1 //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file implements the continuation indenter.
11 ///
12 //===----------------------------------------------------------------------===//
13
14 #include "ContinuationIndenter.h"
15 #include "BreakableToken.h"
16 #include "FormatInternal.h"
17 #include "FormatToken.h"
18 #include "WhitespaceManager.h"
19 #include "clang/Basic/OperatorPrecedence.h"
20 #include "clang/Basic/SourceManager.h"
21 #include "clang/Format/Format.h"
22 #include "llvm/ADT/StringSet.h"
23 #include "llvm/Support/Debug.h"
24 #include <optional>
25
26 #define DEBUG_TYPE "format-indenter"
27
28 namespace clang {
29 namespace format {
30
31 // Returns true if a TT_SelectorName should be indented when wrapped,
32 // false otherwise.
shouldIndentWrappedSelectorName(const FormatStyle & Style,LineType LineType)33 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
34 LineType LineType) {
35 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
36 }
37
38 // Returns the length of everything up to the first possible line break after
39 // the ), ], } or > matching \c Tok.
getLengthToMatchingParen(const FormatToken & Tok,ArrayRef<ParenState> Stack)40 static unsigned getLengthToMatchingParen(const FormatToken &Tok,
41 ArrayRef<ParenState> Stack) {
42 // Normally whether or not a break before T is possible is calculated and
43 // stored in T.CanBreakBefore. Braces, array initializers and text proto
44 // messages like `key: < ... >` are an exception: a break is possible
45 // before a closing brace R if a break was inserted after the corresponding
46 // opening brace. The information about whether or not a break is needed
47 // before a closing brace R is stored in the ParenState field
48 // S.BreakBeforeClosingBrace where S is the state that R closes.
49 //
50 // In order to decide whether there can be a break before encountered right
51 // braces, this implementation iterates over the sequence of tokens and over
52 // the paren stack in lockstep, keeping track of the stack level which visited
53 // right braces correspond to in MatchingStackIndex.
54 //
55 // For example, consider:
56 // L. <- line number
57 // 1. {
58 // 2. {1},
59 // 3. {2},
60 // 4. {{3}}}
61 // ^ where we call this method with this token.
62 // The paren stack at this point contains 3 brace levels:
63 // 0. { at line 1, BreakBeforeClosingBrace: true
64 // 1. first { at line 4, BreakBeforeClosingBrace: false
65 // 2. second { at line 4, BreakBeforeClosingBrace: false,
66 // where there might be fake parens levels in-between these levels.
67 // The algorithm will start at the first } on line 4, which is the matching
68 // brace of the initial left brace and at level 2 of the stack. Then,
69 // examining BreakBeforeClosingBrace: false at level 2, it will continue to
70 // the second } on line 4, and will traverse the stack downwards until it
71 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
72 // false at level 1, it will continue to the third } on line 4 and will
73 // traverse the stack downwards until it finds the matching { on level 0.
74 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
75 // will stop and will use the second } on line 4 to determine the length to
76 // return, as in this example the range will include the tokens: {3}}
77 //
78 // The algorithm will only traverse the stack if it encounters braces, array
79 // initializer squares or text proto angle brackets.
80 if (!Tok.MatchingParen)
81 return 0;
82 FormatToken *End = Tok.MatchingParen;
83 // Maintains a stack level corresponding to the current End token.
84 int MatchingStackIndex = Stack.size() - 1;
85 // Traverses the stack downwards, looking for the level to which LBrace
86 // corresponds. Returns either a pointer to the matching level or nullptr if
87 // LParen is not found in the initial portion of the stack up to
88 // MatchingStackIndex.
89 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
90 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
91 --MatchingStackIndex;
92 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
93 };
94 for (; End->Next; End = End->Next) {
95 if (End->Next->CanBreakBefore)
96 break;
97 if (!End->Next->closesScope())
98 continue;
99 if (End->Next->MatchingParen &&
100 End->Next->MatchingParen->isOneOf(
101 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
102 const ParenState *State = FindParenState(End->Next->MatchingParen);
103 if (State && State->BreakBeforeClosingBrace)
104 break;
105 }
106 }
107 return End->TotalLength - Tok.TotalLength + 1;
108 }
109
getLengthToNextOperator(const FormatToken & Tok)110 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
111 if (!Tok.NextOperator)
112 return 0;
113 return Tok.NextOperator->TotalLength - Tok.TotalLength;
114 }
115
116 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
117 // segment of a builder type call.
startsSegmentOfBuilderTypeCall(const FormatToken & Tok)118 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
119 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
120 }
121
122 // Returns \c true if \c Current starts a new parameter.
startsNextParameter(const FormatToken & Current,const FormatStyle & Style)123 static bool startsNextParameter(const FormatToken &Current,
124 const FormatStyle &Style) {
125 const FormatToken &Previous = *Current.Previous;
126 if (Current.is(TT_CtorInitializerComma) &&
127 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) {
128 return true;
129 }
130 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
131 return true;
132 return Previous.is(tok::comma) && !Current.isTrailingComment() &&
133 ((Previous.isNot(TT_CtorInitializerComma) ||
134 Style.BreakConstructorInitializers !=
135 FormatStyle::BCIS_BeforeComma) &&
136 (Previous.isNot(TT_InheritanceComma) ||
137 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma));
138 }
139
opensProtoMessageField(const FormatToken & LessTok,const FormatStyle & Style)140 static bool opensProtoMessageField(const FormatToken &LessTok,
141 const FormatStyle &Style) {
142 if (LessTok.isNot(tok::less))
143 return false;
144 return Style.Language == FormatStyle::LK_TextProto ||
145 (Style.Language == FormatStyle::LK_Proto &&
146 (LessTok.NestingLevel > 0 ||
147 (LessTok.Previous && LessTok.Previous->is(tok::equal))));
148 }
149
150 // Returns the delimiter of a raw string literal, or std::nullopt if TokenText
151 // is not the text of a raw string literal. The delimiter could be the empty
152 // string. For example, the delimiter of R"deli(cont)deli" is deli.
getRawStringDelimiter(StringRef TokenText)153 static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
154 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
155 || !TokenText.startswith("R\"") || !TokenText.endswith("\"")) {
156 return std::nullopt;
157 }
158
159 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
160 // size at most 16 by the standard, so the first '(' must be among the first
161 // 19 bytes.
162 size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
163 if (LParenPos == StringRef::npos)
164 return std::nullopt;
165 StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
166
167 // Check that the string ends in ')Delimiter"'.
168 size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
169 if (TokenText[RParenPos] != ')')
170 return std::nullopt;
171 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter))
172 return std::nullopt;
173 return Delimiter;
174 }
175
176 // Returns the canonical delimiter for \p Language, or the empty string if no
177 // canonical delimiter is specified.
178 static StringRef
getCanonicalRawStringDelimiter(const FormatStyle & Style,FormatStyle::LanguageKind Language)179 getCanonicalRawStringDelimiter(const FormatStyle &Style,
180 FormatStyle::LanguageKind Language) {
181 for (const auto &Format : Style.RawStringFormats)
182 if (Format.Language == Language)
183 return StringRef(Format.CanonicalDelimiter);
184 return "";
185 }
186
RawStringFormatStyleManager(const FormatStyle & CodeStyle)187 RawStringFormatStyleManager::RawStringFormatStyleManager(
188 const FormatStyle &CodeStyle) {
189 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
190 std::optional<FormatStyle> LanguageStyle =
191 CodeStyle.GetLanguageStyle(RawStringFormat.Language);
192 if (!LanguageStyle) {
193 FormatStyle PredefinedStyle;
194 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
195 RawStringFormat.Language, &PredefinedStyle)) {
196 PredefinedStyle = getLLVMStyle();
197 PredefinedStyle.Language = RawStringFormat.Language;
198 }
199 LanguageStyle = PredefinedStyle;
200 }
201 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
202 for (StringRef Delimiter : RawStringFormat.Delimiters)
203 DelimiterStyle.insert({Delimiter, *LanguageStyle});
204 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions)
205 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
206 }
207 }
208
209 std::optional<FormatStyle>
getDelimiterStyle(StringRef Delimiter) const210 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
211 auto It = DelimiterStyle.find(Delimiter);
212 if (It == DelimiterStyle.end())
213 return std::nullopt;
214 return It->second;
215 }
216
217 std::optional<FormatStyle>
getEnclosingFunctionStyle(StringRef EnclosingFunction) const218 RawStringFormatStyleManager::getEnclosingFunctionStyle(
219 StringRef EnclosingFunction) const {
220 auto It = EnclosingFunctionStyle.find(EnclosingFunction);
221 if (It == EnclosingFunctionStyle.end())
222 return std::nullopt;
223 return It->second;
224 }
225
ContinuationIndenter(const FormatStyle & Style,const AdditionalKeywords & Keywords,const SourceManager & SourceMgr,WhitespaceManager & Whitespaces,encoding::Encoding Encoding,bool BinPackInconclusiveFunctions)226 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
227 const AdditionalKeywords &Keywords,
228 const SourceManager &SourceMgr,
229 WhitespaceManager &Whitespaces,
230 encoding::Encoding Encoding,
231 bool BinPackInconclusiveFunctions)
232 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
233 Whitespaces(Whitespaces), Encoding(Encoding),
234 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
235 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
236
getInitialState(unsigned FirstIndent,unsigned FirstStartColumn,const AnnotatedLine * Line,bool DryRun)237 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
238 unsigned FirstStartColumn,
239 const AnnotatedLine *Line,
240 bool DryRun) {
241 LineState State;
242 State.FirstIndent = FirstIndent;
243 if (FirstStartColumn && Line->First->NewlinesBefore == 0)
244 State.Column = FirstStartColumn;
245 else
246 State.Column = FirstIndent;
247 // With preprocessor directive indentation, the line starts on column 0
248 // since it's indented after the hash, but FirstIndent is set to the
249 // preprocessor indent.
250 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
251 (Line->Type == LT_PreprocessorDirective ||
252 Line->Type == LT_ImportStatement)) {
253 State.Column = 0;
254 }
255 State.Line = Line;
256 State.NextToken = Line->First;
257 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
258 /*AvoidBinPacking=*/false,
259 /*NoLineBreak=*/false));
260 State.NoContinuation = false;
261 State.StartOfStringLiteral = 0;
262 State.StartOfLineLevel = 0;
263 State.LowestLevelOnLine = 0;
264 State.IgnoreStackForComparison = false;
265
266 if (Style.Language == FormatStyle::LK_TextProto) {
267 // We need this in order to deal with the bin packing of text fields at
268 // global scope.
269 auto &CurrentState = State.Stack.back();
270 CurrentState.AvoidBinPacking = true;
271 CurrentState.BreakBeforeParameter = true;
272 CurrentState.AlignColons = false;
273 }
274
275 // The first token has already been indented and thus consumed.
276 moveStateToNextToken(State, DryRun, /*Newline=*/false);
277 return State;
278 }
279
canBreak(const LineState & State)280 bool ContinuationIndenter::canBreak(const LineState &State) {
281 const FormatToken &Current = *State.NextToken;
282 const FormatToken &Previous = *Current.Previous;
283 const auto &CurrentState = State.Stack.back();
284 assert(&Previous == Current.Previous);
285 if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace &&
286 Current.closesBlockOrBlockTypeList(Style))) {
287 return false;
288 }
289 // The opening "{" of a braced list has to be on the same line as the first
290 // element if it is nested in another braced init list or function call.
291 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
292 Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) &&
293 Previous.Previous &&
294 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) {
295 return false;
296 }
297 // This prevents breaks like:
298 // ...
299 // SomeParameter, OtherParameter).DoSomething(
300 // ...
301 // As they hide "DoSomething" and are generally bad for readability.
302 if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
303 State.LowestLevelOnLine < State.StartOfLineLevel &&
304 State.LowestLevelOnLine < Current.NestingLevel) {
305 return false;
306 }
307 if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder)
308 return false;
309
310 // Don't create a 'hanging' indent if there are multiple blocks in a single
311 // statement.
312 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
313 State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
314 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks) {
315 return false;
316 }
317
318 // Don't break after very short return types (e.g. "void") as that is often
319 // unexpected.
320 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) {
321 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None)
322 return false;
323 }
324
325 // If binary operators are moved to the next line (including commas for some
326 // styles of constructor initializers), that's always ok.
