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 "WhitespaceManager.h"
18 #include "clang/Basic/OperatorPrecedence.h"
19 #include "clang/Basic/SourceManager.h"
20 #include "clang/Format/Format.h"
21 #include "llvm/Support/Debug.h"
22
23 #define DEBUG_TYPE "format-indenter"
24
25 namespace clang {
26 namespace format {
27
28 // Returns true if a TT_SelectorName should be indented when wrapped,
29 // false otherwise.
shouldIndentWrappedSelectorName(const FormatStyle & Style,LineType LineType)30 static bool shouldIndentWrappedSelectorName(const FormatStyle &Style,
31 LineType LineType) {
32 return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl;
33 }
34
35 // Returns the length of everything up to the first possible line break after
36 // the ), ], } or > matching \c Tok.
getLengthToMatchingParen(const FormatToken & Tok,const std::vector<ParenState> & Stack)37 static unsigned getLengthToMatchingParen(const FormatToken &Tok,
38 const std::vector<ParenState> &Stack) {
39 // Normally whether or not a break before T is possible is calculated and
40 // stored in T.CanBreakBefore. Braces, array initializers and text proto
41 // messages like `key: < ... >` are an exception: a break is possible
42 // before a closing brace R if a break was inserted after the corresponding
43 // opening brace. The information about whether or not a break is needed
44 // before a closing brace R is stored in the ParenState field
45 // S.BreakBeforeClosingBrace where S is the state that R closes.
46 //
47 // In order to decide whether there can be a break before encountered right
48 // braces, this implementation iterates over the sequence of tokens and over
49 // the paren stack in lockstep, keeping track of the stack level which visited
50 // right braces correspond to in MatchingStackIndex.
51 //
52 // For example, consider:
53 // L. <- line number
54 // 1. {
55 // 2. {1},
56 // 3. {2},
57 // 4. {{3}}}
58 // ^ where we call this method with this token.
59 // The paren stack at this point contains 3 brace levels:
60 // 0. { at line 1, BreakBeforeClosingBrace: true
61 // 1. first { at line 4, BreakBeforeClosingBrace: false
62 // 2. second { at line 4, BreakBeforeClosingBrace: false,
63 // where there might be fake parens levels in-between these levels.
64 // The algorithm will start at the first } on line 4, which is the matching
65 // brace of the initial left brace and at level 2 of the stack. Then,
66 // examining BreakBeforeClosingBrace: false at level 2, it will continue to
67 // the second } on line 4, and will traverse the stack downwards until it
68 // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace:
69 // false at level 1, it will continue to the third } on line 4 and will
70 // traverse the stack downwards until it finds the matching { on level 0.
71 // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm
72 // will stop and will use the second } on line 4 to determine the length to
73 // return, as in this example the range will include the tokens: {3}}
74 //
75 // The algorithm will only traverse the stack if it encounters braces, array
76 // initializer squares or text proto angle brackets.
77 if (!Tok.MatchingParen)
78 return 0;
79 FormatToken *End = Tok.MatchingParen;
80 // Maintains a stack level corresponding to the current End token.
81 int MatchingStackIndex = Stack.size() - 1;
82 // Traverses the stack downwards, looking for the level to which LBrace
83 // corresponds. Returns either a pointer to the matching level or nullptr if
84 // LParen is not found in the initial portion of the stack up to
85 // MatchingStackIndex.
86 auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * {
87 while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace)
88 --MatchingStackIndex;
89 return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr;
90 };
91 for (; End->Next; End = End->Next) {
92 if (End->Next->CanBreakBefore)
93 break;
94 if (!End->Next->closesScope())
95 continue;
96 if (End->Next->MatchingParen &&
97 End->Next->MatchingParen->isOneOf(
98 tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) {
99 const ParenState *State = FindParenState(End->Next->MatchingParen);
100 if (State && State->BreakBeforeClosingBrace)
101 break;
102 }
103 }
104 return End->TotalLength - Tok.TotalLength + 1;
105 }
106
getLengthToNextOperator(const FormatToken & Tok)107 static unsigned getLengthToNextOperator(const FormatToken &Tok) {
108 if (!Tok.NextOperator)
109 return 0;
110 return Tok.NextOperator->TotalLength - Tok.TotalLength;
111 }
112
113 // Returns \c true if \c Tok is the "." or "->" of a call and starts the next
114 // segment of a builder type call.
startsSegmentOfBuilderTypeCall(const FormatToken & Tok)115 static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) {
116 return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope();
117 }
118
119 // Returns \c true if \c Current starts a new parameter.
startsNextParameter(const FormatToken & Current,const FormatStyle & Style)120 static bool startsNextParameter(const FormatToken &Current,
121 const FormatStyle &Style) {
122 const FormatToken &Previous = *Current.Previous;
123 if (Current.is(TT_CtorInitializerComma) &&
124 Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma)
125 return true;
126 if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName))
127 return true;
128 return Previous.is(tok::comma) && !Current.isTrailingComment() &&
129 ((Previous.isNot(TT_CtorInitializerComma) ||
130 Style.BreakConstructorInitializers !=
131 FormatStyle::BCIS_BeforeComma) &&
132 (Previous.isNot(TT_InheritanceComma) ||
133 Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma));
134 }
135
opensProtoMessageField(const FormatToken & LessTok,const FormatStyle & Style)136 static bool opensProtoMessageField(const FormatToken &LessTok,
137 const FormatStyle &Style) {
138 if (LessTok.isNot(tok::less))
139 return false;
140 return Style.Language == FormatStyle::LK_TextProto ||
141 (Style.Language == FormatStyle::LK_Proto &&
142 (LessTok.NestingLevel > 0 ||
143 (LessTok.Previous && LessTok.Previous->is(tok::equal))));
144 }
145
146 // Returns the delimiter of a raw string literal, or None if TokenText is not
147 // the text of a raw string literal. The delimiter could be the empty string.
148 // For example, the delimiter of R"deli(cont)deli" is deli.
getRawStringDelimiter(StringRef TokenText)149 static llvm::Optional<StringRef> getRawStringDelimiter(StringRef TokenText) {
150 if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'.
151 || !TokenText.startswith("R\"") || !TokenText.endswith("\""))
152 return None;
153
154 // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has
155 // size at most 16 by the standard, so the first '(' must be among the first
156 // 19 bytes.
157 size_t LParenPos = TokenText.substr(0, 19).find_first_of('(');
158 if (LParenPos == StringRef::npos)
159 return None;
160 StringRef Delimiter = TokenText.substr(2, LParenPos - 2);
161
162 // Check that the string ends in ')Delimiter"'.
163 size_t RParenPos = TokenText.size() - Delimiter.size() - 2;
164 if (TokenText[RParenPos] != ')')
165 return None;
166 if (!TokenText.substr(RParenPos + 1).startswith(Delimiter))
167 return None;
168 return Delimiter;
169 }
170
171 // Returns the canonical delimiter for \p Language, or the empty string if no
172 // canonical delimiter is specified.
173 static StringRef
getCanonicalRawStringDelimiter(const FormatStyle & Style,FormatStyle::LanguageKind Language)174 getCanonicalRawStringDelimiter(const FormatStyle &Style,
175 FormatStyle::LanguageKind Language) {
176 for (const auto &Format : Style.RawStringFormats) {
177 if (Format.Language == Language)
178 return StringRef(Format.CanonicalDelimiter);
179 }
180 return "";
181 }
182
RawStringFormatStyleManager(const FormatStyle & CodeStyle)183 RawStringFormatStyleManager::RawStringFormatStyleManager(
184 const FormatStyle &CodeStyle) {
185 for (const auto &RawStringFormat : CodeStyle.RawStringFormats) {
186 llvm::Optional<FormatStyle> LanguageStyle =
187 CodeStyle.GetLanguageStyle(RawStringFormat.Language);
188 if (!LanguageStyle) {
189 FormatStyle PredefinedStyle;
190 if (!getPredefinedStyle(RawStringFormat.BasedOnStyle,
191 RawStringFormat.Language, &PredefinedStyle)) {
192 PredefinedStyle = getLLVMStyle();
193 PredefinedStyle.Language = RawStringFormat.Language;
194 }
195 LanguageStyle = PredefinedStyle;
196 }
197 LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit;
198 for (StringRef Delimiter : RawStringFormat.Delimiters) {
199 DelimiterStyle.insert({Delimiter, *LanguageStyle});
200 }
201 for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) {
202 EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle});
203 }
204 }
205 }
206
207 llvm::Optional<FormatStyle>
getDelimiterStyle(StringRef Delimiter) const208 RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const {
209 auto It = DelimiterStyle.find(Delimiter);
210 if (It == DelimiterStyle.end())
211 return None;
212 return It->second;
213 }
214
215 llvm::Optional<FormatStyle>
getEnclosingFunctionStyle(StringRef EnclosingFunction) const216 RawStringFormatStyleManager::getEnclosingFunctionStyle(
217 StringRef EnclosingFunction) const {
218 auto It = EnclosingFunctionStyle.find(EnclosingFunction);
219 if (It == EnclosingFunctionStyle.end())
220 return None;
221 return It->second;
222 }
223
ContinuationIndenter(const FormatStyle & Style,const AdditionalKeywords & Keywords,const SourceManager & SourceMgr,WhitespaceManager & Whitespaces,encoding::Encoding Encoding,bool BinPackInconclusiveFunctions)224 ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style,
225 const AdditionalKeywords &Keywords,
226 const SourceManager &SourceMgr,
227 WhitespaceManager &Whitespaces,
228 encoding::Encoding Encoding,
229 bool BinPackInconclusiveFunctions)
230 : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr),
231 Whitespaces(Whitespaces), Encoding(Encoding),
232 BinPackInconclusiveFunctions(BinPackInconclusiveFunctions),
233 CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {}
234
getInitialState(unsigned FirstIndent,unsigned FirstStartColumn,const AnnotatedLine * Line,bool DryRun)235 LineState ContinuationIndenter::getInitialState(unsigned FirstIndent,
236 unsigned FirstStartColumn,
237 const AnnotatedLine *Line,
238 bool DryRun) {
239 LineState State;
240 State.FirstIndent = FirstIndent;
241 if (FirstStartColumn && Line->First->NewlinesBefore == 0)
242 State.Column = FirstStartColumn;
243 else
244 State.Column = FirstIndent;
245 // With preprocessor directive indentation, the line starts on column 0
246 // since it's indented after the hash, but FirstIndent is set to the
247 // preprocessor indent.
248 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
249 (Line->Type == LT_PreprocessorDirective ||
250 Line->Type == LT_ImportStatement))
251 State.Column = 0;
252 State.Line = Line;
253 State.NextToken = Line->First;
254 State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent,
255 /*AvoidBinPacking=*/false,
256 /*NoLineBreak=*/false));
257 State.LineContainsContinuedForLoopSection = false;
258 State.NoContinuation = false;
259 State.StartOfStringLiteral = 0;
260 State.StartOfLineLevel = 0;
261 State.LowestLevelOnLine = 0;
262 State.IgnoreStackForComparison = false;
263
264 if (Style.Language == FormatStyle::LK_TextProto) {
265 // We need this in order to deal with the bin packing of text fields at
266 // global scope.
267 State.Stack.back().AvoidBinPacking = true;
268 State.Stack.back().BreakBeforeParameter = true;
269 State.Stack.back().AlignColons = false;
270 }
271
272 // The first token has already been indented and thus consumed.
273 moveStateToNextToken(State, DryRun, /*Newline=*/false);
274 return State;
275 }
276
canBreak(const LineState & State)277 bool ContinuationIndenter::canBreak(const LineState &State) {
278 const FormatToken &Current = *State.NextToken;
279 const FormatToken &Previous = *Current.Previous;
280 assert(&Previous == Current.Previous);
281 if (!Current.CanBreakBefore && !(State.Stack.back().BreakBeforeClosingBrace &&
282 Current.closesBlockOrBlockTypeList(Style)))
283 return false;
284 // The opening "{" of a braced list has to be on the same line as the first
285 // element if it is nested in another braced init list or function call.
