/src/llvm-project/clang/lib/Format/ContinuationIndenter.cpp
Line | Count | Source (jump to first uncovered line) |
1 | | //===--- ContinuationIndenter.cpp - Format C++ code -----------------------===// |
2 | | // |
3 | | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | | // See https://llvm.org/LICENSE.txt for license information. |
5 | | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | | // |
7 | | //===----------------------------------------------------------------------===// |
8 | | /// |
9 | | /// \file |
10 | | /// This file implements the continuation indenter. |
11 | | /// |
12 | | //===----------------------------------------------------------------------===// |
13 | | |
14 | | #include "ContinuationIndenter.h" |
15 | | #include "BreakableToken.h" |
16 | | #include "FormatInternal.h" |
17 | | #include "FormatToken.h" |
18 | | #include "WhitespaceManager.h" |
19 | | #include "clang/Basic/OperatorPrecedence.h" |
20 | | #include "clang/Basic/SourceManager.h" |
21 | | #include "clang/Basic/TokenKinds.h" |
22 | | #include "clang/Format/Format.h" |
23 | | #include "llvm/ADT/StringSet.h" |
24 | | #include "llvm/Support/Debug.h" |
25 | | #include <optional> |
26 | | |
27 | | #define DEBUG_TYPE "format-indenter" |
28 | | |
29 | | namespace clang { |
30 | | namespace format { |
31 | | |
32 | | // Returns true if a TT_SelectorName should be indented when wrapped, |
33 | | // false otherwise. |
34 | | static bool shouldIndentWrappedSelectorName(const FormatStyle &Style, |
35 | 5.87k | LineType LineType) { |
36 | 5.87k | return Style.IndentWrappedFunctionNames || LineType == LT_ObjCMethodDecl; |
37 | 5.87k | } |
38 | | |
39 | | // Returns true if a binary operator following \p Tok should be unindented when |
40 | | // the style permits it. |
41 | 405k | static bool shouldUnindentNextOperator(const FormatToken &Tok) { |
42 | 405k | const FormatToken *Previous = Tok.getPreviousNonComment(); |
43 | 405k | return Previous && (Previous->getPrecedence() == prec::Assignment || |
44 | 334k | Previous->isOneOf(tok::kw_return, TT_RequiresClause)); |
45 | 405k | } |
46 | | |
47 | | // Returns the length of everything up to the first possible line break after |
48 | | // the ), ], } or > matching \c Tok. |
49 | | static unsigned getLengthToMatchingParen(const FormatToken &Tok, |
50 | 19.9k | ArrayRef<ParenState> Stack) { |
51 | | // Normally whether or not a break before T is possible is calculated and |
52 | | // stored in T.CanBreakBefore. Braces, array initializers and text proto |
53 | | // messages like `key: < ... >` are an exception: a break is possible |
54 | | // before a closing brace R if a break was inserted after the corresponding |
55 | | // opening brace. The information about whether or not a break is needed |
56 | | // before a closing brace R is stored in the ParenState field |
57 | | // S.BreakBeforeClosingBrace where S is the state that R closes. |
58 | | // |
59 | | // In order to decide whether there can be a break before encountered right |
60 | | // braces, this implementation iterates over the sequence of tokens and over |
61 | | // the paren stack in lockstep, keeping track of the stack level which visited |
62 | | // right braces correspond to in MatchingStackIndex. |
63 | | // |
64 | | // For example, consider: |
65 | | // L. <- line number |
66 | | // 1. { |
67 | | // 2. {1}, |
68 | | // 3. {2}, |
69 | | // 4. {{3}}} |
70 | | // ^ where we call this method with this token. |
71 | | // The paren stack at this point contains 3 brace levels: |
72 | | // 0. { at line 1, BreakBeforeClosingBrace: true |
73 | | // 1. first { at line 4, BreakBeforeClosingBrace: false |
74 | | // 2. second { at line 4, BreakBeforeClosingBrace: false, |
75 | | // where there might be fake parens levels in-between these levels. |
76 | | // The algorithm will start at the first } on line 4, which is the matching |
77 | | // brace of the initial left brace and at level 2 of the stack. Then, |
78 | | // examining BreakBeforeClosingBrace: false at level 2, it will continue to |
79 | | // the second } on line 4, and will traverse the stack downwards until it |
80 | | // finds the matching { on level 1. Then, examining BreakBeforeClosingBrace: |
81 | | // false at level 1, it will continue to the third } on line 4 and will |
82 | | // traverse the stack downwards until it finds the matching { on level 0. |
83 | | // Then, examining BreakBeforeClosingBrace: true at level 0, the algorithm |
84 | | // will stop and will use the second } on line 4 to determine the length to |
85 | | // return, as in this example the range will include the tokens: {3}} |
86 | | // |
87 | | // The algorithm will only traverse the stack if it encounters braces, array |
88 | | // initializer squares or text proto angle brackets. |
89 | 19.9k | if (!Tok.MatchingParen) |
90 | 12 | return 0; |
91 | 19.9k | FormatToken *End = Tok.MatchingParen; |
92 | | // Maintains a stack level corresponding to the current End token. |
93 | 19.9k | int MatchingStackIndex = Stack.size() - 1; |
94 | | // Traverses the stack downwards, looking for the level to which LBrace |
95 | | // corresponds. Returns either a pointer to the matching level or nullptr if |
96 | | // LParen is not found in the initial portion of the stack up to |
97 | | // MatchingStackIndex. |
98 | 19.9k | auto FindParenState = [&](const FormatToken *LBrace) -> const ParenState * { |
99 | 10.7k | while (MatchingStackIndex >= 0 && Stack[MatchingStackIndex].Tok != LBrace) |
100 | 8.81k | --MatchingStackIndex; |
101 | 1.95k | return MatchingStackIndex >= 0 ? &Stack[MatchingStackIndex] : nullptr; |
102 | 1.95k | }; |
103 | 107k | for (; End->Next; End = End->Next) { |
104 | 106k | if (End->Next->CanBreakBefore) |
105 | 18.7k | break; |
106 | 87.3k | if (!End->Next->closesScope()) |
107 | 84.2k | continue; |
108 | 3.11k | if (End->Next->MatchingParen && |
109 | 3.11k | End->Next->MatchingParen->isOneOf( |
110 | 2.92k | tok::l_brace, TT_ArrayInitializerLSquare, tok::less)) { |
111 | 1.95k | const ParenState *State = FindParenState(End->Next->MatchingParen); |
112 | 1.95k | if (State && State->BreakBeforeClosingBrace) |
113 | 212 | break; |
114 | 1.95k | } |
115 | 3.11k | } |
116 | 19.9k | return End->TotalLength - Tok.TotalLength + 1; |
117 | 19.9k | } |
118 | | |
119 | 69.3k | static unsigned getLengthToNextOperator(const FormatToken &Tok) { |
120 | 69.3k | if (!Tok.NextOperator) |
121 | 43.3k | return 0; |
122 | 26.0k | return Tok.NextOperator->TotalLength - Tok.TotalLength; |
123 | 69.3k | } |
124 | | |
125 | | // Returns \c true if \c Tok is the "." or "->" of a call and starts the next |
126 | | // segment of a builder type call. |
127 | 6.95M | static bool startsSegmentOfBuilderTypeCall(const FormatToken &Tok) { |
128 | 6.95M | return Tok.isMemberAccess() && Tok.Previous && Tok.Previous->closesScope(); |
129 | 6.95M | } |
130 | | |
131 | | // Returns \c true if \c Current starts a new parameter. |
132 | | static bool startsNextParameter(const FormatToken &Current, |
133 | 4.29M | const FormatStyle &Style) { |
134 | 4.29M | const FormatToken &Previous = *Current.Previous; |
135 | 4.29M | if (Current.is(TT_CtorInitializerComma) && |
136 | 4.29M | Style.BreakConstructorInitializers == FormatStyle::BCIS_BeforeComma) { |
137 | 0 | return true; |
138 | 0 | } |
139 | 4.29M | if (Style.Language == FormatStyle::LK_Proto && Current.is(TT_SelectorName)) |
140 | 0 | return true; |
141 | 4.29M | return Previous.is(tok::comma) && !Current.isTrailingComment() && |
142 | 4.29M | ((Previous.isNot(TT_CtorInitializerComma) || |
143 | 77.5k | Style.BreakConstructorInitializers != |
144 | 0 | FormatStyle::BCIS_BeforeComma) && |
145 | 77.5k | (Previous.isNot(TT_InheritanceComma) || |
146 | 77.5k | Style.BreakInheritanceList != FormatStyle::BILS_BeforeComma)); |
147 | 4.29M | } |
148 | | |
149 | | static bool opensProtoMessageField(const FormatToken &LessTok, |
150 | 5.20M | const FormatStyle &Style) { |
151 | 5.20M | if (LessTok.isNot(tok::less)) |
152 | 4.73M | return false; |
153 | 471k | return Style.Language == FormatStyle::LK_TextProto || |
154 | 471k | (Style.Language == FormatStyle::LK_Proto && |
155 | 471k | (LessTok.NestingLevel > 0 || |
156 | 0 | (LessTok.Previous && LessTok.Previous->is(tok::equal)))); |
157 | 5.20M | } |
158 | | |
159 | | // Returns the delimiter of a raw string literal, or std::nullopt if TokenText |
160 | | // is not the text of a raw string literal. The delimiter could be the empty |
161 | | // string. For example, the delimiter of R"deli(cont)deli" is deli. |
162 | 106k | static std::optional<StringRef> getRawStringDelimiter(StringRef TokenText) { |
163 | 106k | if (TokenText.size() < 5 // The smallest raw string possible is 'R"()"'. |
164 | 106k | || !TokenText.starts_with("R\"") || !TokenText.ends_with("\"")) { |
165 | 106k | return std::nullopt; |
166 | 106k | } |
167 | | |
168 | | // A raw string starts with 'R"<delimiter>(' and delimiter is ascii and has |
169 | | // size at most 16 by the standard, so the first '(' must be among the first |
170 | | // 19 bytes. |
171 | 0 | size_t LParenPos = TokenText.substr(0, 19).find_first_of('('); |
172 | 0 | if (LParenPos == StringRef::npos) |
173 | 0 | return std::nullopt; |
174 | 0 | StringRef Delimiter = TokenText.substr(2, LParenPos - 2); |
175 | | |
176 | | // Check that the string ends in ')Delimiter"'. |
177 | 0 | size_t RParenPos = TokenText.size() - Delimiter.size() - 2; |
178 | 0 | if (TokenText[RParenPos] != ')') |
179 | 0 | return std::nullopt; |
180 | 0 | if (!TokenText.substr(RParenPos + 1).starts_with(Delimiter)) |
181 | 0 | return std::nullopt; |
182 | 0 | return Delimiter; |
183 | 0 | } |
184 | | |
185 | | // Returns the canonical delimiter for \p Language, or the empty string if no |
186 | | // canonical delimiter is specified. |
187 | | static StringRef |
188 | | getCanonicalRawStringDelimiter(const FormatStyle &Style, |
189 | 0 | FormatStyle::LanguageKind Language) { |
190 | 0 | for (const auto &Format : Style.RawStringFormats) |
191 | 0 | if (Format.Language == Language) |
192 | 0 | return StringRef(Format.CanonicalDelimiter); |
193 | 0 | return ""; |
194 | 0 | } |
195 | | |
196 | | RawStringFormatStyleManager::RawStringFormatStyleManager( |
197 | 351 | const FormatStyle &CodeStyle) { |
198 | 702 | for (const auto &RawStringFormat : CodeStyle.RawStringFormats) { |
199 | 702 | std::optional<FormatStyle> LanguageStyle = |
200 | 702 | CodeStyle.GetLanguageStyle(RawStringFormat.Language); |
201 | 702 | if (!LanguageStyle) { |
202 | 702 | FormatStyle PredefinedStyle; |
203 | 702 | if (!getPredefinedStyle(RawStringFormat.BasedOnStyle, |
204 | 702 | RawStringFormat.Language, &PredefinedStyle)) { |
205 | 0 | PredefinedStyle = getLLVMStyle(); |
206 | 0 | PredefinedStyle.Language = RawStringFormat.Language; |
207 | 0 | } |
208 | 702 | LanguageStyle = PredefinedStyle; |
209 | 702 | } |
210 | 702 | LanguageStyle->ColumnLimit = CodeStyle.ColumnLimit; |
211 | 702 | for (StringRef Delimiter : RawStringFormat.Delimiters) |
212 | 3.86k | DelimiterStyle.insert({Delimiter, *LanguageStyle}); |
213 | 702 | for (StringRef EnclosingFunction : RawStringFormat.EnclosingFunctions) |
214 | 3.15k | EnclosingFunctionStyle.insert({EnclosingFunction, *LanguageStyle}); |
215 | 702 | } |
216 | 351 | } |
217 | | |
218 | | std::optional<FormatStyle> |
219 | 0 | RawStringFormatStyleManager::getDelimiterStyle(StringRef Delimiter) const { |
220 | 0 | auto It = DelimiterStyle.find(Delimiter); |
221 | 0 | if (It == DelimiterStyle.end()) |
222 | 0 | return std::nullopt; |
223 | 0 | return It->second; |
224 | 0 | } |
225 | | |
226 | | std::optional<FormatStyle> |
227 | | RawStringFormatStyleManager::getEnclosingFunctionStyle( |
228 | 0 | StringRef EnclosingFunction) const { |
229 | 0 | auto It = EnclosingFunctionStyle.find(EnclosingFunction); |
230 | 0 | if (It == EnclosingFunctionStyle.end()) |
231 | 0 | return std::nullopt; |
232 | 0 | return It->second; |
233 | 0 | } |
234 | | |
235 | | ContinuationIndenter::ContinuationIndenter(const FormatStyle &Style, |
236 | | const AdditionalKeywords &Keywords, |
237 | | const SourceManager &SourceMgr, |
238 | | WhitespaceManager &Whitespaces, |
239 | | encoding::Encoding Encoding, |
240 | | bool BinPackInconclusiveFunctions) |
241 | | : Style(Style), Keywords(Keywords), SourceMgr(SourceMgr), |
242 | | Whitespaces(Whitespaces), Encoding(Encoding), |
243 | | BinPackInconclusiveFunctions(BinPackInconclusiveFunctions), |
244 | 351 | CommentPragmasRegex(Style.CommentPragmas), RawStringFormats(Style) {} |
245 | | |
246 | | LineState ContinuationIndenter::getInitialState(unsigned FirstIndent, |
247 | | unsigned FirstStartColumn, |
248 | | const AnnotatedLine *Line, |
249 | 409k | bool DryRun) { |
250 | 409k | LineState State; |
251 | 409k | State.FirstIndent = FirstIndent; |
252 | 409k | if (FirstStartColumn && Line->First->NewlinesBefore == 0) |
253 | 0 | State.Column = FirstStartColumn; |
254 | 409k | else |
255 | 409k | State.Column = FirstIndent; |
256 | | // With preprocessor directive indentation, the line starts on column 0 |
257 | | // since it's indented after the hash, but FirstIndent is set to the |
258 | | // preprocessor indent. |
259 | 409k | if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && |
260 | 409k | (Line->Type == LT_PreprocessorDirective || |
261 | 0 | Line->Type == LT_ImportStatement)) { |
262 | 0 | State.Column = 0; |
263 | 0 | } |
264 | 409k | State.Line = Line; |
265 | 409k | State.NextToken = Line->First; |
266 | 409k | State.Stack.push_back(ParenState(/*Tok=*/nullptr, FirstIndent, FirstIndent, |
267 | 409k | /*AvoidBinPacking=*/false, |
268 | 409k | /*NoLineBreak=*/false)); |
269 | 409k | State.NoContinuation = false; |
270 | 409k | State.StartOfStringLiteral = 0; |
271 | 409k | State.NoLineBreak = false; |
272 | 409k | State.StartOfLineLevel = 0; |
273 | 409k | State.LowestLevelOnLine = 0; |
274 | 409k | State.IgnoreStackForComparison = false; |
275 | | |
276 | 409k | if (Style.Language == FormatStyle::LK_TextProto) { |
277 | | // We need this in order to deal with the bin packing of text fields at |
278 | | // global scope. |
279 | 0 | auto &CurrentState = State.Stack.back(); |
280 | 0 | CurrentState.AvoidBinPacking = true; |
281 | 0 | CurrentState.BreakBeforeParameter = true; |
282 | 0 | CurrentState.AlignColons = false; |
283 | 0 | } |
284 | | |
285 | | // The first token has already been indented and thus consumed. |
286 | 409k | moveStateToNextToken(State, DryRun, /*Newline=*/false); |
287 | 409k | return State; |
288 | 409k | } |
289 | | |
290 | 4.24M | bool ContinuationIndenter::canBreak(const LineState &State) { |
291 | 4.24M | const FormatToken &Current = *State.NextToken; |
292 | 4.24M | const FormatToken &Previous = *Current.Previous; |
293 | 4.24M | const auto &CurrentState = State.Stack.back(); |
294 | 4.24M | assert(&Previous == Current.Previous); |
295 | 4.24M | if (!Current.CanBreakBefore && !(CurrentState.BreakBeforeClosingBrace && |
296 | 3.33M | Current.closesBlockOrBlockTypeList(Style))) { |
297 | 3.32M | return false; |
298 | 3.32M | } |
299 | | // The opening "{" of a braced list has to be on the same line as the first |
300 | | // element if it is nested in another braced init list or function call. |
301 | 919k | if (!Current.MustBreakBefore && Previous.is(tok::l_brace) && |
302 | 919k | Previous.isNot(TT_DictLiteral) && Previous.is(BK_BracedInit) && |
303 | 919k | Previous.Previous && |
304 | 919k | Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma)) { |
305 | 90 | return false; |
306 | 90 | } |
307 | | // This prevents breaks like: |
308 | | // ... |
309 | | // SomeParameter, OtherParameter).DoSomething( |
310 | | // ... |
311 | | // As they hide "DoSomething" and are generally bad for readability. |
312 | 919k | if (Previous.opensScope() && Previous.isNot(tok::l_brace) && |
313 | 919k | State.LowestLevelOnLine < State.StartOfLineLevel && |
314 | 919k | State.LowestLevelOnLine < Current.NestingLevel) { |
315 | 7.56k | return false; |
316 | 7.56k | } |
317 | 911k | if (Current.isMemberAccess() && CurrentState.ContainsUnwrappedBuilder) |
318 | 0 | return false; |
319 | | |
320 | | // Don't create a 'hanging' indent if there are multiple blocks in a single |
321 | | // statement and we are aligning lambda blocks to their signatures. |
322 | 911k | if (Previous.is(tok::l_brace) && State.Stack.size() > 1 && |
323 | 911k | State.Stack[State.Stack.size() - 2].NestedBlockInlined && |
324 | 911k | State.Stack[State.Stack.size() - 2].HasMultipleNestedBlocks && |
325 | 911k | Style.LambdaBodyIndentation == FormatStyle::LBI_Signature) { |
326 | 0 | return false; |
327 | 0 | } |
328 | | |
329 | | // Don't break after very short return types (e.g. "void") as that is often |
330 | | // unexpected. |
331 | 911k | if (Current.is(TT_FunctionDeclarationName) && State.Column < 6) { |
332 | 0 | if (Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) |
333 | 0 | return false; |
334 | 0 | } |
335 | | |
336 | | // If binary operators are moved to the next line (including commas for some |
337 | | // styles of constructor initializers), that's always ok. |
338 | 911k | if (!Current.isOneOf(TT_BinaryOperator, tok::comma) && |
339 | | // Allow breaking opening brace of lambdas (when passed as function |
340 | | // arguments) to a new line when BeforeLambdaBody brace wrapping is |
341 | | // enabled. |
342 | 911k | (!Style.BraceWrapping.BeforeLambdaBody || |
343 | 816k | Current.isNot(TT_LambdaLBrace)) && |
344 | 911k | CurrentState.NoLineBreakInOperand) { |
345 | 29.4k | return false; |
346 | 29.4k | } |
347 | | |
348 | 882k | if (Previous.is(tok::l_square) && Previous.is(TT_ObjCMethodExpr)) |
349 | 5.33k | return false; |
350 | | |
351 | 877k | if (Current.is(TT_ConditionalExpr) && Previous.is(tok::r_paren) && |
352 | 877k | Previous.MatchingParen && Previous.MatchingParen->Previous && |
353 | 877k | Previous.MatchingParen->Previous->MatchingParen && |