327 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
328 CurrentState.NoLineBreakInOperand) {
329 return false;
330 }
331
332 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr))
333 return false;
334
335 if (Current.is(TT_ConditionalExpr) && Previous.is(tok::r_paren) &&
336 Previous.MatchingParen && Previous.MatchingParen->Previous &&
337 Previous.MatchingParen->Previous->MatchingParen &&
338 Previous.MatchingParen->Previous->MatchingParen->is(TT_LambdaLBrace)) {
339 // We have a lambda within a conditional expression, allow breaking here.
340 assert(Previous.MatchingParen->Previous->is(tok::r_brace));
341 return true;
342 }
343
344 return !CurrentState.NoLineBreak;
345 }
346
mustBreak(const LineState & State)347 bool ContinuationIndenter::mustBreak(const LineState &State) {
348 const FormatToken &Current = *State.NextToken;
349 const FormatToken &Previous = *Current.Previous;
350 const auto &CurrentState = State.Stack.back();
351 if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore &&
352 Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) {
353 auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack);
354 return LambdaBodyLength > getColumnLimit(State);
355 }
356 if (Current.MustBreakBefore ||
357 (Current.is(TT_InlineASMColon) &&
358 (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always ||
359 Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_OnlyMultiline))) {
360 return true;
361 }
362 if (CurrentState.BreakBeforeClosingBrace &&
363 Current.closesBlockOrBlockTypeList(Style)) {
364 return true;
365 }
366 if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren))
367 return true;
368 if (Style.Language == FormatStyle::LK_ObjC &&
369 Style.ObjCBreakBeforeNestedBlockParam &&
370 Current.ObjCSelectorNameParts > 1 &&
371 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
372 return true;
373 }
374 // Avoid producing inconsistent states by requiring breaks where they are not
375 // permitted for C# generic type constraints.
376 if (CurrentState.IsCSharpGenericTypeConstraint &&
377 Previous.isNot(TT_CSharpGenericTypeConstraintComma)) {
378 return false;
379 }
380 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
381 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
382 Style.isCpp() &&
383 // FIXME: This is a temporary workaround for the case where clang-format
384 // sets BreakBeforeParameter to avoid bin packing and this creates a
385 // completely unnecessary line break after a template type that isn't
386 // line-wrapped.
387 (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
388 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
389 Previous.isNot(tok::question)) ||
390 (!Style.BreakBeforeTernaryOperators &&
391 Previous.is(TT_ConditionalExpr))) &&
392 CurrentState.BreakBeforeParameter && !Current.isTrailingComment() &&
393 !Current.isOneOf(tok::r_paren, tok::r_brace)) {
394 return true;
395 }
396 if (CurrentState.IsChainedConditional &&
397 ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
398 Current.is(tok::colon)) ||
399 (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) &&
400 Previous.is(tok::colon)))) {
401 return true;
402 }
403 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
404 (Previous.is(TT_ArrayInitializerLSquare) &&
405 Previous.ParameterCount > 1) ||
406 opensProtoMessageField(Previous, Style)) &&
407 Style.ColumnLimit > 0 &&
408 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
409 getColumnLimit(State)) {
410 return true;
411 }
412
413 const FormatToken &BreakConstructorInitializersToken =
414 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
415 ? Previous
416 : Current;
417 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
418 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
419 getColumnLimit(State) ||
420 CurrentState.BreakBeforeParameter) &&
421 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
422 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
423 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
424 Style.ColumnLimit != 0)) {
425 return true;
426 }
427
428 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) &&
429 State.Line->startsWith(TT_ObjCMethodSpecifier)) {
430 return true;
431 }
432 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) &&
433 CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter &&
434 (Style.ObjCBreakBeforeNestedBlockParam ||
435 !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) {
436 return true;
437 }
438
439 unsigned NewLineColumn = getNewLineColumn(State);
440 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
441 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
442 (State.Column > NewLineColumn ||
443 Current.NestingLevel < State.StartOfLineLevel)) {
444 return true;
445 }
446
447 if (startsSegmentOfBuilderTypeCall(Current) &&
448 (CurrentState.CallContinuation != 0 ||
449 CurrentState.BreakBeforeParameter) &&
450 // JavaScript is treated different here as there is a frequent pattern:
451 // SomeFunction(function() {
452 // ...
453 // }.bind(...));
454 // FIXME: We should find a more generic solution to this problem.
455 !(State.Column <= NewLineColumn && Style.isJavaScript()) &&
456 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) {
457 return true;
458 }
459
460 // If the template declaration spans multiple lines, force wrap before the
461 // function/class declaration.
462 if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter &&
463 Current.CanBreakBefore) {
464 return true;
465 }
466
467 if (!State.Line->First->is(tok::kw_enum) && State.Column <= NewLineColumn)
468 return false;
469
470 if (Style.AlwaysBreakBeforeMultilineStrings &&
471 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
472 Previous.is(tok::comma) || Current.NestingLevel < 2) &&
473 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at,
474 Keywords.kw_dollar) &&
475 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
476 nextIsMultilineString(State)) {
477 return true;
478 }
479
480 // Using CanBreakBefore here and below takes care of the decision whether the
481 // current style uses wrapping before or after operators for the given
482 // operator.
483 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
484 const auto PreviousPrecedence = Previous.getPrecedence();
485 if (PreviousPrecedence != prec::Assignment &&
486 CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) {
487 const bool LHSIsBinaryExpr =
488 Previous.Previous && Previous.Previous->EndsBinaryExpression;
489 if (LHSIsBinaryExpr)
490 return true;
491 // If we need to break somewhere inside the LHS of a binary expression, we
492 // should also break after the operator. Otherwise, the formatting would
493 // hide the operator precedence, e.g. in:
494 // if (aaaaaaaaaaaaaa ==
495 // bbbbbbbbbbbbbb && c) {..
496 // For comparisons, we only apply this rule, if the LHS is a binary
497 // expression itself as otherwise, the line breaks seem superfluous.
498 // We need special cases for ">>" which we have split into two ">" while
499 // lexing in order to make template parsing easier.
500 const bool IsComparison =
501 (PreviousPrecedence == prec::Relational ||
502 PreviousPrecedence == prec::Equality ||
503 PreviousPrecedence == prec::Spaceship) &&
504 Previous.Previous &&
505 Previous.Previous->isNot(TT_BinaryOperator); // For >>.
506 if (!IsComparison)
507 return true;
508 }
509 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
510 CurrentState.BreakBeforeParameter) {
511 return true;
512 }
513
514 // Same as above, but for the first "<<" operator.
515 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
516 CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) {
517 return true;
518 }
519
520 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
521 // Always break after "template <...>"(*) and leading annotations. This is
522 // only for cases where the entire line does not fit on a single line as a
523 // different LineFormatter would be used otherwise.
524 // *: Except when another option interferes with that, like concepts.
525 if (Previous.ClosesTemplateDeclaration) {
526 if (Current.is(tok::kw_concept)) {
527 switch (Style.BreakBeforeConceptDeclarations) {
528 case FormatStyle::BBCDS_Allowed:
529 break;
530 case FormatStyle::BBCDS_Always:
531 return true;
532 case FormatStyle::BBCDS_Never:
533 return false;
534 }
535 }
536 if (Current.is(TT_RequiresClause)) {
537 switch (Style.RequiresClausePosition) {
538 case FormatStyle::RCPS_SingleLine:
539 case FormatStyle::RCPS_WithPreceding:
540 return false;
541 default:
542 return true;
543 }
544 }
545 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No;
546 }
547 if (Previous.is(TT_FunctionAnnotationRParen) &&
548 State.Line->Type != LT_PreprocessorDirective) {
549 return true;
550 }
551 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
552 Current.isNot(TT_LeadingJavaAnnotation)) {
553 return true;
554 }
555 }
556
557 if (Style.isJavaScript() && Previous.is(tok::r_paren) &&
558 Previous.is(TT_JavaAnnotation)) {
559 // Break after the closing parenthesis of TypeScript decorators before
560 // functions, getters and setters.
561 static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set",
562 "function"};
563 if (BreakBeforeDecoratedTokens.contains(Current.TokenText))
564 return true;
565 }
566
567 // If the return type spans multiple lines, wrap before the function name.
568 if (((Current.is(TT_FunctionDeclarationName) &&
569 !State.Line->ReturnTypeWrapped &&
570 // Don't break before a C# function when no break after return type.
571 (!Style.isCSharp() ||
572 Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) &&
573 // Don't always break between a JavaScript `function` and the function
574 // name.
575 !Style.isJavaScript()) ||
576 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) &&
577 !Previous.is(tok::kw_template) && CurrentState.BreakBeforeParameter) {
578 return true;
579 }
580
581 // The following could be precomputed as they do not depend on the state.
582 // However, as they should take effect only if the UnwrappedLine does not fit
583 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
584 if (Style.ColumnLimit != 0 && Previous.is(BK_Block) &&
585 Previous.is(tok::l_brace) &&
586 !Current.isOneOf(tok::r_brace, tok::comment)) {
587 return true;
588 }
589
590 if (Current.is(tok::lessless) &&
591 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
592 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") ||
593 Previous.TokenText == "\'\\n\'")))) {
594 return true;
595 }
596
597 if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
598 return true;
599
600 if (State.NoContinuation)
601 return true;
602
603 return false;
604 }
605
addTokenToState(LineState & State,bool Newline,bool DryRun,unsigned ExtraSpaces)606 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
607 bool DryRun,
608 unsigned ExtraSpaces) {
609 const FormatToken &Current = *State.NextToken;
610 assert(State.NextToken->Previous);
611 const FormatToken &Previous = *State.NextToken->Previous;
612
613 assert(!State.Stack.empty());
614 State.NoContinuation = false;
615
616 if ((Current.is(TT_ImplicitStringLiteral) &&
617 (Previous.Tok.getIdentifierInfo() == nullptr ||
618 Previous.Tok.getIdentifierInfo()->getPPKeywordID() ==
619 tok::pp_not_keyword))) {
620 unsigned EndColumn =
621 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
622 if (Current.LastNewlineOffset != 0) {
623 // If there is a newline within this token, the final column will solely
624 // determined by the current end column.
625 State.Column = EndColumn;
626 } else {
627 unsigned StartColumn =
628 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
629 assert(EndColumn >= StartColumn);
630 State.Column += EndColumn - StartColumn;
631 }
632 moveStateToNextToken(State, DryRun, /*Newline=*/false);
633 return 0;
634 }
635
636 unsigned Penalty = 0;
637 if (Newline)
638 Penalty = addTokenOnNewLine(State, DryRun);
639 else
640 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
641
642 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
643 }
644
addTokenOnCurrentLine(LineState & State,bool DryRun,unsigned ExtraSpaces)645 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
646 unsigned ExtraSpaces) {
647 FormatToken &Current = *State.NextToken;
648 assert(State.NextToken->Previous);
649 const FormatToken &Previous = *State.NextToken->Previous;
650 auto &CurrentState = State.Stack.back();
651
652 if (Current.is(tok::equal) &&
653 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
654 CurrentState.VariablePos == 0) {
655 CurrentState.VariablePos = State.Column;
656 // Move over * and & if they are bound to the variable name.
657 const FormatToken *Tok = &Previous;
658 while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) {
659 CurrentState.VariablePos -= Tok->ColumnWidth;
660 if (Tok->SpacesRequiredBefore != 0)
661 break;
662 Tok = Tok->Previous;
663 }
664 if (Previous.PartOfMultiVariableDeclStmt)
665 CurrentState.LastSpace = CurrentState.VariablePos;
666 }
667
668 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
669
670 // Indent preprocessor directives after the hash if required.
671 int PPColumnCorrection = 0;
672 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
673 Previous.is(tok::hash) && State.FirstIndent > 0 &&
674 &Previous == State.Line->First &&
675 (State.Line->Type == LT_PreprocessorDirective ||
676 State.Line->Type == LT_ImportStatement)) {
677 Spaces += State.FirstIndent;
678
679 // For preprocessor indent with tabs, State.Column will be 1 because of the
680 // hash. This causes second-level indents onward to have an extra space
681 // after the tabs. We avoid this misalignment by subtracting 1 from the
682 // column value passed to replaceWhitespace().
683 if (Style.UseTab != FormatStyle::UT_Never)
684 PPColumnCorrection = -1;
685 }
686
687 if (!DryRun) {
688 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
689 State.Column + Spaces + PPColumnCorrection);
690 }
691
692 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
693 // declaration unless there is multiple inheritance.