286 if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
287 Previous.isNot(TT_DictLiteral) && Previous.BlockKind == BK_BracedInit &&
288 Previous.Previous &&
289 Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma))
290 return false;
291 // This prevents breaks like:
292 // ...
293 // SomeParameter, OtherParameter).DoSomething(
294 // ...
295 // As they hide "DoSomething" and are generally bad for readability.
296 if (Previous.opensScope() && Previous.isNot(tok::l_brace) &&
297 State.LowestLevelOnLine < State.StartOfLineLevel &&
298 State.LowestLevelOnLine < Current.NestingLevel)
299 return false;
300 if (Current.isMemberAccess() && State.Stack.back().ContainsUnwrappedBuilder)
301 return false;
302
303 // Don't create a 'hanging' indent if there are multiple blocks in a single
304 // statement.
305 if (Previous.is(tok::l_brace) && State.Stack.size() > 1 &&
306 State.Stack[State.Stack.size() - 2].NestedBlockInlined &&
307 State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks)
308 return false;
309
310 // Don't break after very short return types (e.g. "void") as that is often
311 // unexpected.
312 if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) {
313 if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None)
314 return false;
315 }
316
317 // If binary operators are moved to the next line (including commas for some
318 // styles of constructor initializers), that's always ok.
319 if (!Current.isOneOf(TT_BinaryOperator, tok::comma) &&
320 State.Stack.back().NoLineBreakInOperand)
321 return false;
322
323 if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr))
324 return false;
325
326 return !State.Stack.back().NoLineBreak;
327 }
328
mustBreak(const LineState & State)329 bool ContinuationIndenter::mustBreak(const LineState &State) {
330 const FormatToken &Current = *State.NextToken;
331 const FormatToken &Previous = *Current.Previous;
332 if (Current.MustBreakBefore || Current.is(TT_InlineASMColon))
333 return true;
334 if (State.Stack.back().BreakBeforeClosingBrace &&
335 Current.closesBlockOrBlockTypeList(Style))
336 return true;
337 if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection)
338 return true;
339 if (Style.Language == FormatStyle::LK_ObjC &&
340 Current.ObjCSelectorNameParts > 1 &&
341 Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) {
342 return true;
343 }
344 if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
345 (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) &&
346 Style.isCpp() &&
347 // FIXME: This is a temporary workaround for the case where clang-format
348 // sets BreakBeforeParameter to avoid bin packing and this creates a
349 // completely unnecessary line break after a template type that isn't
350 // line-wrapped.
351 (Previous.NestingLevel == 1 || Style.BinPackParameters)) ||
352 (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
353 Previous.isNot(tok::question)) ||
354 (!Style.BreakBeforeTernaryOperators &&
355 Previous.is(TT_ConditionalExpr))) &&
356 State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() &&
357 !Current.isOneOf(tok::r_paren, tok::r_brace))
358 return true;
359 if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) ||
360 (Previous.is(TT_ArrayInitializerLSquare) &&
361 Previous.ParameterCount > 1) ||
362 opensProtoMessageField(Previous, Style)) &&
363 Style.ColumnLimit > 0 &&
364 getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 >
365 getColumnLimit(State))
366 return true;
367
368 const FormatToken &BreakConstructorInitializersToken =
369 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon
370 ? Previous
371 : Current;
372 if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) &&
373 (State.Column + State.Line->Last->TotalLength - Previous.TotalLength >
374 getColumnLimit(State) ||
375 State.Stack.back().BreakBeforeParameter) &&
376 (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All ||
377 Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon ||
378 Style.ColumnLimit != 0))
379 return true;
380
381 if (Current.is(TT_ObjCMethodExpr) && !Previous.is(TT_SelectorName) &&
382 State.Line->startsWith(TT_ObjCMethodSpecifier))
383 return true;
384 if (Current.is(TT_SelectorName) && !Previous.is(tok::at) &&
385 State.Stack.back().ObjCSelectorNameFound &&
386 State.Stack.back().BreakBeforeParameter)
387 return true;
388
389 unsigned NewLineColumn = getNewLineColumn(State);
390 if (Current.isMemberAccess() && Style.ColumnLimit != 0 &&
391 State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit &&
392 (State.Column > NewLineColumn ||
393 Current.NestingLevel < State.StartOfLineLevel))
394 return true;
395
396 if (startsSegmentOfBuilderTypeCall(Current) &&
397 (State.Stack.back().CallContinuation != 0 ||
398 State.Stack.back().BreakBeforeParameter) &&
399 // JavaScript is treated different here as there is a frequent pattern:
400 // SomeFunction(function() {
401 // ...
402 // }.bind(...));
403 // FIXME: We should find a more generic solution to this problem.
404 !(State.Column <= NewLineColumn &&
405 Style.Language == FormatStyle::LK_JavaScript) &&
406 !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn))
407 return true;
408
409 // If the template declaration spans multiple lines, force wrap before the
410 // function/class declaration
411 if (Previous.ClosesTemplateDeclaration &&
412 State.Stack.back().BreakBeforeParameter && Current.CanBreakBefore)
413 return true;
414
415 if (State.Column <= NewLineColumn)
416 return false;
417
418 if (Style.AlwaysBreakBeforeMultilineStrings &&
419 (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth ||
420 Previous.is(tok::comma) || Current.NestingLevel < 2) &&
421 !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at,
422 Keywords.kw_dollar) &&
423 !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) &&
424 nextIsMultilineString(State))
425 return true;
426
427 // Using CanBreakBefore here and below takes care of the decision whether the
428 // current style uses wrapping before or after operators for the given
429 // operator.
430 if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
431 // If we need to break somewhere inside the LHS of a binary expression, we
432 // should also break after the operator. Otherwise, the formatting would
433 // hide the operator precedence, e.g. in:
434 // if (aaaaaaaaaaaaaa ==
435 // bbbbbbbbbbbbbb && c) {..
436 // For comparisons, we only apply this rule, if the LHS is a binary
437 // expression itself as otherwise, the line breaks seem superfluous.
438 // We need special cases for ">>" which we have split into two ">" while
439 // lexing in order to make template parsing easier.
440 bool IsComparison = (Previous.getPrecedence() == prec::Relational ||
441 Previous.getPrecedence() == prec::Equality ||
442 Previous.getPrecedence() == prec::Spaceship) &&
443 Previous.Previous &&
444 Previous.Previous->isNot(TT_BinaryOperator); // For >>.
445 bool LHSIsBinaryExpr =
446 Previous.Previous && Previous.Previous->EndsBinaryExpression;
447 if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() &&
448 Previous.getPrecedence() != prec::Assignment &&
449 State.Stack.back().BreakBeforeParameter)
450 return true;
451 } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
452 State.Stack.back().BreakBeforeParameter) {
453 return true;
454 }
455
456 // Same as above, but for the first "<<" operator.
457 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) &&
458 State.Stack.back().BreakBeforeParameter &&
459 State.Stack.back().FirstLessLess == 0)
460 return true;
461
462 if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
463 // Always break after "template <...>" and leading annotations. This is only
464 // for cases where the entire line does not fit on a single line as a
465 // different LineFormatter would be used otherwise.
466 if (Previous.ClosesTemplateDeclaration)
467 return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No;
468 if (Previous.is(TT_FunctionAnnotationRParen))
469 return true;
470 if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
471 Current.isNot(TT_LeadingJavaAnnotation))
472 return true;
473 }
474
475 // If the return type spans multiple lines, wrap before the function name.
476 if ((Current.is(TT_FunctionDeclarationName) ||
477 (Current.is(tok::kw_operator) && !Previous.is(tok::coloncolon))) &&
478 !Previous.is(tok::kw_template) && State.Stack.back().BreakBeforeParameter)
479 return true;
480
481 // The following could be precomputed as they do not depend on the state.
482 // However, as they should take effect only if the UnwrappedLine does not fit
483 // into the ColumnLimit, they are checked here in the ContinuationIndenter.
484 if (Style.ColumnLimit != 0 && Previous.BlockKind == BK_Block &&
485 Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment))
486 return true;
487
488 if (Current.is(tok::lessless) &&
489 ((Previous.is(tok::identifier) && Previous.TokenText == "endl") ||
490 (Previous.Tok.isLiteral() && (Previous.TokenText.endswith("\\n\"") ||
491 Previous.TokenText == "\'\\n\'"))))
492 return true;
493
494 if (Previous.is(TT_BlockComment) && Previous.IsMultiline)
495 return true;
496
497 if (State.NoContinuation)
498 return true;
499
500 return false;
501 }
502
addTokenToState(LineState & State,bool Newline,bool DryRun,unsigned ExtraSpaces)503 unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
504 bool DryRun,
505 unsigned ExtraSpaces) {
506 const FormatToken &Current = *State.NextToken;
507
508 assert(!State.Stack.empty());
509 State.NoContinuation = false;
510
511 if ((Current.is(TT_ImplicitStringLiteral) &&
512 (Current.Previous->Tok.getIdentifierInfo() == nullptr ||
513 Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() ==
514 tok::pp_not_keyword))) {
515 unsigned EndColumn =
516 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
517 if (Current.LastNewlineOffset != 0) {
518 // If there is a newline within this token, the final column will solely
519 // determined by the current end column.
520 State.Column = EndColumn;
521 } else {
522 unsigned StartColumn =
523 SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
524 assert(EndColumn >= StartColumn);
525 State.Column += EndColumn - StartColumn;
526 }
527 moveStateToNextToken(State, DryRun, /*Newline=*/false);
528 return 0;
529 }
530
531 unsigned Penalty = 0;
532 if (Newline)
533 Penalty = addTokenOnNewLine(State, DryRun);
534 else
535 addTokenOnCurrentLine(State, DryRun, ExtraSpaces);
536
537 return moveStateToNextToken(State, DryRun, Newline) + Penalty;
538 }
539
addTokenOnCurrentLine(LineState & State,bool DryRun,unsigned ExtraSpaces)540 void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
541 unsigned ExtraSpaces) {
542 FormatToken &Current = *State.NextToken;
543 const FormatToken &Previous = *State.NextToken->Previous;
544 if (Current.is(tok::equal) &&
545 (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) &&
546 State.Stack.back().VariablePos == 0) {
547 State.Stack.back().VariablePos = State.Column;
548 // Move over * and & if they are bound to the variable name.
549 const FormatToken *Tok = &Previous;
550 while (Tok && State.Stack.back().VariablePos >= Tok->ColumnWidth) {
551 State.Stack.back().VariablePos -= Tok->ColumnWidth;
552 if (Tok->SpacesRequiredBefore != 0)
553 break;
554 Tok = Tok->Previous;
555 }
556 if (Previous.PartOfMultiVariableDeclStmt)
557 State.Stack.back().LastSpace = State.Stack.back().VariablePos;
558 }
559
560 unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces;
561
562 // Indent preprocessor directives after the hash if required.
563 int PPColumnCorrection = 0;
564 if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash &&
565 Previous.is(tok::hash) && State.FirstIndent > 0 &&
566 (State.Line->Type == LT_PreprocessorDirective ||
567 State.Line->Type == LT_ImportStatement)) {
568 Spaces += State.FirstIndent;
569
570 // For preprocessor indent with tabs, State.Column will be 1 because of the
571 // hash. This causes second-level indents onward to have an extra space
572 // after the tabs. We avoid this misalignment by subtracting 1 from the
573 // column value passed to replaceWhitespace().
574 if (Style.UseTab != FormatStyle::UT_Never)
575 PPColumnCorrection = -1;
576 }
577
578 if (!DryRun)
579 Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces,
580 State.Column + Spaces + PPColumnCorrection);
581
582 // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance
583 // declaration unless there is multiple inheritance.