354 | 877k | Previous.MatchingParen->Previous->MatchingParen->is(TT_LambdaLBrace)) { |
355 | | // We have a lambda within a conditional expression, allow breaking here. |
356 | 0 | assert(Previous.MatchingParen->Previous->is(tok::r_brace)); |
357 | 0 | return true; |
358 | 0 | } |
359 | | |
360 | 877k | return !State.NoLineBreak && !CurrentState.NoLineBreak; |
361 | 877k | } |
362 | | |
363 | 4.18M | bool ContinuationIndenter::mustBreak(const LineState &State) { |
364 | 4.18M | const FormatToken &Current = *State.NextToken; |
365 | 4.18M | const FormatToken &Previous = *Current.Previous; |
366 | 4.18M | const auto &CurrentState = State.Stack.back(); |
367 | 4.18M | if (Style.BraceWrapping.BeforeLambdaBody && Current.CanBreakBefore && |
368 | 4.18M | Current.is(TT_LambdaLBrace) && Previous.isNot(TT_LineComment)) { |
369 | 0 | auto LambdaBodyLength = getLengthToMatchingParen(Current, State.Stack); |
370 | 0 | return LambdaBodyLength > getColumnLimit(State); |
371 | 0 | } |
372 | 4.18M | if (Current.MustBreakBefore || |
373 | 4.18M | (Current.is(TT_InlineASMColon) && |
374 | 4.04M | (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_Always || |
375 | 2.12k | (Style.BreakBeforeInlineASMColon == FormatStyle::BBIAS_OnlyMultiline && |
376 | 135k | Style.ColumnLimit > 0)))) { |
377 | 135k | return true; |
378 | 135k | } |
379 | 4.04M | if (CurrentState.BreakBeforeClosingBrace && |
380 | 4.04M | (Current.closesBlockOrBlockTypeList(Style) || |
381 | 111k | (Current.is(tok::r_brace) && |
382 | 106k | Current.isBlockIndentedInitRBrace(Style)))) { |
383 | 4.91k | return true; |
384 | 4.91k | } |
385 | 4.04M | if (CurrentState.BreakBeforeClosingParen && Current.is(tok::r_paren)) |
386 | 0 | return true; |
387 | 4.04M | if (Style.Language == FormatStyle::LK_ObjC && |
388 | 4.04M | Style.ObjCBreakBeforeNestedBlockParam && |
389 | 4.04M | Current.ObjCSelectorNameParts > 1 && |
390 | 4.04M | Current.startsSequence(TT_SelectorName, tok::colon, tok::caret)) { |
391 | 0 | return true; |
392 | 0 | } |
393 | | // Avoid producing inconsistent states by requiring breaks where they are not |
394 | | // permitted for C# generic type constraints. |
395 | 4.04M | if (CurrentState.IsCSharpGenericTypeConstraint && |
396 | 4.04M | Previous.isNot(TT_CSharpGenericTypeConstraintComma)) { |
397 | 0 | return false; |
398 | 0 | } |
399 | 4.04M | if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) || |
400 | 4.04M | (Previous.is(TT_TemplateCloser) && Current.is(TT_StartOfName) && |
401 | 3.94M | State.Line->First->isNot(TT_AttributeSquare) && Style.isCpp() && |
402 | | // FIXME: This is a temporary workaround for the case where clang-format |
403 | | // sets BreakBeforeParameter to avoid bin packing and this creates a |
404 | | // completely unnecessary line break after a template type that isn't |
405 | | // line-wrapped. |
406 | 3.94M | (Previous.NestingLevel == 1 || Style.BinPackParameters)) || |
407 | 4.04M | (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && |
408 | 3.92M | Previous.isNot(tok::question)) || |
409 | 4.04M | (!Style.BreakBeforeTernaryOperators && |
410 | 3.90M | Previous.is(TT_ConditionalExpr))) && |
411 | 4.04M | CurrentState.BreakBeforeParameter && !Current.isTrailingComment() && |
412 | 4.04M | !Current.isOneOf(tok::r_paren, tok::r_brace)) { |
413 | 65.9k | return true; |
414 | 65.9k | } |
415 | 3.97M | if (CurrentState.IsChainedConditional && |
416 | 3.97M | ((Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) && |
417 | 1.47k | Current.is(tok::colon)) || |
418 | 1.47k | (!Style.BreakBeforeTernaryOperators && Previous.is(TT_ConditionalExpr) && |
419 | 1.45k | Previous.is(tok::colon)))) { |
420 | 20 | return true; |
421 | 20 | } |
422 | 3.97M | if (((Previous.is(TT_DictLiteral) && Previous.is(tok::l_brace)) || |
423 | 3.97M | (Previous.is(TT_ArrayInitializerLSquare) && |
424 | 3.96M | Previous.ParameterCount > 1) || |
425 | 3.97M | opensProtoMessageField(Previous, Style)) && |
426 | 3.97M | Style.ColumnLimit > 0 && |
427 | 3.97M | getLengthToMatchingParen(Previous, State.Stack) + State.Column - 1 > |
428 | 10.3k | getColumnLimit(State)) { |
429 | 4.72k | return true; |
430 | 4.72k | } |
431 | | |
432 | 3.97M | const FormatToken &BreakConstructorInitializersToken = |
433 | 3.97M | Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon |
434 | 3.97M | ? Previous |
435 | 3.97M | : Current; |
436 | 3.97M | if (BreakConstructorInitializersToken.is(TT_CtorInitializerColon) && |
437 | 3.97M | (State.Column + State.Line->Last->TotalLength - Previous.TotalLength > |
438 | 225 | getColumnLimit(State) || |
439 | 225 | CurrentState.BreakBeforeParameter) && |
440 | 3.97M | (!Current.isTrailingComment() || Current.NewlinesBefore > 0) && |
441 | 3.97M | (Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All || |
442 | 225 | Style.BreakConstructorInitializers != FormatStyle::BCIS_BeforeColon || |
443 | 225 | Style.ColumnLimit != 0)) { |
444 | 225 | return true; |
445 | 225 | } |
446 | | |
447 | 3.97M | if (Current.is(TT_ObjCMethodExpr) && Previous.isNot(TT_SelectorName) && |
448 | 3.97M | State.Line->startsWith(TT_ObjCMethodSpecifier)) { |
449 | 117 | return true; |
450 | 117 | } |
451 | 3.97M | if (Current.is(TT_SelectorName) && Previous.isNot(tok::at) && |
452 | 3.97M | CurrentState.ObjCSelectorNameFound && CurrentState.BreakBeforeParameter && |
453 | 3.97M | (Style.ObjCBreakBeforeNestedBlockParam || |
454 | 110 | !Current.startsSequence(TT_SelectorName, tok::colon, tok::caret))) { |
455 | 110 | return true; |
456 | 110 | } |
457 | | |
458 | 3.97M | unsigned NewLineColumn = getNewLineColumn(State); |
459 | 3.97M | if (Current.isMemberAccess() && Style.ColumnLimit != 0 && |
460 | 3.97M | State.Column + getLengthToNextOperator(Current) > Style.ColumnLimit && |
461 | 3.97M | (State.Column > NewLineColumn || |
462 | 13.5k | Current.NestingLevel < State.StartOfLineLevel)) { |
463 | 9.27k | return true; |
464 | 9.27k | } |
465 | | |
466 | 3.96M | if (startsSegmentOfBuilderTypeCall(Current) && |
467 | 3.96M | (CurrentState.CallContinuation != 0 || |
468 | 6.01k | CurrentState.BreakBeforeParameter) && |
469 | | // JavaScript is treated different here as there is a frequent pattern: |
470 | | // SomeFunction(function() { |
471 | | // ... |
472 | | // }.bind(...)); |
473 | | // FIXME: We should find a more generic solution to this problem. |
474 | 3.96M | !(State.Column <= NewLineColumn && Style.isJavaScript()) && |
475 | 3.96M | !(Previous.closesScopeAfterBlock() && State.Column <= NewLineColumn)) { |
476 | 5.82k | return true; |
477 | 5.82k | } |
478 | | |
479 | | // If the template declaration spans multiple lines, force wrap before the |
480 | | // function/class declaration. |
481 | 3.95M | if (Previous.ClosesTemplateDeclaration && CurrentState.BreakBeforeParameter && |
482 | 3.95M | Current.CanBreakBefore) { |
483 | 0 | return true; |
484 | 0 | } |
485 | | |
486 | 3.95M | if (State.Line->First->isNot(tok::kw_enum) && State.Column <= NewLineColumn) |
487 | 414k | return false; |
488 | | |
489 | 3.54M | if (Style.AlwaysBreakBeforeMultilineStrings && |
490 | 3.54M | (NewLineColumn == State.FirstIndent + Style.ContinuationIndentWidth || |
491 | 3.54M | Previous.is(tok::comma) || Current.NestingLevel < 2) && |
492 | 3.54M | !Previous.isOneOf(tok::kw_return, tok::lessless, tok::at, |
493 | 2.81M | Keywords.kw_dollar) && |
494 | 3.54M | !Previous.isOneOf(TT_InlineASMColon, TT_ConditionalExpr) && |
495 | 3.54M | nextIsMultilineString(State)) { |
496 | 9.11k | return true; |
497 | 9.11k | } |
498 | | |
499 | | // Using CanBreakBefore here and below takes care of the decision whether the |
500 | | // current style uses wrapping before or after operators for the given |
501 | | // operator. |
502 | 3.53M | if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) { |
503 | 165k | const auto PreviousPrecedence = Previous.getPrecedence(); |
504 | 165k | if (PreviousPrecedence != prec::Assignment && |
505 | 165k | CurrentState.BreakBeforeParameter && !Current.isTrailingComment()) { |
506 | 33.3k | const bool LHSIsBinaryExpr = |
507 | 33.3k | Previous.Previous && Previous.Previous->EndsBinaryExpression; |
508 | 33.3k | if (LHSIsBinaryExpr) |
509 | 10.2k | return true; |
510 | | // If we need to break somewhere inside the LHS of a binary expression, we |
511 | | // should also break after the operator. Otherwise, the formatting would |
512 | | // hide the operator precedence, e.g. in: |
513 | | // if (aaaaaaaaaaaaaa == |
514 | | // bbbbbbbbbbbbbb && c) {.. |
515 | | // For comparisons, we only apply this rule, if the LHS is a binary |
516 | | // expression itself as otherwise, the line breaks seem superfluous. |
517 | | // We need special cases for ">>" which we have split into two ">" while |
518 | | // lexing in order to make template parsing easier. |
519 | 23.1k | const bool IsComparison = |
520 | 23.1k | (PreviousPrecedence == prec::Relational || |
521 | 23.1k | PreviousPrecedence == prec::Equality || |
522 | 23.1k | PreviousPrecedence == prec::Spaceship) && |
523 | 23.1k | Previous.Previous && |
524 | 23.1k | Previous.Previous->isNot(TT_BinaryOperator); // For >>. |
525 | 23.1k | if (!IsComparison) |
526 | 11.6k | return true; |
527 | 23.1k | } |
528 | 3.36M | } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore && |
529 | 3.36M | CurrentState.BreakBeforeParameter) { |
530 | 330 | return true; |
531 | 330 | } |
532 | | |
533 | | // Same as above, but for the first "<<" operator. |
534 | 3.51M | if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator) && |
535 | 3.51M | CurrentState.BreakBeforeParameter && CurrentState.FirstLessLess == 0) { |
536 | 12 | return true; |
537 | 12 | } |
538 | | |
539 | 3.51M | if (Current.NestingLevel == 0 && !Current.isTrailingComment()) { |
540 | | // Always break after "template <...>"(*) and leading annotations. This is |
541 | | // only for cases where the entire line does not fit on a single line as a |
542 | | // different LineFormatter would be used otherwise. |
543 | | // *: Except when another option interferes with that, like concepts. |
544 | 2.21M | if (Previous.ClosesTemplateDeclaration) { |
545 | 0 | if (Current.is(tok::kw_concept)) { |
546 | 0 | switch (Style.BreakBeforeConceptDeclarations) { |
547 | 0 | case FormatStyle::BBCDS_Allowed: |
548 | 0 | break; |
549 | 0 | case FormatStyle::BBCDS_Always: |
550 | 0 | return true; |
551 | 0 | case FormatStyle::BBCDS_Never: |
552 | 0 | return false; |
553 | 0 | } |
554 | 0 | } |
555 | 0 | if (Current.is(TT_RequiresClause)) { |
556 | 0 | switch (Style.RequiresClausePosition) { |
557 | 0 | case FormatStyle::RCPS_SingleLine: |
558 | 0 | case FormatStyle::RCPS_WithPreceding: |
559 | 0 | return false; |
560 | 0 | default: |
561 | 0 | return true; |
562 | 0 | } |
563 | 0 | } |
564 | 0 | return Style.AlwaysBreakTemplateDeclarations != FormatStyle::BTDS_No; |
565 | 0 | } |
566 | 2.21M | if (Previous.is(TT_FunctionAnnotationRParen) && |
567 | 2.21M | State.Line->Type != LT_PreprocessorDirective) { |
568 | 4 | return true; |
569 | 4 | } |
570 | 2.21M | if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) && |
571 | 2.21M | Current.isNot(TT_LeadingJavaAnnotation)) { |
572 | 0 | return true; |
573 | 0 | } |
574 | 2.21M | } |
575 | | |
576 | 3.51M | if (Style.isJavaScript() && Previous.is(tok::r_paren) && |
577 | 3.51M | Previous.is(TT_JavaAnnotation)) { |
578 | | // Break after the closing parenthesis of TypeScript decorators before |
579 | | // functions, getters and setters. |
580 | 0 | static const llvm::StringSet<> BreakBeforeDecoratedTokens = {"get", "set", |
581 | 0 | "function"}; |
582 | 0 | if (BreakBeforeDecoratedTokens.contains(Current.TokenText)) |
583 | 0 | return true; |
584 | 0 | } |
585 | | |
586 | 3.51M | if (Current.is(TT_FunctionDeclarationName) && |
587 | 3.51M | !State.Line->ReturnTypeWrapped && |
588 | | // Don't break before a C# function when no break after return type. |
589 | 3.51M | (!Style.isCSharp() || |
590 | 330 | Style.AlwaysBreakAfterReturnType != FormatStyle::RTBS_None) && |
591 | | // Don't always break between a JavaScript `function` and the function |
592 | | // name. |
593 | 3.51M | !Style.isJavaScript() && Previous.isNot(tok::kw_template) && |
594 | 3.51M | CurrentState.BreakBeforeParameter) { |
595 | 23 | return true; |
596 | 23 | } |
597 | | |
598 | | // The following could be precomputed as they do not depend on the state. |
599 | | // However, as they should take effect only if the UnwrappedLine does not fit |
600 | | // into the ColumnLimit, they are checked here in the ContinuationIndenter. |
601 | 3.51M | if (Style.ColumnLimit != 0 && Previous.is(BK_Block) && |
602 | 3.51M | Previous.is(tok::l_brace) && |
603 | 3.51M | !Current.isOneOf(tok::r_brace, tok::comment)) { |
604 | 34 | return true; |
605 | 34 | } |
606 | | |
607 | 3.51M | if (Current.is(tok::lessless) && |
608 | 3.51M | ((Previous.is(tok::identifier) && Previous.TokenText == "endl") || |
609 | 278 | (Previous.Tok.isLiteral() && (Previous.TokenText.ends_with("\\n\"") || |
610 | 7 | Previous.TokenText == "\'\\n\'")))) { |
611 | 0 | return true; |
612 | 0 | } |
613 | | |
614 | 3.51M | if (Previous.is(TT_BlockComment) && Previous.IsMultiline) |
615 | 11 | return true; |
616 | | |
617 | 3.51M | if (State.NoContinuation) |
618 | 18 | return true; |
619 | | |
620 | 3.51M | return false; |
621 | 3.51M | } |
622 | | |
623 | | unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline, |
624 | | bool DryRun, |
625 | 6.63M | unsigned ExtraSpaces) { |
626 | 6.63M | const FormatToken &Current = *State.NextToken; |
627 | 6.63M | assert(State.NextToken->Previous); |
628 | 0 | const FormatToken &Previous = *State.NextToken->Previous; |
629 | | |
630 | 6.63M | assert(!State.Stack.empty()); |
631 | 0 | State.NoContinuation = false; |
632 | | |
633 | 6.63M | if (Current.is(TT_ImplicitStringLiteral) && |
634 | 6.63M | (!Previous.Tok.getIdentifierInfo() || |
635 | 2.65M | Previous.Tok.getIdentifierInfo()->getPPKeywordID() == |
636 | 2.65M | tok::pp_not_keyword)) { |
637 | 2.65M | unsigned EndColumn = |
638 | 2.65M | SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd()); |
639 | 2.65M | if (Current.LastNewlineOffset != 0) { |
640 | | // If there is a newline within this token, the final column will solely |
641 | | // determined by the current end column. |
642 | 39.6k | State.Column = EndColumn; |
643 | 2.61M | } else { |
644 | 2.61M | unsigned StartColumn = |
645 | 2.61M | SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin()); |
646 | 2.61M | assert(EndColumn >= StartColumn); |
647 | 0 | State.Column += EndColumn - StartColumn; |
648 | 2.61M | } |
649 | 0 | moveStateToNextToken(State, DryRun, /*Newline=*/false); |
650 | 2.65M | return 0; |
651 | 2.65M | } |
652 | | |
653 | 3.98M | unsigned Penalty = 0; |
654 | 3.98M | if (Newline) |
655 | 992k | Penalty = addTokenOnNewLine(State, DryRun); |
656 | 2.99M | else |
657 | 2.99M | addTokenOnCurrentLine(State, DryRun, ExtraSpaces); |
658 | | |
659 | 3.98M | return moveStateToNextToken(State, DryRun, Newline) + Penalty; |
660 | 6.63M | } |
661 | | |
662 | | void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun, |
663 | 2.99M | unsigned ExtraSpaces) { |
664 | 2.99M | FormatToken &Current = *State.NextToken; |
665 | 2.99M | assert(State.NextToken->Previous); |
666 | 0 | const FormatToken &Previous = *State.NextToken->Previous; |
667 | 2.99M | auto &CurrentState = State.Stack.back(); |
668 | | |
669 | 2.99M | bool DisallowLineBreaksOnThisLine = |
670 | 2.99M | Style.LambdaBodyIndentation == FormatStyle::LBI_Signature && |
671 | 2.99M | Style.isCpp() && [&Current] { |
672 | | // Deal with lambda arguments in C++. The aim here is to ensure that we |
673 | | // don't over-indent lambda function bodies when lambdas are passed as |
674 | | // arguments to function calls. We do this by ensuring that either all |
675 | | // arguments (including any lambdas) go on the same line as the function |
676 | | // call, or we break before the first argument. |
677 | 2.99M | auto PrevNonComment = Current.getPreviousNonComment(); |
678 | 2.99M | if (!PrevNonComment || PrevNonComment->isNot(tok::l_paren)) |
679 | 2.96M | return false; |
680 | 25.6k | if (Current.isOneOf(tok::comment, tok::l_paren, TT_LambdaLSquare)) |
681 | 5.72k | return false; |
682 | 19.9k | auto BlockParameterCount = PrevNonComment->BlockParameterCount; |
683 | 19.9k | if (BlockParameterCount == 0) |
684 | 19.9k | return false; |
685 | | |
686 | | // Multiple lambdas in the same function call. |
687 | 4 | if (BlockParameterCount > 1) |
688 | 0 | return true; |
689 | | |
690 | | // A lambda followed by another arg. |
691 | 4 | if (!PrevNonComment->Role) |
692 | 0 | return false; |
693 | 4 | auto Comma = PrevNonComment->Role->lastComma(); |
694 | 4 | if (!Comma) |
695 | 0 | return false; |
696 | 4 | auto Next = Comma->getNextNonComment(); |
697 | 4 | return Next && |
698 | 4 | !Next->isOneOf(TT_LambdaLSquare, tok::l_brace, tok::caret); |
699 | 4 | }(); |
700 | | |
701 | 2.99M | if (DisallowLineBreaksOnThisLine) |
702 | 4 | State.NoLineBreak = true; |
703 | | |
704 | 2.99M | if (Current.is(tok::equal) && |
705 | 2.99M | (State.Line->First->is(tok::kw_for) || Current.NestingLevel == 0) && |
706 | 2.99M | CurrentState.VariablePos == 0) { |
707 | 7.42k | CurrentState.VariablePos = State.Column; |
708 | | // Move over * and & if they are bound to the variable name. |
709 | 7.42k | const FormatToken *Tok = &Previous; |
710 | 35.9k | while (Tok && CurrentState.VariablePos >= Tok->ColumnWidth) { |
711 | 35.9k | CurrentState.VariablePos -= Tok->ColumnWidth; |
712 | 35.9k | if (Tok->SpacesRequiredBefore != 0) |
713 | 7.37k | break; |
714 | 28.5k | Tok = Tok->Previous; |
715 | 28.5k | } |
716 | 7.42k | if (Previous.PartOfMultiVariableDeclStmt) |
717 | 173 | CurrentState.LastSpace = CurrentState.VariablePos; |
718 | 7.42k | } |
719 | | |
720 | 2.99M | unsigned Spaces = Current.SpacesRequiredBefore + ExtraSpaces; |