694 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
695 Current.is(TT_InheritanceColon)) {
696 CurrentState.NoLineBreak = true;
697 }
698 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
699 Previous.is(TT_InheritanceColon)) {
700 CurrentState.NoLineBreak = true;
701 }
702
703 if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) {
704 unsigned MinIndent = std::max(
705 State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent);
706 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
707 if (Current.LongestObjCSelectorName == 0)
708 CurrentState.AlignColons = false;
709 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
710 CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName;
711 else
712 CurrentState.ColonPos = FirstColonPos;
713 }
714
715 // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the
716 // parenthesis by disallowing any further line breaks if there is no line
717 // break after the opening parenthesis. Don't break if it doesn't conserve
718 // columns.
719 if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak ||
720 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) &&
721 (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) ||
722 (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) &&
723 Style.Cpp11BracedListStyle)) &&
724 State.Column > getNewLineColumn(State) &&
725 (!Previous.Previous ||
726 !Previous.Previous->isOneOf(TT_CastRParen, tok::kw_for, tok::kw_while,
727 tok::kw_switch)) &&
728 // Don't do this for simple (no expressions) one-argument function calls
729 // as that feels like needlessly wasting whitespace, e.g.:
730 //
731 // caaaaaaaaaaaall(
732 // caaaaaaaaaaaall(
733 // caaaaaaaaaaaall(
734 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
735 Current.FakeLParens.size() > 0 &&
736 Current.FakeLParens.back() > prec::Unknown) {
737 CurrentState.NoLineBreak = true;
738 }
739 if (Previous.is(TT_TemplateString) && Previous.opensScope())
740 CurrentState.NoLineBreak = true;
741
742 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
743 !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
744 Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) &&
745 (Current.isNot(TT_LineComment) || Previous.is(BK_BracedInit))) {
746 CurrentState.Indent = State.Column + Spaces;
747 CurrentState.IsAligned = true;
748 }
749 if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style))
750 CurrentState.NoLineBreak = true;
751 if (startsSegmentOfBuilderTypeCall(Current) &&
752 State.Column > getNewLineColumn(State)) {
753 CurrentState.ContainsUnwrappedBuilder = true;
754 }
755
756 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java)
757 CurrentState.NoLineBreak = true;
758 if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
759 (Previous.MatchingParen &&
760 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
761 // If there is a function call with long parameters, break before trailing
762 // calls. This prevents things like:
763 // EXPECT_CALL(SomeLongParameter).Times(
764 // 2);
765 // We don't want to do this for short parameters as they can just be
766 // indexes.
767 CurrentState.NoLineBreak = true;
768 }
769
770 // Don't allow the RHS of an operator to be split over multiple lines unless
771 // there is a line-break right after the operator.
772 // Exclude relational operators, as there, it is always more desirable to
773 // have the LHS 'left' of the RHS.
774 const FormatToken *P = Current.getPreviousNonComment();
775 if (!Current.is(tok::comment) && P &&
776 (P->isOneOf(TT_BinaryOperator, tok::comma) ||
777 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
778 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
779 P->getPrecedence() != prec::Assignment &&
780 P->getPrecedence() != prec::Relational &&
781 P->getPrecedence() != prec::Spaceship) {
782 bool BreakBeforeOperator =
783 P->MustBreakBefore || P->is(tok::lessless) ||
784 (P->is(TT_BinaryOperator) &&
785 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
786 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
787 // Don't do this if there are only two operands. In these cases, there is
788 // always a nice vertical separation between them and the extra line break
789 // does not help.
790 bool HasTwoOperands =
791 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr);
792 if ((!BreakBeforeOperator &&
793 !(HasTwoOperands &&
794 Style.AlignOperands != FormatStyle::OAS_DontAlign)) ||
795 (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) {
796 CurrentState.NoLineBreakInOperand = true;
797 }
798 }
799
800 State.Column += Spaces;
801 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
802 Previous.Previous &&
803 (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) {
804 // Treat the condition inside an if as if it was a second function
805 // parameter, i.e. let nested calls have a continuation indent.
806 CurrentState.LastSpace = State.Column;
807 CurrentState.NestedBlockIndent = State.Column;
808 } else if (!Current.isOneOf(tok::comment, tok::caret) &&
809 ((Previous.is(tok::comma) &&
810 !Previous.is(TT_OverloadedOperator)) ||
811 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
812 CurrentState.LastSpace = State.Column;
813 } else if (Previous.is(TT_CtorInitializerColon) &&
814 (!Current.isTrailingComment() || Current.NewlinesBefore > 0) &&
815 Style.BreakConstructorInitializers ==
816 FormatStyle::BCIS_AfterColon) {
817 CurrentState.Indent = State.Column;
818 CurrentState.LastSpace = State.Column;
819 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
820 TT_CtorInitializerColon)) &&
821 ((Previous.getPrecedence() != prec::Assignment &&
822 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
823 Previous.NextOperator)) ||
824 Current.StartsBinaryExpression)) {
825 // Indent relative to the RHS of the expression unless this is a simple
826 // assignment without binary expression on the RHS. Also indent relative to
827 // unary operators and the colons of constructor initializers.
828 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
829 CurrentState.LastSpace = State.Column;
830 } else if (Previous.is(TT_InheritanceColon)) {
831 CurrentState.Indent = State.Column;
832 CurrentState.LastSpace = State.Column;
833 } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) {
834 CurrentState.ColonPos = State.Column;
835 } else if (Previous.opensScope()) {
836 // If a function has a trailing call, indent all parameters from the
837 // opening parenthesis. This avoids confusing indents like:
838 // OuterFunction(InnerFunctionCall( // break
839 // ParameterToInnerFunction)) // break
840 // .SecondInnerFunctionCall();
841 if (Previous.MatchingParen) {
842 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
843 if (Next && Next->isMemberAccess() && State.Stack.size() > 1 &&
844 State.Stack[State.Stack.size() - 2].CallContinuation == 0) {
845 CurrentState.LastSpace = State.Column;
846 }
847 }
848 }
849 }
850
addTokenOnNewLine(LineState & State,bool DryRun)851 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
852 bool DryRun) {
853 FormatToken &Current = *State.NextToken;
854 assert(State.NextToken->Previous);
855 const FormatToken &Previous = *State.NextToken->Previous;
856 auto &CurrentState = State.Stack.back();
857
858 // Extra penalty that needs to be added because of the way certain line
859 // breaks are chosen.
860 unsigned Penalty = 0;
861
862 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
863 const FormatToken *NextNonComment = Previous.getNextNonComment();
864 if (!NextNonComment)
865 NextNonComment = &Current;
866 // The first line break on any NestingLevel causes an extra penalty in order
867 // prefer similar line breaks.
868 if (!CurrentState.ContainsLineBreak)
869 Penalty += 15;
870 CurrentState.ContainsLineBreak = true;
871
872 Penalty += State.NextToken->SplitPenalty;
873
874 // Breaking before the first "<<" is generally not desirable if the LHS is
875 // short. Also always add the penalty if the LHS is split over multiple lines
876 // to avoid unnecessary line breaks that just work around this penalty.
877 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 &&
878 (State.Column <= Style.ColumnLimit / 3 ||
879 CurrentState.BreakBeforeParameter)) {
880 Penalty += Style.PenaltyBreakFirstLessLess;
881 }
882
883 State.Column = getNewLineColumn(State);
884
885 // Add Penalty proportional to amount of whitespace away from FirstColumn
886 // This tends to penalize several lines that are far-right indented,
887 // and prefers a line-break prior to such a block, e.g:
888 //
889 // Constructor() :
890 // member(value), looooooooooooooooong_member(
891 // looooooooooong_call(param_1, param_2, param_3))
892 // would then become
893 // Constructor() :
894 // member(value),
895 // looooooooooooooooong_member(
896 // looooooooooong_call(param_1, param_2, param_3))
897 if (State.Column > State.FirstIndent) {
898 Penalty +=
899 Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent);
900 }
901
902 // Indent nested blocks relative to this column, unless in a very specific
903 // JavaScript special case where:
904 //
905 // var loooooong_name =
906 // function() {
907 // // code
908 // }
909 //
910 // is common and should be formatted like a free-standing function. The same
911 // goes for wrapping before the lambda return type arrow.
912 if (!Current.is(TT_LambdaArrow) &&
913 (!Style.isJavaScript() || Current.NestingLevel != 0 ||
914 !PreviousNonComment || !PreviousNonComment->is(tok::equal) ||
915 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) {
916 CurrentState.NestedBlockIndent = State.Column;
917 }
918
919 if (NextNonComment->isMemberAccess()) {
920 if (CurrentState.CallContinuation == 0)
921 CurrentState.CallContinuation = State.Column;
922 } else if (NextNonComment->is(TT_SelectorName)) {
923 if (!CurrentState.ObjCSelectorNameFound) {
924 if (NextNonComment->LongestObjCSelectorName == 0) {
925 CurrentState.AlignColons = false;
926 } else {
927 CurrentState.ColonPos =
928 (shouldIndentWrappedSelectorName(Style, State.Line->Type)
929 ? std::max(CurrentState.Indent,
930 State.FirstIndent + Style.ContinuationIndentWidth)
931 : CurrentState.Indent) +
932 std::max(NextNonComment->LongestObjCSelectorName,
933 NextNonComment->ColumnWidth);
934 }
935 } else if (CurrentState.AlignColons &&
936 CurrentState.ColonPos <= NextNonComment->ColumnWidth) {
937 CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth;
938 }
939 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
940 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
941 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
942 // method expression, the block should be aligned to the line starting it,
943 // e.g.:
944 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
945 // ^(int *i) {
946 // // ...
947 // }];
948 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
949 // when we consume all of the "}"'s FakeRParens at the "{".
950 if (State.Stack.size() > 1) {
951 State.Stack[State.Stack.size() - 2].LastSpace =
952 std::max(CurrentState.LastSpace, CurrentState.Indent) +
953 Style.ContinuationIndentWidth;
954 }
955 }
956
957 if ((PreviousNonComment &&
958 PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
959 !CurrentState.AvoidBinPacking) ||
960 Previous.is(TT_BinaryOperator)) {
961 CurrentState.BreakBeforeParameter = false;
962 }
963 if (PreviousNonComment &&
964 (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) ||
965 PreviousNonComment->ClosesRequiresClause) &&
966 Current.NestingLevel == 0) {
967 CurrentState.BreakBeforeParameter = false;
968 }
969 if (NextNonComment->is(tok::question) ||
970 (PreviousNonComment && PreviousNonComment->is(tok::question))) {
971 CurrentState.BreakBeforeParameter = true;
972 }
973 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
974 CurrentState.BreakBeforeParameter = false;
975
976 if (!DryRun) {
977 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
978 if (Current.is(tok::r_brace) && Current.MatchingParen &&
979 // Only strip trailing empty lines for l_braces that have children, i.e.
980 // for function expressions (lambdas, arrows, etc).
981 !Current.MatchingParen->Children.empty()) {
982 // lambdas and arrow functions are expressions, thus their r_brace is not
983 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
984 // about removing empty lines on closing blocks. Special case them here.
985 MaxEmptyLinesToKeep = 1;
986 }
987 unsigned Newlines =
988 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
989 bool ContinuePPDirective =
990 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
991 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
992 CurrentState.IsAligned, ContinuePPDirective);
993 }
994
995 if (!Current.isTrailingComment())
996 CurrentState.LastSpace = State.Column;
997 if (Current.is(tok::lessless)) {
998 // If we are breaking before a "<<", we always want to indent relative to
999 // RHS. This is necessary only for "<<", as we special-case it and don't
1000 // always indent relative to the RHS.
1001 CurrentState.LastSpace += 3; // 3 -> width of "<< ".
1002 }
1003
1004 State.StartOfLineLevel = Current.NestingLevel;
1005 State.LowestLevelOnLine = Current.NestingLevel;
1006
1007 // Any break on this level means that the parent level has been broken
1008 // and we need to avoid bin packing there.
1009 bool NestedBlockSpecialCase =
1010 (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
1011 State.Stack[State.Stack.size() - 2].NestedBlockInlined) ||
1012 (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) &&
1013 State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam);
1014 // Do not force parameter break for statements with requires expressions.