584 if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma &&
585 Current.is(TT_InheritanceColon))
586 State.Stack.back().NoLineBreak = true;
587 if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon &&
588 Previous.is(TT_InheritanceColon))
589 State.Stack.back().NoLineBreak = true;
590
591 if (Current.is(TT_SelectorName) &&
592 !State.Stack.back().ObjCSelectorNameFound) {
593 unsigned MinIndent =
594 std::max(State.FirstIndent + Style.ContinuationIndentWidth,
595 State.Stack.back().Indent);
596 unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth;
597 if (Current.LongestObjCSelectorName == 0)
598 State.Stack.back().AlignColons = false;
599 else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos)
600 State.Stack.back().ColonPos = MinIndent + Current.LongestObjCSelectorName;
601 else
602 State.Stack.back().ColonPos = FirstColonPos;
603 }
604
605 // In "AlwaysBreak" mode, enforce wrapping directly after the parenthesis by
606 // disallowing any further line breaks if there is no line break after the
607 // opening parenthesis. Don't break if it doesn't conserve columns.
608 if (Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak &&
609 Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) &&
610 State.Column > getNewLineColumn(State) &&
611 (!Previous.Previous || !Previous.Previous->isOneOf(
612 tok::kw_for, tok::kw_while, tok::kw_switch)) &&
613 // Don't do this for simple (no expressions) one-argument function calls
614 // as that feels like needlessly wasting whitespace, e.g.:
615 //
616 // caaaaaaaaaaaall(
617 // caaaaaaaaaaaall(
618 // caaaaaaaaaaaall(
619 // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa))));
620 Current.FakeLParens.size() > 0 &&
621 Current.FakeLParens.back() > prec::Unknown)
622 State.Stack.back().NoLineBreak = true;
623 if (Previous.is(TT_TemplateString) && Previous.opensScope())
624 State.Stack.back().NoLineBreak = true;
625
626 if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign &&
627 Previous.opensScope() && Previous.isNot(TT_ObjCMethodExpr) &&
628 (Current.isNot(TT_LineComment) || Previous.BlockKind == BK_BracedInit))
629 State.Stack.back().Indent = State.Column + Spaces;
630 if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style))
631 State.Stack.back().NoLineBreak = true;
632 if (startsSegmentOfBuilderTypeCall(Current) &&
633 State.Column > getNewLineColumn(State))
634 State.Stack.back().ContainsUnwrappedBuilder = true;
635
636 if (Current.is(TT_LambdaArrow) && Style.Language == FormatStyle::LK_Java)
637 State.Stack.back().NoLineBreak = true;
638 if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
639 (Previous.MatchingParen &&
640 (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10)))
641 // If there is a function call with long parameters, break before trailing
642 // calls. This prevents things like:
643 // EXPECT_CALL(SomeLongParameter).Times(
644 // 2);
645 // We don't want to do this for short parameters as they can just be
646 // indexes.
647 State.Stack.back().NoLineBreak = true;
648
649 // Don't allow the RHS of an operator to be split over multiple lines unless
650 // there is a line-break right after the operator.
651 // Exclude relational operators, as there, it is always more desirable to
652 // have the LHS 'left' of the RHS.
653 const FormatToken *P = Current.getPreviousNonComment();
654 if (!Current.is(tok::comment) && P &&
655 (P->isOneOf(TT_BinaryOperator, tok::comma) ||
656 (P->is(TT_ConditionalExpr) && P->is(tok::colon))) &&
657 !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) &&
658 P->getPrecedence() != prec::Assignment &&
659 P->getPrecedence() != prec::Relational &&
660 P->getPrecedence() != prec::Spaceship) {
661 bool BreakBeforeOperator =
662 P->MustBreakBefore || P->is(tok::lessless) ||
663 (P->is(TT_BinaryOperator) &&
664 Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) ||
665 (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators);
666 // Don't do this if there are only two operands. In these cases, there is
667 // always a nice vertical separation between them and the extra line break
668 // does not help.
669 bool HasTwoOperands =
670 P->OperatorIndex == 0 && !P->NextOperator && !P->is(TT_ConditionalExpr);
671 if ((!BreakBeforeOperator && !(HasTwoOperands && Style.AlignOperands)) ||
672 (!State.Stack.back().LastOperatorWrapped && BreakBeforeOperator))
673 State.Stack.back().NoLineBreakInOperand = true;
674 }
675
676 State.Column += Spaces;
677 if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) &&
678 Previous.Previous &&
679 (Previous.Previous->isOneOf(tok::kw_if, tok::kw_for) ||
680 Previous.Previous->endsSequence(tok::kw_constexpr, tok::kw_if))) {
681 // Treat the condition inside an if as if it was a second function
682 // parameter, i.e. let nested calls have a continuation indent.
683 State.Stack.back().LastSpace = State.Column;
684 State.Stack.back().NestedBlockIndent = State.Column;
685 } else if (!Current.isOneOf(tok::comment, tok::caret) &&
686 ((Previous.is(tok::comma) &&
687 !Previous.is(TT_OverloadedOperator)) ||
688 (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) {
689 State.Stack.back().LastSpace = State.Column;
690 } else if (Previous.is(TT_CtorInitializerColon) &&
691 Style.BreakConstructorInitializers ==
692 FormatStyle::BCIS_AfterColon) {
693 State.Stack.back().Indent = State.Column;
694 State.Stack.back().LastSpace = State.Column;
695 } else if ((Previous.isOneOf(TT_BinaryOperator, TT_ConditionalExpr,
696 TT_CtorInitializerColon)) &&
697 ((Previous.getPrecedence() != prec::Assignment &&
698 (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 ||
699 Previous.NextOperator)) ||
700 Current.StartsBinaryExpression)) {
701 // Indent relative to the RHS of the expression unless this is a simple
702 // assignment without binary expression on the RHS. Also indent relative to
703 // unary operators and the colons of constructor initializers.
704 if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None)
705 State.Stack.back().LastSpace = State.Column;
706 } else if (Previous.is(TT_InheritanceColon)) {
707 State.Stack.back().Indent = State.Column;
708 State.Stack.back().LastSpace = State.Column;
709 } else if (Previous.opensScope()) {
710 // If a function has a trailing call, indent all parameters from the
711 // opening parenthesis. This avoids confusing indents like:
712 // OuterFunction(InnerFunctionCall( // break
713 // ParameterToInnerFunction)) // break
714 // .SecondInnerFunctionCall();
715 bool HasTrailingCall = false;
716 if (Previous.MatchingParen) {
717 const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
718 HasTrailingCall = Next && Next->isMemberAccess();
719 }
720 if (HasTrailingCall && State.Stack.size() > 1 &&
721 State.Stack[State.Stack.size() - 2].CallContinuation == 0)
722 State.Stack.back().LastSpace = State.Column;
723 }
724 }
725
addTokenOnNewLine(LineState & State,bool DryRun)726 unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
727 bool DryRun) {
728 FormatToken &Current = *State.NextToken;
729 const FormatToken &Previous = *State.NextToken->Previous;
730
731 // Extra penalty that needs to be added because of the way certain line
732 // breaks are chosen.
733 unsigned Penalty = 0;
734
735 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
736 const FormatToken *NextNonComment = Previous.getNextNonComment();
737 if (!NextNonComment)
738 NextNonComment = &Current;
739 // The first line break on any NestingLevel causes an extra penalty in order
740 // prefer similar line breaks.
741 if (!State.Stack.back().ContainsLineBreak)
742 Penalty += 15;
743 State.Stack.back().ContainsLineBreak = true;
744
745 Penalty += State.NextToken->SplitPenalty;
746
747 // Breaking before the first "<<" is generally not desirable if the LHS is
748 // short. Also always add the penalty if the LHS is split over multiple lines
749 // to avoid unnecessary line breaks that just work around this penalty.
750 if (NextNonComment->is(tok::lessless) &&
751 State.Stack.back().FirstLessLess == 0 &&
752 (State.Column <= Style.ColumnLimit / 3 ||
753 State.Stack.back().BreakBeforeParameter))
754 Penalty += Style.PenaltyBreakFirstLessLess;
755
756 State.Column = getNewLineColumn(State);
757
758 // Indent nested blocks relative to this column, unless in a very specific
759 // JavaScript special case where:
760 //
761 // var loooooong_name =
762 // function() {
763 // // code
764 // }
765 //
766 // is common and should be formatted like a free-standing function. The same
767 // goes for wrapping before the lambda return type arrow.
768 if (!Current.is(TT_LambdaArrow) &&
769 (Style.Language != FormatStyle::LK_JavaScript ||
770 Current.NestingLevel != 0 || !PreviousNonComment ||
771 !PreviousNonComment->is(tok::equal) ||
772 !Current.isOneOf(Keywords.kw_async, Keywords.kw_function)))
773 State.Stack.back().NestedBlockIndent = State.Column;
774
775 if (NextNonComment->isMemberAccess()) {
776 if (State.Stack.back().CallContinuation == 0)
777 State.Stack.back().CallContinuation = State.Column;
778 } else if (NextNonComment->is(TT_SelectorName)) {
779 if (!State.Stack.back().ObjCSelectorNameFound) {
780 if (NextNonComment->LongestObjCSelectorName == 0) {
781 State.Stack.back().AlignColons = false;
782 } else {
783 State.Stack.back().ColonPos =
784 (shouldIndentWrappedSelectorName(Style, State.Line->Type)
785 ? std::max(State.Stack.back().Indent,
786 State.FirstIndent + Style.ContinuationIndentWidth)
787 : State.Stack.back().Indent) +
788 std::max(NextNonComment->LongestObjCSelectorName,
789 NextNonComment->ColumnWidth);
790 }
791 } else if (State.Stack.back().AlignColons &&
792 State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) {
793 State.Stack.back().ColonPos = State.Column + NextNonComment->ColumnWidth;
794 }
795 } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
796 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) {
797 // FIXME: This is hacky, find a better way. The problem is that in an ObjC
798 // method expression, the block should be aligned to the line starting it,
799 // e.g.:
800 // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason
801 // ^(int *i) {
802 // // ...
803 // }];
804 // Thus, we set LastSpace of the next higher NestingLevel, to which we move
805 // when we consume all of the "}"'s FakeRParens at the "{".
806 if (State.Stack.size() > 1)
807 State.Stack[State.Stack.size() - 2].LastSpace =
808 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
809 Style.ContinuationIndentWidth;
810 }
811
812 if ((PreviousNonComment &&
813 PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
814 !State.Stack.back().AvoidBinPacking) ||
815 Previous.is(TT_BinaryOperator))
816 State.Stack.back().BreakBeforeParameter = false;
817 if (PreviousNonComment &&
818 PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) &&
819 Current.NestingLevel == 0)
820 State.Stack.back().BreakBeforeParameter = false;
821 if (NextNonComment->is(tok::question) ||
822 (PreviousNonComment && PreviousNonComment->is(tok::question)))
823 State.Stack.back().BreakBeforeParameter = true;
824 if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore)
825 State.Stack.back().BreakBeforeParameter = false;
826
827 if (!DryRun) {
828 unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1;
829 if (Current.is(tok::r_brace) && Current.MatchingParen &&
830 // Only strip trailing empty lines for l_braces that have children, i.e.
831 // for function expressions (lambdas, arrows, etc).
832 !Current.MatchingParen->Children.empty()) {
833 // lambdas and arrow functions are expressions, thus their r_brace is not
834 // on its own line, and thus not covered by UnwrappedLineFormatter's logic
835 // about removing empty lines on closing blocks. Special case them here.
836 MaxEmptyLinesToKeep = 1;
837 }
838 unsigned Newlines =
839 std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep));
840 bool ContinuePPDirective =
841 State.Line->InPPDirective && State.Line->Type != LT_ImportStatement;
842 Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column,
843 ContinuePPDirective);
844 }
845
846 if (!Current.isTrailingComment())
847 State.Stack.back().LastSpace = State.Column;
848 if (Current.is(tok::lessless))
849 // If we are breaking before a "<<", we always want to indent relative to
850 // RHS. This is necessary only for "<<", as we special-case it and don't
851 // always indent relative to the RHS.
852 State.Stack.back().LastSpace += 3; // 3 -> width of "<< ".
853
854 State.StartOfLineLevel = Current.NestingLevel;
855 State.LowestLevelOnLine = Current.NestingLevel;
856
857 // Any break on this level means that the parent level has been broken
858 // and we need to avoid bin packing there.
