1//===--- Format.cpp - Format C++ code -------------------------------------===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9///
10/// \file
11/// \brief This file implements functions declared in Format.h. This will be
12/// split into separate files as we go.
13///
14//===----------------------------------------------------------------------===//
15
16#define DEBUG_TYPE "format-formatter"
17
18#include "BreakableToken.h"
19#include "TokenAnnotator.h"
20#include "UnwrappedLineParser.h"
21#include "WhitespaceManager.h"
22#include "clang/Basic/Diagnostic.h"
23#include "clang/Basic/OperatorPrecedence.h"
24#include "clang/Basic/SourceManager.h"
25#include "clang/Format/Format.h"
26#include "clang/Lex/Lexer.h"
27#include "llvm/ADT/STLExtras.h"
28#include "llvm/Support/Allocator.h"
29#include "llvm/Support/Debug.h"
30#include "llvm/Support/YAMLTraits.h"
31#include <queue>
32#include <string>
33
34namespace llvm {
35namespace yaml {
36template <>
37struct ScalarEnumerationTraits<clang::format::FormatStyle::LanguageStandard> {
38  static void enumeration(IO &IO,
39                          clang::format::FormatStyle::LanguageStandard &Value) {
40    IO.enumCase(Value, "C++03", clang::format::FormatStyle::LS_Cpp03);
41    IO.enumCase(Value, "C++11", clang::format::FormatStyle::LS_Cpp11);
42    IO.enumCase(Value, "Auto", clang::format::FormatStyle::LS_Auto);
43  }
44};
45
46template <>
47struct ScalarEnumerationTraits<clang::format::FormatStyle::BraceBreakingStyle> {
48  static void
49  enumeration(IO &IO, clang::format::FormatStyle::BraceBreakingStyle &Value) {
50    IO.enumCase(Value, "Attach", clang::format::FormatStyle::BS_Attach);
51    IO.enumCase(Value, "Linux", clang::format::FormatStyle::BS_Linux);
52    IO.enumCase(Value, "Stroustrup", clang::format::FormatStyle::BS_Stroustrup);
53    IO.enumCase(Value, "Allman", clang::format::FormatStyle::BS_Allman);
54  }
55};
56
57template <>
58struct ScalarEnumerationTraits<
59    clang::format::FormatStyle::NamespaceIndentationKind> {
60  static void
61  enumeration(IO &IO,
62              clang::format::FormatStyle::NamespaceIndentationKind &Value) {
63    IO.enumCase(Value, "None", clang::format::FormatStyle::NI_None);
64    IO.enumCase(Value, "Inner", clang::format::FormatStyle::NI_Inner);
65    IO.enumCase(Value, "All", clang::format::FormatStyle::NI_All);
66  }
67};
68
69template <> struct MappingTraits<clang::format::FormatStyle> {
70  static void mapping(llvm::yaml::IO &IO, clang::format::FormatStyle &Style) {
71    if (IO.outputting()) {
72      StringRef StylesArray[] = { "LLVM", "Google", "Chromium", "Mozilla" };
73      ArrayRef<StringRef> Styles(StylesArray);
74      for (size_t i = 0, e = Styles.size(); i < e; ++i) {
75        StringRef StyleName(Styles[i]);
76        clang::format::FormatStyle PredefinedStyle;
77        if (clang::format::getPredefinedStyle(StyleName, &PredefinedStyle) &&
78            Style == PredefinedStyle) {
79          IO.mapOptional("# BasedOnStyle", StyleName);
80          break;
81        }
82      }
83    } else {
84      StringRef BasedOnStyle;
85      IO.mapOptional("BasedOnStyle", BasedOnStyle);
86      if (!BasedOnStyle.empty())
87        if (!clang::format::getPredefinedStyle(BasedOnStyle, &Style)) {
88          IO.setError(Twine("Unknown value for BasedOnStyle: ", BasedOnStyle));
89          return;
90        }
91    }
92
93    IO.mapOptional("AccessModifierOffset", Style.AccessModifierOffset);
94    IO.mapOptional("AlignEscapedNewlinesLeft", Style.AlignEscapedNewlinesLeft);
95    IO.mapOptional("AlignTrailingComments", Style.AlignTrailingComments);
96    IO.mapOptional("AllowAllParametersOfDeclarationOnNextLine",
97                   Style.AllowAllParametersOfDeclarationOnNextLine);
98    IO.mapOptional("AllowShortIfStatementsOnASingleLine",
99                   Style.AllowShortIfStatementsOnASingleLine);
100    IO.mapOptional("AllowShortLoopsOnASingleLine",
101                   Style.AllowShortLoopsOnASingleLine);
102    IO.mapOptional("AlwaysBreakTemplateDeclarations",
103                   Style.AlwaysBreakTemplateDeclarations);
104    IO.mapOptional("AlwaysBreakBeforeMultilineStrings",
105                   Style.AlwaysBreakBeforeMultilineStrings);
106    IO.mapOptional("BreakBeforeBinaryOperators",
107                   Style.BreakBeforeBinaryOperators);
108    IO.mapOptional("BreakConstructorInitializersBeforeComma",
109                   Style.BreakConstructorInitializersBeforeComma);
110    IO.mapOptional("BinPackParameters", Style.BinPackParameters);
111    IO.mapOptional("ColumnLimit", Style.ColumnLimit);
112    IO.mapOptional("ConstructorInitializerAllOnOneLineOrOnePerLine",