1015 NestedBlockSpecialCase =
1016 NestedBlockSpecialCase ||
1017 (Current.MatchingParen &&
1018 Current.MatchingParen->is(TT_RequiresExpressionLBrace));
1019 if (!NestedBlockSpecialCase)
1020 for (ParenState &PState : llvm::drop_end(State.Stack))
1021 PState.BreakBeforeParameter = true;
1022
1023 if (PreviousNonComment &&
1024 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
1025 ((PreviousNonComment->isNot(TT_TemplateCloser) &&
1026 !PreviousNonComment->ClosesRequiresClause) ||
1027 Current.NestingLevel != 0) &&
1028 !PreviousNonComment->isOneOf(
1029 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
1030 TT_LeadingJavaAnnotation) &&
1031 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope()) {
1032 CurrentState.BreakBeforeParameter = true;
1033 }
1034
1035 // If we break after { or the [ of an array initializer, we should also break
1036 // before the corresponding } or ].
1037 if (PreviousNonComment &&
1038 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1039 opensProtoMessageField(*PreviousNonComment, Style))) {
1040 CurrentState.BreakBeforeClosingBrace = true;
1041 }
1042
1043 if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) {
1044 CurrentState.BreakBeforeClosingParen =
1045 Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent;
1046 }
1047
1048 if (CurrentState.AvoidBinPacking) {
1049 // If we are breaking after '(', '{', '<', or this is the break after a ':'
1050 // to start a member initializater list in a constructor, this should not
1051 // be considered bin packing unless the relevant AllowAll option is false or
1052 // this is a dict/object literal.
1053 bool PreviousIsBreakingCtorInitializerColon =
1054 PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1055 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
1056 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
1057 PreviousIsBreakingCtorInitializerColon) ||
1058 (!Style.AllowAllParametersOfDeclarationOnNextLine &&
1059 State.Line->MustBeDeclaration) ||
1060 (!Style.AllowAllArgumentsOnNextLine &&
1061 !State.Line->MustBeDeclaration) ||
1062 (Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine &&
1063 PreviousIsBreakingCtorInitializerColon) ||
1064 Previous.is(TT_DictLiteral)) {
1065 CurrentState.BreakBeforeParameter = true;
1066 }
1067
1068 // If we are breaking after a ':' to start a member initializer list,
1069 // and we allow all arguments on the next line, we should not break
1070 // before the next parameter.
1071 if (PreviousIsBreakingCtorInitializerColon &&
1072 Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine) {
1073 CurrentState.BreakBeforeParameter = false;
1074 }
1075 }
1076
1077 return Penalty;
1078 }
1079
getNewLineColumn(const LineState & State)1080 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
1081 if (!State.NextToken || !State.NextToken->Previous)
1082 return 0;
1083
1084 FormatToken &Current = *State.NextToken;
1085 const auto &CurrentState = State.Stack.back();
1086
1087 if (CurrentState.IsCSharpGenericTypeConstraint &&
1088 Current.isNot(TT_CSharpGenericTypeConstraint)) {
1089 return CurrentState.ColonPos + 2;
1090 }
1091
1092 const FormatToken &Previous = *Current.Previous;
1093 // If we are continuing an expression, we want to use the continuation indent.
1094 unsigned ContinuationIndent =
1095 std::max(CurrentState.LastSpace, CurrentState.Indent) +
1096 Style.ContinuationIndentWidth;
1097 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
1098 const FormatToken *NextNonComment = Previous.getNextNonComment();
1099 if (!NextNonComment)
1100 NextNonComment = &Current;
1101
1102 // Java specific bits.
1103 if (Style.Language == FormatStyle::LK_Java &&
1104 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) {
1105 return std::max(CurrentState.LastSpace,
1106 CurrentState.Indent + Style.ContinuationIndentWidth);
1107 }
1108
1109 // After a goto label. Usually labels are on separate lines. However
1110 // for Verilog the labels may be only recognized by the annotator and
1111 // thus are on the same line as the current token.
1112 if ((Style.isVerilog() && Keywords.isVerilogEndOfLabel(Previous)) ||
1113 (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths &&
1114 State.Line->First->is(tok::kw_enum))) {
1115 return (Style.IndentWidth * State.Line->First->IndentLevel) +
1116 Style.IndentWidth;
1117 }
1118
1119 if (NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block))
1120 return Current.NestingLevel == 0 ? State.FirstIndent : CurrentState.Indent;
1121 if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
1122 (Current.is(tok::greater) &&
1123 (Style.Language == FormatStyle::LK_Proto ||
1124 Style.Language == FormatStyle::LK_TextProto))) &&
1125 State.Stack.size() > 1) {
1126 if (Current.closesBlockOrBlockTypeList(Style))
1127 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
1128 if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit))
1129 return State.Stack[State.Stack.size() - 2].LastSpace;
1130 return State.FirstIndent;
1131 }
1132 // Indent a closing parenthesis at the previous level if followed by a semi,
1133 // const, or opening brace. This allows indentations such as:
1134 // foo(
1135 // a,
1136 // );
1137 // int Foo::getter(
1138 // //
1139 // ) const {
1140 // return foo;
1141 // }
1142 // function foo(
1143 // a,
1144 // ) {
1145 // code(); //
1146 // }
1147 if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
1148 (!Current.Next ||
1149 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) {
1150 return State.Stack[State.Stack.size() - 2].LastSpace;
1151 }
1152 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent &&
1153 Current.is(tok::r_paren) && State.Stack.size() > 1) {
1154 return State.Stack[State.Stack.size() - 2].LastSpace;
1155 }
1156 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
1157 return State.Stack[State.Stack.size() - 2].LastSpace;
1158 if (Current.is(tok::identifier) && Current.Next &&
1159 (Current.Next->is(TT_DictLiteral) ||
1160 ((Style.Language == FormatStyle::LK_Proto ||
1161 Style.Language == FormatStyle::LK_TextProto) &&
1162 Current.Next->isOneOf(tok::less, tok::l_brace)))) {
1163 return CurrentState.Indent;
1164 }
1165 if (NextNonComment->is(TT_ObjCStringLiteral) &&
1166 State.StartOfStringLiteral != 0) {
1167 return State.StartOfStringLiteral - 1;
1168 }
1169 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
1170 return State.StartOfStringLiteral;
1171 if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0)
1172 return CurrentState.FirstLessLess;
1173 if (NextNonComment->isMemberAccess()) {
1174 if (CurrentState.CallContinuation == 0)
1175 return ContinuationIndent;
1176 return CurrentState.CallContinuation;
1177 }
1178 if (CurrentState.QuestionColumn != 0 &&
1179 ((NextNonComment->is(tok::colon) &&
1180 NextNonComment->is(TT_ConditionalExpr)) ||
1181 Previous.is(TT_ConditionalExpr))) {
1182 if (((NextNonComment->is(tok::colon) && NextNonComment->Next &&
1183 !NextNonComment->Next->FakeLParens.empty() &&
1184 NextNonComment->Next->FakeLParens.back() == prec::Conditional) ||
1185 (Previous.is(tok::colon) && !Current.FakeLParens.empty() &&
1186 Current.FakeLParens.back() == prec::Conditional)) &&
1187 !CurrentState.IsWrappedConditional) {
1188 // NOTE: we may tweak this slightly:
1189 // * not remove the 'lead' ContinuationIndentWidth
1190 // * always un-indent by the operator when
1191 // BreakBeforeTernaryOperators=true
1192 unsigned Indent = CurrentState.Indent;
1193 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1194 Indent -= Style.ContinuationIndentWidth;
1195 if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator)
1196 Indent -= 2;
1197 return Indent;
1198 }
1199 return CurrentState.QuestionColumn;
1200 }
1201 if (Previous.is(tok::comma) && CurrentState.VariablePos != 0)
1202 return CurrentState.VariablePos;
1203 if (Current.is(TT_RequiresClause)) {
1204 if (Style.IndentRequiresClause)
1205 return CurrentState.Indent + Style.IndentWidth;
1206 switch (Style.RequiresClausePosition) {
1207 case FormatStyle::RCPS_OwnLine:
1208 case FormatStyle::RCPS_WithFollowing:
1209 return CurrentState.Indent;
1210 default:
1211 break;
1212 }
1213 }
1214 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1215 TT_InheritanceComma)) {
1216 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1217 }
1218 if ((PreviousNonComment &&
1219 (PreviousNonComment->ClosesTemplateDeclaration ||
1220 PreviousNonComment->ClosesRequiresClause ||
1221 PreviousNonComment->isOneOf(
1222 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
1223 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
1224 (!Style.IndentWrappedFunctionNames &&
1225 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) {
1226 return std::max(CurrentState.LastSpace, CurrentState.Indent);
1227 }
1228 if (NextNonComment->is(TT_SelectorName)) {
1229 if (!CurrentState.ObjCSelectorNameFound) {
1230 unsigned MinIndent = CurrentState.Indent;
1231 if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) {
1232 MinIndent = std::max(MinIndent,
1233 State.FirstIndent + Style.ContinuationIndentWidth);
1234 }
1235 // If LongestObjCSelectorName is 0, we are indenting the first
1236 // part of an ObjC selector (or a selector component which is
1237 // not colon-aligned due to block formatting).
1238 //
1239 // Otherwise, we are indenting a subsequent part of an ObjC
1240 // selector which should be colon-aligned to the longest
1241 // component of the ObjC selector.
1242 //
1243 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1244 return MinIndent +
1245 std::max(NextNonComment->LongestObjCSelectorName,
1246 NextNonComment->ColumnWidth) -
1247 NextNonComment->ColumnWidth;
1248 }
1249 if (!CurrentState.AlignColons)
1250 return CurrentState.Indent;
1251 if (CurrentState.ColonPos > NextNonComment->ColumnWidth)
1252 return CurrentState.ColonPos - NextNonComment->ColumnWidth;
1253 return CurrentState.Indent;
1254 }
1255 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1256 return CurrentState.ColonPos;
1257 if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1258 if (CurrentState.StartOfArraySubscripts != 0) {
1259 return CurrentState.StartOfArraySubscripts;
1260 } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object
1261 // initializers.
1262 return CurrentState.Indent;
1263 }
1264 return ContinuationIndent;
1265 }
1266
1267 if (State.Line->InPragmaDirective)
1268 return CurrentState.Indent + Style.ContinuationIndentWidth;
1269
1270 // This ensure that we correctly format ObjC methods calls without inputs,
1271 // i.e. where the last element isn't selector like: [callee method];
1272 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1273 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) {
1274 return CurrentState.Indent;
1275 }
1276
1277 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1278 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) {
1279 return ContinuationIndent;
1280 }
1281 if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1282 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
1283 return ContinuationIndent;
1284 }
1285 if (NextNonComment->is(TT_CtorInitializerComma))
1286 return CurrentState.Indent;
1287 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1288 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1289 return CurrentState.Indent;
1290 }
1291 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1292 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
1293 return CurrentState.Indent;
1294 }
1295 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
1296 !Current.isOneOf(tok::colon, tok::comment)) {
1297 return ContinuationIndent;
1298 }
1299 if (Current.is(TT_ProtoExtensionLSquare))
1300 return CurrentState.Indent;
1301 if (Current.isBinaryOperator() && CurrentState.UnindentOperator) {
1302 return CurrentState.Indent - Current.Tok.getLength() -
1303 Current.SpacesRequiredBefore;
1304 }
1305 if (Current.isOneOf(tok::comment, TT_BlockComment, TT_LineComment) &&
1306 NextNonComment->isBinaryOperator() && CurrentState.UnindentOperator) {
1307 return CurrentState.Indent - NextNonComment->Tok.getLength() -
1308 NextNonComment->SpacesRequiredBefore;
1309 }
1310 if (CurrentState.Indent == State.FirstIndent && PreviousNonComment &&
1311 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) {
1312 // Ensure that we fall back to the continuation indent width instead of
1313 // just flushing continuations left.
1314 return CurrentState.Indent + Style.ContinuationIndentWidth;
1315 }
1316 return CurrentState.Indent;
1317 }
1318
hasNestedBlockInlined(const FormatToken * Previous,const FormatToken & Current,const FormatStyle & Style)1319 static bool hasNestedBlockInlined(const FormatToken *Previous,
1320 const FormatToken &Current,
1321 const FormatStyle &Style) {
1322 if (Previous->isNot(tok::l_paren))
1323 return true;
1324 if (Previous->ParameterCount > 1)
1325 return true;
1326
1327 // Also a nested block if contains a lambda inside function with 1 parameter.