859 bool NestedBlockSpecialCase =
860 !Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 &&
861 State.Stack[State.Stack.size() - 2].NestedBlockInlined;
862 if (!NestedBlockSpecialCase)
863 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
864 State.Stack[i].BreakBeforeParameter = true;
865
866 if (PreviousNonComment &&
867 !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) &&
868 (PreviousNonComment->isNot(TT_TemplateCloser) ||
869 Current.NestingLevel != 0) &&
870 !PreviousNonComment->isOneOf(
871 TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
872 TT_LeadingJavaAnnotation) &&
873 Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope())
874 State.Stack.back().BreakBeforeParameter = true;
875
876 // If we break after { or the [ of an array initializer, we should also break
877 // before the corresponding } or ].
878 if (PreviousNonComment &&
879 (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
880 opensProtoMessageField(*PreviousNonComment, Style)))
881 State.Stack.back().BreakBeforeClosingBrace = true;
882
883 if (State.Stack.back().AvoidBinPacking) {
884 // If we are breaking after '(', '{', '<', or this is the break after a ':'
885 // to start a member initializater list in a constructor, this should not
886 // be considered bin packing unless the relevant AllowAll option is false or
887 // this is a dict/object literal.
888 bool PreviousIsBreakingCtorInitializerColon =
889 Previous.is(TT_CtorInitializerColon) &&
890 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon;
891 if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) ||
892 PreviousIsBreakingCtorInitializerColon) ||
893 (!Style.AllowAllParametersOfDeclarationOnNextLine &&
894 State.Line->MustBeDeclaration) ||
895 (!Style.AllowAllArgumentsOnNextLine &&
896 !State.Line->MustBeDeclaration) ||
897 (!Style.AllowAllConstructorInitializersOnNextLine &&
898 PreviousIsBreakingCtorInitializerColon) ||
899 Previous.is(TT_DictLiteral))
900 State.Stack.back().BreakBeforeParameter = true;
901
902 // If we are breaking after a ':' to start a member initializer list,
903 // and we allow all arguments on the next line, we should not break
904 // before the next parameter.
905 if (PreviousIsBreakingCtorInitializerColon &&
906 Style.AllowAllConstructorInitializersOnNextLine)
907 State.Stack.back().BreakBeforeParameter = false;
908 }
909
910 return Penalty;
911 }
912
getNewLineColumn(const LineState & State)913 unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
914 if (!State.NextToken || !State.NextToken->Previous)
915 return 0;
916 FormatToken &Current = *State.NextToken;
917 const FormatToken &Previous = *Current.Previous;
918 // If we are continuing an expression, we want to use the continuation indent.
919 unsigned ContinuationIndent =
920 std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) +
921 Style.ContinuationIndentWidth;
922 const FormatToken *PreviousNonComment = Current.getPreviousNonComment();
923 const FormatToken *NextNonComment = Previous.getNextNonComment();
924 if (!NextNonComment)
925 NextNonComment = &Current;
926
927 // Java specific bits.
928 if (Style.Language == FormatStyle::LK_Java &&
929 Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends))
930 return std::max(State.Stack.back().LastSpace,
931 State.Stack.back().Indent + Style.ContinuationIndentWidth);
932
933 if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block)
934 return Current.NestingLevel == 0 ? State.FirstIndent
935 : State.Stack.back().Indent;
936 if ((Current.isOneOf(tok::r_brace, tok::r_square) ||
937 (Current.is(tok::greater) &&
938 (Style.Language == FormatStyle::LK_Proto ||
939 Style.Language == FormatStyle::LK_TextProto))) &&
940 State.Stack.size() > 1) {
941 if (Current.closesBlockOrBlockTypeList(Style))
942 return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
943 if (Current.MatchingParen &&
944 Current.MatchingParen->BlockKind == BK_BracedInit)
945 return State.Stack[State.Stack.size() - 2].LastSpace;
946 return State.FirstIndent;
947 }
948 // Indent a closing parenthesis at the previous level if followed by a semi,
949 // const, or opening brace. This allows indentations such as:
950 // foo(
951 // a,
952 // );
953 // int Foo::getter(
954 // //
955 // ) const {
956 // return foo;
957 // }
958 // function foo(
959 // a,
960 // ) {
961 // code(); //
962 // }
963 if (Current.is(tok::r_paren) && State.Stack.size() > 1 &&
964 (!Current.Next ||
965 Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace)))
966 return State.Stack[State.Stack.size() - 2].LastSpace;
967 if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope())
968 return State.Stack[State.Stack.size() - 2].LastSpace;
969 if (Current.is(tok::identifier) && Current.Next &&
970 (Current.Next->is(TT_DictLiteral) ||
971 ((Style.Language == FormatStyle::LK_Proto ||
972 Style.Language == FormatStyle::LK_TextProto) &&
973 Current.Next->isOneOf(tok::less, tok::l_brace))))
974 return State.Stack.back().Indent;
975 if (NextNonComment->is(TT_ObjCStringLiteral) &&
976 State.StartOfStringLiteral != 0)
977 return State.StartOfStringLiteral - 1;
978 if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
979 return State.StartOfStringLiteral;
980 if (NextNonComment->is(tok::lessless) &&
981 State.Stack.back().FirstLessLess != 0)
982 return State.Stack.back().FirstLessLess;
983 if (NextNonComment->isMemberAccess()) {
984 if (State.Stack.back().CallContinuation == 0)
985 return ContinuationIndent;
986 return State.Stack.back().CallContinuation;
987 }
988 if (State.Stack.back().QuestionColumn != 0 &&
989 ((NextNonComment->is(tok::colon) &&
990 NextNonComment->is(TT_ConditionalExpr)) ||
991 Previous.is(TT_ConditionalExpr)))
992 return State.Stack.back().QuestionColumn;
993 if (Previous.is(tok::comma) && State.Stack.back().VariablePos != 0)
994 return State.Stack.back().VariablePos;
995 if ((PreviousNonComment &&
996 (PreviousNonComment->ClosesTemplateDeclaration ||
997 PreviousNonComment->isOneOf(
998 TT_AttributeParen, TT_AttributeSquare, TT_FunctionAnnotationRParen,
999 TT_JavaAnnotation, TT_LeadingJavaAnnotation))) ||
1000 (!Style.IndentWrappedFunctionNames &&
1001 NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName)))
1002 return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent);
1003 if (NextNonComment->is(TT_SelectorName)) {
1004 if (!State.Stack.back().ObjCSelectorNameFound) {
1005 unsigned MinIndent = State.Stack.back().Indent;
1006 if (shouldIndentWrappedSelectorName(Style, State.Line->Type))
1007 MinIndent = std::max(MinIndent,
1008 State.FirstIndent + Style.ContinuationIndentWidth);
1009 // If LongestObjCSelectorName is 0, we are indenting the first
1010 // part of an ObjC selector (or a selector component which is
1011 // not colon-aligned due to block formatting).
1012 //
1013 // Otherwise, we are indenting a subsequent part of an ObjC
1014 // selector which should be colon-aligned to the longest
1015 // component of the ObjC selector.
1016 //
1017 // In either case, we want to respect Style.IndentWrappedFunctionNames.
1018 return MinIndent +
1019 std::max(NextNonComment->LongestObjCSelectorName,
1020 NextNonComment->ColumnWidth) -
1021 NextNonComment->ColumnWidth;
1022 }
1023 if (!State.Stack.back().AlignColons)
1024 return State.Stack.back().Indent;
1025 if (State.Stack.back().ColonPos > NextNonComment->ColumnWidth)
1026 return State.Stack.back().ColonPos - NextNonComment->ColumnWidth;
1027 return State.Stack.back().Indent;
1028 }
1029 if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr))
1030 return State.Stack.back().ColonPos;
1031 if (NextNonComment->is(TT_ArraySubscriptLSquare)) {
1032 if (State.Stack.back().StartOfArraySubscripts != 0)
1033 return State.Stack.back().StartOfArraySubscripts;
1034 return ContinuationIndent;
1035 }
1036
1037 // This ensure that we correctly format ObjC methods calls without inputs,
1038 // i.e. where the last element isn't selector like: [callee method];
1039 if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
1040 NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr))
1041 return State.Stack.back().Indent;
1042
1043 if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
1044 Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon))
1045 return ContinuationIndent;
1046 if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
1047 PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))
1048 return ContinuationIndent;
1049 if (NextNonComment->is(TT_CtorInitializerComma))
1050 return State.Stack.back().Indent;
1051 if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) &&
1052 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon)
1053 return State.Stack.back().Indent;
1054 if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) &&
1055 Style.BreakInheritanceList == FormatStyle::BILS_AfterColon)
1056 return State.Stack.back().Indent;
1057 if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon,
1058 TT_InheritanceComma))
1059 return State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1060 if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
1061 !Current.isOneOf(tok::colon, tok::comment))
1062 return ContinuationIndent;
1063 if (Current.is(TT_ProtoExtensionLSquare))
1064 return State.Stack.back().Indent;
1065 if (State.Stack.back().Indent == State.FirstIndent && PreviousNonComment &&
1066 !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma))
1067 // Ensure that we fall back to the continuation indent width instead of
1068 // just flushing continuations left.
1069 return State.Stack.back().Indent + Style.ContinuationIndentWidth;
1070 return State.Stack.back().Indent;
1071 }
1072
moveStateToNextToken(LineState & State,bool DryRun,bool Newline)1073 unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
1074 bool DryRun, bool Newline) {
1075 assert(State.Stack.size());
1076 const FormatToken &Current = *State.NextToken;
1077
1078 if (Current.isOneOf(tok::comma, TT_BinaryOperator))
1079 State.Stack.back().NoLineBreakInOperand = false;
1080 if (Current.is(TT_InheritanceColon))
1081 State.Stack.back().AvoidBinPacking = true;
1082 if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) {
1083 if (State.Stack.back().FirstLessLess == 0)
1084 State.Stack.back().FirstLessLess = State.Column;
1085 else
1086 State.Stack.back().LastOperatorWrapped = Newline;
1087 }
1088 if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless))
1089 State.Stack.back().LastOperatorWrapped = Newline;
1090 if (Current.is(TT_ConditionalExpr) && Current.Previous &&
1091 !Current.Previous->is(TT_ConditionalExpr))
1092 State.Stack.back().LastOperatorWrapped = Newline;
1093 if (Current.is(TT_ArraySubscriptLSquare) &&
1094 State.Stack.back().StartOfArraySubscripts == 0)
1095 State.Stack.back().StartOfArraySubscripts = State.Column;
1096 if (Style.BreakBeforeTernaryOperators && Current.is(tok::question))
1097 State.Stack.back().QuestionColumn = State.Column;
1098 if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) {
1099 const FormatToken *Previous = Current.Previous;
1100 while (Previous && Previous->isTrailingComment())
1101 Previous = Previous->Previous;
1102 if (Previous && Previous->is(tok::question))
1103 State.Stack.back().QuestionColumn = State.Column;
1104 }
1105 if (!Current.opensScope() && !Current.closesScope() &&
1106 !Current.is(TT_PointerOrReference))
1107 State.LowestLevelOnLine =
1108 std::min(State.LowestLevelOnLine, Current.NestingLevel);
1109 if (Current.isMemberAccess())
1110 State.Stack.back().StartOfFunctionCall =
1111 !Current.NextOperator ? 0 : State.Column;
1112 if (Current.is(TT_SelectorName))
1113 State.Stack.back().ObjCSelectorNameFound = true;
1114 if (Current.is(TT_CtorInitializerColon) &&
1115 Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) {
1116 // Indent 2 from the column, so:
1117 // SomeClass::SomeClass()
1118 // : First(...), ...
1119 // Next(...)
1120 // ^ line up here.
