format.h:3483
@ RTBS_ExceptShortType
Same as Automatic above, except that there is no break after short return types.
Definition: Format.h:976
@ RTBS_None
This is deprecated. See Automatic below.
Definition: Format.h:953
UseTabStyle UseTab
The way to use tab characters in the resulting file.
Definition: Format.h:4868
@ OAS_AlignAfterOperator
Horizontally align operands of binary and ternary expressions.
Definition: Format.h:518
@ OAS_DontAlign
Do not align operands of binary and ternary expressions.
Definition: Format.h:492
bool BreakBeforeTernaryOperators
If true, ternary operators will be placed after line breaks.
Definition: Format.h:2194
unsigned ColumnLimit
The column limit.
Definition: Format.h:2299
A wrapper around a Token storing information about the whitespace characters preceding it.
Definition: FormatToken.h:287
unsigned NestingLevel
The nesting level of this token, i.e.
Definition: FormatToken.h:510
bool isMemberAccess() const
Returns true if this is a "." or "->" accessing a member.
Definition: FormatToken.h:715
bool isNot(T Kind) const
Definition: FormatToken.h:618
StringRef TokenText
The raw text of the token.
Definition: FormatToken.h:307
FormatToken * getPreviousNonComment() const
Returns the previous token ignoring comments.
Definition: FormatToken.h:804
bool closesScope() const
Returns whether Tok is )]} or a closing > of a template or in protos.
Definition: FormatToken.h:705
bool is(tok::TokenKind Kind) const
Definition: FormatToken.h:599
unsigned TotalLength
The total length of the unwrapped line up to and including this token.
Definition: FormatToken.h:493
bool isTrailingComment() const
Definition: FormatToken.h:746
FormatToken * NextOperator
If this is an operator (or "."/"->") in a sequence of operators with the same precedence,...
Definition: FormatToken.h:550
FormatToken * MatchingParen
If this is a bracket, this points to the matching one.
Definition: FormatToken.h:553
FormatToken * Previous
The previous token in the unwrapped line.
Definition: FormatToken.h:556
The current state when indenting a unwrapped line.
llvm::StringMap< FormatStyle > EnclosingFunctionStyle
std::optional< FormatStyle > getDelimiterStyle(StringRef Delimiter) const
std::optional< FormatStyle > getEnclosingFunctionStyle(StringRef EnclosingFunction) const
RawStringFormatStyleManager(const FormatStyle &CodeStyle)
llvm::StringMap< FormatStyle > DelimiterStyle