1328 return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare);
1329 }
1330
moveStateToNextToken(LineState & State,bool DryRun,bool Newline)1331 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1332 bool DryRun, bool Newline) {
1333 assert(State.Stack.size());
1334 const FormatToken &Current = *State.NextToken;
1335 auto &CurrentState = State.Stack.back();
1336
1337 if (Current.is(TT_CSharpGenericTypeConstraint))
1338 CurrentState.IsCSharpGenericTypeConstraint = true;
1339 if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1340 CurrentState.NoLineBreakInOperand = false;
1341 if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon))
1342 CurrentState.AvoidBinPacking = true;
1343 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1344 if (CurrentState.FirstLessLess == 0)
1345 CurrentState.FirstLessLess = State.Column;
1346 else
1347 CurrentState.LastOperatorWrapped = Newline;
1348 }
1349 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1350 CurrentState.LastOperatorWrapped = Newline;
1351 if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1352 !Current.Previous->is(TT_ConditionalExpr)) {
1353 CurrentState.LastOperatorWrapped = Newline;
1354 }
1355 if (Current.is(TT_ArraySubscriptLSquare) &&
1356 CurrentState.StartOfArraySubscripts == 0) {
1357 CurrentState.StartOfArraySubscripts = State.Column;
1358 }
1359
1360 auto IsWrappedConditional = [](const FormatToken &Tok) {
1361 if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question)))
1362 return false;
1363 if (Tok.MustBreakBefore)
1364 return true;
1365
1366 const FormatToken *Next = Tok.getNextNonComment();
1367 return Next && Next->MustBreakBefore;
1368 };
1369 if (IsWrappedConditional(Current))
1370 CurrentState.IsWrappedConditional = true;
1371 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1372 CurrentState.QuestionColumn = State.Column;
1373 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1374 const FormatToken *Previous = Current.Previous;
1375 while (Previous && Previous->isTrailingComment())
1376 Previous = Previous->Previous;
1377 if (Previous && Previous->is(tok::question))
1378 CurrentState.QuestionColumn = State.Column;
1379 }
1380 if (!Current.opensScope() && !Current.closesScope() &&
1381 !Current.is(TT_PointerOrReference)) {
1382 State.LowestLevelOnLine =
1383 std::min(State.LowestLevelOnLine, Current.NestingLevel);
1384 }
1385 if (Current.isMemberAccess())
1386 CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column;
1387 if (Current.is(TT_SelectorName))
1388 CurrentState.ObjCSelectorNameFound = true;
1389 if (Current.is(TT_CtorInitializerColon) &&
1390 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1391 // Indent 2 from the column, so:
1392 // SomeClass::SomeClass()
1393 // : First(...), ...
1394 // Next(...)
1395 // ^ line up here.
1396 CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers ==
1397 FormatStyle::BCIS_BeforeComma
1398 ? 0
1399 : 2);
1400 CurrentState.NestedBlockIndent = CurrentState.Indent;
1401 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) {
1402 CurrentState.AvoidBinPacking = true;
1403 CurrentState.BreakBeforeParameter =
1404 Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine;
1405 } else {
1406 CurrentState.BreakBeforeParameter = false;
1407 }
1408 }
1409 if (Current.is(TT_CtorInitializerColon) &&
1410 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1411 CurrentState.Indent =
1412 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1413 CurrentState.NestedBlockIndent = CurrentState.Indent;
1414 if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack)
1415 CurrentState.AvoidBinPacking = true;
1416 }
1417 if (Current.is(TT_InheritanceColon)) {
1418 CurrentState.Indent =
1419 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1420 }
1421 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1422 CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1;
1423 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow))
1424 CurrentState.LastSpace = State.Column;
1425 if (Current.is(TT_RequiresExpression) &&
1426 Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) {
1427 CurrentState.NestedBlockIndent = State.Column;
1428 }
1429
1430 // Insert scopes created by fake parenthesis.
1431 const FormatToken *Previous = Current.getPreviousNonComment();
1432
1433 // Add special behavior to support a format commonly used for JavaScript
1434 // closures:
1435 // SomeFunction(function() {
1436 // foo();
1437 // bar();
1438 // }, a, b, c);
1439 if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause &&
1440 Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1441 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 &&
1442 !CurrentState.HasMultipleNestedBlocks) {
1443 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1444 for (ParenState &PState : llvm::drop_end(State.Stack))
1445 PState.NoLineBreak = true;
1446 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1447 }
1448 if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) ||
1449 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) &&
1450 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) {
1451 CurrentState.NestedBlockInlined =
1452 !Newline && hasNestedBlockInlined(Previous, Current, Style);
1453 }
1454
1455 moveStatePastFakeLParens(State, Newline);
1456 moveStatePastScopeCloser(State);
1457 // Do not use CurrentState here, since the two functions before may change the
1458 // Stack.
1459 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1460 !State.Stack.back().NoLineBreakInOperand;
1461 moveStatePastScopeOpener(State, Newline);
1462 moveStatePastFakeRParens(State);
1463
1464 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1465 State.StartOfStringLiteral = State.Column + 1;
1466 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) {
1467 State.StartOfStringLiteral = State.Column + 1;
1468 } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
1469 State.StartOfStringLiteral = State.Column;
1470 } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1471 !Current.isStringLiteral()) {
1472 State.StartOfStringLiteral = 0;
1473 }
1474
1475 State.Column += Current.ColumnWidth;
1476 State.NextToken = State.NextToken->Next;
1477
1478 unsigned Penalty =
1479 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1480
1481 if (Current.Role)
1482 Current.Role->formatFromToken(State, this, DryRun);
1483 // If the previous has a special role, let it consume tokens as appropriate.
1484 // It is necessary to start at the previous token for the only implemented
1485 // role (comma separated list). That way, the decision whether or not to break
1486 // after the "{" is already done and both options are tried and evaluated.
1487 // FIXME: This is ugly, find a better way.
1488 if (Previous && Previous->Role)
1489 Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1490
1491 return Penalty;
1492 }
1493
moveStatePastFakeLParens(LineState & State,bool Newline)1494 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1495 bool Newline) {
1496 const FormatToken &Current = *State.NextToken;
1497 if (Current.FakeLParens.empty())
1498 return;
1499
1500 const FormatToken *Previous = Current.getPreviousNonComment();
1501
1502 // Don't add extra indentation for the first fake parenthesis after
1503 // 'return', assignments, opening <({[, or requires clauses. The indentation
1504 // for these cases is special cased.
1505 bool SkipFirstExtraIndent =
1506 Previous &&
1507 (Previous->opensScope() ||
1508 Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) ||
1509 (Previous->getPrecedence() == prec::Assignment &&
1510 Style.AlignOperands != FormatStyle::OAS_DontAlign) ||
1511 Previous->is(TT_ObjCMethodExpr));
1512 for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) {
1513 const auto &CurrentState = State.Stack.back();
1514 ParenState NewParenState = CurrentState;
1515 NewParenState.Tok = nullptr;
1516 NewParenState.ContainsLineBreak = false;
1517 NewParenState.LastOperatorWrapped = true;
1518 NewParenState.IsChainedConditional = false;
1519 NewParenState.IsWrappedConditional = false;
1520 NewParenState.UnindentOperator = false;
1521 NewParenState.NoLineBreak =
1522 NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand;
1523
1524 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1525 if (PrecedenceLevel > prec::Comma)
1526 NewParenState.AvoidBinPacking = false;
1527
1528 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1529 // a builder type call after 'return' or, if the alignment after opening
1530 // brackets is disabled.
1531 if (!Current.isTrailingComment() &&
1532 (Style.AlignOperands != FormatStyle::OAS_DontAlign ||
1533 PrecedenceLevel < prec::Assignment) &&
1534 (!Previous || Previous->isNot(tok::kw_return) ||
1535 (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) &&
1536 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1537 PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) {
1538 NewParenState.Indent = std::max(
1539 std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace);
1540 }
1541
1542 // Special case for generic selection expressions, its comma-separated
1543 // expressions are not aligned to the opening paren like regular calls, but
1544 // rather continuation-indented relative to the _Generic keyword.
1545 if (Previous && Previous->endsSequence(tok::l_paren, tok::kw__Generic))
1546 NewParenState.Indent = CurrentState.LastSpace;
1547
1548 if (Previous &&
1549 (Previous->getPrecedence() == prec::Assignment ||
1550 Previous->isOneOf(tok::kw_return, TT_RequiresClause) ||
1551 (PrecedenceLevel == prec::Conditional && Previous->is(tok::question) &&
1552 Previous->is(TT_ConditionalExpr))) &&
1553 !Newline) {
1554 // If BreakBeforeBinaryOperators is set, un-indent a bit to account for
1555 // the operator and keep the operands aligned.
1556 if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator)
1557 NewParenState.UnindentOperator = true;
1558 // Mark indentation as alignment if the expression is aligned.
1559 if (Style.AlignOperands != FormatStyle::OAS_DontAlign)
1560 NewParenState.IsAligned = true;
1561 }
1562
1563 // Do not indent relative to the fake parentheses inserted for "." or "->".
1564 // This is a special case to make the following to statements consistent:
1565 // OuterFunction(InnerFunctionCall( // break
1566 // ParameterToInnerFunction));
1567 // OuterFunction(SomeObject.InnerFunctionCall( // break
1568 // ParameterToInnerFunction));
1569 if (PrecedenceLevel > prec::Unknown)
1570 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1571 if (PrecedenceLevel != prec::Conditional && !Current.is(TT_UnaryOperator) &&
1572 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) {
1573 NewParenState.StartOfFunctionCall = State.Column;
1574 }
1575
1576 // Indent conditional expressions, unless they are chained "else-if"
1577 // conditionals. Never indent expression where the 'operator' is ',', ';' or
1578 // an assignment (i.e. *I <= prec::Assignment) as those have different
1579 // indentation rules. Indent other expression, unless the indentation needs
1580 // to be skipped.
1581 if (PrecedenceLevel == prec::Conditional && Previous &&
1582 Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) &&
1583 &PrecedenceLevel == &Current.FakeLParens.back() &&
1584 !CurrentState.IsWrappedConditional) {
1585 NewParenState.IsChainedConditional = true;
1586 NewParenState.UnindentOperator = State.Stack.back().UnindentOperator;
1587 } else if (PrecedenceLevel == prec::Conditional ||
1588 (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment &&
1589 !Current.isTrailingComment())) {
1590 NewParenState.Indent += Style.ContinuationIndentWidth;
1591 }
1592 if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma)
1593 NewParenState.BreakBeforeParameter = false;
1594 State.Stack.push_back(NewParenState);
1595 SkipFirstExtraIndent = false;
1596 }
1597 }
1598
moveStatePastFakeRParens(LineState & State)1599 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1600 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1601 unsigned VariablePos = State.Stack.back().VariablePos;
1602 if (State.Stack.size() == 1) {
1603 // Do not pop the last element.
1604 break;
1605 }
1606 State.Stack.pop_back();
1607 State.Stack.back().VariablePos = VariablePos;
1608 }
1609
1610 if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) {
1611 // Remove the indentation of the requires clauses (which is not in Indent,
1612 // but in LastSpace).
1613 State.Stack.back().LastSpace -= Style.IndentWidth;
1614 }
1615 }
1616
moveStatePastScopeOpener(LineState & State,bool Newline)1617 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1618 bool Newline) {
1619 const FormatToken &Current = *State.NextToken;
1620 if (!Current.opensScope())
1621 return;
1622
1623 const auto &CurrentState = State.Stack.back();
1624
1625 // Don't allow '<' or '(' in C# generic type constraints to start new scopes.