1121 State.Stack.back().Indent =
1122 State.Column +
1123 (Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma
1124 ? 0
1125 : 2);
1126 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1127 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine) {
1128 State.Stack.back().AvoidBinPacking = true;
1129 State.Stack.back().BreakBeforeParameter =
1130 !Style.AllowAllConstructorInitializersOnNextLine;
1131 } else {
1132 State.Stack.back().BreakBeforeParameter = false;
1133 }
1134 }
1135 if (Current.is(TT_CtorInitializerColon) &&
1136 Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) {
1137 State.Stack.back().Indent =
1138 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1139 State.Stack.back().NestedBlockIndent = State.Stack.back().Indent;
1140 if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
1141 State.Stack.back().AvoidBinPacking = true;
1142 }
1143 if (Current.is(TT_InheritanceColon))
1144 State.Stack.back().Indent =
1145 State.FirstIndent + Style.ConstructorInitializerIndentWidth;
1146 if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
1147 State.Stack.back().NestedBlockIndent =
1148 State.Column + Current.ColumnWidth + 1;
1149 if (Current.isOneOf(TT_LambdaLSquare, TT_LambdaArrow))
1150 State.Stack.back().LastSpace = State.Column;
1151
1152 // Insert scopes created by fake parenthesis.
1153 const FormatToken *Previous = Current.getPreviousNonComment();
1154
1155 // Add special behavior to support a format commonly used for JavaScript
1156 // closures:
1157 // SomeFunction(function() {
1158 // foo();
1159 // bar();
1160 // }, a, b, c);
1161 if (Current.isNot(tok::comment) && Previous &&
1162 Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
1163 !Previous->is(TT_DictLiteral) && State.Stack.size() > 1 &&
1164 !State.Stack.back().HasMultipleNestedBlocks) {
1165 if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline)
1166 for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i)
1167 State.Stack[i].NoLineBreak = true;
1168 State.Stack[State.Stack.size() - 2].NestedBlockInlined = false;
1169 }
1170 if (Previous &&
1171 (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) ||
1172 Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr)) &&
1173 !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)) {
1174 State.Stack.back().NestedBlockInlined =
1175 !Newline &&
1176 (Previous->isNot(tok::l_paren) || Previous->ParameterCount > 1);
1177 }
1178
1179 moveStatePastFakeLParens(State, Newline);
1180 moveStatePastScopeCloser(State);
1181 bool AllowBreak = !State.Stack.back().NoLineBreak &&
1182 !State.Stack.back().NoLineBreakInOperand;
1183 moveStatePastScopeOpener(State, Newline);
1184 moveStatePastFakeRParens(State);
1185
1186 if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
1187 State.StartOfStringLiteral = State.Column + 1;
1188 if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0)
1189 State.StartOfStringLiteral = State.Column + 1;
1190 else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0)
1191 State.StartOfStringLiteral = State.Column;
1192 else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
1193 !Current.isStringLiteral())
1194 State.StartOfStringLiteral = 0;
1195
1196 State.Column += Current.ColumnWidth;
1197 State.NextToken = State.NextToken->Next;
1198
1199 unsigned Penalty =
1200 handleEndOfLine(Current, State, DryRun, AllowBreak, Newline);
1201
1202 if (Current.Role)
1203 Current.Role->formatFromToken(State, this, DryRun);
1204 // If the previous has a special role, let it consume tokens as appropriate.
1205 // It is necessary to start at the previous token for the only implemented
1206 // role (comma separated list). That way, the decision whether or not to break
1207 // after the "{" is already done and both options are tried and evaluated.
1208 // FIXME: This is ugly, find a better way.
1209 if (Previous && Previous->Role)
1210 Penalty += Previous->Role->formatAfterToken(State, this, DryRun);
1211
1212 return Penalty;
1213 }
1214
moveStatePastFakeLParens(LineState & State,bool Newline)1215 void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
1216 bool Newline) {
1217 const FormatToken &Current = *State.NextToken;
1218 const FormatToken *Previous = Current.getPreviousNonComment();
1219
1220 // Don't add extra indentation for the first fake parenthesis after
1221 // 'return', assignments or opening <({[. The indentation for these cases
1222 // is special cased.
1223 bool SkipFirstExtraIndent =
1224 (Previous && (Previous->opensScope() ||
1225 Previous->isOneOf(tok::semi, tok::kw_return) ||
1226 (Previous->getPrecedence() == prec::Assignment &&
1227 Style.AlignOperands) ||
1228 Previous->is(TT_ObjCMethodExpr)));
1229 for (SmallVectorImpl<prec::Level>::const_reverse_iterator
1230 I = Current.FakeLParens.rbegin(),
1231 E = Current.FakeLParens.rend();
1232 I != E; ++I) {
1233 ParenState NewParenState = State.Stack.back();
1234 NewParenState.Tok = nullptr;
1235 NewParenState.ContainsLineBreak = false;
1236 NewParenState.LastOperatorWrapped = true;
1237 NewParenState.NoLineBreak =
1238 NewParenState.NoLineBreak || State.Stack.back().NoLineBreakInOperand;
1239
1240 // Don't propagate AvoidBinPacking into subexpressions of arg/param lists.
1241 if (*I > prec::Comma)
1242 NewParenState.AvoidBinPacking = false;
1243
1244 // Indent from 'LastSpace' unless these are fake parentheses encapsulating
1245 // a builder type call after 'return' or, if the alignment after opening
1246 // brackets is disabled.
1247 if (!Current.isTrailingComment() &&
1248 (Style.AlignOperands || *I < prec::Assignment) &&
1249 (!Previous || Previous->isNot(tok::kw_return) ||
1250 (Style.Language != FormatStyle::LK_Java && *I > 0)) &&
1251 (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
1252 *I != prec::Comma || Current.NestingLevel == 0))
1253 NewParenState.Indent =
1254 std::max(std::max(State.Column, NewParenState.Indent),
1255 State.Stack.back().LastSpace);
1256
1257 // Do not indent relative to the fake parentheses inserted for "." or "->".
1258 // This is a special case to make the following to statements consistent:
1259 // OuterFunction(InnerFunctionCall( // break
1260 // ParameterToInnerFunction));
1261 // OuterFunction(SomeObject.InnerFunctionCall( // break
1262 // ParameterToInnerFunction));
1263 if (*I > prec::Unknown)
1264 NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column);
1265 if (*I != prec::Conditional && !Current.is(TT_UnaryOperator) &&
1266 Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign)
1267 NewParenState.StartOfFunctionCall = State.Column;
1268
1269 // Always indent conditional expressions. Never indent expression where
1270 // the 'operator' is ',', ';' or an assignment (i.e. *I <=
1271 // prec::Assignment) as those have different indentation rules. Indent
1272 // other expression, unless the indentation needs to be skipped.
1273 if (*I == prec::Conditional ||
1274 (!SkipFirstExtraIndent && *I > prec::Assignment &&
1275 !Current.isTrailingComment()))
1276 NewParenState.Indent += Style.ContinuationIndentWidth;
1277 if ((Previous && !Previous->opensScope()) || *I != prec::Comma)
1278 NewParenState.BreakBeforeParameter = false;
1279 State.Stack.push_back(NewParenState);
1280 SkipFirstExtraIndent = false;
1281 }
1282 }
1283
moveStatePastFakeRParens(LineState & State)1284 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
1285 for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
1286 unsigned VariablePos = State.Stack.back().VariablePos;
1287 if (State.Stack.size() == 1) {
1288 // Do not pop the last element.
1289 break;
1290 }
1291 State.Stack.pop_back();
1292 State.Stack.back().VariablePos = VariablePos;
1293 }
1294 }
1295
moveStatePastScopeOpener(LineState & State,bool Newline)1296 void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
1297 bool Newline) {
1298 const FormatToken &Current = *State.NextToken;
1299 if (!Current.opensScope())
1300 return;
1301
1302 if (Current.MatchingParen && Current.BlockKind == BK_Block) {
1303 moveStateToNewBlock(State);
1304 return;
1305 }
1306
1307 unsigned NewIndent;
1308 unsigned LastSpace = State.Stack.back().LastSpace;
1309 bool AvoidBinPacking;
1310 bool BreakBeforeParameter = false;
1311 unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall,
1312 State.Stack.back().NestedBlockIndent);
1313 if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) ||
1314 opensProtoMessageField(Current, Style)) {
1315 if (Current.opensBlockOrBlockTypeList(Style)) {
1316 NewIndent = Style.IndentWidth +
1317 std::min(State.Column, State.Stack.back().NestedBlockIndent);
1318 } else {
1319 NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth;
1320 }
1321 const FormatToken *NextNoComment = Current.getNextNonComment();
1322 bool EndsInComma = Current.MatchingParen &&
1323 Current.MatchingParen->Previous &&
1324 Current.MatchingParen->Previous->is(tok::comma);
1325 AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) ||
1326 Style.Language == FormatStyle::LK_Proto ||
1327 Style.Language == FormatStyle::LK_TextProto ||
1328 !Style.BinPackArguments ||
1329 (NextNoComment &&
1330 NextNoComment->isOneOf(TT_DesignatedInitializerPeriod,
1331 TT_DesignatedInitializerLSquare));
1332 BreakBeforeParameter = EndsInComma;
1333 if (Current.ParameterCount > 1)
1334 NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1);
1335 } else {
1336 NewIndent = Style.ContinuationIndentWidth +
1337 std::max(State.Stack.back().LastSpace,
1338 State.Stack.back().StartOfFunctionCall);
1339
1340 // Ensure that different different brackets force relative alignment, e.g.:
1341 // void SomeFunction(vector< // break
1342 // int> v);
1343 // FIXME: We likely want to do this for more combinations of brackets.
1344 if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) {
1345 NewIndent = std::max(NewIndent, State.Stack.back().Indent);
1346 LastSpace = std::max(LastSpace, State.Stack.back().Indent);
1347 }
1348
1349 bool EndsInComma =
1350 Current.MatchingParen &&
1351 Current.MatchingParen->getPreviousNonComment() &&
1352 Current.MatchingParen->getPreviousNonComment()->is(tok::comma);
1353
1354 // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters
1355 // for backwards compatibility.
1356 bool ObjCBinPackProtocolList =
1357 (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto &&
1358 Style.BinPackParameters) ||
1359 Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always;
1360
1361 bool BinPackDeclaration =
1362 (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) ||
1363 (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList);
1364
1365 AvoidBinPacking =
1366 (Style.Language == FormatStyle::LK_JavaScript && EndsInComma) ||
1367 (State.Line->MustBeDeclaration && !BinPackDeclaration) ||
1368 (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
1369 (Style.ExperimentalAutoDetectBinPacking &&
1370 (Current.PackingKind == PPK_OnePerLine ||
1371 (!BinPackInconclusiveFunctions &&
1372 Current.PackingKind == PPK_Inconclusive)));
1373
1374 if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) {
1375 if (Style.ColumnLimit) {
1376 // If this '[' opens an ObjC call, determine whether all parameters fit
1377 // into one line and put one per line if they don't.
1378 if (getLengthToMatchingParen(Current, State.Stack) + State.Column >
1379 getColumnLimit(State))
1380 BreakBeforeParameter = true;
1381 } else {
1382 // For ColumnLimit = 0, we have to figure out whether there is or has to
1383 // be a line break within this call.
1384 for (const FormatToken *Tok = &Current;
1385 Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
1386 if (Tok->MustBreakBefore ||
1387 (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
1388 BreakBeforeParameter = true;
1389 break;
1390 }
1391 }
1392 }
1393 }
1394
1395 if (Style.Language == FormatStyle::LK_JavaScript && EndsInComma)
1396 BreakBeforeParameter = true;
1397 }
1398 // Generally inherit NoLineBreak from the current scope to nested scope.
1399 // However, don't do this for non-empty nested blocks, dict literals and
1400 // array literals as these follow different indentation rules.