721 | | |
722 | | // Indent preprocessor directives after the hash if required. |
723 | 2.99M | int PPColumnCorrection = 0; |
724 | 2.99M | if (Style.IndentPPDirectives == FormatStyle::PPDIS_AfterHash && |
725 | 2.99M | Previous.is(tok::hash) && State.FirstIndent > 0 && |
726 | 2.99M | &Previous == State.Line->First && |
727 | 2.99M | (State.Line->Type == LT_PreprocessorDirective || |
728 | 0 | State.Line->Type == LT_ImportStatement)) { |
729 | 0 | Spaces += State.FirstIndent; |
730 | | |
731 | | // For preprocessor indent with tabs, State.Column will be 1 because of the |
732 | | // hash. This causes second-level indents onward to have an extra space |
733 | | // after the tabs. We avoid this misalignment by subtracting 1 from the |
734 | | // column value passed to replaceWhitespace(). |
735 | 0 | if (Style.UseTab != FormatStyle::UT_Never) |
736 | 0 | PPColumnCorrection = -1; |
737 | 0 | } |
738 | | |
739 | 2.99M | if (!DryRun) { |
740 | 1.00M | Whitespaces.replaceWhitespace(Current, /*Newlines=*/0, Spaces, |
741 | 1.00M | State.Column + Spaces + PPColumnCorrection); |
742 | 1.00M | } |
743 | | |
744 | | // If "BreakBeforeInheritanceComma" mode, don't break within the inheritance |
745 | | // declaration unless there is multiple inheritance. |
746 | 2.99M | if (Style.BreakInheritanceList == FormatStyle::BILS_BeforeComma && |
747 | 2.99M | Current.is(TT_InheritanceColon)) { |
748 | 0 | CurrentState.NoLineBreak = true; |
749 | 0 | } |
750 | 2.99M | if (Style.BreakInheritanceList == FormatStyle::BILS_AfterColon && |
751 | 2.99M | Previous.is(TT_InheritanceColon)) { |
752 | 0 | CurrentState.NoLineBreak = true; |
753 | 0 | } |
754 | | |
755 | 2.99M | if (Current.is(TT_SelectorName) && !CurrentState.ObjCSelectorNameFound) { |
756 | 4.89k | unsigned MinIndent = std::max( |
757 | 4.89k | State.FirstIndent + Style.ContinuationIndentWidth, CurrentState.Indent); |
758 | 4.89k | unsigned FirstColonPos = State.Column + Spaces + Current.ColumnWidth; |
759 | 4.89k | if (Current.LongestObjCSelectorName == 0) |
760 | 4.87k | CurrentState.AlignColons = false; |
761 | 24 | else if (MinIndent + Current.LongestObjCSelectorName > FirstColonPos) |
762 | 6 | CurrentState.ColonPos = MinIndent + Current.LongestObjCSelectorName; |
763 | 18 | else |
764 | 18 | CurrentState.ColonPos = FirstColonPos; |
765 | 4.89k | } |
766 | | |
767 | | // In "AlwaysBreak" or "BlockIndent" mode, enforce wrapping directly after the |
768 | | // parenthesis by disallowing any further line breaks if there is no line |
769 | | // break after the opening parenthesis. Don't break if it doesn't conserve |
770 | | // columns. |
771 | 2.99M | if ((Style.AlignAfterOpenBracket == FormatStyle::BAS_AlwaysBreak || |
772 | 2.99M | Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent) && |
773 | 2.99M | (Previous.isOneOf(tok::l_paren, TT_TemplateOpener, tok::l_square) || |
774 | 0 | (Previous.is(tok::l_brace) && Previous.isNot(BK_Block) && |
775 | 0 | Style.Cpp11BracedListStyle)) && |
776 | 2.99M | State.Column > getNewLineColumn(State) && |
777 | 2.99M | (!Previous.Previous || |
778 | 0 | !Previous.Previous->isOneOf(TT_CastRParen, tok::kw_for, tok::kw_while, |
779 | 0 | tok::kw_switch)) && |
780 | | // Don't do this for simple (no expressions) one-argument function calls |
781 | | // as that feels like needlessly wasting whitespace, e.g.: |
782 | | // |
783 | | // caaaaaaaaaaaall( |
784 | | // caaaaaaaaaaaall( |
785 | | // caaaaaaaaaaaall( |
786 | | // caaaaaaaaaaaaaaaaaaaaaaall(aaaaaaaaaaaaaa, aaaaaaaaa)))); |
787 | 2.99M | Current.FakeLParens.size() > 0 && |
788 | 2.99M | Current.FakeLParens.back() > prec::Unknown) { |
789 | 0 | CurrentState.NoLineBreak = true; |
790 | 0 | } |
791 | 2.99M | if (Previous.is(TT_TemplateString) && Previous.opensScope()) |
792 | 0 | CurrentState.NoLineBreak = true; |
793 | | |
794 | | // Align following lines within parentheses / brackets if configured. |
795 | | // Note: This doesn't apply to macro expansion lines, which are MACRO( , , ) |
796 | | // with args as children of the '(' and ',' tokens. It does not make sense to |
797 | | // align the commas with the opening paren. |
798 | 2.99M | if (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign && |
799 | 2.99M | !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() && |
800 | 2.99M | Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) && |
801 | 2.99M | !(Current.MacroParent && Previous.MacroParent) && |
802 | 2.99M | (Current.isNot(TT_LineComment) || |
803 | 286k | Previous.isOneOf(BK_BracedInit, TT_VerilogMultiLineListLParen))) { |
804 | 286k | CurrentState.Indent = State.Column + Spaces; |
805 | 286k | CurrentState.IsAligned = true; |
806 | 286k | } |
807 | 2.99M | if (CurrentState.AvoidBinPacking && startsNextParameter(Current, Style)) |
808 | 588 | CurrentState.NoLineBreak = true; |
809 | 2.99M | if (startsSegmentOfBuilderTypeCall(Current) && |
810 | 2.99M | State.Column > getNewLineColumn(State)) { |
811 | 104 | CurrentState.ContainsUnwrappedBuilder = true; |
812 | 104 | } |
813 | | |
814 | 2.99M | if (Current.is(TT_TrailingReturnArrow) && |
815 | 2.99M | Style.Language == FormatStyle::LK_Java) { |
816 | 0 | CurrentState.NoLineBreak = true; |
817 | 0 | } |
818 | 2.99M | if (Current.isMemberAccess() && Previous.is(tok::r_paren) && |
819 | 2.99M | (Previous.MatchingParen && |
820 | 0 | (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) { |
821 | | // If there is a function call with long parameters, break before trailing |
822 | | // calls. This prevents things like: |
823 | | // EXPECT_CALL(SomeLongParameter).Times( |
824 | | // 2); |
825 | | // We don't want to do this for short parameters as they can just be |
826 | | // indexes. |
827 | 0 | CurrentState.NoLineBreak = true; |
828 | 0 | } |
829 | | |
830 | | // Don't allow the RHS of an operator to be split over multiple lines unless |
831 | | // there is a line-break right after the operator. |
832 | | // Exclude relational operators, as there, it is always more desirable to |
833 | | // have the LHS 'left' of the RHS. |
834 | 2.99M | const FormatToken *P = Current.getPreviousNonComment(); |
835 | 2.99M | if (Current.isNot(tok::comment) && P && |
836 | 2.99M | (P->isOneOf(TT_BinaryOperator, tok::comma) || |
837 | 2.97M | (P->is(TT_ConditionalExpr) && P->is(tok::colon))) && |
838 | 2.99M | !P->isOneOf(TT_OverloadedOperator, TT_CtorInitializerComma) && |
839 | 2.99M | P->getPrecedence() != prec::Assignment && |
840 | 2.99M | P->getPrecedence() != prec::Relational && |
841 | 2.99M | P->getPrecedence() != prec::Spaceship) { |
842 | 222k | bool BreakBeforeOperator = |
843 | 222k | P->MustBreakBefore || P->is(tok::lessless) || |
844 | 222k | (P->is(TT_BinaryOperator) && |
845 | 213k | Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None) || |
846 | 222k | (P->is(TT_ConditionalExpr) && Style.BreakBeforeTernaryOperators); |
847 | | // Don't do this if there are only two operands. In these cases, there is |
848 | | // always a nice vertical separation between them and the extra line break |
849 | | // does not help. |
850 | 222k | bool HasTwoOperands = P->OperatorIndex == 0 && !P->NextOperator && |
851 | 222k | P->isNot(TT_ConditionalExpr); |
852 | 222k | if ((!BreakBeforeOperator && |
853 | 222k | !(HasTwoOperands && |
854 | 210k | Style.AlignOperands != FormatStyle::OAS_DontAlign)) || |
855 | 222k | (!CurrentState.LastOperatorWrapped && BreakBeforeOperator)) { |
856 | 131k | CurrentState.NoLineBreakInOperand = true; |
857 | 131k | } |
858 | 222k | } |
859 | | |
860 | 2.99M | State.Column += Spaces; |
861 | 2.99M | if (Current.isNot(tok::comment) && Previous.is(tok::l_paren) && |
862 | 2.99M | Previous.Previous && |
863 | 2.99M | (Previous.Previous->is(tok::kw_for) || Previous.Previous->isIf())) { |
864 | | // Treat the condition inside an if as if it was a second function |
865 | | // parameter, i.e. let nested calls have a continuation indent. |
866 | 2.84k | CurrentState.LastSpace = State.Column; |
867 | 2.84k | CurrentState.NestedBlockIndent = State.Column; |
868 | 2.98M | } else if (!Current.isOneOf(tok::comment, tok::caret) && |
869 | 2.98M | ((Previous.is(tok::comma) && |
870 | 2.96M | Previous.isNot(TT_OverloadedOperator)) || |
871 | 2.96M | (Previous.is(tok::colon) && Previous.is(TT_ObjCMethodExpr)))) { |
872 | 71.4k | CurrentState.LastSpace = State.Column; |
873 | 2.91M | } else if (Previous.is(TT_CtorInitializerColon) && |
874 | 2.91M | (!Current.isTrailingComment() || Current.NewlinesBefore > 0) && |
875 | 2.91M | Style.BreakConstructorInitializers == |
876 | 195 | FormatStyle::BCIS_AfterColon) { |
877 | 0 | CurrentState.Indent = State.Column; |
878 | 0 | CurrentState.LastSpace = State.Column; |
879 | 2.91M | } else if (Previous.isOneOf(TT_ConditionalExpr, TT_CtorInitializerColon)) { |
880 | 49.3k | CurrentState.LastSpace = State.Column; |
881 | 2.86M | } else if (Previous.is(TT_BinaryOperator) && |
882 | 2.86M | ((Previous.getPrecedence() != prec::Assignment && |
883 | 299k | (Previous.isNot(tok::lessless) || Previous.OperatorIndex != 0 || |
884 | 262k | Previous.NextOperator)) || |
885 | 299k | Current.StartsBinaryExpression)) { |
886 | | // Indent relative to the RHS of the expression unless this is a simple |
887 | | // assignment without binary expression on the RHS. |
888 | 269k | if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) |
889 | 269k | CurrentState.LastSpace = State.Column; |
890 | 2.59M | } else if (Previous.is(TT_InheritanceColon)) { |
891 | 37.9k | CurrentState.Indent = State.Column; |
892 | 37.9k | CurrentState.LastSpace = State.Column; |
893 | 2.56M | } else if (Current.is(TT_CSharpGenericTypeConstraintColon)) { |
894 | 0 | CurrentState.ColonPos = State.Column; |
895 | 2.56M | } else if (Previous.opensScope()) { |
896 | | // If a function has a trailing call, indent all parameters from the |
897 | | // opening parenthesis. This avoids confusing indents like: |
898 | | // OuterFunction(InnerFunctionCall( // break |
899 | | // ParameterToInnerFunction)) // break |
900 | | // .SecondInnerFunctionCall(); |
901 | 293k | if (Previous.MatchingParen) { |
902 | 239k | const FormatToken *Next = Previous.MatchingParen->getNextNonComment(); |
903 | 239k | if (Next && Next->isMemberAccess() && State.Stack.size() > 1 && |
904 | 239k | State.Stack[State.Stack.size() - 2].CallContinuation == 0) { |
905 | 2.33k | CurrentState.LastSpace = State.Column; |
906 | 2.33k | } |
907 | 239k | } |
908 | 293k | } |
909 | 2.99M | } |
910 | | |
911 | | unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State, |
912 | 992k | bool DryRun) { |
913 | 992k | FormatToken &Current = *State.NextToken; |
914 | 992k | assert(State.NextToken->Previous); |
915 | 0 | const FormatToken &Previous = *State.NextToken->Previous; |
916 | 992k | auto &CurrentState = State.Stack.back(); |
917 | | |
918 | | // Extra penalty that needs to be added because of the way certain line |
919 | | // breaks are chosen. |
920 | 992k | unsigned Penalty = 0; |
921 | | |
922 | 992k | const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); |
923 | 992k | const FormatToken *NextNonComment = Previous.getNextNonComment(); |
924 | 992k | if (!NextNonComment) |
925 | 57 | NextNonComment = &Current; |
926 | | // The first line break on any NestingLevel causes an extra penalty in order |
927 | | // prefer similar line breaks. |
928 | 992k | if (!CurrentState.ContainsLineBreak) |
929 | 403k | Penalty += 15; |
930 | 992k | CurrentState.ContainsLineBreak = true; |
931 | | |
932 | 992k | Penalty += State.NextToken->SplitPenalty; |
933 | | |
934 | | // Breaking before the first "<<" is generally not desirable if the LHS is |
935 | | // short. Also always add the penalty if the LHS is split over multiple lines |
936 | | // to avoid unnecessary line breaks that just work around this penalty. |
937 | 992k | if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess == 0 && |
938 | 992k | (State.Column <= Style.ColumnLimit / 3 || |
939 | 151 | CurrentState.BreakBeforeParameter)) { |
940 | 73 | Penalty += Style.PenaltyBreakFirstLessLess; |
941 | 73 | } |
942 | | |
943 | 992k | State.Column = getNewLineColumn(State); |
944 | | |
945 | | // Add Penalty proportional to amount of whitespace away from FirstColumn |
946 | | // This tends to penalize several lines that are far-right indented, |
947 | | // and prefers a line-break prior to such a block, e.g: |
948 | | // |
949 | | // Constructor() : |
950 | | // member(value), looooooooooooooooong_member( |
951 | | // looooooooooong_call(param_1, param_2, param_3)) |
952 | | // would then become |
953 | | // Constructor() : |
954 | | // member(value), |
955 | | // looooooooooooooooong_member( |
956 | | // looooooooooong_call(param_1, param_2, param_3)) |
957 | 992k | if (State.Column > State.FirstIndent) { |
958 | 972k | Penalty += |
959 | 972k | Style.PenaltyIndentedWhitespace * (State.Column - State.FirstIndent); |
960 | 972k | } |
961 | | |
962 | | // Indent nested blocks relative to this column, unless in a very specific |
963 | | // JavaScript special case where: |
964 | | // |
965 | | // var loooooong_name = |
966 | | // function() { |
967 | | // // code |
968 | | // } |
969 | | // |
970 | | // is common and should be formatted like a free-standing function. The same |
971 | | // goes for wrapping before the lambda return type arrow. |
972 | 992k | if (Current.isNot(TT_TrailingReturnArrow) && |
973 | 992k | (!Style.isJavaScript() || Current.NestingLevel != 0 || |
974 | 992k | !PreviousNonComment || PreviousNonComment->isNot(tok::equal) || |
975 | 992k | !Current.isOneOf(Keywords.kw_async, Keywords.kw_function))) { |
976 | 992k | CurrentState.NestedBlockIndent = State.Column; |
977 | 992k | } |
978 | | |
979 | 992k | if (NextNonComment->isMemberAccess()) { |
980 | 53.6k | if (CurrentState.CallContinuation == 0) |
981 | 20.0k | CurrentState.CallContinuation = State.Column; |
982 | 938k | } else if (NextNonComment->is(TT_SelectorName)) { |
983 | 1.41k | if (!CurrentState.ObjCSelectorNameFound) { |
984 | 1.19k | if (NextNonComment->LongestObjCSelectorName == 0) { |
985 | 1.17k | CurrentState.AlignColons = false; |
986 | 1.17k | } else { |
987 | 24 | CurrentState.ColonPos = |
988 | 24 | (shouldIndentWrappedSelectorName(Style, State.Line->Type) |
989 | 24 | ? std::max(CurrentState.Indent, |
990 | 0 | State.FirstIndent + Style.ContinuationIndentWidth) |
991 | 24 | : CurrentState.Indent) + |
992 | 24 | std::max(NextNonComment->LongestObjCSelectorName, |
993 | 24 | NextNonComment->ColumnWidth); |
994 | 24 | } |
995 | 1.19k | } else if (CurrentState.AlignColons && |
996 | 213 | CurrentState.ColonPos <= NextNonComment->ColumnWidth) { |
997 | 0 | CurrentState.ColonPos = State.Column + NextNonComment->ColumnWidth; |
998 | 0 | } |
999 | 937k | } else if (PreviousNonComment && PreviousNonComment->is(tok::colon) && |
1000 | 937k | PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { |
1001 | | // FIXME: This is hacky, find a better way. The problem is that in an ObjC |
1002 | | // method expression, the block should be aligned to the line starting it, |
1003 | | // e.g.: |
1004 | | // [aaaaaaaaaaaaaaa aaaaaaaaa: \\ break for some reason |
1005 | | // ^(int *i) { |
1006 | | // // ... |
1007 | | // }]; |
1008 | | // Thus, we set LastSpace of the next higher NestingLevel, to which we move |
1009 | | // when we consume all of the "}"'s FakeRParens at the "{". |
1010 | 15.1k | if (State.Stack.size() > 1) { |
1011 | 15.1k | State.Stack[State.Stack.size() - 2].LastSpace = |
1012 | 15.1k | std::max(CurrentState.LastSpace, CurrentState.Indent) + |
1013 | 15.1k | Style.ContinuationIndentWidth; |
1014 | 15.1k | } |
1015 | 15.1k | } |
1016 | | |
1017 | 992k | if ((PreviousNonComment && |
1018 | 992k | PreviousNonComment->isOneOf(tok::comma, tok::semi) && |
1019 | 992k | !CurrentState.AvoidBinPacking) || |
1020 | 992k | Previous.is(TT_BinaryOperator)) { |
1021 | 283k | CurrentState.BreakBeforeParameter = false; |
1022 | 283k | } |
1023 | 992k | if (PreviousNonComment && |
1024 | 992k | (PreviousNonComment->isOneOf(TT_TemplateCloser, TT_JavaAnnotation) || |
1025 | 992k | PreviousNonComment->ClosesRequiresClause) && |
1026 | 992k | Current.NestingLevel == 0) { |
1027 | 19.2k | CurrentState.BreakBeforeParameter = false; |
1028 | 19.2k | } |
1029 | 992k | if (NextNonComment->is(tok::question) || |
1030 | 992k | (PreviousNonComment && PreviousNonComment->is(tok::question))) { |
1031 | 48.6k | CurrentState.BreakBeforeParameter = true; |
1032 | 48.6k | } |
1033 | 992k | if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore) |
1034 | 77.7k | CurrentState.BreakBeforeParameter = false; |
1035 | | |
1036 | 992k | if (!DryRun) { |
1037 | 139k | unsigned MaxEmptyLinesToKeep = Style.MaxEmptyLinesToKeep + 1; |
1038 | 139k | if (Current.is(tok::r_brace) && Current.MatchingParen && |
1039 | | // Only strip trailing empty lines for l_braces that have children, i.e. |
1040 | | // for function expressions (lambdas, arrows, etc). |
1041 | 139k | !Current.MatchingParen->Children.empty()) { |
1042 | | // lambdas and arrow functions are expressions, thus their r_brace is not |
1043 | | // on its own line, and thus not covered by UnwrappedLineFormatter's logic |
1044 | | // about removing empty lines on closing blocks. Special case them here. |
1045 | 5 | MaxEmptyLinesToKeep = 1; |
1046 | 5 | } |
1047 | 139k | unsigned Newlines = |
1048 | 139k | std::max(1u, std::min(Current.NewlinesBefore, MaxEmptyLinesToKeep)); |
1049 | 139k | bool ContinuePPDirective = |
1050 | 139k | State.Line->InPPDirective && State.Line->Type != LT_ImportStatement; |
1051 | 139k | Whitespaces.replaceWhitespace(Current, Newlines, State.Column, State.Column, |
1052 | 139k | CurrentState.IsAligned, ContinuePPDirective); |
1053 | 139k | } |
1054 | | |
1055 | 992k | if (!Current.isTrailingComment()) |
1056 | 987k | CurrentState.LastSpace = State.Column; |
1057 | 992k | if (Current.is(tok::lessless)) { |