1626 if (Current.isOneOf(tok::less, tok::l_paren) &&
1627 CurrentState.IsCSharpGenericTypeConstraint) {
1628 return;
1629 }
1630
1631 if (Current.MatchingParen && Current.is(BK_Block)) {
1632 moveStateToNewBlock(State);
1633 return;
1634 }
1635
1636 unsigned NewIndent;
1637 unsigned LastSpace = CurrentState.LastSpace;
1638 bool AvoidBinPacking;
1639 bool BreakBeforeParameter = false;
1640 unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall,
1641 CurrentState.NestedBlockIndent);
1642 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1643 opensProtoMessageField(Current, Style)) {
1644 if (Current.opensBlockOrBlockTypeList(Style)) {
1645 NewIndent = Style.IndentWidth +
1646 std::min(State.Column, CurrentState.NestedBlockIndent);
1647 } else {
1648 NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth;
1649 }
1650 const FormatToken *NextNoComment = Current.getNextNonComment();
1651 bool EndsInComma = Current.MatchingParen &&
1652 Current.MatchingParen->Previous &&
1653 Current.MatchingParen->Previous->is(tok::comma);
1654 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1655 Style.Language == FormatStyle::LK_Proto ||
1656 Style.Language == FormatStyle::LK_TextProto ||
1657 !Style.BinPackArguments ||
1658 (NextNoComment &&
1659 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod,
1660 TT_DesignatedInitializerLSquare));
1661 BreakBeforeParameter = EndsInComma;
1662 if (Current.ParameterCount > 1)
1663 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1664 } else {
1665 NewIndent =
1666 Style.ContinuationIndentWidth +
1667 std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall);
1668
1669 // Ensure that different different brackets force relative alignment, e.g.:
1670 // void SomeFunction(vector< // break
1671 // int> v);
1672 // FIXME: We likely want to do this for more combinations of brackets.
1673 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1674 NewIndent = std::max(NewIndent, CurrentState.Indent);
1675 LastSpace = std::max(LastSpace, CurrentState.Indent);
1676 }
1677
1678 bool EndsInComma =
1679 Current.MatchingParen &&
1680 Current.MatchingParen->getPreviousNonComment() &&
1681 Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1682
1683 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1684 // for backwards compatibility.
1685 bool ObjCBinPackProtocolList =
1686 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1687 Style.BinPackParameters) ||
1688 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1689
1690 bool BinPackDeclaration =
1691 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1692 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1693
1694 bool GenericSelection =
1695 Current.getPreviousNonComment() &&
1696 Current.getPreviousNonComment()->is(tok::kw__Generic);
1697
1698 AvoidBinPacking =
1699 (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection ||
1700 (Style.isJavaScript() && EndsInComma) ||
1701 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1702 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1703 (Style.ExperimentalAutoDetectBinPacking &&
1704 (Current.is(PPK_OnePerLine) ||
1705 (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive))));
1706
1707 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen &&
1708 Style.ObjCBreakBeforeNestedBlockParam) {
1709 if (Style.ColumnLimit) {
1710 // If this '[' opens an ObjC call, determine whether all parameters fit
1711 // into one line and put one per line if they don't.
1712 if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1713 getColumnLimit(State)) {
1714 BreakBeforeParameter = true;
1715 }
1716 } else {
1717 // For ColumnLimit = 0, we have to figure out whether there is or has to
1718 // be a line break within this call.
1719 for (const FormatToken *Tok = &Current;
1720 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1721 if (Tok->MustBreakBefore ||
1722 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1723 BreakBeforeParameter = true;
1724 break;
1725 }
1726 }
1727 }
1728 }
1729
1730 if (Style.isJavaScript() && EndsInComma)
1731 BreakBeforeParameter = true;
1732 }
1733 // Generally inherit NoLineBreak from the current scope to nested scope.
1734 // However, don't do this for non-empty nested blocks, dict literals and
1735 // array literals as these follow different indentation rules.
1736 bool NoLineBreak =
1737 Current.Children.empty() &&
1738 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1739 (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand ||
1740 (Current.is(TT_TemplateOpener) &&
1741 CurrentState.ContainsUnwrappedBuilder));
1742 State.Stack.push_back(
1743 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1744 auto &NewState = State.Stack.back();
1745 NewState.NestedBlockIndent = NestedBlockIndent;
1746 NewState.BreakBeforeParameter = BreakBeforeParameter;
1747 NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1);
1748
1749 if (Style.BraceWrapping.BeforeLambdaBody && Current.Next != nullptr &&
1750 Current.is(tok::l_paren)) {
1751 // Search for any parameter that is a lambda.
1752 FormatToken const *next = Current.Next;
1753 while (next != nullptr) {
1754 if (next->is(TT_LambdaLSquare)) {
1755 NewState.HasMultipleNestedBlocks = true;
1756 break;
1757 }
1758 next = next->Next;
1759 }
1760 }
1761
1762 NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) &&
1763 Current.Previous &&
1764 Current.Previous->is(tok::at);
1765 }
1766
moveStatePastScopeCloser(LineState & State)1767 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1768 const FormatToken &Current = *State.NextToken;
1769 if (!Current.closesScope())
1770 return;
1771
1772 // If we encounter a closing ), ], } or >, we can remove a level from our
1773 // stacks.
1774 if (State.Stack.size() > 1 &&
1775 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1776 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1777 State.NextToken->is(TT_TemplateCloser) ||
1778 (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) {
1779 State.Stack.pop_back();
1780 }
1781
1782 auto &CurrentState = State.Stack.back();
1783
1784 // Reevaluate whether ObjC message arguments fit into one line.
1785 // If a receiver spans multiple lines, e.g.:
1786 // [[object block:^{
1787 // return 42;
1788 // }] a:42 b:42];
1789 // BreakBeforeParameter is calculated based on an incorrect assumption
1790 // (it is checked whether the whole expression fits into one line without
1791 // considering a line break inside a message receiver).
1792 // We check whether arguments fit after receiver scope closer (into the same
1793 // line).
1794 if (CurrentState.BreakBeforeParameter && Current.MatchingParen &&
1795 Current.MatchingParen->Previous) {
1796 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
1797 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) &&
1798 CurrentScopeOpener.MatchingParen) {
1799 int NecessarySpaceInLine =
1800 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) +
1801 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
1802 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
1803 Style.ColumnLimit) {
1804 CurrentState.BreakBeforeParameter = false;
1805 }
1806 }
1807 }
1808
1809 if (Current.is(tok::r_square)) {
1810 // If this ends the array subscript expr, reset the corresponding value.
1811 const FormatToken *NextNonComment = Current.getNextNonComment();
1812 if (NextNonComment && NextNonComment->isNot(tok::l_square))
1813 CurrentState.StartOfArraySubscripts = 0;
1814 }
1815 }
1816
moveStateToNewBlock(LineState & State)1817 void ContinuationIndenter::moveStateToNewBlock(LineState &State) {
1818 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
1819 // ObjC block sometimes follow special indentation rules.
1820 unsigned NewIndent =
1821 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
1822 ? Style.ObjCBlockIndentWidth
1823 : Style.IndentWidth);
1824 State.Stack.push_back(ParenState(State.NextToken, NewIndent,
1825 State.Stack.back().LastSpace,
1826 /*AvoidBinPacking=*/true,
1827 /*NoLineBreak=*/false));
1828 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1829 State.Stack.back().BreakBeforeParameter = true;
1830 }
1831
getLastLineEndColumn(StringRef Text,unsigned StartColumn,unsigned TabWidth,encoding::Encoding Encoding)1832 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
1833 unsigned TabWidth,
1834 encoding::Encoding Encoding) {
1835 size_t LastNewlinePos = Text.find_last_of("\n");
1836 if (LastNewlinePos == StringRef::npos) {
1837 return StartColumn +
1838 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
1839 } else {
1840 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
1841 /*StartColumn=*/0, TabWidth, Encoding);
1842 }
1843 }
1844
reformatRawStringLiteral(const FormatToken & Current,LineState & State,const FormatStyle & RawStringStyle,bool DryRun,bool Newline)1845 unsigned ContinuationIndenter::reformatRawStringLiteral(
1846 const FormatToken &Current, LineState &State,
1847 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
1848 unsigned StartColumn = State.Column - Current.ColumnWidth;
1849 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
1850 StringRef NewDelimiter =
1851 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
1852 if (NewDelimiter.empty())
1853 NewDelimiter = OldDelimiter;
1854 // The text of a raw string is between the leading 'R"delimiter(' and the
1855 // trailing 'delimiter)"'.
1856 unsigned OldPrefixSize = 3 + OldDelimiter.size();
1857 unsigned OldSuffixSize = 2 + OldDelimiter.size();
1858 // We create a virtual text environment which expects a null-terminated
1859 // string, so we cannot use StringRef.
1860 std::string RawText = std::string(
1861 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize));
1862 if (NewDelimiter != OldDelimiter) {
1863 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
1864 // raw string.
1865 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
1866 if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
1867 NewDelimiter = OldDelimiter;
1868 }
1869
1870 unsigned NewPrefixSize = 3 + NewDelimiter.size();
1871 unsigned NewSuffixSize = 2 + NewDelimiter.size();
1872
1873 // The first start column is the column the raw text starts after formatting.
1874 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
1875
1876 // The next start column is the intended indentation a line break inside
1877 // the raw string at level 0. It is determined by the following rules:
1878 // - if the content starts on newline, it is one level more than the current
1879 // indent, and
1880 // - if the content does not start on a newline, it is the first start
1881 // column.
1882 // These rules have the advantage that the formatted content both does not
1883 // violate the rectangle rule and visually flows within the surrounding
1884 // source.
1885 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
1886 // If this token is the last parameter (checked by looking if it's followed by
1887 // `)` and is not on a newline, the base the indent off the line's nested
1888 // block indent. Otherwise, base the indent off the arguments indent, so we
1889 // can achieve:
1890 //
1891 // fffffffffff(1, 2, 3, R"pb(
1892 // key1: 1 #
1893 // key2: 2)pb");
1894 //
1895 // fffffffffff(1, 2, 3,
1896 // R"pb(
1897 // key1: 1 #
1898 // key2: 2
1899 // )pb");
1900 //
1901 // fffffffffff(1, 2, 3,
1902 // R"pb(
1903 // key1: 1 #
1904 // key2: 2
1905 // )pb",
1906 // 5);
1907 unsigned CurrentIndent =
1908 (!Newline && Current.Next && Current.Next->is(tok::r_paren))
1909 ? State.Stack.back().NestedBlockIndent
1910 : State.Stack.back().Indent;
1911 unsigned NextStartColumn = ContentStartsOnNewline
1912 ? CurrentIndent + Style.IndentWidth
1913 : FirstStartColumn;
1914
1915 // The last start column is the column the raw string suffix starts if it is
1916 // put on a newline.
1917 // The last start column is the intended indentation of the raw string postfix
1918 // if it is put on a newline. It is determined by the following rules:
1919 // - if the raw string prefix starts on a newline, it is the column where
1920 // that raw string prefix starts, and
1921 // - if the raw string prefix does not start on a newline, it is the current
1922 // indent.
1923 unsigned LastStartColumn =
1924 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
1925
1926 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
1927 RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
1928 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
1929 /*Status=*/nullptr);
1930
1931 auto NewCode = applyAllReplacements(RawText, Fixes.first);
1932 tooling::Replacements NoFixes;
1933 if (!NewCode)
1934 return addMultilineToken(Current, State);
1935 if (!DryRun) {
1936 if (NewDelimiter != OldDelimiter) {
1937 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
1938 // of the token.
1939 SourceLocation PrefixDelimiterStart =
1940 Current.Tok.getLocation().getLocWithOffset(2);
1941 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
1942 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1943 if (PrefixErr) {
1944 llvm::errs()
1945 << "Failed to update the prefix delimiter of a raw string: "
1946 << llvm::toString(std::move(PrefixErr)) << "\n";
1947 }
1948 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
1949 // position length - 1 - |delimiter|.
1950 SourceLocation SuffixDelimiterStart =
1951 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
1952 1 - OldDelimiter.size());
1953 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
1954 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1955 if (SuffixErr) {
1956 llvm::errs()
1957 << "Failed to update the suffix delimiter of a raw string: "
1958 << llvm::toString(std::move(SuffixErr)) << "\n";
1959 }
1960 }
1961 SourceLocation OriginLoc =
1962 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
1963 for (const tooling::Replacement &Fix : Fixes.first) {
1964 auto Err = Whitespaces.addReplacement(tooling::Replacement(
1965 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
1966 Fix.getLength(), Fix.getReplacementText()));
1967 if (Err) {
1968 llvm::errs() << "Failed to reformat raw string: "
1969 << llvm::toString(std::move(Err)) << "\n";
1970 }
1971 }
1972 }
1973 unsigned RawLastLineEndColumn = getLastLineEndColumn(
1974 *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
1975 State.Column = RawLastLineEndColumn + NewSuffixSize;
1976 // Since we're updating the column to after the raw string literal here, we
1977 // have to manually add the penalty for the prefix R"delim( over the column
1978 // limit.