1401 bool NoLineBreak =
1402 Current.Children.empty() &&
1403 !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) &&
1404 (State.Stack.back().NoLineBreak ||
1405 State.Stack.back().NoLineBreakInOperand ||
1406 (Current.is(TT_TemplateOpener) &&
1407 State.Stack.back().ContainsUnwrappedBuilder));
1408 State.Stack.push_back(
1409 ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak));
1410 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1411 State.Stack.back().BreakBeforeParameter = BreakBeforeParameter;
1412 State.Stack.back().HasMultipleNestedBlocks = Current.BlockParameterCount > 1;
1413 State.Stack.back().IsInsideObjCArrayLiteral =
1414 Current.is(TT_ArrayInitializerLSquare) && Current.Previous &&
1415 Current.Previous->is(tok::at);
1416 }
1417
moveStatePastScopeCloser(LineState & State)1418 void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) {
1419 const FormatToken &Current = *State.NextToken;
1420 if (!Current.closesScope())
1421 return;
1422
1423 // If we encounter a closing ), ], } or >, we can remove a level from our
1424 // stacks.
1425 if (State.Stack.size() > 1 &&
1426 (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) ||
1427 (Current.is(tok::r_brace) && State.NextToken != State.Line->First) ||
1428 State.NextToken->is(TT_TemplateCloser) ||
1429 (Current.is(tok::greater) && Current.is(TT_DictLiteral))))
1430 State.Stack.pop_back();
1431
1432 // Reevaluate whether ObjC message arguments fit into one line.
1433 // If a receiver spans multiple lines, e.g.:
1434 // [[object block:^{
1435 // return 42;
1436 // }] a:42 b:42];
1437 // BreakBeforeParameter is calculated based on an incorrect assumption
1438 // (it is checked whether the whole expression fits into one line without
1439 // considering a line break inside a message receiver).
1440 // We check whether arguements fit after receiver scope closer (into the same
1441 // line).
1442 if (State.Stack.back().BreakBeforeParameter && Current.MatchingParen &&
1443 Current.MatchingParen->Previous) {
1444 const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous;
1445 if (CurrentScopeOpener.is(TT_ObjCMethodExpr) &&
1446 CurrentScopeOpener.MatchingParen) {
1447 int NecessarySpaceInLine =
1448 getLengthToMatchingParen(CurrentScopeOpener, State.Stack) +
1449 CurrentScopeOpener.TotalLength - Current.TotalLength - 1;
1450 if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <=
1451 Style.ColumnLimit)
1452 State.Stack.back().BreakBeforeParameter = false;
1453 }
1454 }
1455
1456 if (Current.is(tok::r_square)) {
1457 // If this ends the array subscript expr, reset the corresponding value.
1458 const FormatToken *NextNonComment = Current.getNextNonComment();
1459 if (NextNonComment && NextNonComment->isNot(tok::l_square))
1460 State.Stack.back().StartOfArraySubscripts = 0;
1461 }
1462 }
1463
moveStateToNewBlock(LineState & State)1464 void ContinuationIndenter::moveStateToNewBlock(LineState &State) {
1465 unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
1466 // ObjC block sometimes follow special indentation rules.
1467 unsigned NewIndent =
1468 NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace)
1469 ? Style.ObjCBlockIndentWidth
1470 : Style.IndentWidth);
1471 State.Stack.push_back(ParenState(State.NextToken, NewIndent,
1472 State.Stack.back().LastSpace,
1473 /*AvoidBinPacking=*/true,
1474 /*NoLineBreak=*/false));
1475 State.Stack.back().NestedBlockIndent = NestedBlockIndent;
1476 State.Stack.back().BreakBeforeParameter = true;
1477 }
1478
getLastLineEndColumn(StringRef Text,unsigned StartColumn,unsigned TabWidth,encoding::Encoding Encoding)1479 static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn,
1480 unsigned TabWidth,
1481 encoding::Encoding Encoding) {
1482 size_t LastNewlinePos = Text.find_last_of("\n");
1483 if (LastNewlinePos == StringRef::npos) {
1484 return StartColumn +
1485 encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding);
1486 } else {
1487 return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos),
1488 /*StartColumn=*/0, TabWidth, Encoding);
1489 }
1490 }
1491
reformatRawStringLiteral(const FormatToken & Current,LineState & State,const FormatStyle & RawStringStyle,bool DryRun,bool Newline)1492 unsigned ContinuationIndenter::reformatRawStringLiteral(
1493 const FormatToken &Current, LineState &State,
1494 const FormatStyle &RawStringStyle, bool DryRun, bool Newline) {
1495 unsigned StartColumn = State.Column - Current.ColumnWidth;
1496 StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText);
1497 StringRef NewDelimiter =
1498 getCanonicalRawStringDelimiter(Style, RawStringStyle.Language);
1499 if (NewDelimiter.empty() || OldDelimiter.empty())
1500 NewDelimiter = OldDelimiter;
1501 // The text of a raw string is between the leading 'R"delimiter(' and the
1502 // trailing 'delimiter)"'.
1503 unsigned OldPrefixSize = 3 + OldDelimiter.size();
1504 unsigned OldSuffixSize = 2 + OldDelimiter.size();
1505 // We create a virtual text environment which expects a null-terminated
1506 // string, so we cannot use StringRef.
1507 std::string RawText =
1508 Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize);
1509 if (NewDelimiter != OldDelimiter) {
1510 // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the
1511 // raw string.
1512 std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str();
1513 if (StringRef(RawText).contains(CanonicalDelimiterSuffix))
1514 NewDelimiter = OldDelimiter;
1515 }
1516
1517 unsigned NewPrefixSize = 3 + NewDelimiter.size();
1518 unsigned NewSuffixSize = 2 + NewDelimiter.size();
1519
1520 // The first start column is the column the raw text starts after formatting.
1521 unsigned FirstStartColumn = StartColumn + NewPrefixSize;
1522
1523 // The next start column is the intended indentation a line break inside
1524 // the raw string at level 0. It is determined by the following rules:
1525 // - if the content starts on newline, it is one level more than the current
1526 // indent, and
1527 // - if the content does not start on a newline, it is the first start
1528 // column.
1529 // These rules have the advantage that the formatted content both does not
1530 // violate the rectangle rule and visually flows within the surrounding
1531 // source.
1532 bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n';
1533 // If this token is the last parameter (checked by looking if it's followed by
1534 // `)` and is not on a newline, the base the indent off the line's nested
1535 // block indent. Otherwise, base the indent off the arguments indent, so we
1536 // can achieve:
1537 //
1538 // fffffffffff(1, 2, 3, R"pb(
1539 // key1: 1 #
1540 // key2: 2)pb");
1541 //
1542 // fffffffffff(1, 2, 3,
1543 // R"pb(
1544 // key1: 1 #
1545 // key2: 2
1546 // )pb");
1547 //
1548 // fffffffffff(1, 2, 3,
1549 // R"pb(
1550 // key1: 1 #
1551 // key2: 2
1552 // )pb",
1553 // 5);
1554 unsigned CurrentIndent =
1555 (!Newline && Current.Next && Current.Next->is(tok::r_paren))
1556 ? State.Stack.back().NestedBlockIndent
1557 : State.Stack.back().Indent;
1558 unsigned NextStartColumn = ContentStartsOnNewline
1559 ? CurrentIndent + Style.IndentWidth
1560 : FirstStartColumn;
1561
1562 // The last start column is the column the raw string suffix starts if it is
1563 // put on a newline.
1564 // The last start column is the intended indentation of the raw string postfix
1565 // if it is put on a newline. It is determined by the following rules:
1566 // - if the raw string prefix starts on a newline, it is the column where
1567 // that raw string prefix starts, and
1568 // - if the raw string prefix does not start on a newline, it is the current
1569 // indent.
1570 unsigned LastStartColumn =
1571 Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent;
1572
1573 std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat(
1574 RawStringStyle, RawText, {tooling::Range(0, RawText.size())},
1575 FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>",
1576 /*Status=*/nullptr);
1577
1578 auto NewCode = applyAllReplacements(RawText, Fixes.first);
1579 tooling::Replacements NoFixes;
1580 if (!NewCode) {
1581 return addMultilineToken(Current, State);
1582 }
1583 if (!DryRun) {
1584 if (NewDelimiter != OldDelimiter) {
1585 // In 'R"delimiter(...', the delimiter starts 2 characters after the start
1586 // of the token.
1587 SourceLocation PrefixDelimiterStart =
1588 Current.Tok.getLocation().getLocWithOffset(2);
1589 auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement(
1590 SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1591 if (PrefixErr) {
1592 llvm::errs()
1593 << "Failed to update the prefix delimiter of a raw string: "
1594 << llvm::toString(std::move(PrefixErr)) << "\n";
1595 }
1596 // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at
1597 // position length - 1 - |delimiter|.
1598 SourceLocation SuffixDelimiterStart =
1599 Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() -
1600 1 - OldDelimiter.size());
1601 auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement(
1602 SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter));
1603 if (SuffixErr) {
1604 llvm::errs()
1605 << "Failed to update the suffix delimiter of a raw string: "
1606 << llvm::toString(std::move(SuffixErr)) << "\n";
1607 }
1608 }
1609 SourceLocation OriginLoc =
1610 Current.Tok.getLocation().getLocWithOffset(OldPrefixSize);
1611 for (const tooling::Replacement &Fix : Fixes.first) {
1612 auto Err = Whitespaces.addReplacement(tooling::Replacement(
1613 SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()),
1614 Fix.getLength(), Fix.getReplacementText()));
1615 if (Err) {
1616 llvm::errs() << "Failed to reformat raw string: "
1617 << llvm::toString(std::move(Err)) << "\n";
1618 }
1619 }
1620 }
1621 unsigned RawLastLineEndColumn = getLastLineEndColumn(
1622 *NewCode, FirstStartColumn, Style.TabWidth, Encoding);
1623 State.Column = RawLastLineEndColumn + NewSuffixSize;
1624 // Since we're updating the column to after the raw string literal here, we
1625 // have to manually add the penalty for the prefix R"delim( over the column
1626 // limit.
1627 unsigned PrefixExcessCharacters =
1628 StartColumn + NewPrefixSize > Style.ColumnLimit
1629 ? StartColumn + NewPrefixSize - Style.ColumnLimit
1630 : 0;
1631 bool IsMultiline =
1632 ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos);
1633 if (IsMultiline) {
1634 // Break before further function parameters on all levels.
1635 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1636 State.Stack[i].BreakBeforeParameter = true;
1637 }
1638 return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter;
1639 }
1640
addMultilineToken(const FormatToken & Current,LineState & State)1641 unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current,
1642 LineState &State) {
1643 // Break before further function parameters on all levels.
1644 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1645 State.Stack[i].BreakBeforeParameter = true;
1646
1647 unsigned ColumnsUsed = State.Column;
1648 // We can only affect layout of the first and the last line, so the penalty
1649 // for all other lines is constant, and we ignore it.
1650 State.Column = Current.LastLineColumnWidth;
1651
1652 if (ColumnsUsed > getColumnLimit(State))
1653 return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State));
1654 return 0;
1655 }
1656
handleEndOfLine(const FormatToken & Current,LineState & State,bool DryRun,bool AllowBreak,bool Newline)1657 unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current,
1658 LineState &State, bool DryRun,
1659 bool AllowBreak, bool Newline) {
1660 unsigned Penalty = 0;
1661 // Compute the raw string style to use in case this is a raw string literal
1662 // that can be reformatted.
1663 auto RawStringStyle = getRawStringStyle(Current, State);
1664 if (RawStringStyle && !Current.Finalized) {
1665 Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun,
1666 Newline);
1667 } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) {
1668 // Don't break multi-line tokens other than block comments and raw string
1669 // literals. Instead, just update the state.
1670 Penalty = addMultilineToken(Current, State);
1671 } else if (State.Line->Type != LT_ImportStatement) {
1672 // We generally don't break import statements.
1673 LineState OriginalState = State;
1674
1675 // Whether we force the reflowing algorithm to stay strictly within the
1676 // column limit.
1677 bool Strict = false;
1678 // Whether the first non-strict attempt at reflowing did intentionally
1679 // exceed the column limit.
1680 bool Exceeded = false;
1681 std::tie(Penalty, Exceeded) = breakProtrudingToken(
1682 Current, State, AllowBreak, /*DryRun=*/true, Strict);
1683 if (Exceeded) {
1684 // If non-strict reflowing exceeds the column limit, try whether strict
1685 // reflowing leads to an overall lower penalty.