1058 | | // If we are breaking before a "<<", we always want to indent relative to |
1059 | | // RHS. This is necessary only for "<<", as we special-case it and don't |
1060 | | // always indent relative to the RHS. |
1061 | 355 | CurrentState.LastSpace += 3; // 3 -> width of "<< ". |
1062 | 355 | } |
1063 | | |
1064 | 992k | State.StartOfLineLevel = Current.NestingLevel; |
1065 | 992k | State.LowestLevelOnLine = Current.NestingLevel; |
1066 | | |
1067 | | // Any break on this level means that the parent level has been broken |
1068 | | // and we need to avoid bin packing there. |
1069 | 992k | bool NestedBlockSpecialCase = |
1070 | 992k | (!Style.isCpp() && Current.is(tok::r_brace) && State.Stack.size() > 1 && |
1071 | 992k | State.Stack[State.Stack.size() - 2].NestedBlockInlined) || |
1072 | 992k | (Style.Language == FormatStyle::LK_ObjC && Current.is(tok::r_brace) && |
1073 | 992k | State.Stack.size() > 1 && !Style.ObjCBreakBeforeNestedBlockParam); |
1074 | | // Do not force parameter break for statements with requires expressions. |
1075 | 992k | NestedBlockSpecialCase = |
1076 | 992k | NestedBlockSpecialCase || |
1077 | 992k | (Current.MatchingParen && |
1078 | 992k | Current.MatchingParen->is(TT_RequiresExpressionLBrace)); |
1079 | 992k | if (!NestedBlockSpecialCase) { |
1080 | 992k | auto ParentLevelIt = std::next(State.Stack.rbegin()); |
1081 | 992k | if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && |
1082 | 992k | Current.MatchingParen && Current.MatchingParen->is(TT_LambdaLBrace)) { |
1083 | | // If the first character on the new line is a lambda's closing brace, the |
1084 | | // stack still contains that lambda's parenthesis. As such, we need to |
1085 | | // recurse further down the stack than usual to find the parenthesis level |
1086 | | // containing the lambda, which is where we want to set |
1087 | | // BreakBeforeParameter. |
1088 | | // |
1089 | | // We specifically special case "OuterScope"-formatted lambdas here |
1090 | | // because, when using that setting, breaking before the parameter |
1091 | | // directly following the lambda is particularly unsightly. However, when |
1092 | | // "OuterScope" is not set, the logic to find the parent parenthesis level |
1093 | | // still appears to be sometimes incorrect. It has not been fixed yet |
1094 | | // because it would lead to significant changes in existing behaviour. |
1095 | | // |
1096 | | // TODO: fix the non-"OuterScope" case too. |
1097 | 0 | auto FindCurrentLevel = [&](const auto &It) { |
1098 | 0 | return std::find_if(It, State.Stack.rend(), [](const auto &PState) { |
1099 | 0 | return PState.Tok != nullptr; // Ignore fake parens. |
1100 | 0 | }); |
1101 | 0 | }; |
1102 | 0 | auto MaybeIncrement = [&](const auto &It) { |
1103 | 0 | return It != State.Stack.rend() ? std::next(It) : It; |
1104 | 0 | }; |
1105 | 0 | auto LambdaLevelIt = FindCurrentLevel(State.Stack.rbegin()); |
1106 | 0 | auto LevelContainingLambdaIt = |
1107 | 0 | FindCurrentLevel(MaybeIncrement(LambdaLevelIt)); |
1108 | 0 | ParentLevelIt = MaybeIncrement(LevelContainingLambdaIt); |
1109 | 0 | } |
1110 | 5.68M | for (auto I = ParentLevelIt, E = State.Stack.rend(); I != E; ++I) |
1111 | 4.69M | I->BreakBeforeParameter = true; |
1112 | 992k | } |
1113 | | |
1114 | 992k | if (PreviousNonComment && |
1115 | 992k | !PreviousNonComment->isOneOf(tok::comma, tok::colon, tok::semi) && |
1116 | 992k | ((PreviousNonComment->isNot(TT_TemplateCloser) && |
1117 | 853k | !PreviousNonComment->ClosesRequiresClause) || |
1118 | 853k | Current.NestingLevel != 0) && |
1119 | 992k | !PreviousNonComment->isOneOf( |
1120 | 834k | TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation, |
1121 | 834k | TT_LeadingJavaAnnotation) && |
1122 | 992k | Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope() && |
1123 | | // We don't want to enforce line breaks for subsequent arguments just |
1124 | | // because we have been forced to break before a lambda body. |
1125 | 992k | (!Style.BraceWrapping.BeforeLambdaBody || |
1126 | 463k | Current.isNot(TT_LambdaLBrace))) { |
1127 | 463k | CurrentState.BreakBeforeParameter = true; |
1128 | 463k | } |
1129 | | |
1130 | | // If we break after { or the [ of an array initializer, we should also break |
1131 | | // before the corresponding } or ]. |
1132 | 992k | if (PreviousNonComment && |
1133 | 992k | (PreviousNonComment->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || |
1134 | 992k | opensProtoMessageField(*PreviousNonComment, Style))) { |
1135 | 13.0k | CurrentState.BreakBeforeClosingBrace = true; |
1136 | 13.0k | } |
1137 | | |
1138 | 992k | if (PreviousNonComment && PreviousNonComment->is(tok::l_paren)) { |
1139 | 13.3k | CurrentState.BreakBeforeClosingParen = |
1140 | 13.3k | Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent; |
1141 | 13.3k | } |
1142 | | |
1143 | 992k | if (CurrentState.AvoidBinPacking) { |
1144 | | // If we are breaking after '(', '{', '<', or this is the break after a ':' |
1145 | | // to start a member initializer list in a constructor, this should not |
1146 | | // be considered bin packing unless the relevant AllowAll option is false or |
1147 | | // this is a dict/object literal. |
1148 | 203k | bool PreviousIsBreakingCtorInitializerColon = |
1149 | 203k | PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && |
1150 | 203k | Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon; |
1151 | 203k | bool AllowAllConstructorInitializersOnNextLine = |
1152 | 203k | Style.PackConstructorInitializers == FormatStyle::PCIS_NextLine || |
1153 | 203k | Style.PackConstructorInitializers == FormatStyle::PCIS_NextLineOnly; |
1154 | 203k | if (!(Previous.isOneOf(tok::l_paren, tok::l_brace, TT_BinaryOperator) || |
1155 | 203k | PreviousIsBreakingCtorInitializerColon) || |
1156 | 203k | (!Style.AllowAllParametersOfDeclarationOnNextLine && |
1157 | 12.9k | State.Line->MustBeDeclaration) || |
1158 | 203k | (!Style.AllowAllArgumentsOnNextLine && |
1159 | 12.9k | !State.Line->MustBeDeclaration) || |
1160 | 203k | (!AllowAllConstructorInitializersOnNextLine && |
1161 | 12.9k | PreviousIsBreakingCtorInitializerColon) || |
1162 | 203k | Previous.is(TT_DictLiteral)) { |
1163 | 200k | CurrentState.BreakBeforeParameter = true; |
1164 | 200k | } |
1165 | | |
1166 | | // If we are breaking after a ':' to start a member initializer list, |
1167 | | // and we allow all arguments on the next line, we should not break |
1168 | | // before the next parameter. |
1169 | 203k | if (PreviousIsBreakingCtorInitializerColon && |
1170 | 203k | AllowAllConstructorInitializersOnNextLine) { |
1171 | 0 | CurrentState.BreakBeforeParameter = false; |
1172 | 0 | } |
1173 | 203k | } |
1174 | | |
1175 | 992k | return Penalty; |
1176 | 992k | } |
1177 | | |
1178 | 4.96M | unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) { |
1179 | 4.96M | if (!State.NextToken || !State.NextToken->Previous) |
1180 | 0 | return 0; |
1181 | | |
1182 | 4.96M | FormatToken &Current = *State.NextToken; |
1183 | 4.96M | const auto &CurrentState = State.Stack.back(); |
1184 | | |
1185 | 4.96M | if (CurrentState.IsCSharpGenericTypeConstraint && |
1186 | 4.96M | Current.isNot(TT_CSharpGenericTypeConstraint)) { |
1187 | 0 | return CurrentState.ColonPos + 2; |
1188 | 0 | } |
1189 | | |
1190 | 4.96M | const FormatToken &Previous = *Current.Previous; |
1191 | | // If we are continuing an expression, we want to use the continuation indent. |
1192 | 4.96M | unsigned ContinuationIndent = |
1193 | 4.96M | std::max(CurrentState.LastSpace, CurrentState.Indent) + |
1194 | 4.96M | Style.ContinuationIndentWidth; |
1195 | 4.96M | const FormatToken *PreviousNonComment = Current.getPreviousNonComment(); |
1196 | 4.96M | const FormatToken *NextNonComment = Previous.getNextNonComment(); |
1197 | 4.96M | if (!NextNonComment) |
1198 | 430 | NextNonComment = &Current; |
1199 | | |
1200 | | // Java specific bits. |
1201 | 4.96M | if (Style.Language == FormatStyle::LK_Java && |
1202 | 4.96M | Current.isOneOf(Keywords.kw_implements, Keywords.kw_extends)) { |
1203 | 0 | return std::max(CurrentState.LastSpace, |
1204 | 0 | CurrentState.Indent + Style.ContinuationIndentWidth); |
1205 | 0 | } |
1206 | | |
1207 | | // Indentation of the statement following a Verilog case label is taken care |
1208 | | // of in moveStateToNextToken. |
1209 | 4.96M | if (Style.isVerilog() && PreviousNonComment && |
1210 | 4.96M | Keywords.isVerilogEndOfLabel(*PreviousNonComment)) { |
1211 | 0 | return State.FirstIndent; |
1212 | 0 | } |
1213 | | |
1214 | 4.96M | if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths && |
1215 | 4.96M | State.Line->First->is(tok::kw_enum)) { |
1216 | 0 | return (Style.IndentWidth * State.Line->First->IndentLevel) + |
1217 | 0 | Style.IndentWidth; |
1218 | 0 | } |
1219 | | |
1220 | 4.96M | if ((NextNonComment->is(tok::l_brace) && NextNonComment->is(BK_Block)) || |
1221 | 4.96M | (Style.isVerilog() && Keywords.isVerilogBegin(*NextNonComment))) { |
1222 | 42.0k | if (Current.NestingLevel == 0 || |
1223 | 42.0k | (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && |
1224 | 41.9k | State.NextToken->is(TT_LambdaLBrace))) { |
1225 | 41.9k | return State.FirstIndent; |
1226 | 41.9k | } |
1227 | 76 | return CurrentState.Indent; |
1228 | 42.0k | } |
1229 | 4.92M | if ((Current.isOneOf(tok::r_brace, tok::r_square) || |
1230 | 4.92M | (Current.is(tok::greater) && Style.isProto())) && |
1231 | 4.92M | State.Stack.size() > 1) { |
1232 | 80.2k | if (Current.closesBlockOrBlockTypeList(Style)) |
1233 | 12.2k | return State.Stack[State.Stack.size() - 2].NestedBlockIndent; |
1234 | 67.9k | if (Current.MatchingParen && Current.MatchingParen->is(BK_BracedInit)) |
1235 | 26.7k | return State.Stack[State.Stack.size() - 2].LastSpace; |
1236 | 41.2k | return State.FirstIndent; |
1237 | 67.9k | } |
1238 | | // Indent a closing parenthesis at the previous level if followed by a semi, |
1239 | | // const, or opening brace. This allows indentations such as: |
1240 | | // foo( |
1241 | | // a, |
1242 | | // ); |
1243 | | // int Foo::getter( |
1244 | | // // |
1245 | | // ) const { |
1246 | | // return foo; |
1247 | | // } |
1248 | | // function foo( |
1249 | | // a, |
1250 | | // ) { |
1251 | | // code(); // |
1252 | | // } |
1253 | 4.84M | if (Current.is(tok::r_paren) && State.Stack.size() > 1 && |
1254 | 4.84M | (!Current.Next || |
1255 | 16.4k | Current.Next->isOneOf(tok::semi, tok::kw_const, tok::l_brace))) { |
1256 | 6.92k | return State.Stack[State.Stack.size() - 2].LastSpace; |
1257 | 6.92k | } |
1258 | 4.83M | if (Style.AlignAfterOpenBracket == FormatStyle::BAS_BlockIndent && |
1259 | 4.83M | (Current.is(tok::r_paren) || |
1260 | 0 | (Current.is(tok::r_brace) && Current.MatchingParen && |
1261 | 0 | Current.MatchingParen->is(BK_BracedInit))) && |
1262 | 4.83M | State.Stack.size() > 1) { |
1263 | 0 | return State.Stack[State.Stack.size() - 2].LastSpace; |
1264 | 0 | } |
1265 | 4.83M | if (NextNonComment->is(TT_TemplateString) && NextNonComment->closesScope()) |
1266 | 0 | return State.Stack[State.Stack.size() - 2].LastSpace; |
1267 | | // Field labels in a nested type should be aligned to the brace. For example |
1268 | | // in ProtoBuf: |
1269 | | // optional int32 b = 2 [(foo_options) = {aaaaaaaaaaaaaaaaaaa: 123, |
1270 | | // bbbbbbbbbbbbbbbbbbbbbbbb:"baz"}]; |
1271 | | // For Verilog, a quote following a brace is treated as an identifier. And |
1272 | | // Both braces and colons get annotated as TT_DictLiteral. So we have to |
1273 | | // check. |
1274 | 4.83M | if (Current.is(tok::identifier) && Current.Next && |
1275 | 4.83M | (!Style.isVerilog() || Current.Next->is(tok::colon)) && |
1276 | 4.83M | (Current.Next->is(TT_DictLiteral) || |
1277 | 894k | (Style.isProto() && Current.Next->isOneOf(tok::less, tok::l_brace)))) { |
1278 | 28.8k | return CurrentState.Indent; |
1279 | 28.8k | } |
1280 | 4.80M | if (NextNonComment->is(TT_ObjCStringLiteral) && |
1281 | 4.80M | State.StartOfStringLiteral != 0) { |
1282 | 2 | return State.StartOfStringLiteral - 1; |
1283 | 2 | } |
1284 | 4.80M | if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0) |
1285 | 12.6k | return State.StartOfStringLiteral; |
1286 | 4.79M | if (NextNonComment->is(tok::lessless) && CurrentState.FirstLessLess != 0) |
1287 | 378 | return CurrentState.FirstLessLess; |
1288 | 4.79M | if (NextNonComment->isMemberAccess()) { |
1289 | 123k | if (CurrentState.CallContinuation == 0) |
1290 | 43.5k | return ContinuationIndent; |
1291 | 79.9k | return CurrentState.CallContinuation; |
1292 | 123k | } |
1293 | 4.66M | if (CurrentState.QuestionColumn != 0 && |
1294 | 4.66M | ((NextNonComment->is(tok::colon) && |
1295 | 707k | NextNonComment->is(TT_ConditionalExpr)) || |
1296 | 707k | Previous.is(TT_ConditionalExpr))) { |
1297 | 73.7k | if (((NextNonComment->is(tok::colon) && NextNonComment->Next && |
1298 | 73.7k | !NextNonComment->Next->FakeLParens.empty() && |
1299 | 73.7k | NextNonComment->Next->FakeLParens.back() == prec::Conditional) || |
1300 | 73.7k | (Previous.is(tok::colon) && !Current.FakeLParens.empty() && |
1301 | 73.6k | Current.FakeLParens.back() == prec::Conditional)) && |
1302 | 73.7k | !CurrentState.IsWrappedConditional) { |
1303 | | // NOTE: we may tweak this slightly: |
1304 | | // * not remove the 'lead' ContinuationIndentWidth |
1305 | | // * always un-indent by the operator when |
1306 | | // BreakBeforeTernaryOperators=true |
1307 | 238 | unsigned Indent = CurrentState.Indent; |
1308 | 238 | if (Style.AlignOperands != FormatStyle::OAS_DontAlign) |
1309 | 238 | Indent -= Style.ContinuationIndentWidth; |
1310 | 238 | if (Style.BreakBeforeTernaryOperators && CurrentState.UnindentOperator) |
1311 | 0 | Indent -= 2; |
1312 | 238 | return Indent; |
1313 | 238 | } |
1314 | 73.5k | return CurrentState.QuestionColumn; |
1315 | 73.7k | } |
1316 | 4.59M | if (Previous.is(tok::comma) && CurrentState.VariablePos != 0) |
1317 | 3.13k | return CurrentState.VariablePos; |
1318 | 4.59M | if (Current.is(TT_RequiresClause)) { |
1319 | 0 | if (Style.IndentRequiresClause) |
1320 | 0 | return CurrentState.Indent + Style.IndentWidth; |
1321 | 0 | switch (Style.RequiresClausePosition) { |
1322 | 0 | case FormatStyle::RCPS_OwnLine: |
1323 | 0 | case FormatStyle::RCPS_WithFollowing: |
1324 | 0 | return CurrentState.Indent; |
1325 | 0 | default: |
1326 | 0 | break; |
1327 | 0 | } |
1328 | 0 | } |
1329 | 4.59M | if (NextNonComment->isOneOf(TT_CtorInitializerColon, TT_InheritanceColon, |
1330 | 4.59M | TT_InheritanceComma)) { |
1331 | 41.0k | return State.FirstIndent + Style.ConstructorInitializerIndentWidth; |
1332 | 41.0k | } |
1333 | 4.55M | if ((PreviousNonComment && |
1334 | 4.55M | (PreviousNonComment->ClosesTemplateDeclaration || |
1335 | 4.55M | PreviousNonComment->ClosesRequiresClause || |
1336 | 4.55M | (PreviousNonComment->is(TT_AttributeMacro) && |
1337 | 4.55M | Current.isNot(tok::l_paren)) || |
1338 | 4.55M | PreviousNonComment->isOneOf( |
1339 | 4.55M | TT_AttributeRParen, TT_AttributeSquare, TT_FunctionAnnotationRParen, |
1340 | 4.55M | TT_JavaAnnotation, TT_LeadingJavaAnnotation))) || |
1341 | 4.55M | (!Style.IndentWrappedFunctionNames && |
1342 | 4.55M | NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName))) { |
1343 | 588 | return std::max(CurrentState.LastSpace, CurrentState.Indent); |
1344 | 588 | } |
1345 | 4.55M | if (NextNonComment->is(TT_SelectorName)) { |
1346 | 6.12k | if (!CurrentState.ObjCSelectorNameFound) { |
1347 | 5.85k | unsigned MinIndent = CurrentState.Indent; |
1348 | 5.85k | if (shouldIndentWrappedSelectorName(Style, State.Line->Type)) { |
1349 | 21 | MinIndent = std::max(MinIndent, |
1350 | 21 | State.FirstIndent + Style.ContinuationIndentWidth); |
1351 | 21 | } |
1352 | | // If LongestObjCSelectorName is 0, we are indenting the first |
1353 | | // part of an ObjC selector (or a selector component which is |
1354 | | // not colon-aligned due to block formatting). |
1355 | | // |
1356 | | // Otherwise, we are indenting a subsequent part of an ObjC |
1357 | | // selector which should be colon-aligned to the longest |
1358 | | // component of the ObjC selector. |
1359 | | // |
1360 | | // In either case, we want to respect Style.IndentWrappedFunctionNames. |
1361 | 5.85k | return MinIndent + |
1362 | 5.85k | std::max(NextNonComment->LongestObjCSelectorName, |
1363 | 5.85k | NextNonComment->ColumnWidth) - |
1364 | 5.85k | NextNonComment->ColumnWidth; |
1365 | 5.85k | } |
1366 | 269 | if (!CurrentState.AlignColons) |
1367 | 247 | return CurrentState.Indent; |
1368 | 22 | if (CurrentState.ColonPos > NextNonComment->ColumnWidth) |
1369 | 22 | return CurrentState.ColonPos - NextNonComment->ColumnWidth; |
1370 | 0 | return CurrentState.Indent; |
1371 | 22 | } |
1372 | 4.54M | if (NextNonComment->is(tok::colon) && NextNonComment->is(TT_ObjCMethodExpr)) |
1373 | 5.93k | return CurrentState.ColonPos; |
1374 | 4.53M | if (NextNonComment->is(TT_ArraySubscriptLSquare)) { |
1375 | 17.5k | if (CurrentState.StartOfArraySubscripts != 0) { |
1376 | 643 | return CurrentState.StartOfArraySubscripts; |
1377 | 16.8k | } else if (Style.isCSharp()) { // C# allows `["key"] = value` inside object |
1378 | | // initializers. |
1379 | 0 | return CurrentState.Indent; |
1380 | 0 | } |
1381 | 16.8k | return ContinuationIndent; |
1382 | 17.5k | } |
1383 | | |
1384 | | // OpenMP clauses want to get additional indentation when they are pushed onto |
1385 | | // the next line. |
1386 | 4.52M | if (State.Line->InPragmaDirective) { |
1387 | 60 | FormatToken *PragmaType = State.Line->First->Next->Next; |
1388 | 60 | if (PragmaType && PragmaType->TokenText.equals("omp")) |
1389 | 0 | return CurrentState.Indent + Style.ContinuationIndentWidth; |
1390 | 60 | } |
1391 | | |
1392 | | // This ensure that we correctly format ObjC methods calls without inputs, |
1393 | | // i.e. where the last element isn't selector like: [callee method]; |