1979 unsigned PrefixExcessCharacters =
1980 StartColumn + NewPrefixSize > Style.ColumnLimit
1981 ? StartColumn + NewPrefixSize - Style.ColumnLimit
1982 : 0;
1983 bool IsMultiline =
1984 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos);
1985 if (IsMultiline) {
1986 // Break before further function parameters on all levels.
1987 for (ParenState &Paren : State.Stack)
1988 Paren.BreakBeforeParameter = true;
1989 }
1990 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
1991 }
1992
addMultilineToken(const FormatToken & Current,LineState & State)1993 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
1994 LineState &State) {
1995 // Break before further function parameters on all levels.
1996 for (ParenState &Paren : State.Stack)
1997 Paren.BreakBeforeParameter = true;
1998
1999 unsigned ColumnsUsed = State.Column;
2000 // We can only affect layout of the first and the last line, so the penalty
2001 // for all other lines is constant, and we ignore it.
2002 State.Column = Current.LastLineColumnWidth;
2003
2004 if (ColumnsUsed > getColumnLimit(State))
2005 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
2006 return 0;
2007 }
2008
handleEndOfLine(const FormatToken & Current,LineState & State,bool DryRun,bool AllowBreak,bool Newline)2009 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
2010 LineState &State, bool DryRun,
2011 bool AllowBreak, bool Newline) {
2012 unsigned Penalty = 0;
2013 // Compute the raw string style to use in case this is a raw string literal
2014 // that can be reformatted.
2015 auto RawStringStyle = getRawStringStyle(Current, State);
2016 if (RawStringStyle && !Current.Finalized) {
2017 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun,
2018 Newline);
2019 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
2020 // Don't break multi-line tokens other than block comments and raw string
2021 // literals. Instead, just update the state.
2022 Penalty = addMultilineToken(Current, State);
2023 } else if (State.Line->Type != LT_ImportStatement) {
2024 // We generally don't break import statements.
2025 LineState OriginalState = State;
2026
2027 // Whether we force the reflowing algorithm to stay strictly within the
2028 // column limit.
2029 bool Strict = false;
2030 // Whether the first non-strict attempt at reflowing did intentionally
2031 // exceed the column limit.
2032 bool Exceeded = false;
2033 std::tie(Penalty, Exceeded) = breakProtrudingToken(
2034 Current, State, AllowBreak, /*DryRun=*/true, Strict);
2035 if (Exceeded) {
2036 // If non-strict reflowing exceeds the column limit, try whether strict
2037 // reflowing leads to an overall lower penalty.
2038 LineState StrictState = OriginalState;
2039 unsigned StrictPenalty =
2040 breakProtrudingToken(Current, StrictState, AllowBreak,
2041 /*DryRun=*/true, /*Strict=*/true)
2042 .first;
2043 Strict = StrictPenalty <= Penalty;
2044 if (Strict) {
2045 Penalty = StrictPenalty;
2046 State = StrictState;
2047 }
2048 }
2049 if (!DryRun) {
2050 // If we're not in dry-run mode, apply the changes with the decision on
2051 // strictness made above.
2052 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
2053 Strict);
2054 }
2055 }
2056 if (State.Column > getColumnLimit(State)) {
2057 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
2058 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
2059 }
2060 return Penalty;
2061 }
2062
2063 // Returns the enclosing function name of a token, or the empty string if not
2064 // found.
getEnclosingFunctionName(const FormatToken & Current)2065 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
2066 // Look for: 'function(' or 'function<templates>(' before Current.
2067 auto Tok = Current.getPreviousNonComment();
2068 if (!Tok || !Tok->is(tok::l_paren))
2069 return "";
2070 Tok = Tok->getPreviousNonComment();
2071 if (!Tok)
2072 return "";
2073 if (Tok->is(TT_TemplateCloser)) {
2074 Tok = Tok->MatchingParen;
2075 if (Tok)
2076 Tok = Tok->getPreviousNonComment();
2077 }
2078 if (!Tok || !Tok->is(tok::identifier))
2079 return "";
2080 return Tok->TokenText;
2081 }
2082
2083 std::optional<FormatStyle>
getRawStringStyle(const FormatToken & Current,const LineState & State)2084 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
2085 const LineState &State) {
2086 if (!Current.isStringLiteral())
2087 return std::nullopt;
2088 auto Delimiter = getRawStringDelimiter(Current.TokenText);
2089 if (!Delimiter)
2090 return std::nullopt;
2091 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
2092 if (!RawStringStyle && Delimiter->empty()) {
2093 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
2094 getEnclosingFunctionName(Current));
2095 }
2096 if (!RawStringStyle)
2097 return std::nullopt;
2098 RawStringStyle->ColumnLimit = getColumnLimit(State);
2099 return RawStringStyle;
2100 }
2101
2102 std::unique_ptr<BreakableToken>
createBreakableToken(const FormatToken & Current,LineState & State,bool AllowBreak)2103 ContinuationIndenter::createBreakableToken(const FormatToken &Current,
2104 LineState &State, bool AllowBreak) {
2105 unsigned StartColumn = State.Column - Current.ColumnWidth;
2106 if (Current.isStringLiteral()) {
2107 // FIXME: String literal breaking is currently disabled for C#, Java, Json
2108 // and JavaScript, as it requires strings to be merged using "+" which we
2109 // don't support.
2110 if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
2111 Style.isCSharp() || Style.isJson() || !Style.BreakStringLiterals ||
2112 !AllowBreak) {
2113 return nullptr;
2114 }
2115
2116 // Don't break string literals inside preprocessor directives (except for
2117 // #define directives, as their contents are stored in separate lines and
2118 // are not affected by this check).
2119 // This way we avoid breaking code with line directives and unknown
2120 // preprocessor directives that contain long string literals.
2121 if (State.Line->Type == LT_PreprocessorDirective)
2122 return nullptr;
2123 // Exempts unterminated string literals from line breaking. The user will
2124 // likely want to terminate the string before any line breaking is done.
2125 if (Current.IsUnterminatedLiteral)
2126 return nullptr;
2127 // Don't break string literals inside Objective-C array literals (doing so
2128 // raises the warning -Wobjc-string-concatenation).
2129 if (State.Stack.back().IsInsideObjCArrayLiteral)
2130 return nullptr;
2131
2132 StringRef Text = Current.TokenText;
2133 StringRef Prefix;
2134 StringRef Postfix;
2135 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
2136 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
2137 // reduce the overhead) for each FormatToken, which is a string, so that we
2138 // don't run multiple checks here on the hot path.
2139 if ((Text.endswith(Postfix = "\"") &&
2140 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") ||
2141 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") ||
2142 Text.startswith(Prefix = "u8\"") ||
2143 Text.startswith(Prefix = "L\""))) ||
2144 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) {
2145 // We need this to address the case where there is an unbreakable tail
2146 // only if certain other formatting decisions have been taken. The
2147 // UnbreakableTailLength of Current is an overapproximation is that case
2148 // and we need to be correct here.
2149 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
2150 ? 0
2151 : Current.UnbreakableTailLength;
2152 return std::make_unique<BreakableStringLiteral>(
2153 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
2154 State.Line->InPPDirective, Encoding, Style);
2155 }
2156 } else if (Current.is(TT_BlockComment)) {
2157 if (!Style.ReflowComments ||
2158 // If a comment token switches formatting, like
2159 // /* clang-format on */, we don't want to break it further,
2160 // but we may still want to adjust its indentation.
2161 switchesFormatting(Current)) {
2162 return nullptr;
2163 }
2164 return std::make_unique<BreakableBlockComment>(
2165 Current, StartColumn, Current.OriginalColumn, !Current.Previous,
2166 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF());
2167 } else if (Current.is(TT_LineComment) &&
2168 (Current.Previous == nullptr ||
2169 Current.Previous->isNot(TT_ImplicitStringLiteral))) {
2170 bool RegularComments = [&]() {
2171 for (const FormatToken *T = &Current; T && T->is(TT_LineComment);
2172 T = T->Next) {
2173 if (!(T->TokenText.startswith("//") || T->TokenText.startswith("#")))
2174 return false;
2175 }
2176 return true;
2177 }();
2178 if (!Style.ReflowComments ||
2179 CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
2180 switchesFormatting(Current) || !RegularComments) {
2181 return nullptr;
2182 }
2183 return std::make_unique<BreakableLineCommentSection>(
2184 Current, StartColumn, /*InPPDirective=*/false, Encoding, Style);
2185 }
2186 return nullptr;
2187 }
2188
2189 std::pair<unsigned, bool>
breakProtrudingToken(const FormatToken & Current,LineState & State,bool AllowBreak,bool DryRun,bool Strict)2190 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
2191 LineState &State, bool AllowBreak,
2192 bool DryRun, bool Strict) {
2193 std::unique_ptr<const BreakableToken> Token =
2194 createBreakableToken(Current, State, AllowBreak);
2195 if (!Token)
2196 return {0, false};
2197 assert(Token->getLineCount() > 0);
2198 unsigned ColumnLimit = getColumnLimit(State);
2199 if (Current.is(TT_LineComment)) {
2200 // We don't insert backslashes when breaking line comments.
2201 ColumnLimit = Style.ColumnLimit;
2202 }
2203 if (ColumnLimit == 0) {
2204 // To make the rest of the function easier set the column limit to the
2205 // maximum, if there should be no limit.
2206 ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max();
2207 }
2208 if (Current.UnbreakableTailLength >= ColumnLimit)
2209 return {0, false};
2210 // ColumnWidth was already accounted into State.Column before calling
2211 // breakProtrudingToken.
2212 unsigned StartColumn = State.Column - Current.ColumnWidth;
2213 unsigned NewBreakPenalty = Current.isStringLiteral()
2214 ? Style.PenaltyBreakString
2215 : Style.PenaltyBreakComment;
2216 // Stores whether we intentionally decide to let a line exceed the column
2217 // limit.
2218 bool Exceeded = false;
2219 // Stores whether we introduce a break anywhere in the token.
2220 bool BreakInserted = Token->introducesBreakBeforeToken();
2221 // Store whether we inserted a new line break at the end of the previous
2222 // logical line.
2223 bool NewBreakBefore = false;
2224 // We use a conservative reflowing strategy. Reflow starts after a line is
2225 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
2226 // line that doesn't get reflown with the previous line is reached.
2227 bool Reflow = false;
2228 // Keep track of where we are in the token:
2229 // Where we are in the content of the current logical line.
2230 unsigned TailOffset = 0;
2231 // The column number we're currently at.
2232 unsigned ContentStartColumn =
2233 Token->getContentStartColumn(0, /*Break=*/false);
2234 // The number of columns left in the current logical line after TailOffset.
2235 unsigned RemainingTokenColumns =
2236 Token->getRemainingLength(0, TailOffset, ContentStartColumn);
2237 // Adapt the start of the token, for example indent.
2238 if (!DryRun)
2239 Token->adaptStartOfLine(0, Whitespaces);
2240
2241 unsigned ContentIndent = 0;
2242 unsigned Penalty = 0;
2243 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
2244 << StartColumn << ".\n");
2245 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
2246 LineIndex != EndIndex; ++LineIndex) {
2247 LLVM_DEBUG(llvm::dbgs()
2248 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
2249 NewBreakBefore = false;
2250 // If we did reflow the previous line, we'll try reflowing again. Otherwise
2251 // we'll start reflowing if the current line is broken or whitespace is
2252 // compressed.
2253 bool TryReflow = Reflow;
2254 // Break the current token until we can fit the rest of the line.
2255 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2256 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
2257 << (ContentStartColumn + RemainingTokenColumns)
2258 << ", space: " << ColumnLimit
2259 << ", reflown prefix: " << ContentStartColumn
2260 << ", offset in line: " << TailOffset << "\n");
2261 // If the current token doesn't fit, find the latest possible split in the
2262 // current line so that breaking at it will be under the column limit.
2263 // FIXME: Use the earliest possible split while reflowing to correctly
2264 // compress whitespace within a line.
2265 BreakableToken::Split Split =
2266 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
2267 ContentStartColumn, CommentPragmasRegex);
2268 if (Split.first == StringRef::npos) {
2269 // No break opportunity - update the penalty and continue with the next
2270 // logical line.