1686 LineState StrictState = OriginalState;
1687 unsigned StrictPenalty =
1688 breakProtrudingToken(Current, StrictState, AllowBreak,
1689 /*DryRun=*/true, /*Strict=*/true)
1690 .first;
1691 Strict = StrictPenalty <= Penalty;
1692 if (Strict) {
1693 Penalty = StrictPenalty;
1694 State = StrictState;
1695 }
1696 }
1697 if (!DryRun) {
1698 // If we're not in dry-run mode, apply the changes with the decision on
1699 // strictness made above.
1700 breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false,
1701 Strict);
1702 }
1703 }
1704 if (State.Column > getColumnLimit(State)) {
1705 unsigned ExcessCharacters = State.Column - getColumnLimit(State);
1706 Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
1707 }
1708 return Penalty;
1709 }
1710
1711 // Returns the enclosing function name of a token, or the empty string if not
1712 // found.
getEnclosingFunctionName(const FormatToken & Current)1713 static StringRef getEnclosingFunctionName(const FormatToken &Current) {
1714 // Look for: 'function(' or 'function<templates>(' before Current.
1715 auto Tok = Current.getPreviousNonComment();
1716 if (!Tok || !Tok->is(tok::l_paren))
1717 return "";
1718 Tok = Tok->getPreviousNonComment();
1719 if (!Tok)
1720 return "";
1721 if (Tok->is(TT_TemplateCloser)) {
1722 Tok = Tok->MatchingParen;
1723 if (Tok)
1724 Tok = Tok->getPreviousNonComment();
1725 }
1726 if (!Tok || !Tok->is(tok::identifier))
1727 return "";
1728 return Tok->TokenText;
1729 }
1730
1731 llvm::Optional<FormatStyle>
getRawStringStyle(const FormatToken & Current,const LineState & State)1732 ContinuationIndenter::getRawStringStyle(const FormatToken &Current,
1733 const LineState &State) {
1734 if (!Current.isStringLiteral())
1735 return None;
1736 auto Delimiter = getRawStringDelimiter(Current.TokenText);
1737 if (!Delimiter)
1738 return None;
1739 auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter);
1740 if (!RawStringStyle && Delimiter->empty())
1741 RawStringStyle = RawStringFormats.getEnclosingFunctionStyle(
1742 getEnclosingFunctionName(Current));
1743 if (!RawStringStyle)
1744 return None;
1745 RawStringStyle->ColumnLimit = getColumnLimit(State);
1746 return RawStringStyle;
1747 }
1748
1749 std::unique_ptr<BreakableToken>
createBreakableToken(const FormatToken & Current,LineState & State,bool AllowBreak)1750 ContinuationIndenter::createBreakableToken(const FormatToken &Current,
1751 LineState &State, bool AllowBreak) {
1752 unsigned StartColumn = State.Column - Current.ColumnWidth;
1753 if (Current.isStringLiteral()) {
1754 // FIXME: String literal breaking is currently disabled for C#,Java and
1755 // JavaScript, as it requires strings to be merged using "+" which we
1756 // don't support.
1757 if (Style.Language == FormatStyle::LK_Java ||
1758 Style.Language == FormatStyle::LK_JavaScript || Style.isCSharp() ||
1759 !Style.BreakStringLiterals || !AllowBreak)
1760 return nullptr;
1761
1762 // Don't break string literals inside preprocessor directives (except for
1763 // #define directives, as their contents are stored in separate lines and
1764 // are not affected by this check).
1765 // This way we avoid breaking code with line directives and unknown
1766 // preprocessor directives that contain long string literals.
1767 if (State.Line->Type == LT_PreprocessorDirective)
1768 return nullptr;
1769 // Exempts unterminated string literals from line breaking. The user will
1770 // likely want to terminate the string before any line breaking is done.
1771 if (Current.IsUnterminatedLiteral)
1772 return nullptr;
1773 // Don't break string literals inside Objective-C array literals (doing so
1774 // raises the warning -Wobjc-string-concatenation).
1775 if (State.Stack.back().IsInsideObjCArrayLiteral) {
1776 return nullptr;
1777 }
1778
1779 StringRef Text = Current.TokenText;
1780 StringRef Prefix;
1781 StringRef Postfix;
1782 // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'.
1783 // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to
1784 // reduce the overhead) for each FormatToken, which is a string, so that we
1785 // don't run multiple checks here on the hot path.
1786 if ((Text.endswith(Postfix = "\"") &&
1787 (Text.startswith(Prefix = "@\"") || Text.startswith(Prefix = "\"") ||
1788 Text.startswith(Prefix = "u\"") || Text.startswith(Prefix = "U\"") ||
1789 Text.startswith(Prefix = "u8\"") ||
1790 Text.startswith(Prefix = "L\""))) ||
1791 (Text.startswith(Prefix = "_T(\"") && Text.endswith(Postfix = "\")"))) {
1792 // We need this to address the case where there is an unbreakable tail
1793 // only if certain other formatting decisions have been taken. The
1794 // UnbreakableTailLength of Current is an overapproximation is that case
1795 // and we need to be correct here.
1796 unsigned UnbreakableTailLength = (State.NextToken && canBreak(State))
1797 ? 0
1798 : Current.UnbreakableTailLength;
1799 return llvm::make_unique<BreakableStringLiteral>(
1800 Current, StartColumn, Prefix, Postfix, UnbreakableTailLength,
1801 State.Line->InPPDirective, Encoding, Style);
1802 }
1803 } else if (Current.is(TT_BlockComment)) {
1804 if (!Style.ReflowComments ||
1805 // If a comment token switches formatting, like
1806 // /* clang-format on */, we don't want to break it further,
1807 // but we may still want to adjust its indentation.
1808 switchesFormatting(Current)) {
1809 return nullptr;
1810 }
1811 return llvm::make_unique<BreakableBlockComment>(
1812 Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1813 State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF());
1814 } else if (Current.is(TT_LineComment) &&
1815 (Current.Previous == nullptr ||
1816 Current.Previous->isNot(TT_ImplicitStringLiteral))) {
1817 if (!Style.ReflowComments ||
1818 CommentPragmasRegex.match(Current.TokenText.substr(2)) ||
1819 switchesFormatting(Current))
1820 return nullptr;
1821 return llvm::make_unique<BreakableLineCommentSection>(
1822 Current, StartColumn, Current.OriginalColumn, !Current.Previous,
1823 /*InPPDirective=*/false, Encoding, Style);
1824 }
1825 return nullptr;
1826 }
1827
1828 std::pair<unsigned, bool>
breakProtrudingToken(const FormatToken & Current,LineState & State,bool AllowBreak,bool DryRun,bool Strict)1829 ContinuationIndenter::breakProtrudingToken(const FormatToken &Current,
1830 LineState &State, bool AllowBreak,
1831 bool DryRun, bool Strict) {
1832 std::unique_ptr<const BreakableToken> Token =
1833 createBreakableToken(Current, State, AllowBreak);
1834 if (!Token)
1835 return {0, false};
1836 assert(Token->getLineCount() > 0);
1837 unsigned ColumnLimit = getColumnLimit(State);
1838 if (Current.is(TT_LineComment)) {
1839 // We don't insert backslashes when breaking line comments.
1840 ColumnLimit = Style.ColumnLimit;
1841 }
1842 if (Current.UnbreakableTailLength >= ColumnLimit)
1843 return {0, false};
1844 // ColumnWidth was already accounted into State.Column before calling
1845 // breakProtrudingToken.
1846 unsigned StartColumn = State.Column - Current.ColumnWidth;
1847 unsigned NewBreakPenalty = Current.isStringLiteral()
1848 ? Style.PenaltyBreakString
1849 : Style.PenaltyBreakComment;
1850 // Stores whether we intentionally decide to let a line exceed the column
1851 // limit.
1852 bool Exceeded = false;
1853 // Stores whether we introduce a break anywhere in the token.
1854 bool BreakInserted = Token->introducesBreakBeforeToken();
1855 // Store whether we inserted a new line break at the end of the previous
1856 // logical line.
1857 bool NewBreakBefore = false;
1858 // We use a conservative reflowing strategy. Reflow starts after a line is
1859 // broken or the corresponding whitespace compressed. Reflow ends as soon as a
1860 // line that doesn't get reflown with the previous line is reached.
1861 bool Reflow = false;
1862 // Keep track of where we are in the token:
1863 // Where we are in the content of the current logical line.
1864 unsigned TailOffset = 0;
1865 // The column number we're currently at.
1866 unsigned ContentStartColumn =
1867 Token->getContentStartColumn(0, /*Break=*/false);
1868 // The number of columns left in the current logical line after TailOffset.
1869 unsigned RemainingTokenColumns =
1870 Token->getRemainingLength(0, TailOffset, ContentStartColumn);
1871 // Adapt the start of the token, for example indent.
1872 if (!DryRun)
1873 Token->adaptStartOfLine(0, Whitespaces);
1874
1875 unsigned ContentIndent = 0;
1876 unsigned Penalty = 0;
1877 LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column "
1878 << StartColumn << ".\n");
1879 for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
1880 LineIndex != EndIndex; ++LineIndex) {
1881 LLVM_DEBUG(llvm::dbgs()
1882 << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n");
1883 NewBreakBefore = false;
1884 // If we did reflow the previous line, we'll try reflowing again. Otherwise
1885 // we'll start reflowing if the current line is broken or whitespace is
1886 // compressed.
1887 bool TryReflow = Reflow;
1888 // Break the current token until we can fit the rest of the line.
1889 while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
1890 LLVM_DEBUG(llvm::dbgs() << " Over limit, need: "
1891 << (ContentStartColumn + RemainingTokenColumns)
1892 << ", space: " << ColumnLimit
1893 << ", reflown prefix: " << ContentStartColumn
1894 << ", offset in line: " << TailOffset << "\n");
1895 // If the current token doesn't fit, find the latest possible split in the
1896 // current line so that breaking at it will be under the column limit.
1897 // FIXME: Use the earliest possible split while reflowing to correctly
1898 // compress whitespace within a line.
1899 BreakableToken::Split Split =
1900 Token->getSplit(LineIndex, TailOffset, ColumnLimit,
1901 ContentStartColumn, CommentPragmasRegex);
1902 if (Split.first == StringRef::npos) {
1903 // No break opportunity - update the penalty and continue with the next
1904 // logical line.
1905 if (LineIndex < EndIndex - 1)
1906 // The last line's penalty is handled in addNextStateToQueue() or when
1907 // calling replaceWhitespaceAfterLastLine below.
1908 Penalty += Style.PenaltyExcessCharacter *
1909 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
1910 LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n");
1911 break;
1912 }
1913 assert(Split.first != 0);
1914
1915 if (Token->supportsReflow()) {
1916 // Check whether the next natural split point after the current one can
1917 // still fit the line, either because we can compress away whitespace,
1918 // or because the penalty the excess characters introduce is lower than
1919 // the break penalty.
1920 // We only do this for tokens that support reflowing, and thus allow us
1921 // to change the whitespace arbitrarily (e.g. comments).
1922 // Other tokens, like string literals, can be broken on arbitrary
1923 // positions.
1924
1925 // First, compute the columns from TailOffset to the next possible split
1926 // position.
1927 // For example:
1928 // ColumnLimit: |
1929 // // Some text that breaks
1930 // ^ tail offset
1931 // ^-- split
1932 // ^-------- to split columns
1933 // ^--- next split
1934 // ^--------------- to next split columns
1935 unsigned ToSplitColumns = Token->getRangeLength(
1936 LineIndex, TailOffset, Split.first, ContentStartColumn);
1937 LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n");
1938
1939 BreakableToken::Split NextSplit = Token->getSplit(
1940 LineIndex, TailOffset + Split.first + Split.second, ColumnLimit,
1941 ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex);
1942 // Compute the columns necessary to fit the next non-breakable sequence
1943 // into the current line.