1394 | 4.52M | if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 && |
1395 | 4.52M | NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr)) { |
1396 | 2.92k | return CurrentState.Indent; |
1397 | 2.92k | } |
1398 | | |
1399 | 4.51M | if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) || |
1400 | 4.51M | Previous.isOneOf(tok::coloncolon, tok::equal, TT_JsTypeColon)) { |
1401 | 309k | return ContinuationIndent; |
1402 | 309k | } |
1403 | 4.20M | if (PreviousNonComment && PreviousNonComment->is(tok::colon) && |
1404 | 4.20M | PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral)) { |
1405 | 30.2k | return ContinuationIndent; |
1406 | 30.2k | } |
1407 | 4.17M | if (NextNonComment->is(TT_CtorInitializerComma)) |
1408 | 0 | return CurrentState.Indent; |
1409 | 4.17M | if (PreviousNonComment && PreviousNonComment->is(TT_CtorInitializerColon) && |
1410 | 4.17M | Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { |
1411 | 0 | return CurrentState.Indent; |
1412 | 0 | } |
1413 | 4.17M | if (PreviousNonComment && PreviousNonComment->is(TT_InheritanceColon) && |
1414 | 4.17M | Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) { |
1415 | 0 | return CurrentState.Indent; |
1416 | 0 | } |
1417 | 4.17M | if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() && |
1418 | 4.17M | !Current.isOneOf(tok::colon, tok::comment)) { |
1419 | 23.6k | return ContinuationIndent; |
1420 | 23.6k | } |
1421 | 4.15M | if (Current.is(TT_ProtoExtensionLSquare)) |
1422 | 0 | return CurrentState.Indent; |
1423 | 4.15M | if (Current.isBinaryOperator() && CurrentState.UnindentOperator) { |
1424 | 0 | return CurrentState.Indent - Current.Tok.getLength() - |
1425 | 0 | Current.SpacesRequiredBefore; |
1426 | 0 | } |
1427 | 4.15M | if (Current.is(tok::comment) && NextNonComment->isBinaryOperator() && |
1428 | 4.15M | CurrentState.UnindentOperator) { |
1429 | 0 | return CurrentState.Indent - NextNonComment->Tok.getLength() - |
1430 | 0 | NextNonComment->SpacesRequiredBefore; |
1431 | 0 | } |
1432 | 4.15M | if (CurrentState.Indent == State.FirstIndent && PreviousNonComment && |
1433 | 4.15M | !PreviousNonComment->isOneOf(tok::r_brace, TT_CtorInitializerComma)) { |
1434 | | // Ensure that we fall back to the continuation indent width instead of |
1435 | | // just flushing continuations left. |
1436 | 652k | return CurrentState.Indent + Style.ContinuationIndentWidth; |
1437 | 652k | } |
1438 | 3.50M | return CurrentState.Indent; |
1439 | 4.15M | } |
1440 | | |
1441 | | static bool hasNestedBlockInlined(const FormatToken *Previous, |
1442 | | const FormatToken &Current, |
1443 | 595k | const FormatStyle &Style) { |
1444 | 595k | if (Previous->isNot(tok::l_paren)) |
1445 | 567k | return true; |
1446 | 27.8k | if (Previous->ParameterCount > 1) |
1447 | 7.01k | return true; |
1448 | | |
1449 | | // Also a nested block if contains a lambda inside function with 1 parameter. |
1450 | 20.8k | return Style.BraceWrapping.BeforeLambdaBody && Current.is(TT_LambdaLSquare); |
1451 | 27.8k | } |
1452 | | |
1453 | | unsigned ContinuationIndenter::moveStateToNextToken(LineState &State, |
1454 | 7.04M | bool DryRun, bool Newline) { |
1455 | 7.04M | assert(State.Stack.size()); |
1456 | 0 | const FormatToken &Current = *State.NextToken; |
1457 | 7.04M | auto &CurrentState = State.Stack.back(); |
1458 | | |
1459 | 7.04M | if (Current.is(TT_CSharpGenericTypeConstraint)) |
1460 | 0 | CurrentState.IsCSharpGenericTypeConstraint = true; |
1461 | 7.04M | if (Current.isOneOf(tok::comma, TT_BinaryOperator)) |
1462 | 623k | CurrentState.NoLineBreakInOperand = false; |
1463 | 7.04M | if (Current.isOneOf(TT_InheritanceColon, TT_CSharpGenericTypeConstraintColon)) |
1464 | 47.5k | CurrentState.AvoidBinPacking = true; |
1465 | 7.04M | if (Current.is(tok::lessless) && Current.isNot(TT_OverloadedOperator)) { |
1466 | 773 | if (CurrentState.FirstLessLess == 0) |
1467 | 392 | CurrentState.FirstLessLess = State.Column; |
1468 | 381 | else |
1469 | 381 | CurrentState.LastOperatorWrapped = Newline; |
1470 | 773 | } |
1471 | 7.04M | if (Current.is(TT_BinaryOperator) && Current.isNot(tok::lessless)) |
1472 | 474k | CurrentState.LastOperatorWrapped = Newline; |
1473 | 7.04M | if (Current.is(TT_ConditionalExpr) && Current.Previous && |
1474 | 7.04M | Current.Previous->isNot(TT_ConditionalExpr)) { |
1475 | 33.3k | CurrentState.LastOperatorWrapped = Newline; |
1476 | 33.3k | } |
1477 | 7.04M | if (Current.is(TT_ArraySubscriptLSquare) && |
1478 | 7.04M | CurrentState.StartOfArraySubscripts == 0) { |
1479 | 18.5k | CurrentState.StartOfArraySubscripts = State.Column; |
1480 | 18.5k | } |
1481 | | |
1482 | 7.04M | auto IsWrappedConditional = [](const FormatToken &Tok) { |
1483 | 7.04M | if (!(Tok.is(TT_ConditionalExpr) && Tok.is(tok::question))) |
1484 | 6.96M | return false; |
1485 | 83.0k | if (Tok.MustBreakBefore) |
1486 | 21 | return true; |
1487 | | |
1488 | 82.9k | const FormatToken *Next = Tok.getNextNonComment(); |
1489 | 82.9k | return Next && Next->MustBreakBefore; |
1490 | 83.0k | }; |
1491 | 7.04M | if (IsWrappedConditional(Current)) |
1492 | 36 | CurrentState.IsWrappedConditional = true; |
1493 | 7.04M | if (Style.BreakBeforeTernaryOperators && Current.is(tok::question)) |
1494 | 83.0k | CurrentState.QuestionColumn = State.Column; |
1495 | 7.04M | if (!Style.BreakBeforeTernaryOperators && Current.isNot(tok::colon)) { |
1496 | 0 | const FormatToken *Previous = Current.Previous; |
1497 | 0 | while (Previous && Previous->isTrailingComment()) |
1498 | 0 | Previous = Previous->Previous; |
1499 | 0 | if (Previous && Previous->is(tok::question)) |
1500 | 0 | CurrentState.QuestionColumn = State.Column; |
1501 | 0 | } |
1502 | 7.04M | if (!Current.opensScope() && !Current.closesScope() && |
1503 | 7.04M | Current.isNot(TT_PointerOrReference)) { |
1504 | 6.00M | State.LowestLevelOnLine = |
1505 | 6.00M | std::min(State.LowestLevelOnLine, Current.NestingLevel); |
1506 | 6.00M | } |
1507 | 7.04M | if (Current.isMemberAccess()) |
1508 | 123k | CurrentState.StartOfFunctionCall = !Current.NextOperator ? 0 : State.Column; |
1509 | 7.04M | if (Current.is(TT_SelectorName)) |
1510 | 6.37k | CurrentState.ObjCSelectorNameFound = true; |
1511 | 7.04M | if (Current.is(TT_CtorInitializerColon) && |
1512 | 7.04M | Style.BreakConstructorInitializers != FormatStyle::BCIS_AfterColon) { |
1513 | | // Indent 2 from the column, so: |
1514 | | // SomeClass::SomeClass() |
1515 | | // : First(...), ... |
1516 | | // Next(...) |
1517 | | // ^ line up here. |
1518 | 310 | CurrentState.Indent = State.Column + (Style.BreakConstructorInitializers == |
1519 | 310 | FormatStyle::BCIS_BeforeComma |
1520 | 310 | ? 0 |
1521 | 310 | : 2); |
1522 | 310 | CurrentState.NestedBlockIndent = CurrentState.Indent; |
1523 | 310 | if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) { |
1524 | 310 | CurrentState.AvoidBinPacking = true; |
1525 | 310 | CurrentState.BreakBeforeParameter = |
1526 | 310 | Style.ColumnLimit > 0 && |
1527 | 310 | Style.PackConstructorInitializers != FormatStyle::PCIS_NextLine && |
1528 | 310 | Style.PackConstructorInitializers != FormatStyle::PCIS_NextLineOnly; |
1529 | 310 | } else { |
1530 | 0 | CurrentState.BreakBeforeParameter = false; |
1531 | 0 | } |
1532 | 310 | } |
1533 | 7.04M | if (Current.is(TT_CtorInitializerColon) && |
1534 | 7.04M | Style.BreakConstructorInitializers == FormatStyle::BCIS_AfterColon) { |
1535 | 0 | CurrentState.Indent = |
1536 | 0 | State.FirstIndent + Style.ConstructorInitializerIndentWidth; |
1537 | 0 | CurrentState.NestedBlockIndent = CurrentState.Indent; |
1538 | 0 | if (Style.PackConstructorInitializers > FormatStyle::PCIS_BinPack) |
1539 | 0 | CurrentState.AvoidBinPacking = true; |
1540 | 0 | else |
1541 | 0 | CurrentState.BreakBeforeParameter = false; |
1542 | 0 | } |
1543 | 7.04M | if (Current.is(TT_InheritanceColon)) { |
1544 | 47.5k | CurrentState.Indent = |
1545 | 47.5k | State.FirstIndent + Style.ConstructorInitializerIndentWidth; |
1546 | 47.5k | } |
1547 | 7.04M | if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline) |
1548 | 128k | CurrentState.NestedBlockIndent = State.Column + Current.ColumnWidth + 1; |
1549 | 7.04M | if (Current.isOneOf(TT_LambdaLSquare, TT_TrailingReturnArrow)) |
1550 | 20 | CurrentState.LastSpace = State.Column; |
1551 | 7.04M | if (Current.is(TT_RequiresExpression) && |
1552 | 7.04M | Style.RequiresExpressionIndentation == FormatStyle::REI_Keyword) { |
1553 | 0 | CurrentState.NestedBlockIndent = State.Column; |
1554 | 0 | } |
1555 | | |
1556 | | // Insert scopes created by fake parenthesis. |
1557 | 7.04M | const FormatToken *Previous = Current.getPreviousNonComment(); |
1558 | | |
1559 | | // Add special behavior to support a format commonly used for JavaScript |
1560 | | // closures: |
1561 | | // SomeFunction(function() { |
1562 | | // foo(); |
1563 | | // bar(); |
1564 | | // }, a, b, c); |
1565 | 7.04M | if (Current.isNot(tok::comment) && !Current.ClosesRequiresClause && |
1566 | 7.04M | Previous && Previous->isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) && |
1567 | 7.04M | Previous->isNot(TT_DictLiteral) && State.Stack.size() > 1 && |
1568 | 7.04M | !CurrentState.HasMultipleNestedBlocks) { |
1569 | 109k | if (State.Stack[State.Stack.size() - 2].NestedBlockInlined && Newline) |
1570 | 88 | for (ParenState &PState : llvm::drop_end(State.Stack)) |
1571 | 273 | PState.NoLineBreak = true; |
1572 | 109k | State.Stack[State.Stack.size() - 2].NestedBlockInlined = false; |
1573 | 109k | } |
1574 | 7.04M | if (Previous && (Previous->isOneOf(TT_BinaryOperator, TT_ConditionalExpr) || |
1575 | 6.63M | (Previous->isOneOf(tok::l_paren, tok::comma, tok::colon) && |
1576 | 5.97M | !Previous->isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))) { |
1577 | 966k | CurrentState.NestedBlockInlined = |
1578 | 966k | !Newline && hasNestedBlockInlined(Previous, Current, Style); |
1579 | 966k | } |
1580 | | |
1581 | 7.04M | moveStatePastFakeLParens(State, Newline); |
1582 | 7.04M | moveStatePastScopeCloser(State); |
1583 | | // Do not use CurrentState here, since the two functions before may change the |
1584 | | // Stack. |
1585 | 7.04M | bool AllowBreak = !State.Stack.back().NoLineBreak && |
1586 | 7.04M | !State.Stack.back().NoLineBreakInOperand; |
1587 | 7.04M | moveStatePastScopeOpener(State, Newline); |
1588 | 7.04M | moveStatePastFakeRParens(State); |
1589 | | |
1590 | 7.04M | if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0) |
1591 | 8 | State.StartOfStringLiteral = State.Column + 1; |
1592 | 7.04M | if (Current.is(TT_CSharpStringLiteral) && State.StartOfStringLiteral == 0) { |
1593 | 0 | State.StartOfStringLiteral = State.Column + 1; |
1594 | 7.04M | } else if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) { |
1595 | 95.1k | State.StartOfStringLiteral = State.Column; |
1596 | 6.95M | } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) && |
1597 | 6.95M | !Current.isStringLiteral()) { |
1598 | 5.51M | State.StartOfStringLiteral = 0; |
1599 | 5.51M | } |
1600 | | |
1601 | 7.04M | State.Column += Current.ColumnWidth; |
1602 | 7.04M | State.NextToken = State.NextToken->Next; |
1603 | | // Verilog case labels are on the same unwrapped lines as the statements that |
1604 | | // follow. TokenAnnotator identifies them and sets MustBreakBefore. |
1605 | | // Indentation is taken care of here. A case label can only have 1 statement |
1606 | | // in Verilog, so we don't have to worry about lines that follow. |
1607 | 7.04M | if (Style.isVerilog() && State.NextToken && |
1608 | 7.04M | State.NextToken->MustBreakBefore && |
1609 | 7.04M | Keywords.isVerilogEndOfLabel(Current)) { |
1610 | 0 | State.FirstIndent += Style.IndentWidth; |
1611 | 0 | CurrentState.Indent = State.FirstIndent; |
1612 | 0 | } |
1613 | | |
1614 | 7.04M | unsigned Penalty = |
1615 | 7.04M | handleEndOfLine(Current, State, DryRun, AllowBreak, Newline); |
1616 | | |
1617 | 7.04M | if (Current.Role) |
1618 | 21.4k | Current.Role->formatFromToken(State, this, DryRun); |
1619 | | // If the previous has a special role, let it consume tokens as appropriate. |
1620 | | // It is necessary to start at the previous token for the only implemented |
1621 | | // role (comma separated list). That way, the decision whether or not to break |
1622 | | // after the "{" is already done and both options are tried and evaluated. |
1623 | | // FIXME: This is ugly, find a better way. |
1624 | 7.04M | if (Previous && Previous->Role) |
1625 | 25.2k | Penalty += Previous->Role->formatAfterToken(State, this, DryRun); |
1626 | | |
1627 | 7.04M | return Penalty; |
1628 | 7.04M | } |
1629 | | |
1630 | | void ContinuationIndenter::moveStatePastFakeLParens(LineState &State, |
1631 | 7.04M | bool Newline) { |
1632 | 7.04M | const FormatToken &Current = *State.NextToken; |
1633 | 7.04M | if (Current.FakeLParens.empty()) |
1634 | 6.67M | return; |
1635 | | |
1636 | 372k | const FormatToken *Previous = Current.getPreviousNonComment(); |
1637 | | |
1638 | | // Don't add extra indentation for the first fake parenthesis after |
1639 | | // 'return', assignments, opening <({[, or requires clauses. The indentation |
1640 | | // for these cases is special cased. |
1641 | 372k | bool SkipFirstExtraIndent = |
1642 | 372k | Previous && |
1643 | 372k | (Previous->opensScope() || |
1644 | 310k | Previous->isOneOf(tok::semi, tok::kw_return, TT_RequiresClause) || |
1645 | 310k | (Previous->getPrecedence() == prec::Assignment && |
1646 | 172k | Style.AlignOperands != FormatStyle::OAS_DontAlign) || |
1647 | 310k | Previous->is(TT_ObjCMethodExpr)); |
1648 | 405k | for (const auto &PrecedenceLevel : llvm::reverse(Current.FakeLParens)) { |
1649 | 405k | const auto &CurrentState = State.Stack.back(); |
1650 | 405k | ParenState NewParenState = CurrentState; |
1651 | 405k | NewParenState.Tok = nullptr; |
1652 | 405k | NewParenState.ContainsLineBreak = false; |
1653 | 405k | NewParenState.LastOperatorWrapped = true; |
1654 | 405k | NewParenState.IsChainedConditional = false; |
1655 | 405k | NewParenState.IsWrappedConditional = false; |
1656 | 405k | NewParenState.UnindentOperator = false; |
1657 | 405k | NewParenState.NoLineBreak = |
1658 | 405k | NewParenState.NoLineBreak || CurrentState.NoLineBreakInOperand; |
1659 | | |
1660 | | // Don't propagate AvoidBinPacking into subexpressions of arg/param lists. |
1661 | 405k | if (PrecedenceLevel > prec::Comma) |
1662 | 233k | NewParenState.AvoidBinPacking = false; |
1663 | | |
1664 | | // Indent from 'LastSpace' unless these are fake parentheses encapsulating |
1665 | | // a builder type call after 'return' or, if the alignment after opening |
1666 | | // brackets is disabled. |
1667 | 405k | if (!Current.isTrailingComment() && |
1668 | 405k | (Style.AlignOperands != FormatStyle::OAS_DontAlign || |
1669 | 405k | PrecedenceLevel < prec::Assignment) && |
1670 | 405k | (!Previous || Previous->isNot(tok::kw_return) || |
1671 | 405k | (Style.Language != FormatStyle::LK_Java && PrecedenceLevel > 0)) && |
1672 | 405k | (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign || |
1673 | 405k | PrecedenceLevel != prec::Comma || Current.NestingLevel == 0)) { |
1674 | 405k | NewParenState.Indent = std::max( |
1675 | 405k | std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace); |
1676 | 405k | } |
1677 | | |
1678 | | // Special case for generic selection expressions, its comma-separated |
1679 | | // expressions are not aligned to the opening paren like regular calls, but |
1680 | | // rather continuation-indented relative to the _Generic keyword. |
1681 | 405k | if (Previous && Previous->endsSequence(tok::l_paren, tok::kw__Generic)) |
1682 | 0 | NewParenState.Indent = CurrentState.LastSpace; |
1683 | | |
1684 | 405k | if ((shouldUnindentNextOperator(Current) || |
1685 | 405k | (Previous && |
1686 | 387k | (PrecedenceLevel == prec::Conditional && |
1687 | 316k | Previous->is(tok::question) && Previous->is(TT_ConditionalExpr)))) && |
1688 | 405k | !Newline) { |
1689 | | // If BreakBeforeBinaryOperators is set, un-indent a bit to account for |
1690 | | // the operator and keep the operands aligned. |
1691 | 12.5k | if (Style.AlignOperands == FormatStyle::OAS_AlignAfterOperator) |
1692 | 0 | NewParenState.UnindentOperator = true; |
1693 | | // Mark indentation as alignment if the expression is aligned. |
1694 | 12.5k | if (Style.AlignOperands != FormatStyle::OAS_DontAlign) |
1695 | 12.5k | NewParenState.IsAligned = true; |
1696 | 12.5k | } |
1697 | | |
1698 | | // Do not indent relative to the fake parentheses inserted for "." or "->". |
1699 | | // This is a special case to make the following to statements consistent: |
1700 | | // OuterFunction(InnerFunctionCall( // break |
1701 | | // ParameterToInnerFunction)); |
1702 | | // OuterFunction(SomeObject.InnerFunctionCall( // break |
1703 | | // ParameterToInnerFunction)); |
1704 | 405k | if (PrecedenceLevel > prec::Unknown) |
1705 | 264k | NewParenState.LastSpace = std::max(NewParenState.LastSpace, State.Column); |
1706 | 405k | if (PrecedenceLevel != prec::Conditional && |
1707 | 405k | Current.isNot(TT_UnaryOperator) && |
1708 | 405k | Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign) { |
1709 | 343k | NewParenState.StartOfFunctionCall = State.Column; |
1710 | 343k | } |
1711 | | |
1712 | | // Indent conditional expressions, unless they are chained "else-if" |
1713 | | // conditionals. Never indent expression where the 'operator' is ',', ';' or |
1714 | | // an assignment (i.e. *I <= prec::Assignment) as those have different |
1715 | | // indentation rules. Indent other expression, unless the indentation needs |
1716 | | // to be skipped. |
1717 | 405k | if (PrecedenceLevel == prec::Conditional && Previous && |
1718 | 405k | Previous->is(tok::colon) && Previous->is(TT_ConditionalExpr) && |
1719 | 405k | &PrecedenceLevel == &Current.FakeLParens.back() && |
1720 | 405k | !CurrentState.IsWrappedConditional) { |
1721 | 73 | NewParenState.IsChainedConditional = true; |