2271 if (LineIndex < EndIndex - 1) {
2272 // The last line's penalty is handled in addNextStateToQueue() or when
2273 // calling replaceWhitespaceAfterLastLine below.
2274 Penalty += Style.PenaltyExcessCharacter *
2275 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2276 }
2277 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
2278 break;
2279 }
2280 assert(Split.first != 0);
2281
2282 if (Token->supportsReflow()) {
2283 // Check whether the next natural split point after the current one can
2284 // still fit the line, either because we can compress away whitespace,
2285 // or because the penalty the excess characters introduce is lower than
2286 // the break penalty.
2287 // We only do this for tokens that support reflowing, and thus allow us
2288 // to change the whitespace arbitrarily (e.g. comments).
2289 // Other tokens, like string literals, can be broken on arbitrary
2290 // positions.
2291
2292 // First, compute the columns from TailOffset to the next possible split
2293 // position.
2294 // For example:
2295 // ColumnLimit: |
2296 // // Some text that breaks
2297 // ^ tail offset
2298 // ^-- split
2299 // ^-------- to split columns
2300 // ^--- next split
2301 // ^--------------- to next split columns
2302 unsigned ToSplitColumns = Token->getRangeLength(
2303 LineIndex, TailOffset, Split.first, ContentStartColumn);
2304 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
2305
2306 BreakableToken::Split NextSplit = Token->getSplit(
2307 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
2308 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
2309 // Compute the columns necessary to fit the next non-breakable sequence
2310 // into the current line.
2311 unsigned ToNextSplitColumns = 0;
2312 if (NextSplit.first == StringRef::npos) {
2313 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
2314 ContentStartColumn);
2315 } else {
2316 ToNextSplitColumns = Token->getRangeLength(
2317 LineIndex, TailOffset,
2318 Split.first + Split.second + NextSplit.first, ContentStartColumn);
2319 }
2320 // Compress the whitespace between the break and the start of the next
2321 // unbreakable sequence.
2322 ToNextSplitColumns =
2323 Token->getLengthAfterCompression(ToNextSplitColumns, Split);
2324 LLVM_DEBUG(llvm::dbgs()
2325 << " ContentStartColumn: " << ContentStartColumn << "\n");
2326 LLVM_DEBUG(llvm::dbgs()
2327 << " ToNextSplit: " << ToNextSplitColumns << "\n");
2328 // If the whitespace compression makes us fit, continue on the current
2329 // line.
2330 bool ContinueOnLine =
2331 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
2332 unsigned ExcessCharactersPenalty = 0;
2333 if (!ContinueOnLine && !Strict) {
2334 // Similarly, if the excess characters' penalty is lower than the
2335 // penalty of introducing a new break, continue on the current line.
2336 ExcessCharactersPenalty =
2337 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
2338 Style.PenaltyExcessCharacter;
2339 LLVM_DEBUG(llvm::dbgs()
2340 << " Penalty excess: " << ExcessCharactersPenalty
2341 << "\n break : " << NewBreakPenalty << "\n");
2342 if (ExcessCharactersPenalty < NewBreakPenalty) {
2343 Exceeded = true;
2344 ContinueOnLine = true;
2345 }
2346 }
2347 if (ContinueOnLine) {
2348 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
2349 // The current line fits after compressing the whitespace - reflow
2350 // the next line into it if possible.
2351 TryReflow = true;
2352 if (!DryRun) {
2353 Token->compressWhitespace(LineIndex, TailOffset, Split,
2354 Whitespaces);
2355 }
2356 // When we continue on the same line, leave one space between content.
2357 ContentStartColumn += ToSplitColumns + 1;
2358 Penalty += ExcessCharactersPenalty;
2359 TailOffset += Split.first + Split.second;
2360 RemainingTokenColumns = Token->getRemainingLength(
2361 LineIndex, TailOffset, ContentStartColumn);
2362 continue;
2363 }
2364 }
2365 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
2366 // Update the ContentIndent only if the current line was not reflown with
2367 // the previous line, since in that case the previous line should still
2368 // determine the ContentIndent. Also never intent the last line.
2369 if (!Reflow)
2370 ContentIndent = Token->getContentIndent(LineIndex);
2371 LLVM_DEBUG(llvm::dbgs()
2372 << " ContentIndent: " << ContentIndent << "\n");
2373 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2374 LineIndex, /*Break=*/true);
2375
2376 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2377 LineIndex, TailOffset + Split.first + Split.second,
2378 ContentStartColumn);
2379 if (NewRemainingTokenColumns == 0) {
2380 // No content to indent.
2381 ContentIndent = 0;
2382 ContentStartColumn =
2383 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2384 NewRemainingTokenColumns = Token->getRemainingLength(
2385 LineIndex, TailOffset + Split.first + Split.second,
2386 ContentStartColumn);
2387 }
2388
2389 // When breaking before a tab character, it may be moved by a few columns,
2390 // but will still be expanded to the next tab stop, so we don't save any
2391 // columns.
2392 if (NewRemainingTokenColumns >= RemainingTokenColumns) {
2393 // FIXME: Do we need to adjust the penalty?
2394 break;
2395 }
2396
2397 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2398 << ", " << Split.second << "\n");
2399 if (!DryRun) {
2400 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2401 Whitespaces);
2402 }
2403
2404 Penalty += NewBreakPenalty;
2405 TailOffset += Split.first + Split.second;
2406 RemainingTokenColumns = NewRemainingTokenColumns;
2407 BreakInserted = true;
2408 NewBreakBefore = true;
2409 }
2410 // In case there's another line, prepare the state for the start of the next
2411 // line.
2412 if (LineIndex + 1 != EndIndex) {
2413 unsigned NextLineIndex = LineIndex + 1;
2414 if (NewBreakBefore) {
2415 // After breaking a line, try to reflow the next line into the current
2416 // one once RemainingTokenColumns fits.
2417 TryReflow = true;
2418 }
2419 if (TryReflow) {
2420 // We decided that we want to try reflowing the next line into the
2421 // current one.
2422 // We will now adjust the state as if the reflow is successful (in
2423 // preparation for the next line), and see whether that works. If we
2424 // decide that we cannot reflow, we will later reset the state to the
2425 // start of the next line.
2426 Reflow = false;
2427 // As we did not continue breaking the line, RemainingTokenColumns is
2428 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2429 // the position at which we want to format the next line if we do
2430 // actually reflow.
2431 // When we reflow, we need to add a space between the end of the current
2432 // line and the next line's start column.
2433 ContentStartColumn += RemainingTokenColumns + 1;
2434 // Get the split that we need to reflow next logical line into the end
2435 // of the current one; the split will include any leading whitespace of
2436 // the next logical line.
2437 BreakableToken::Split SplitBeforeNext =
2438 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
2439 LLVM_DEBUG(llvm::dbgs()
2440 << " Size of reflown text: " << ContentStartColumn
2441 << "\n Potential reflow split: ");
2442 if (SplitBeforeNext.first != StringRef::npos) {
2443 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2444 << SplitBeforeNext.second << "\n");
2445 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2446 // If the rest of the next line fits into the current line below the
2447 // column limit, we can safely reflow.
2448 RemainingTokenColumns = Token->getRemainingLength(
2449 NextLineIndex, TailOffset, ContentStartColumn);
2450 Reflow = true;
2451 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2452 LLVM_DEBUG(llvm::dbgs()
2453 << " Over limit after reflow, need: "
2454 << (ContentStartColumn + RemainingTokenColumns)
2455 << ", space: " << ColumnLimit
2456 << ", reflown prefix: " << ContentStartColumn
2457 << ", offset in line: " << TailOffset << "\n");
2458 // If the whole next line does not fit, try to find a point in
2459 // the next line at which we can break so that attaching the part
2460 // of the next line to that break point onto the current line is
2461 // below the column limit.
2462 BreakableToken::Split Split =
2463 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
2464 ContentStartColumn, CommentPragmasRegex);
2465 if (Split.first == StringRef::npos) {
2466 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
2467 Reflow = false;
2468 } else {
2469 // Check whether the first split point gets us below the column
2470 // limit. Note that we will execute this split below as part of
2471 // the normal token breaking and reflow logic within the line.
2472 unsigned ToSplitColumns = Token->getRangeLength(
2473 NextLineIndex, TailOffset, Split.first, ContentStartColumn);
2474 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2475 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
2476 << (ContentStartColumn + ToSplitColumns)
2477 << ", space: " << ColumnLimit);
2478 unsigned ExcessCharactersPenalty =
2479 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2480 Style.PenaltyExcessCharacter;
2481 if (NewBreakPenalty < ExcessCharactersPenalty)
2482 Reflow = false;
2483 }
2484 }
2485 }
2486 } else {
2487 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2488 }
2489 }
2490 if (!Reflow) {
2491 // If we didn't reflow into the next line, the only space to consider is
2492 // the next logical line. Reset our state to match the start of the next
2493 // line.
2494 TailOffset = 0;
2495 ContentStartColumn =
2496 Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2497 RemainingTokenColumns = Token->getRemainingLength(
2498 NextLineIndex, TailOffset, ContentStartColumn);
2499 // Adapt the start of the token, for example indent.
2500 if (!DryRun)
2501 Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2502 } else {
2503 // If we found a reflow split and have added a new break before the next
2504 // line, we are going to remove the line break at the start of the next
2505 // logical line. For example, here we'll add a new line break after
2506 // 'text', and subsequently delete the line break between 'that' and
2507 // 'reflows'.
2508 // // some text that
2509 // // reflows
2510 // ->
2511 // // some text
2512 // // that reflows
2513 // When adding the line break, we also added the penalty for it, so we
2514 // need to subtract that penalty again when we remove the line break due
2515 // to reflowing.
2516 if (NewBreakBefore) {
2517 assert(Penalty >= NewBreakPenalty);
2518 Penalty -= NewBreakPenalty;
2519 }
2520 if (!DryRun)
2521 Token->reflow(NextLineIndex, Whitespaces);
2522 }
2523 }
2524 }
2525
2526 BreakableToken::Split SplitAfterLastLine =
2527 Token->getSplitAfterLastLine(TailOffset);
2528 if (SplitAfterLastLine.first != StringRef::npos) {
2529 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2530
2531 // We add the last line's penalty here, since that line is going to be split
2532 // now.
2533 Penalty += Style.PenaltyExcessCharacter *
2534 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2535
2536 if (!DryRun) {
2537 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2538 Whitespaces);
2539 }
2540 ContentStartColumn =
2541 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2542 RemainingTokenColumns = Token->getRemainingLength(
2543 Token->getLineCount() - 1,
2544 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2545 ContentStartColumn);
2546 }
2547
2548 State.Column = ContentStartColumn + RemainingTokenColumns -
2549 Current.UnbreakableTailLength;
2550
2551 if (BreakInserted) {
2552 // If we break the token inside a parameter list, we need to break before
2553 // the next parameter on all levels, so that the next parameter is clearly
2554 // visible. Line comments already introduce a break.
2555 if (Current.isNot(TT_LineComment))
2556 for (ParenState &Paren : State.Stack)
2557 Paren.BreakBeforeParameter = true;
2558
2559 if (Current.is(TT_BlockComment))
2560 State.NoContinuation = true;
2561
2562 State.Stack.back().LastSpace = StartColumn;
2563 }
2564
2565 Token->updateNextToken(State);
2566
2567 return {Penalty, Exceeded};
2568 }
2569
getColumnLimit(const LineState & State) const2570 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2571 // In preprocessor directives reserve two chars for trailing " \".
2572 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2573 }
2574
nextIsMultilineString(const LineState & State)2575 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2576 const FormatToken &Current = *State.NextToken;
2577 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2578 return false;
2579 // We never consider raw string literals "multiline" for the purpose of
2580 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2581 // (see TokenAnnotator::mustBreakBefore().
2582 if (Current.TokenText.startswith("R\""))
2583 return false;
2584 if (Current.IsMultiline)
2585 return true;
2586 if (Current.getNextNonComment() &&
2587 Current.getNextNonComment()->isStringLiteral()) {
2588 return true; // Implicit concatenation.
2589 }
2590 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2591 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2592 Style.ColumnLimit) {
2593 return true; // String will be split.
2594 }
2595 return false;
2596 }
2597
2598 } // namespace format
2599 } // namespace clang
2600