1944 unsigned ToNextSplitColumns = 0;
1945 if (NextSplit.first == StringRef::npos) {
1946 ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset,
1947 ContentStartColumn);
1948 } else {
1949 ToNextSplitColumns = Token->getRangeLength(
1950 LineIndex, TailOffset,
1951 Split.first + Split.second + NextSplit.first, ContentStartColumn);
1952 }
1953 // Compress the whitespace between the break and the start of the next
1954 // unbreakable sequence.
1955 ToNextSplitColumns =
1956 Token->getLengthAfterCompression(ToNextSplitColumns, Split);
1957 LLVM_DEBUG(llvm::dbgs()
1958 << " ContentStartColumn: " << ContentStartColumn << "\n");
1959 LLVM_DEBUG(llvm::dbgs()
1960 << " ToNextSplit: " << ToNextSplitColumns << "\n");
1961 // If the whitespace compression makes us fit, continue on the current
1962 // line.
1963 bool ContinueOnLine =
1964 ContentStartColumn + ToNextSplitColumns <= ColumnLimit;
1965 unsigned ExcessCharactersPenalty = 0;
1966 if (!ContinueOnLine && !Strict) {
1967 // Similarly, if the excess characters' penalty is lower than the
1968 // penalty of introducing a new break, continue on the current line.
1969 ExcessCharactersPenalty =
1970 (ContentStartColumn + ToNextSplitColumns - ColumnLimit) *
1971 Style.PenaltyExcessCharacter;
1972 LLVM_DEBUG(llvm::dbgs()
1973 << " Penalty excess: " << ExcessCharactersPenalty
1974 << "\n break : " << NewBreakPenalty << "\n");
1975 if (ExcessCharactersPenalty < NewBreakPenalty) {
1976 Exceeded = true;
1977 ContinueOnLine = true;
1978 }
1979 }
1980 if (ContinueOnLine) {
1981 LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n");
1982 // The current line fits after compressing the whitespace - reflow
1983 // the next line into it if possible.
1984 TryReflow = true;
1985 if (!DryRun)
1986 Token->compressWhitespace(LineIndex, TailOffset, Split,
1987 Whitespaces);
1988 // When we continue on the same line, leave one space between content.
1989 ContentStartColumn += ToSplitColumns + 1;
1990 Penalty += ExcessCharactersPenalty;
1991 TailOffset += Split.first + Split.second;
1992 RemainingTokenColumns = Token->getRemainingLength(
1993 LineIndex, TailOffset, ContentStartColumn);
1994 continue;
1995 }
1996 }
1997 LLVM_DEBUG(llvm::dbgs() << " Breaking...\n");
1998 // Update the ContentIndent only if the current line was not reflown with
1999 // the previous line, since in that case the previous line should still
2000 // determine the ContentIndent. Also never intent the last line.
2001 if (!Reflow)
2002 ContentIndent = Token->getContentIndent(LineIndex);
2003 LLVM_DEBUG(llvm::dbgs()
2004 << " ContentIndent: " << ContentIndent << "\n");
2005 ContentStartColumn = ContentIndent + Token->getContentStartColumn(
2006 LineIndex, /*Break=*/true);
2007
2008 unsigned NewRemainingTokenColumns = Token->getRemainingLength(
2009 LineIndex, TailOffset + Split.first + Split.second,
2010 ContentStartColumn);
2011 if (NewRemainingTokenColumns == 0) {
2012 // No content to indent.
2013 ContentIndent = 0;
2014 ContentStartColumn =
2015 Token->getContentStartColumn(LineIndex, /*Break=*/true);
2016 NewRemainingTokenColumns = Token->getRemainingLength(
2017 LineIndex, TailOffset + Split.first + Split.second,
2018 ContentStartColumn);
2019 }
2020
2021 // When breaking before a tab character, it may be moved by a few columns,
2022 // but will still be expanded to the next tab stop, so we don't save any
2023 // columns.
2024 if (NewRemainingTokenColumns == RemainingTokenColumns) {
2025 // FIXME: Do we need to adjust the penalty?
2026 break;
2027 }
2028 assert(NewRemainingTokenColumns < RemainingTokenColumns);
2029
2030 LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first
2031 << ", " << Split.second << "\n");
2032 if (!DryRun)
2033 Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent,
2034 Whitespaces);
2035
2036 Penalty += NewBreakPenalty;
2037 TailOffset += Split.first + Split.second;
2038 RemainingTokenColumns = NewRemainingTokenColumns;
2039 BreakInserted = true;
2040 NewBreakBefore = true;
2041 }
2042 // In case there's another line, prepare the state for the start of the next
2043 // line.
2044 if (LineIndex + 1 != EndIndex) {
2045 unsigned NextLineIndex = LineIndex + 1;
2046 if (NewBreakBefore)
2047 // After breaking a line, try to reflow the next line into the current
2048 // one once RemainingTokenColumns fits.
2049 TryReflow = true;
2050 if (TryReflow) {
2051 // We decided that we want to try reflowing the next line into the
2052 // current one.
2053 // We will now adjust the state as if the reflow is successful (in
2054 // preparation for the next line), and see whether that works. If we
2055 // decide that we cannot reflow, we will later reset the state to the
2056 // start of the next line.
2057 Reflow = false;
2058 // As we did not continue breaking the line, RemainingTokenColumns is
2059 // known to fit after ContentStartColumn. Adapt ContentStartColumn to
2060 // the position at which we want to format the next line if we do
2061 // actually reflow.
2062 // When we reflow, we need to add a space between the end of the current
2063 // line and the next line's start column.
2064 ContentStartColumn += RemainingTokenColumns + 1;
2065 // Get the split that we need to reflow next logical line into the end
2066 // of the current one; the split will include any leading whitespace of
2067 // the next logical line.
2068 BreakableToken::Split SplitBeforeNext =
2069 Token->getReflowSplit(NextLineIndex, CommentPragmasRegex);
2070 LLVM_DEBUG(llvm::dbgs()
2071 << " Size of reflown text: " << ContentStartColumn
2072 << "\n Potential reflow split: ");
2073 if (SplitBeforeNext.first != StringRef::npos) {
2074 LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", "
2075 << SplitBeforeNext.second << "\n");
2076 TailOffset = SplitBeforeNext.first + SplitBeforeNext.second;
2077 // If the rest of the next line fits into the current line below the
2078 // column limit, we can safely reflow.
2079 RemainingTokenColumns = Token->getRemainingLength(
2080 NextLineIndex, TailOffset, ContentStartColumn);
2081 Reflow = true;
2082 if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) {
2083 LLVM_DEBUG(llvm::dbgs()
2084 << " Over limit after reflow, need: "
2085 << (ContentStartColumn + RemainingTokenColumns)
2086 << ", space: " << ColumnLimit
2087 << ", reflown prefix: " << ContentStartColumn
2088 << ", offset in line: " << TailOffset << "\n");
2089 // If the whole next line does not fit, try to find a point in
2090 // the next line at which we can break so that attaching the part
2091 // of the next line to that break point onto the current line is
2092 // below the column limit.
2093 BreakableToken::Split Split =
2094 Token->getSplit(NextLineIndex, TailOffset, ColumnLimit,
2095 ContentStartColumn, CommentPragmasRegex);
2096 if (Split.first == StringRef::npos) {
2097 LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n");
2098 Reflow = false;
2099 } else {
2100 // Check whether the first split point gets us below the column
2101 // limit. Note that we will execute this split below as part of
2102 // the normal token breaking and reflow logic within the line.
2103 unsigned ToSplitColumns = Token->getRangeLength(
2104 NextLineIndex, TailOffset, Split.first, ContentStartColumn);
2105 if (ContentStartColumn + ToSplitColumns > ColumnLimit) {
2106 LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: "
2107 << (ContentStartColumn + ToSplitColumns)
2108 << ", space: " << ColumnLimit);
2109 unsigned ExcessCharactersPenalty =
2110 (ContentStartColumn + ToSplitColumns - ColumnLimit) *
2111 Style.PenaltyExcessCharacter;
2112 if (NewBreakPenalty < ExcessCharactersPenalty) {
2113 Reflow = false;
2114 }
2115 }
2116 }
2117 }
2118 } else {
2119 LLVM_DEBUG(llvm::dbgs() << "not found.\n");
2120 }
2121 }
2122 if (!Reflow) {
2123 // If we didn't reflow into the next line, the only space to consider is
2124 // the next logical line. Reset our state to match the start of the next
2125 // line.
2126 TailOffset = 0;
2127 ContentStartColumn =
2128 Token->getContentStartColumn(NextLineIndex, /*Break=*/false);
2129 RemainingTokenColumns = Token->getRemainingLength(
2130 NextLineIndex, TailOffset, ContentStartColumn);
2131 // Adapt the start of the token, for example indent.
2132 if (!DryRun)
2133 Token->adaptStartOfLine(NextLineIndex, Whitespaces);
2134 } else {
2135 // If we found a reflow split and have added a new break before the next
2136 // line, we are going to remove the line break at the start of the next
2137 // logical line. For example, here we'll add a new line break after
2138 // 'text', and subsequently delete the line break between 'that' and
2139 // 'reflows'.
2140 // // some text that
2141 // // reflows
2142 // ->
2143 // // some text
2144 // // that reflows
2145 // When adding the line break, we also added the penalty for it, so we
2146 // need to subtract that penalty again when we remove the line break due
2147 // to reflowing.
2148 if (NewBreakBefore) {
2149 assert(Penalty >= NewBreakPenalty);
2150 Penalty -= NewBreakPenalty;
2151 }
2152 if (!DryRun)
2153 Token->reflow(NextLineIndex, Whitespaces);
2154 }
2155 }
2156 }
2157
2158 BreakableToken::Split SplitAfterLastLine =
2159 Token->getSplitAfterLastLine(TailOffset);
2160 if (SplitAfterLastLine.first != StringRef::npos) {
2161 LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n");
2162
2163 // We add the last line's penalty here, since that line is going to be split
2164 // now.
2165 Penalty += Style.PenaltyExcessCharacter *
2166 (ContentStartColumn + RemainingTokenColumns - ColumnLimit);
2167
2168 if (!DryRun)
2169 Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine,
2170 Whitespaces);
2171 ContentStartColumn =
2172 Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true);
2173 RemainingTokenColumns = Token->getRemainingLength(
2174 Token->getLineCount() - 1,
2175 TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second,
2176 ContentStartColumn);
2177 }
2178
2179 State.Column = ContentStartColumn + RemainingTokenColumns -
2180 Current.UnbreakableTailLength;
2181
2182 if (BreakInserted) {
2183 // If we break the token inside a parameter list, we need to break before
2184 // the next parameter on all levels, so that the next parameter is clearly
2185 // visible. Line comments already introduce a break.
2186 if (Current.isNot(TT_LineComment)) {
2187 for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
2188 State.Stack[i].BreakBeforeParameter = true;
2189 }
2190
2191 if (Current.is(TT_BlockComment))
2192 State.NoContinuation = true;
2193
2194 State.Stack.back().LastSpace = StartColumn;
2195 }
2196
2197 Token->updateNextToken(State);
2198
2199 return {Penalty, Exceeded};
2200 }
2201
getColumnLimit(const LineState & State) const2202 unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const {
2203 // In preprocessor directives reserve two chars for trailing " \"
2204 return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0);
2205 }
2206
nextIsMultilineString(const LineState & State)2207 bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
2208 const FormatToken &Current = *State.NextToken;
2209 if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral))
2210 return false;
2211 // We never consider raw string literals "multiline" for the purpose of
2212 // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased
2213 // (see TokenAnnotator::mustBreakBefore().
2214 if (Current.TokenText.startswith("R\""))
2215 return false;
2216 if (Current.IsMultiline)
2217 return true;
2218 if (Current.getNextNonComment() &&
2219 Current.getNextNonComment()->isStringLiteral())
2220 return true; // Implicit concatenation.
2221 if (Style.ColumnLimit != 0 && Style.BreakStringLiterals &&
2222 State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
2223 Style.ColumnLimit)
2224 return true; // String will be split.
2225 return false;
2226 }
2227
2228 } // namespace format
2229 } // namespace clang
2230