1722 | 73 | NewParenState.UnindentOperator = State.Stack.back().UnindentOperator; |
1723 | 405k | } else if (PrecedenceLevel == prec::Conditional || |
1724 | 405k | (!SkipFirstExtraIndent && PrecedenceLevel > prec::Assignment && |
1725 | 399k | !Current.isTrailingComment())) { |
1726 | 100k | NewParenState.Indent += Style.ContinuationIndentWidth; |
1727 | 100k | } |
1728 | 405k | if ((Previous && !Previous->opensScope()) || PrecedenceLevel != prec::Comma) |
1729 | 375k | NewParenState.BreakBeforeParameter = false; |
1730 | 405k | State.Stack.push_back(NewParenState); |
1731 | 405k | SkipFirstExtraIndent = false; |
1732 | 405k | } |
1733 | 372k | } |
1734 | | |
1735 | 7.04M | void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) { |
1736 | 7.49M | for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) { |
1737 | 447k | unsigned VariablePos = State.Stack.back().VariablePos; |
1738 | 447k | if (State.Stack.size() == 1) { |
1739 | | // Do not pop the last element. |
1740 | 604 | break; |
1741 | 604 | } |
1742 | 446k | State.Stack.pop_back(); |
1743 | 446k | State.Stack.back().VariablePos = VariablePos; |
1744 | 446k | } |
1745 | | |
1746 | 7.04M | if (State.NextToken->ClosesRequiresClause && Style.IndentRequiresClause) { |
1747 | | // Remove the indentation of the requires clauses (which is not in Indent, |
1748 | | // but in LastSpace). |
1749 | 0 | State.Stack.back().LastSpace -= Style.IndentWidth; |
1750 | 0 | } |
1751 | 7.04M | } |
1752 | | |
1753 | | void ContinuationIndenter::moveStatePastScopeOpener(LineState &State, |
1754 | 7.04M | bool Newline) { |
1755 | 7.04M | const FormatToken &Current = *State.NextToken; |
1756 | 7.04M | if (!Current.opensScope()) |
1757 | 6.52M | return; |
1758 | | |
1759 | 518k | const auto &CurrentState = State.Stack.back(); |
1760 | | |
1761 | | // Don't allow '<' or '(' in C# generic type constraints to start new scopes. |
1762 | 518k | if (Current.isOneOf(tok::less, tok::l_paren) && |
1763 | 518k | CurrentState.IsCSharpGenericTypeConstraint) { |
1764 | 0 | return; |
1765 | 0 | } |
1766 | | |
1767 | 518k | if (Current.MatchingParen && Current.is(BK_Block)) { |
1768 | 167 | moveStateToNewBlock(State); |
1769 | 167 | return; |
1770 | 167 | } |
1771 | | |
1772 | 518k | unsigned NewIndent; |
1773 | 518k | unsigned LastSpace = CurrentState.LastSpace; |
1774 | 518k | bool AvoidBinPacking; |
1775 | 518k | bool BreakBeforeParameter = false; |
1776 | 518k | unsigned NestedBlockIndent = std::max(CurrentState.StartOfFunctionCall, |
1777 | 518k | CurrentState.NestedBlockIndent); |
1778 | 518k | if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) || |
1779 | 518k | opensProtoMessageField(Current, Style)) { |
1780 | 260k | if (Current.opensBlockOrBlockTypeList(Style)) { |
1781 | 200k | NewIndent = Style.IndentWidth + |
1782 | 200k | std::min(State.Column, CurrentState.NestedBlockIndent); |
1783 | 200k | } else if (Current.is(tok::l_brace)) { |
1784 | 59.8k | NewIndent = |
1785 | 59.8k | CurrentState.LastSpace + Style.BracedInitializerIndentWidth.value_or( |
1786 | 59.8k | Style.ContinuationIndentWidth); |
1787 | 59.8k | } else { |
1788 | 0 | NewIndent = CurrentState.LastSpace + Style.ContinuationIndentWidth; |
1789 | 0 | } |
1790 | 260k | const FormatToken *NextNonComment = Current.getNextNonComment(); |
1791 | 260k | bool EndsInComma = Current.MatchingParen && |
1792 | 260k | Current.MatchingParen->Previous && |
1793 | 260k | Current.MatchingParen->Previous->is(tok::comma); |
1794 | 260k | AvoidBinPacking = EndsInComma || Current.is(TT_DictLiteral) || |
1795 | 260k | Style.isProto() || !Style.BinPackArguments || |
1796 | 260k | (NextNonComment && NextNonComment->isOneOf( |
1797 | 107k | TT_DesignatedInitializerPeriod, |
1798 | 107k | TT_DesignatedInitializerLSquare)); |
1799 | 260k | BreakBeforeParameter = EndsInComma; |
1800 | 260k | if (Current.ParameterCount > 1) |
1801 | 6.91k | NestedBlockIndent = std::max(NestedBlockIndent, State.Column + 1); |
1802 | 260k | } else { |
1803 | 258k | NewIndent = |
1804 | 258k | Style.ContinuationIndentWidth + |
1805 | 258k | std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall); |
1806 | | |
1807 | | // Ensure that different different brackets force relative alignment, e.g.: |
1808 | | // void SomeFunction(vector< // break |
1809 | | // int> v); |
1810 | | // FIXME: We likely want to do this for more combinations of brackets. |
1811 | 258k | if (Current.is(tok::less) && Current.ParentBracket == tok::l_paren) { |
1812 | 0 | NewIndent = std::max(NewIndent, CurrentState.Indent); |
1813 | 0 | LastSpace = std::max(LastSpace, CurrentState.Indent); |
1814 | 0 | } |
1815 | | |
1816 | 258k | bool EndsInComma = |
1817 | 258k | Current.MatchingParen && |
1818 | 258k | Current.MatchingParen->getPreviousNonComment() && |
1819 | 258k | Current.MatchingParen->getPreviousNonComment()->is(tok::comma); |
1820 | | |
1821 | | // If ObjCBinPackProtocolList is unspecified, fall back to BinPackParameters |
1822 | | // for backwards compatibility. |
1823 | 258k | bool ObjCBinPackProtocolList = |
1824 | 258k | (Style.ObjCBinPackProtocolList == FormatStyle::BPS_Auto && |
1825 | 258k | Style.BinPackParameters) || |
1826 | 258k | Style.ObjCBinPackProtocolList == FormatStyle::BPS_Always; |
1827 | | |
1828 | 258k | bool BinPackDeclaration = |
1829 | 258k | (State.Line->Type != LT_ObjCDecl && Style.BinPackParameters) || |
1830 | 258k | (State.Line->Type == LT_ObjCDecl && ObjCBinPackProtocolList); |
1831 | | |
1832 | 258k | bool GenericSelection = |
1833 | 258k | Current.getPreviousNonComment() && |
1834 | 258k | Current.getPreviousNonComment()->is(tok::kw__Generic); |
1835 | | |
1836 | 258k | AvoidBinPacking = |
1837 | 258k | (CurrentState.IsCSharpGenericTypeConstraint) || GenericSelection || |
1838 | 258k | (Style.isJavaScript() && EndsInComma) || |
1839 | 258k | (State.Line->MustBeDeclaration && !BinPackDeclaration) || |
1840 | 258k | (!State.Line->MustBeDeclaration && !Style.BinPackArguments) || |
1841 | 258k | (Style.ExperimentalAutoDetectBinPacking && |
1842 | 258k | (Current.is(PPK_OnePerLine) || |
1843 | 0 | (!BinPackInconclusiveFunctions && Current.is(PPK_Inconclusive)))); |
1844 | | |
1845 | 258k | if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen && |
1846 | 258k | Style.ObjCBreakBeforeNestedBlockParam) { |
1847 | 9.61k | if (Style.ColumnLimit) { |
1848 | | // If this '[' opens an ObjC call, determine whether all parameters fit |
1849 | | // into one line and put one per line if they don't. |
1850 | 9.61k | if (getLengthToMatchingParen(Current, State.Stack) + State.Column > |
1851 | 9.61k | getColumnLimit(State)) { |
1852 | 7.32k | BreakBeforeParameter = true; |
1853 | 7.32k | } |
1854 | 9.61k | } else { |
1855 | | // For ColumnLimit = 0, we have to figure out whether there is or has to |
1856 | | // be a line break within this call. |
1857 | 0 | for (const FormatToken *Tok = &Current; |
1858 | 0 | Tok && Tok != Current.MatchingParen; Tok = Tok->Next) { |
1859 | 0 | if (Tok->MustBreakBefore || |
1860 | 0 | (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) { |
1861 | 0 | BreakBeforeParameter = true; |
1862 | 0 | break; |
1863 | 0 | } |
1864 | 0 | } |
1865 | 0 | } |
1866 | 9.61k | } |
1867 | | |
1868 | 258k | if (Style.isJavaScript() && EndsInComma) |
1869 | 0 | BreakBeforeParameter = true; |
1870 | 258k | } |
1871 | | // Generally inherit NoLineBreak from the current scope to nested scope. |
1872 | | // However, don't do this for non-empty nested blocks, dict literals and |
1873 | | // array literals as these follow different indentation rules. |
1874 | 518k | bool NoLineBreak = |
1875 | 518k | Current.Children.empty() && |
1876 | 518k | !Current.isOneOf(TT_DictLiteral, TT_ArrayInitializerLSquare) && |
1877 | 518k | (CurrentState.NoLineBreak || CurrentState.NoLineBreakInOperand || |
1878 | 489k | (Current.is(TT_TemplateOpener) && |
1879 | 481k | CurrentState.ContainsUnwrappedBuilder)); |
1880 | 518k | State.Stack.push_back( |
1881 | 518k | ParenState(&Current, NewIndent, LastSpace, AvoidBinPacking, NoLineBreak)); |
1882 | 518k | auto &NewState = State.Stack.back(); |
1883 | 518k | NewState.NestedBlockIndent = NestedBlockIndent; |
1884 | 518k | NewState.BreakBeforeParameter = BreakBeforeParameter; |
1885 | 518k | NewState.HasMultipleNestedBlocks = (Current.BlockParameterCount > 1); |
1886 | | |
1887 | 518k | if (Style.BraceWrapping.BeforeLambdaBody && Current.Next && |
1888 | 518k | Current.is(tok::l_paren)) { |
1889 | | // Search for any parameter that is a lambda. |
1890 | 0 | FormatToken const *next = Current.Next; |
1891 | 0 | while (next) { |
1892 | 0 | if (next->is(TT_LambdaLSquare)) { |
1893 | 0 | NewState.HasMultipleNestedBlocks = true; |
1894 | 0 | break; |
1895 | 0 | } |
1896 | 0 | next = next->Next; |
1897 | 0 | } |
1898 | 0 | } |
1899 | | |
1900 | 518k | NewState.IsInsideObjCArrayLiteral = Current.is(TT_ArrayInitializerLSquare) && |
1901 | 518k | Current.Previous && |
1902 | 518k | Current.Previous->is(tok::at); |
1903 | 518k | } |
1904 | | |
1905 | 7.04M | void ContinuationIndenter::moveStatePastScopeCloser(LineState &State) { |
1906 | 7.04M | const FormatToken &Current = *State.NextToken; |
1907 | 7.04M | if (!Current.closesScope()) |
1908 | 6.56M | return; |
1909 | | |
1910 | | // If we encounter a closing ), ], } or >, we can remove a level from our |
1911 | | // stacks. |
1912 | 478k | if (State.Stack.size() > 1 && |
1913 | 478k | (Current.isOneOf(tok::r_paren, tok::r_square, TT_TemplateString) || |
1914 | 353k | (Current.is(tok::r_brace) && State.NextToken != State.Line->First) || |
1915 | 353k | State.NextToken->is(TT_TemplateCloser) || |
1916 | 353k | (Current.is(tok::greater) && Current.is(TT_DictLiteral)))) { |
1917 | 353k | State.Stack.pop_back(); |
1918 | 353k | } |
1919 | | |
1920 | 478k | auto &CurrentState = State.Stack.back(); |
1921 | | |
1922 | | // Reevaluate whether ObjC message arguments fit into one line. |
1923 | | // If a receiver spans multiple lines, e.g.: |
1924 | | // [[object block:^{ |
1925 | | // return 42; |
1926 | | // }] a:42 b:42]; |
1927 | | // BreakBeforeParameter is calculated based on an incorrect assumption |
1928 | | // (it is checked whether the whole expression fits into one line without |
1929 | | // considering a line break inside a message receiver). |
1930 | | // We check whether arguments fit after receiver scope closer (into the same |
1931 | | // line). |
1932 | 478k | if (CurrentState.BreakBeforeParameter && Current.MatchingParen && |
1933 | 478k | Current.MatchingParen->Previous) { |
1934 | 233k | const FormatToken &CurrentScopeOpener = *Current.MatchingParen->Previous; |
1935 | 233k | if (CurrentScopeOpener.is(TT_ObjCMethodExpr) && |
1936 | 233k | CurrentScopeOpener.MatchingParen) { |
1937 | 0 | int NecessarySpaceInLine = |
1938 | 0 | getLengthToMatchingParen(CurrentScopeOpener, State.Stack) + |
1939 | 0 | CurrentScopeOpener.TotalLength - Current.TotalLength - 1; |
1940 | 0 | if (State.Column + Current.ColumnWidth + NecessarySpaceInLine <= |
1941 | 0 | Style.ColumnLimit) { |
1942 | 0 | CurrentState.BreakBeforeParameter = false; |
1943 | 0 | } |
1944 | 0 | } |
1945 | 233k | } |
1946 | | |
1947 | 478k | if (Current.is(tok::r_square)) { |
1948 | | // If this ends the array subscript expr, reset the corresponding value. |
1949 | 45.1k | const FormatToken *NextNonComment = Current.getNextNonComment(); |
1950 | 45.1k | if (NextNonComment && NextNonComment->isNot(tok::l_square)) |
1951 | 44.1k | CurrentState.StartOfArraySubscripts = 0; |
1952 | 45.1k | } |
1953 | 478k | } |
1954 | | |
1955 | 167 | void ContinuationIndenter::moveStateToNewBlock(LineState &State) { |
1956 | 167 | if (Style.LambdaBodyIndentation == FormatStyle::LBI_OuterScope && |
1957 | 167 | State.NextToken->is(TT_LambdaLBrace) && |
1958 | 167 | !State.Line->MightBeFunctionDecl) { |
1959 | 0 | State.Stack.back().NestedBlockIndent = State.FirstIndent; |
1960 | 0 | } |
1961 | 167 | unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent; |
1962 | | // ObjC block sometimes follow special indentation rules. |
1963 | 167 | unsigned NewIndent = |
1964 | 167 | NestedBlockIndent + (State.NextToken->is(TT_ObjCBlockLBrace) |
1965 | 167 | ? Style.ObjCBlockIndentWidth |
1966 | 167 | : Style.IndentWidth); |
1967 | 167 | State.Stack.push_back(ParenState(State.NextToken, NewIndent, |
1968 | 167 | State.Stack.back().LastSpace, |
1969 | 167 | /*AvoidBinPacking=*/true, |
1970 | 167 | /*NoLineBreak=*/false)); |
1971 | 167 | State.Stack.back().NestedBlockIndent = NestedBlockIndent; |
1972 | 167 | State.Stack.back().BreakBeforeParameter = true; |
1973 | 167 | } |
1974 | | |
1975 | | static unsigned getLastLineEndColumn(StringRef Text, unsigned StartColumn, |
1976 | | unsigned TabWidth, |
1977 | 0 | encoding::Encoding Encoding) { |
1978 | 0 | size_t LastNewlinePos = Text.find_last_of("\n"); |
1979 | 0 | if (LastNewlinePos == StringRef::npos) { |
1980 | 0 | return StartColumn + |
1981 | 0 | encoding::columnWidthWithTabs(Text, StartColumn, TabWidth, Encoding); |
1982 | 0 | } else { |
1983 | 0 | return encoding::columnWidthWithTabs(Text.substr(LastNewlinePos), |
1984 | 0 | /*StartColumn=*/0, TabWidth, Encoding); |
1985 | 0 | } |
1986 | 0 | } |
1987 | | |
1988 | | unsigned ContinuationIndenter::reformatRawStringLiteral( |
1989 | | const FormatToken &Current, LineState &State, |
1990 | 0 | const FormatStyle &RawStringStyle, bool DryRun, bool Newline) { |
1991 | 0 | unsigned StartColumn = State.Column - Current.ColumnWidth; |
1992 | 0 | StringRef OldDelimiter = *getRawStringDelimiter(Current.TokenText); |
1993 | 0 | StringRef NewDelimiter = |
1994 | 0 | getCanonicalRawStringDelimiter(Style, RawStringStyle.Language); |
1995 | 0 | if (NewDelimiter.empty()) |
1996 | 0 | NewDelimiter = OldDelimiter; |
1997 | | // The text of a raw string is between the leading 'R"delimiter(' and the |
1998 | | // trailing 'delimiter)"'. |
1999 | 0 | unsigned OldPrefixSize = 3 + OldDelimiter.size(); |
2000 | 0 | unsigned OldSuffixSize = 2 + OldDelimiter.size(); |
2001 | | // We create a virtual text environment which expects a null-terminated |
2002 | | // string, so we cannot use StringRef. |
2003 | 0 | std::string RawText = std::string( |
2004 | 0 | Current.TokenText.substr(OldPrefixSize).drop_back(OldSuffixSize)); |
2005 | 0 | if (NewDelimiter != OldDelimiter) { |
2006 | | // Don't update to the canonical delimiter 'deli' if ')deli"' occurs in the |
2007 | | // raw string. |
2008 | 0 | std::string CanonicalDelimiterSuffix = (")" + NewDelimiter + "\"").str(); |
2009 | 0 | if (StringRef(RawText).contains(CanonicalDelimiterSuffix)) |
2010 | 0 | NewDelimiter = OldDelimiter; |
2011 | 0 | } |
2012 | |
|
2013 | 0 | unsigned NewPrefixSize = 3 + NewDelimiter.size(); |
2014 | 0 | unsigned NewSuffixSize = 2 + NewDelimiter.size(); |
2015 | | |
2016 | | // The first start column is the column the raw text starts after formatting. |
2017 | 0 | unsigned FirstStartColumn = StartColumn + NewPrefixSize; |
2018 | | |
2019 | | // The next start column is the intended indentation a line break inside |
2020 | | // the raw string at level 0. It is determined by the following rules: |
2021 | | // - if the content starts on newline, it is one level more than the current |
2022 | | // indent, and |
2023 | | // - if the content does not start on a newline, it is the first start |
2024 | | // column. |
2025 | | // These rules have the advantage that the formatted content both does not |
2026 | | // violate the rectangle rule and visually flows within the surrounding |
2027 | | // source. |
2028 | 0 | bool ContentStartsOnNewline = Current.TokenText[OldPrefixSize] == '\n'; |
2029 | | // If this token is the last parameter (checked by looking if it's followed by |
2030 | | // `)` and is not on a newline, the base the indent off the line's nested |
2031 | | // block indent. Otherwise, base the indent off the arguments indent, so we |
2032 | | // can achieve: |
2033 | | // |
2034 | | // fffffffffff(1, 2, 3, R"pb( |
2035 | | // key1: 1 # |
2036 | | // key2: 2)pb"); |
2037 | | // |
2038 | | // fffffffffff(1, 2, 3, |
2039 | | // R"pb( |
2040 | | // key1: 1 # |
2041 | | // key2: 2 |
2042 | | // )pb"); |
2043 | | // |
2044 | | // fffffffffff(1, 2, 3, |
2045 | | // R"pb( |
2046 | | // key1: 1 # |
2047 | | // key2: 2 |
2048 | | // )pb", |
2049 | | // 5); |
2050 | 0 | unsigned CurrentIndent = |
2051 | 0 | (!Newline && Current.Next && Current.Next->is(tok::r_paren)) |
2052 | 0 | ? State.Stack.back().NestedBlockIndent |
2053 | 0 | : State.Stack.back().Indent; |
2054 | 0 | unsigned NextStartColumn = ContentStartsOnNewline |
2055 | 0 | ? CurrentIndent + Style.IndentWidth |
2056 | 0 | : FirstStartColumn; |
2057 | | |
2058 | | // The last start column is the column the raw string suffix starts if it is |
2059 | | // put on a newline. |
2060 | | // The last start column is the intended indentation of the raw string postfix |
2061 | | // if it is put on a newline. It is determined by the following rules: |
2062 | | // - if the raw string prefix starts on a newline, it is the column where |
2063 | | // that raw string prefix starts, and |
2064 | | // - if the raw string prefix does not start on a newline, it is the current |
2065 | | // indent. |
2066 | 0 | unsigned LastStartColumn = |
2067 | 0 | Current.NewlinesBefore ? FirstStartColumn - NewPrefixSize : CurrentIndent; |
2068 | |
|
2069 | 0 | std::pair<tooling::Replacements, unsigned> Fixes = internal::reformat( |
2070 | 0 | RawStringStyle, RawText, {tooling::Range(0, RawText.size())}, |
2071 | 0 | FirstStartColumn, NextStartColumn, LastStartColumn, "<stdin>", |
2072 | 0 | /*Status=*/nullptr); |
2073 | |
|
2074 | 0 | auto NewCode = applyAllReplacements(RawText, Fixes.first); |
2075 | 0 | tooling::Replacements NoFixes; |
2076 | 0 | if (!NewCode) |
2077 | 0 | return addMultilineToken(Current, State); |
2078 | 0 | if (!DryRun) { |
2079 | 0 | if (NewDelimiter != OldDelimiter) { |
2080 | | // In 'R"delimiter(...', the delimiter starts 2 characters after the start |
2081 | | // of the token. |
2082 | 0 | SourceLocation PrefixDelimiterStart = |
2083 | 0 | Current.Tok.getLocation().getLocWithOffset(2); |
2084 | 0 | auto PrefixErr = Whitespaces.addReplacement(tooling::Replacement( |
2085 | 0 | SourceMgr, PrefixDelimiterStart, OldDelimiter.size(), NewDelimiter)); |
2086 | 0 | if (PrefixErr) { |
2087 | 0 | llvm::errs() |
2088 | 0 | << "Failed to update the prefix delimiter of a raw string: " |
2089 | 0 | << llvm::toString(std::move(PrefixErr)) << "\n"; |
2090 | 0 | } |
2091 | | // In 'R"delimiter(...)delimiter"', the suffix delimiter starts at |
2092 | | // position length - 1 - |delimiter|. |
2093 | 0 | SourceLocation SuffixDelimiterStart = |
2094 | 0 | Current.Tok.getLocation().getLocWithOffset(Current.TokenText.size() - |
2095 | 0 | 1 - OldDelimiter.size()); |
2096 | 0 | auto SuffixErr = Whitespaces.addReplacement(tooling::Replacement( |
2097 | 0 | SourceMgr, SuffixDelimiterStart, OldDelimiter.size(), NewDelimiter)); |
2098 | 0 | if (SuffixErr) { |
2099 | 0 | llvm::errs() |
2100 | 0 | << "Failed to update the suffix delimiter of a raw string: " |
2101 | 0 | << llvm::toString(std::move(SuffixErr)) << "\n"; |
2102 | 0 | } |
2103 | 0 | } |
2104 | 0 | SourceLocation OriginLoc = |
2105 | 0 | Current.Tok.getLocation().getLocWithOffset(OldPrefixSize); |
2106 | 0 | for (const tooling::Replacement &Fix : Fixes.first) { |
2107 | 0 | auto Err = Whitespaces.addReplacement(tooling::Replacement( |
2108 | 0 | SourceMgr, OriginLoc.getLocWithOffset(Fix.getOffset()), |
2109 | 0 | Fix.getLength(), Fix.getReplacementText())); |
2110 | 0 | if (Err) { |
2111 | 0 | llvm::errs() << "Failed to reformat raw string: " |
2112 | 0 | << llvm::toString(std::move(Err)) << "\n"; |
2113 | 0 | } |
2114 | 0 | } |
2115 | 0 | } |
2116 | 0 | unsigned RawLastLineEndColumn = getLastLineEndColumn( |
2117 | 0 | *NewCode, FirstStartColumn, Style.TabWidth, Encoding); |
2118 | 0 | State.Column = RawLastLineEndColumn + NewSuffixSize; |
2119 | | // Since we're updating the column to after the raw string literal here, we |
2120 | | // have to manually add the penalty for the prefix R"delim( over the column |
2121 | | // limit. |
2122 | 0 | unsigned PrefixExcessCharacters = |
2123 | 0 | StartColumn + NewPrefixSize > Style.ColumnLimit |
2124 | 0 | ? StartColumn + NewPrefixSize - Style.ColumnLimit |
2125 | 0 | : 0; |
2126 | 0 | bool IsMultiline = |
2127 | 0 | ContentStartsOnNewline || (NewCode->find('\n') != std::string::npos); |
2128 | 0 | if (IsMultiline) { |
2129 | | // Break before further function parameters on all levels. |
2130 | 0 | for (ParenState &Paren : State.Stack) |
2131 | 0 | Paren.BreakBeforeParameter = true; |
2132 | 0 | } |
2133 | 0 | return Fixes.second + PrefixExcessCharacters * Style.PenaltyExcessCharacter; |
2134 | 0 | } |
2135 | | |
2136 | | unsigned ContinuationIndenter::addMultilineToken(const FormatToken &Current, |
2137 | 7.78k | LineState &State) { |
2138 | | // Break before further function parameters on all levels. |
2139 | 7.78k | for (ParenState &Paren : State.Stack) |
2140 | 64.8k | Paren.BreakBeforeParameter = true; |
2141 | | |
2142 | 7.78k | unsigned ColumnsUsed = State.Column; |
2143 | | // We can only affect layout of the first and the last line, so the penalty |
2144 | | // for all other lines is constant, and we ignore it. |
2145 | 7.78k | State.Column = Current.LastLineColumnWidth; |
2146 | | |
2147 | 7.78k | if (ColumnsUsed > getColumnLimit(State)) |
2148 | 2.72k | return Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit(State)); |
2149 | 5.06k | return 0; |
2150 | 7.78k | } |
2151 | | |
2152 | | unsigned ContinuationIndenter::handleEndOfLine(const FormatToken &Current, |
2153 | | LineState &State, bool DryRun, |
2154 | 7.04M | bool AllowBreak, bool Newline) { |
2155 | 7.04M | unsigned Penalty = 0; |
2156 | | // Compute the raw string style to use in case this is a raw string literal |
2157 | | // that can be reformatted. |
2158 | 7.04M | auto RawStringStyle = getRawStringStyle(Current, State); |
2159 | 7.04M | if (RawStringStyle && !Current.Finalized) { |
2160 | 0 | Penalty = reformatRawStringLiteral(Current, State, *RawStringStyle, DryRun, |
2161 | 0 | Newline); |
2162 | 7.04M | } else if (Current.IsMultiline && Current.isNot(TT_BlockComment)) { |
2163 | | // Don't break multi-line tokens other than block comments and raw string |
2164 | | // literals. Instead, just update the state. |
2165 | 7.78k | Penalty = addMultilineToken(Current, State); |
2166 | 7.03M | } else if (State.Line->Type != LT_ImportStatement) { |
2167 | | // We generally don't break import statements. |
2168 | 7.03M | LineState OriginalState = State; |
2169 | | |
2170 | | // Whether we force the reflowing algorithm to stay strictly within the |
2171 | | // column limit. |
2172 | 7.03M | bool Strict = false; |
2173 | | // Whether the first non-strict attempt at reflowing did intentionally |
2174 | | // exceed the column limit. |
2175 | 7.03M | bool Exceeded = false; |
2176 | 7.03M | std::tie(Penalty, Exceeded) = breakProtrudingToken( |
2177 | 7.03M | Current, State, AllowBreak, /*DryRun=*/true, Strict); |
2178 | 7.03M | if (Exceeded) { |
2179 | | // If non-strict reflowing exceeds the column limit, try whether strict |
2180 | | // reflowing leads to an overall lower penalty. |
2181 | 0 | LineState StrictState = OriginalState; |
2182 | 0 | unsigned StrictPenalty = |
2183 | 0 | breakProtrudingToken(Current, StrictState, AllowBreak, |
2184 | 0 | /*DryRun=*/true, /*Strict=*/true) |
2185 | 0 | .first; |
2186 | 0 | Strict = StrictPenalty <= Penalty; |
2187 | 0 | if (Strict) { |
2188 | 0 | Penalty = StrictPenalty; |
2189 | 0 | State = StrictState; |
2190 | 0 | } |
2191 | 0 | } |
2192 | 7.03M | if (!DryRun) { |
2193 | | // If we're not in dry-run mode, apply the changes with the decision on |
2194 | | // strictness made above. |
2195 | 2.24M | breakProtrudingToken(Current, OriginalState, AllowBreak, /*DryRun=*/false, |
2196 | 2.24M | Strict); |
2197 | 2.24M | } |
2198 | 7.03M | } |
2199 | 7.04M | if (State.Column > getColumnLimit(State)) { |
2200 | 2.83M | unsigned ExcessCharacters = State.Column - getColumnLimit(State); |
2201 | 2.83M | Penalty += Style.PenaltyExcessCharacter * ExcessCharacters; |
2202 | 2.83M | } |
2203 | 7.04M | return Penalty; |
2204 | 7.04M | } |
2205 | | |
2206 | | // Returns the enclosing function name of a token, or the empty string if not |
2207 | | // found. |
2208 | 0 | static StringRef getEnclosingFunctionName(const FormatToken &Current) { |
2209 | | // Look for: 'function(' or 'function<templates>(' before Current. |
2210 | 0 | auto Tok = Current.getPreviousNonComment(); |
2211 | 0 | if (!Tok || Tok->isNot(tok::l_paren)) |
2212 | 0 | return ""; |
2213 | 0 | Tok = Tok->getPreviousNonComment(); |
2214 | 0 | if (!Tok) |
2215 | 0 | return ""; |
2216 | 0 | if (Tok->is(TT_TemplateCloser)) { |
2217 | 0 | Tok = Tok->MatchingParen; |
2218 | 0 | if (Tok) |
2219 | 0 | Tok = Tok->getPreviousNonComment(); |
2220 | 0 | } |
2221 | 0 | if (!Tok || Tok->isNot(tok::identifier)) |
2222 | 0 | return ""; |
2223 | 0 | return Tok->TokenText; |
2224 | 0 | } |
2225 | | |
2226 | | std::optional<FormatStyle> |
2227 | | ContinuationIndenter::getRawStringStyle(const FormatToken &Current, |
2228 | 7.04M | const LineState &State) { |
2229 | 7.04M | if (!Current.isStringLiteral()) |
2230 | 6.94M | return std::nullopt; |
2231 | 106k | auto Delimiter = getRawStringDelimiter(Current.TokenText); |
2232 | 106k | if (!Delimiter) |
2233 | 106k | return std::nullopt; |
2234 | 0 | auto RawStringStyle = RawStringFormats.getDelimiterStyle(*Delimiter); |
2235 | 0 | if (!RawStringStyle && Delimiter->empty()) { |
2236 | 0 | RawStringStyle = RawStringFormats.getEnclosingFunctionStyle( |
2237 | 0 | getEnclosingFunctionName(Current)); |
2238 | 0 | } |
2239 | 0 | if (!RawStringStyle) |
2240 | 0 | return std::nullopt; |
2241 | 0 | RawStringStyle->ColumnLimit = getColumnLimit(State); |
2242 | 0 | return RawStringStyle; |
2243 | 0 | } |
2244 | | |
2245 | | std::unique_ptr<BreakableToken> |
2246 | | ContinuationIndenter::createBreakableToken(const FormatToken &Current, |
2247 | 9.28M | LineState &State, bool AllowBreak) { |
2248 | 9.28M | unsigned StartColumn = State.Column - Current.ColumnWidth; |
2249 | 9.28M | if (Current.isStringLiteral()) { |
2250 | | // Strings in JSON cannot be broken. Breaking strings in JavaScript is |
2251 | | // disabled for now. |
2252 | 124k | if (Style.isJson() || Style.isJavaScript() || !Style.BreakStringLiterals || |
2253 | 124k | !AllowBreak) { |
2254 | 567 | return nullptr; |
2255 | 567 | } |
2256 | | |
2257 | | // Don't break string literals inside preprocessor directives (except for |
2258 | | // #define directives, as their contents are stored in separate lines and |
2259 | | // are not affected by this check). |
2260 | | // This way we avoid breaking code with line directives and unknown |
2261 | | // preprocessor directives that contain long string literals. |
2262 | 124k | if (State.Line->Type == LT_PreprocessorDirective) |
2263 | 8.44k | return nullptr; |
2264 | | // Exempts unterminated string literals from line breaking. The user will |
2265 | | // likely want to terminate the string before any line breaking is done. |
2266 | 115k | if (Current.IsUnterminatedLiteral) |
2267 | 26.7k | return nullptr; |
2268 | | // Don't break string literals inside Objective-C array literals (doing so |
2269 | | // raises the warning -Wobjc-string-concatenation). |
2270 | 88.7k | if (State.Stack.back().IsInsideObjCArrayLiteral) |
2271 | 0 | return nullptr; |
2272 | | |
2273 | | // The "DPI"/"DPI-C" in SystemVerilog direct programming interface |
2274 | | // imports/exports cannot be split, e.g. |
2275 | | // `import "DPI" function foo();` |
2276 | | // FIXME: make this use same infra as C++ import checks |
2277 | 88.7k | if (Style.isVerilog() && Current.Previous && |
2278 | 88.7k | Current.Previous->isOneOf(tok::kw_export, Keywords.kw_import)) { |
2279 | 0 | return nullptr; |
2280 | 0 | } |
2281 | 88.7k | StringRef Text = Current.TokenText; |
2282 | | |
2283 | | // We need this to address the case where there is an unbreakable tail only |
2284 | | // if certain other formatting decisions have been taken. The |
2285 | | // UnbreakableTailLength of Current is an overapproximation in that case and |
2286 | | // we need to be correct here. |
2287 | 88.7k | unsigned UnbreakableTailLength = (State.NextToken && canBreak(State)) |
2288 | 88.7k | ? 0 |
2289 | 88.7k | : Current.UnbreakableTailLength; |
2290 | | |
2291 | 88.7k | if (Style.isVerilog() || Style.Language == FormatStyle::LK_Java || |
2292 | 88.7k | Style.isJavaScript() || Style.isCSharp()) { |
2293 | 0 | BreakableStringLiteralUsingOperators::QuoteStyleType QuoteStyle; |
2294 | 0 | if (Style.isJavaScript() && Text.starts_with("'") && |
2295 | 0 | Text.ends_with("'")) { |
2296 | 0 | QuoteStyle = BreakableStringLiteralUsingOperators::SingleQuotes; |
2297 | 0 | } else if (Style.isCSharp() && Text.starts_with("@\"") && |
2298 | 0 | Text.ends_with("\"")) { |
2299 | 0 | QuoteStyle = BreakableStringLiteralUsingOperators::AtDoubleQuotes; |
2300 | 0 | } else if (Text.starts_with("\"") && Text.ends_with("\"")) { |
2301 | 0 | QuoteStyle = BreakableStringLiteralUsingOperators::DoubleQuotes; |
2302 | 0 | } else { |
2303 | 0 | return nullptr; |
2304 | 0 | } |
2305 | 0 | return std::make_unique<BreakableStringLiteralUsingOperators>( |
2306 | 0 | Current, QuoteStyle, |
2307 | 0 | /*UnindentPlus=*/shouldUnindentNextOperator(Current), StartColumn, |
2308 | 0 | UnbreakableTailLength, State.Line->InPPDirective, Encoding, Style); |
2309 | 0 | } |
2310 | | |
2311 | 88.7k | StringRef Prefix; |
2312 | 88.7k | StringRef Postfix; |
2313 | | // FIXME: Handle whitespace between '_T', '(', '"..."', and ')'. |
2314 | | // FIXME: Store Prefix and Suffix (or PrefixLength and SuffixLength to |
2315 | | // reduce the overhead) for each FormatToken, which is a string, so that we |
2316 | | // don't run multiple checks here on the hot path. |
2317 | 88.7k | if ((Text.ends_with(Postfix = "\"") && |
2318 | 88.7k | (Text.starts_with(Prefix = "@\"") || Text.starts_with(Prefix = "\"") || |
2319 | 88.7k | Text.starts_with(Prefix = "u\"") || |
2320 | 88.7k | Text.starts_with(Prefix = "U\"") || |
2321 | 88.7k | Text.starts_with(Prefix = "u8\"") || |
2322 | 88.7k | Text.starts_with(Prefix = "L\""))) || |
2323 | 88.7k | (Text.starts_with(Prefix = "_T(\"") && |
2324 | 88.7k | Text.ends_with(Postfix = "\")"))) { |
2325 | 88.7k | return std::make_unique<BreakableStringLiteral>( |
2326 | 88.7k | Current, StartColumn, Prefix, Postfix, UnbreakableTailLength, |
2327 | 88.7k | State.Line->InPPDirective, Encoding, Style); |
2328 | 88.7k | } |
2329 | 9.15M | } else if (Current.is(TT_BlockComment)) { |
2330 | 15.0k | if (!Style.ReflowComments || |
2331 | | // If a comment token switches formatting, like |
2332 | | // /* clang-format on */, we don't want to break it further, |
2333 | | // but we may still want to adjust its indentation. |
2334 | 15.0k | switchesFormatting(Current)) { |
2335 | 0 | return nullptr; |
2336 | 0 | } |
2337 | 15.0k | return std::make_unique<BreakableBlockComment>( |
2338 | 15.0k | Current, StartColumn, Current.OriginalColumn, !Current.Previous, |
2339 | 15.0k | State.Line->InPPDirective, Encoding, Style, Whitespaces.useCRLF()); |
2340 | 9.14M | } else if (Current.is(TT_LineComment) && |
2341 | 9.14M | (!Current.Previous || |
2342 | 21.8k | Current.Previous->isNot(TT_ImplicitStringLiteral))) { |
2343 | 21.7k | bool RegularComments = [&]() { |
2344 | 45.5k | for (const FormatToken *T = &Current; T && T->is(TT_LineComment); |
2345 | 23.8k | T = T->Next) { |
2346 | 23.8k | if (!(T->TokenText.starts_with("//") || T->TokenText.starts_with("#"))) |
2347 | 0 | return false; |
2348 | 23.8k | } |
2349 | 21.7k | return true; |
2350 | 21.7k | }(); |
2351 | 21.7k | if (!Style.ReflowComments || |
2352 | 21.7k | CommentPragmasRegex.match(Current.TokenText.substr(2)) || |
2353 | 21.7k | switchesFormatting(Current) || !RegularComments) { |
2354 | 0 | return nullptr; |
2355 | 0 | } |
2356 | 21.7k | return std::make_unique<BreakableLineCommentSection>( |
2357 | 21.7k | Current, StartColumn, /*InPPDirective=*/false, Encoding, Style); |
2358 | 21.7k | } |
2359 | 9.12M | return nullptr; |
2360 | 9.28M | } |
2361 | | |
2362 | | std::pair<unsigned, bool> |
2363 | | ContinuationIndenter::breakProtrudingToken(const FormatToken &Current, |
2364 | | LineState &State, bool AllowBreak, |
2365 | 9.28M | bool DryRun, bool Strict) { |
2366 | 9.28M | std::unique_ptr<const BreakableToken> Token = |
2367 | 9.28M | createBreakableToken(Current, State, AllowBreak); |
2368 | 9.28M | if (!Token) |
2369 | 9.15M | return {0, false}; |
2370 | 125k | assert(Token->getLineCount() > 0); |
2371 | 0 | unsigned ColumnLimit = getColumnLimit(State); |
2372 | 125k | if (Current.is(TT_LineComment)) { |
2373 | | // We don't insert backslashes when breaking line comments. |
2374 | 21.7k | ColumnLimit = Style.ColumnLimit; |
2375 | 21.7k | } |
2376 | 125k | if (ColumnLimit == 0) { |
2377 | | // To make the rest of the function easier set the column limit to the |
2378 | | // maximum, if there should be no limit. |
2379 | 0 | ColumnLimit = std::numeric_limits<decltype(ColumnLimit)>::max(); |
2380 | 0 | } |
2381 | 125k | if (Current.UnbreakableTailLength >= ColumnLimit) |
2382 | 442 | return {0, false}; |
2383 | | // ColumnWidth was already accounted into State.Column before calling |
2384 | | // breakProtrudingToken. |
2385 | 125k | unsigned StartColumn = State.Column - Current.ColumnWidth; |
2386 | 125k | unsigned NewBreakPenalty = Current.isStringLiteral() |
2387 | 125k | ? Style.PenaltyBreakString |
2388 | 125k | : Style.PenaltyBreakComment; |
2389 | | // Stores whether we intentionally decide to let a line exceed the column |
2390 | | // limit. |
2391 | 125k | bool Exceeded = false; |
2392 | | // Stores whether we introduce a break anywhere in the token. |
2393 | 125k | bool BreakInserted = Token->introducesBreakBeforeToken(); |
2394 | | // Store whether we inserted a new line break at the end of the previous |
2395 | | // logical line. |
2396 | 125k | bool NewBreakBefore = false; |
2397 | | // We use a conservative reflowing strategy. Reflow starts after a line is |
2398 | | // broken or the corresponding whitespace compressed. Reflow ends as soon as a |
2399 | | // line that doesn't get reflown with the previous line is reached. |
2400 | 125k | bool Reflow = false; |
2401 | | // Keep track of where we are in the token: |
2402 | | // Where we are in the content of the current logical line. |
2403 | 125k | unsigned TailOffset = 0; |
2404 | | // The column number we're currently at. |
2405 | 125k | unsigned ContentStartColumn = |
2406 | 125k | Token->getContentStartColumn(0, /*Break=*/false); |
2407 | | // The number of columns left in the current logical line after TailOffset. |
2408 | 125k | unsigned RemainingTokenColumns = |
2409 | 125k | Token->getRemainingLength(0, TailOffset, ContentStartColumn); |
2410 | | // Adapt the start of the token, for example indent. |
2411 | 125k | if (!DryRun) |
2412 | 23.4k | Token->adaptStartOfLine(0, Whitespaces); |
2413 | | |
2414 | 125k | unsigned ContentIndent = 0; |
2415 | 125k | unsigned Penalty = 0; |
2416 | 125k | LLVM_DEBUG(llvm::dbgs() << "Breaking protruding token at column " |
2417 | 125k | << StartColumn << ".\n"); |
2418 | 125k | for (unsigned LineIndex = 0, EndIndex = Token->getLineCount(); |
2419 | 558k | LineIndex != EndIndex; ++LineIndex) { |
2420 | 433k | LLVM_DEBUG(llvm::dbgs() |
2421 | 433k | << " Line: " << LineIndex << " (Reflow: " << Reflow << ")\n"); |
2422 | 433k | NewBreakBefore = false; |
2423 | | // If we did reflow the previous line, we'll try reflowing again. Otherwise |
2424 | | // we'll start reflowing if the current line is broken or whitespace is |
2425 | | // compressed. |
2426 | 433k | bool TryReflow = Reflow; |
2427 | | // Break the current token until we can fit the rest of the line. |
2428 | 1.03M | while (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { |
2429 | 616k | LLVM_DEBUG(llvm::dbgs() << " Over limit, need: " |
2430 | 616k | << (ContentStartColumn + RemainingTokenColumns) |
2431 | 616k | << ", space: " << ColumnLimit |
2432 | 616k | << ", reflown prefix: " << ContentStartColumn |
2433 | 616k | << ", offset in line: " << TailOffset << "\n"); |
2434 | | // If the current token doesn't fit, find the latest possible split in the |
2435 | | // current line so that breaking at it will be under the column limit. |
2436 | | // FIXME: Use the earliest possible split while reflowing to correctly |
2437 | | // compress whitespace within a line. |
2438 | 616k | BreakableToken::Split Split = |
2439 | 616k | Token->getSplit(LineIndex, TailOffset, ColumnLimit, |
2440 | 616k | ContentStartColumn, CommentPragmasRegex); |
2441 | 616k | if (Split.first == StringRef::npos) { |
2442 | | // No break opportunity - update the penalty and continue with the next |
2443 | | // logical line. |
2444 | 18.7k | if (LineIndex < EndIndex - 1) { |
2445 | | // The last line's penalty is handled in addNextStateToQueue() or when |
2446 | | // calling replaceWhitespaceAfterLastLine below. |
2447 | 627 | Penalty += Style.PenaltyExcessCharacter * |
2448 | 627 | (ContentStartColumn + RemainingTokenColumns - ColumnLimit); |
2449 | 627 | } |
2450 | 18.7k | LLVM_DEBUG(llvm::dbgs() << " No break opportunity.\n"); |
2451 | 18.7k | break; |
2452 | 18.7k | } |
2453 | 597k | assert(Split.first != 0); |
2454 | | |
2455 | 597k | if (Token->supportsReflow()) { |
2456 | | // Check whether the next natural split point after the current one can |
2457 | | // still fit the line, either because we can compress away whitespace, |
2458 | | // or because the penalty the excess characters introduce is lower than |
2459 | | // the break penalty. |
2460 | | // We only do this for tokens that support reflowing, and thus allow us |
2461 | | // to change the whitespace arbitrarily (e.g. comments). |
2462 | | // Other tokens, like string literals, can be broken on arbitrary |
2463 | | // positions. |
2464 | | |
2465 | | // First, compute the columns from TailOffset to the next possible split |
2466 | | // position. |
2467 | | // For example: |
2468 | | // ColumnLimit: | |
2469 | | // // Some text that breaks |
2470 | | // ^ tail offset |
2471 | | // ^-- split |
2472 | | // ^-------- to split columns |
2473 | | // ^--- next split |
2474 | | // ^--------------- to next split columns |
2475 | 29.1k | unsigned ToSplitColumns = Token->getRangeLength( |
2476 | 29.1k | LineIndex, TailOffset, Split.first, ContentStartColumn); |
2477 | 29.1k | LLVM_DEBUG(llvm::dbgs() << " ToSplit: " << ToSplitColumns << "\n"); |
2478 | | |
2479 | 29.1k | BreakableToken::Split NextSplit = Token->getSplit( |
2480 | 29.1k | LineIndex, TailOffset + Split.first + Split.second, ColumnLimit, |
2481 | 29.1k | ContentStartColumn + ToSplitColumns + 1, CommentPragmasRegex); |
2482 | | // Compute the columns necessary to fit the next non-breakable sequence |
2483 | | // into the current line. |
2484 | 29.1k | unsigned ToNextSplitColumns = 0; |
2485 | 29.1k | if (NextSplit.first == StringRef::npos) { |
2486 | 16.1k | ToNextSplitColumns = Token->getRemainingLength(LineIndex, TailOffset, |
2487 | 16.1k | ContentStartColumn); |
2488 | 16.1k | } else { |
2489 | 12.9k | ToNextSplitColumns = Token->getRangeLength( |
2490 | 12.9k | LineIndex, TailOffset, |
2491 | 12.9k | Split.first + Split.second + NextSplit.first, ContentStartColumn); |
2492 | 12.9k | } |
2493 | | // Compress the whitespace between the break and the start of the next |
2494 | | // unbreakable sequence. |
2495 | 29.1k | ToNextSplitColumns = |
2496 | 29.1k | Token->getLengthAfterCompression(ToNextSplitColumns, Split); |
2497 | 29.1k | LLVM_DEBUG(llvm::dbgs() |
2498 | 29.1k | << " ContentStartColumn: " << ContentStartColumn << "\n"); |
2499 | 29.1k | LLVM_DEBUG(llvm::dbgs() |
2500 | 29.1k | << " ToNextSplit: " << ToNextSplitColumns << "\n"); |
2501 | | // If the whitespace compression makes us fit, continue on the current |
2502 | | // line. |
2503 | 29.1k | bool ContinueOnLine = |
2504 | 29.1k | ContentStartColumn + ToNextSplitColumns <= ColumnLimit; |
2505 | 29.1k | unsigned ExcessCharactersPenalty = 0; |
2506 | 29.1k | if (!ContinueOnLine && !Strict) { |
2507 | | // Similarly, if the excess characters' penalty is lower than the |
2508 | | // penalty of introducing a new break, continue on the current line. |
2509 | 28.9k | ExcessCharactersPenalty = |
2510 | 28.9k | (ContentStartColumn + ToNextSplitColumns - ColumnLimit) * |
2511 | 28.9k | Style.PenaltyExcessCharacter; |
2512 | 28.9k | LLVM_DEBUG(llvm::dbgs() |
2513 | 28.9k | << " Penalty excess: " << ExcessCharactersPenalty |
2514 | 28.9k | << "\n break : " << NewBreakPenalty << "\n"); |
2515 | 28.9k | if (ExcessCharactersPenalty < NewBreakPenalty) { |
2516 | 0 | Exceeded = true; |
2517 | 0 | ContinueOnLine = true; |
2518 | 0 | } |
2519 | 28.9k | } |
2520 | 29.1k | if (ContinueOnLine) { |
2521 | 180 | LLVM_DEBUG(llvm::dbgs() << " Continuing on line...\n"); |
2522 | | // The current line fits after compressing the whitespace - reflow |
2523 | | // the next line into it if possible. |
2524 | 180 | TryReflow = true; |
2525 | 180 | if (!DryRun) { |
2526 | 83 | Token->compressWhitespace(LineIndex, TailOffset, Split, |
2527 | 83 | Whitespaces); |
2528 | 83 | } |
2529 | | // When we continue on the same line, leave one space between content. |
2530 | 180 | ContentStartColumn += ToSplitColumns + 1; |
2531 | 180 | Penalty += ExcessCharactersPenalty; |
2532 | 180 | TailOffset += Split.first + Split.second; |
2533 | 180 | RemainingTokenColumns = Token->getRemainingLength( |
2534 | 180 | LineIndex, TailOffset, ContentStartColumn); |
2535 | 180 | continue; |
2536 | 180 | } |
2537 | 29.1k | } |
2538 | 597k | LLVM_DEBUG(llvm::dbgs() << " Breaking...\n"); |
2539 | | // Update the ContentIndent only if the current line was not reflown with |
2540 | | // the previous line, since in that case the previous line should still |
2541 | | // determine the ContentIndent. Also never intent the last line. |
2542 | 597k | if (!Reflow) |
2543 | 595k | ContentIndent = Token->getContentIndent(LineIndex); |
2544 | 597k | LLVM_DEBUG(llvm::dbgs() |
2545 | 597k | << " ContentIndent: " << ContentIndent << "\n"); |
2546 | 597k | ContentStartColumn = ContentIndent + Token->getContentStartColumn( |
2547 | 597k | LineIndex, /*Break=*/true); |
2548 | | |
2549 | 597k | unsigned NewRemainingTokenColumns = Token->getRemainingLength( |
2550 | 597k | LineIndex, TailOffset + Split.first + Split.second, |
2551 | 597k | ContentStartColumn); |
2552 | 597k | if (NewRemainingTokenColumns == 0) { |
2553 | | // No content to indent. |
2554 | 0 | ContentIndent = 0; |
2555 | 0 | ContentStartColumn = |
2556 | 0 | Token->getContentStartColumn(LineIndex, /*Break=*/true); |
2557 | 0 | NewRemainingTokenColumns = Token->getRemainingLength( |
2558 | 0 | LineIndex, TailOffset + Split.first + Split.second, |
2559 | 0 | ContentStartColumn); |
2560 | 0 | } |
2561 | | |
2562 | | // When breaking before a tab character, it may be moved by a few columns, |
2563 | | // but will still be expanded to the next tab stop, so we don't save any |
2564 | | // columns. |
2565 | 597k | if (NewRemainingTokenColumns >= RemainingTokenColumns) { |
2566 | | // FIXME: Do we need to adjust the penalty? |
2567 | 218 | break; |
2568 | 218 | } |
2569 | | |
2570 | 597k | LLVM_DEBUG(llvm::dbgs() << " Breaking at: " << TailOffset + Split.first |
2571 | 597k | << ", " << Split.second << "\n"); |
2572 | 597k | if (!DryRun) { |
2573 | 83.1k | Token->insertBreak(LineIndex, TailOffset, Split, ContentIndent, |
2574 | 83.1k | Whitespaces); |
2575 | 83.1k | } |
2576 | | |
2577 | 597k | Penalty += NewBreakPenalty; |
2578 | 597k | TailOffset += Split.first + Split.second; |
2579 | 597k | RemainingTokenColumns = NewRemainingTokenColumns; |
2580 | 597k | BreakInserted = true; |
2581 | 597k | NewBreakBefore = true; |
2582 | 597k | } |
2583 | | // In case there's another line, prepare the state for the start of the next |
2584 | | // line. |
2585 | 433k | if (LineIndex + 1 != EndIndex) { |
2586 | 308k | unsigned NextLineIndex = LineIndex + 1; |
2587 | 308k | if (NewBreakBefore) { |
2588 | | // After breaking a line, try to reflow the next line into the current |
2589 | | // one once RemainingTokenColumns fits. |
2590 | 3.55k | TryReflow = true; |
2591 | 3.55k | } |
2592 | 308k | if (TryReflow) { |
2593 | | // We decided that we want to try reflowing the next line into the |
2594 | | // current one. |
2595 | | // We will now adjust the state as if the reflow is successful (in |
2596 | | // preparation for the next line), and see whether that works. If we |
2597 | | // decide that we cannot reflow, we will later reset the state to the |
2598 | | // start of the next line. |
2599 | 3.90k | Reflow = false; |
2600 | | // As we did not continue breaking the line, RemainingTokenColumns is |
2601 | | // known to fit after ContentStartColumn. Adapt ContentStartColumn to |
2602 | | // the position at which we want to format the next line if we do |
2603 | | // actually reflow. |
2604 | | // When we reflow, we need to add a space between the end of the current |
2605 | | // line and the next line's start column. |
2606 | 3.90k | ContentStartColumn += RemainingTokenColumns + 1; |
2607 | | // Get the split that we need to reflow next logical line into the end |
2608 | | // of the current one; the split will include any leading whitespace of |
2609 | | // the next logical line. |
2610 | 3.90k | BreakableToken::Split SplitBeforeNext = |
2611 | 3.90k | Token->getReflowSplit(NextLineIndex, CommentPragmasRegex); |
2612 | 3.90k | LLVM_DEBUG(llvm::dbgs() |
2613 | 3.90k | << " Size of reflown text: " << ContentStartColumn |
2614 | 3.90k | << "\n Potential reflow split: "); |
2615 | 3.90k | if (SplitBeforeNext.first != StringRef::npos) { |
2616 | 2.11k | LLVM_DEBUG(llvm::dbgs() << SplitBeforeNext.first << ", " |
2617 | 2.11k | << SplitBeforeNext.second << "\n"); |
2618 | 2.11k | TailOffset = SplitBeforeNext.first + SplitBeforeNext.second; |
2619 | | // If the rest of the next line fits into the current line below the |
2620 | | // column limit, we can safely reflow. |
2621 | 2.11k | RemainingTokenColumns = Token->getRemainingLength( |
2622 | 2.11k | NextLineIndex, TailOffset, ContentStartColumn); |
2623 | 2.11k | Reflow = true; |
2624 | 2.11k | if (ContentStartColumn + RemainingTokenColumns > ColumnLimit) { |
2625 | 1.91k | LLVM_DEBUG(llvm::dbgs() |
2626 | 1.91k | << " Over limit after reflow, need: " |
2627 | 1.91k | << (ContentStartColumn + RemainingTokenColumns) |
2628 | 1.91k | << ", space: " << ColumnLimit |
2629 | 1.91k | << ", reflown prefix: " << ContentStartColumn |
2630 | 1.91k | << ", offset in line: " << TailOffset << "\n"); |
2631 | | // If the whole next line does not fit, try to find a point in |
2632 | | // the next line at which we can break so that attaching the part |
2633 | | // of the next line to that break point onto the current line is |
2634 | | // below the column limit. |
2635 | 1.91k | BreakableToken::Split Split = |
2636 | 1.91k | Token->getSplit(NextLineIndex, TailOffset, ColumnLimit, |
2637 | 1.91k | ContentStartColumn, CommentPragmasRegex); |
2638 | 1.91k | if (Split.first == StringRef::npos) { |
2639 | 626 | LLVM_DEBUG(llvm::dbgs() << " Did not find later break\n"); |
2640 | 626 | Reflow = false; |
2641 | 1.28k | } else { |
2642 | | // Check whether the first split point gets us below the column |
2643 | | // limit. Note that we will execute this split below as part of |
2644 | | // the normal token breaking and reflow logic within the line. |
2645 | 1.28k | unsigned ToSplitColumns = Token->getRangeLength( |
2646 | 1.28k | NextLineIndex, TailOffset, Split.first, ContentStartColumn); |
2647 | 1.28k | if (ContentStartColumn + ToSplitColumns > ColumnLimit) { |
2648 | 522 | LLVM_DEBUG(llvm::dbgs() << " Next split protrudes, need: " |
2649 | 522 | << (ContentStartColumn + ToSplitColumns) |
2650 | 522 | << ", space: " << ColumnLimit); |
2651 | 522 | unsigned ExcessCharactersPenalty = |
2652 | 522 | (ContentStartColumn + ToSplitColumns - ColumnLimit) * |
2653 | 522 | Style.PenaltyExcessCharacter; |
2654 | 522 | if (NewBreakPenalty < ExcessCharactersPenalty) |
2655 | 522 | Reflow = false; |
2656 | 522 | } |
2657 | 1.28k | } |
2658 | 1.91k | } |
2659 | 2.11k | } else { |
2660 | 1.79k | LLVM_DEBUG(llvm::dbgs() << "not found.\n"); |
2661 | 1.79k | } |
2662 | 3.90k | } |
2663 | 308k | if (!Reflow) { |
2664 | | // If we didn't reflow into the next line, the only space to consider is |
2665 | | // the next logical line. Reset our state to match the start of the next |
2666 | | // line. |
2667 | 307k | TailOffset = 0; |
2668 | 307k | ContentStartColumn = |
2669 | 307k | Token->getContentStartColumn(NextLineIndex, /*Break=*/false); |
2670 | 307k | RemainingTokenColumns = Token->getRemainingLength( |
2671 | 307k | NextLineIndex, TailOffset, ContentStartColumn); |
2672 | | // Adapt the start of the token, for example indent. |
2673 | 307k | if (!DryRun) |
2674 | 19.2k | Token->adaptStartOfLine(NextLineIndex, Whitespaces); |
2675 | 307k | } else { |
2676 | | // If we found a reflow split and have added a new break before the next |
2677 | | // line, we are going to remove the line break at the start of the next |
2678 | | // logical line. For example, here we'll add a new line break after |
2679 | | // 'text', and subsequently delete the line break between 'that' and |
2680 | | // 'reflows'. |
2681 | | // // some text that |
2682 | | // // reflows |
2683 | | // -> |
2684 | | // // some text |
2685 | | // // that reflows |
2686 | | // When adding the line break, we also added the penalty for it, so we |
2687 | | // need to subtract that penalty again when we remove the line break due |
2688 | | // to reflowing. |
2689 | 965 | if (NewBreakBefore) { |
2690 | 900 | assert(Penalty >= NewBreakPenalty); |
2691 | 0 | Penalty -= NewBreakPenalty; |
2692 | 900 | } |
2693 | 965 | if (!DryRun) |
2694 | 325 | Token->reflow(NextLineIndex, Whitespaces); |
2695 | 965 | } |
2696 | 308k | } |
2697 | 433k | } |
2698 | | |
2699 | 125k | BreakableToken::Split SplitAfterLastLine = |
2700 | 125k | Token->getSplitAfterLastLine(TailOffset); |
2701 | 125k | if (SplitAfterLastLine.first != StringRef::npos) { |
2702 | 0 | LLVM_DEBUG(llvm::dbgs() << "Replacing whitespace after last line.\n"); |
2703 | | |
2704 | | // We add the last line's penalty here, since that line is going to be split |
2705 | | // now. |
2706 | 0 | Penalty += Style.PenaltyExcessCharacter * |
2707 | 0 | (ContentStartColumn + RemainingTokenColumns - ColumnLimit); |
2708 | |
|
2709 | 0 | if (!DryRun) { |
2710 | 0 | Token->replaceWhitespaceAfterLastLine(TailOffset, SplitAfterLastLine, |
2711 | 0 | Whitespaces); |
2712 | 0 | } |
2713 | 0 | ContentStartColumn = |
2714 | 0 | Token->getContentStartColumn(Token->getLineCount() - 1, /*Break=*/true); |
2715 | 0 | RemainingTokenColumns = Token->getRemainingLength( |
2716 | 0 | Token->getLineCount() - 1, |
2717 | 0 | TailOffset + SplitAfterLastLine.first + SplitAfterLastLine.second, |
2718 | 0 | ContentStartColumn); |
2719 | 0 | } |
2720 | | |
2721 | 125k | State.Column = ContentStartColumn + RemainingTokenColumns - |
2722 | 125k | Current.UnbreakableTailLength; |
2723 | | |
2724 | 125k | if (BreakInserted) { |
2725 | 10.5k | if (!DryRun) |
2726 | 2.52k | Token->updateAfterBroken(Whitespaces); |
2727 | | |
2728 | | // If we break the token inside a parameter list, we need to break before |
2729 | | // the next parameter on all levels, so that the next parameter is clearly |
2730 | | // visible. Line comments already introduce a break. |
2731 | 10.5k | if (Current.isNot(TT_LineComment)) |
2732 | 9.25k | for (ParenState &Paren : State.Stack) |
2733 | 35.2k | Paren.BreakBeforeParameter = true; |
2734 | | |
2735 | 10.5k | if (Current.is(TT_BlockComment)) |
2736 | 6.59k | State.NoContinuation = true; |
2737 | | |
2738 | 10.5k | State.Stack.back().LastSpace = StartColumn; |
2739 | 10.5k | } |
2740 | | |
2741 | 125k | Token->updateNextToken(State); |
2742 | | |
2743 | 125k | return {Penalty, Exceeded}; |
2744 | 125k | } |
2745 | | |
2746 | 10.0M | unsigned ContinuationIndenter::getColumnLimit(const LineState &State) const { |
2747 | | // In preprocessor directives reserve two chars for trailing " \". |
2748 | 10.0M | return Style.ColumnLimit - (State.Line->InPPDirective ? 2 : 0); |
2749 | 10.0M | } |
2750 | | |
2751 | 2.76M | bool ContinuationIndenter::nextIsMultilineString(const LineState &State) { |
2752 | 2.76M | const FormatToken &Current = *State.NextToken; |
2753 | 2.76M | if (!Current.isStringLiteral() || Current.is(TT_ImplicitStringLiteral)) |
2754 | 2.72M | return false; |
2755 | | // We never consider raw string literals "multiline" for the purpose of |
2756 | | // AlwaysBreakBeforeMultilineStrings implementation as they are special-cased |
2757 | | // (see TokenAnnotator::mustBreakBefore(). |
2758 | 33.2k | if (Current.TokenText.starts_with("R\"")) |
2759 | 0 | return false; |
2760 | 33.2k | if (Current.IsMultiline) |
2761 | 4 | return true; |
2762 | 33.2k | if (Current.getNextNonComment() && |
2763 | 33.2k | Current.getNextNonComment()->isStringLiteral()) { |
2764 | 1.44k | return true; // Implicit concatenation. |
2765 | 1.44k | } |
2766 | 31.8k | if (Style.ColumnLimit != 0 && Style.BreakStringLiterals && |
2767 | 31.8k | State.Column + Current.ColumnWidth + Current.UnbreakableTailLength > |
2768 | 31.8k | Style.ColumnLimit) { |
2769 | 7.67k | return true; // String will be split. |
2770 | 7.67k | } |
2771 | 24.1k | return false; |
2772 | 31.8k | } |
2773 | | |
2774 | | } // namespace format |
2775 | | } // namespace clang |