113                   Style.ConstructorInitializerAllOnOneLineOrOnePerLine);
114    IO.mapOptional("DerivePointerBinding", Style.DerivePointerBinding);
115    IO.mapOptional("ExperimentalAutoDetectBinPacking",
116                   Style.ExperimentalAutoDetectBinPacking);
117    IO.mapOptional("IndentCaseLabels", Style.IndentCaseLabels);
118    IO.mapOptional("MaxEmptyLinesToKeep", Style.MaxEmptyLinesToKeep);
119    IO.mapOptional("NamespaceIndentation", Style.NamespaceIndentation);
120    IO.mapOptional("ObjCSpaceBeforeProtocolList",
121                   Style.ObjCSpaceBeforeProtocolList);
122    IO.mapOptional("PenaltyBreakComment", Style.PenaltyBreakComment);
123    IO.mapOptional("PenaltyBreakString", Style.PenaltyBreakString);
124    IO.mapOptional("PenaltyBreakFirstLessLess",
125                   Style.PenaltyBreakFirstLessLess);
126    IO.mapOptional("PenaltyExcessCharacter", Style.PenaltyExcessCharacter);
127    IO.mapOptional("PenaltyReturnTypeOnItsOwnLine",
128                   Style.PenaltyReturnTypeOnItsOwnLine);
129    IO.mapOptional("PointerBindsToType", Style.PointerBindsToType);
130    IO.mapOptional("SpacesBeforeTrailingComments",
131                   Style.SpacesBeforeTrailingComments);
132    IO.mapOptional("Cpp11BracedListStyle", Style.Cpp11BracedListStyle);
133    IO.mapOptional("Standard", Style.Standard);
134    IO.mapOptional("IndentWidth", Style.IndentWidth);
135    IO.mapOptional("UseTab", Style.UseTab);
136    IO.mapOptional("BreakBeforeBraces", Style.BreakBeforeBraces);
137    IO.mapOptional("IndentFunctionDeclarationAfterType",
138                   Style.IndentFunctionDeclarationAfterType);
139  }
140};
141}
142}
143
144namespace clang {
145namespace format {
146
147void setDefaultPenalties(FormatStyle &Style) {
148  Style.PenaltyBreakComment = 45;
149  Style.PenaltyBreakFirstLessLess = 120;
150  Style.PenaltyBreakString = 1000;
151  Style.PenaltyExcessCharacter = 1000000;
152}
153
154FormatStyle getLLVMStyle() {
155  FormatStyle LLVMStyle;
156  LLVMStyle.AccessModifierOffset = -2;
157  LLVMStyle.AlignEscapedNewlinesLeft = false;
158  LLVMStyle.AlignTrailingComments = true;
159  LLVMStyle.AllowAllParametersOfDeclarationOnNextLine = true;
160  LLVMStyle.AllowShortIfStatementsOnASingleLine = false;
161  LLVMStyle.AllowShortLoopsOnASingleLine = false;
162  LLVMStyle.AlwaysBreakBeforeMultilineStrings = false;
163  LLVMStyle.AlwaysBreakTemplateDeclarations = false;
164  LLVMStyle.BinPackParameters = true;
165  LLVMStyle.BreakBeforeBinaryOperators = false;
166  LLVMStyle.BreakBeforeBraces = FormatStyle::BS_Attach;
167  LLVMStyle.BreakConstructorInitializersBeforeComma = false;
168  LLVMStyle.ColumnLimit = 80;
169  LLVMStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = false;
170  LLVMStyle.Cpp11BracedListStyle = false;
171  LLVMStyle.DerivePointerBinding = false;
172  LLVMStyle.ExperimentalAutoDetectBinPacking = false;
173  LLVMStyle.IndentCaseLabels = false;
174  LLVMStyle.IndentFunctionDeclarationAfterType = false;
175  LLVMStyle.IndentWidth = 2;
176  LLVMStyle.MaxEmptyLinesToKeep = 1;
177  LLVMStyle.NamespaceIndentation = FormatStyle::NI_None;
178  LLVMStyle.ObjCSpaceBeforeProtocolList = true;
179  LLVMStyle.PointerBindsToType = false;
180  LLVMStyle.SpacesBeforeTrailingComments = 1;
181  LLVMStyle.Standard = FormatStyle::LS_Cpp03;
182  LLVMStyle.UseTab = false;
183
184  setDefaultPenalties(LLVMStyle);
185  LLVMStyle.PenaltyReturnTypeOnItsOwnLine = 60;
186
187  return LLVMStyle;
188}
189
190FormatStyle getGoogleStyle() {
191  FormatStyle GoogleStyle;
192  GoogleStyle.AccessModifierOffset = -1;
193  GoogleStyle.AlignEscapedNewlinesLeft = true;
194  GoogleStyle.AlignTrailingComments = true;
195  GoogleStyle.AllowAllParametersOfDeclarationOnNextLine = true;
196  GoogleStyle.AllowShortIfStatementsOnASingleLine = true;
197  GoogleStyle.AllowShortLoopsOnASingleLine = true;
198  GoogleStyle.AlwaysBreakBeforeMultilineStrings = true;
199  GoogleStyle.AlwaysBreakTemplateDeclarations = true;
200  GoogleStyle.BinPackParameters = true;
201  GoogleStyle.BreakBeforeBinaryOperators = false;
202  GoogleStyle.BreakBeforeBraces = FormatStyle::BS_Attach;
203  GoogleStyle.BreakConstructorInitializersBeforeComma = false;
204  GoogleStyle.ColumnLimit = 80;
205  GoogleStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = true;
206  GoogleStyle.Cpp11BracedListStyle = true;
207  GoogleStyle.DerivePointerBinding = true;
208  GoogleStyle.ExperimentalAutoDetectBinPacking = false;
209  GoogleStyle.IndentCaseLabels = true;
210  GoogleStyle.IndentFunctionDeclarationAfterType = true;
211  GoogleStyle.IndentWidth = 2;
212  GoogleStyle.MaxEmptyLinesToKeep = 1;
213  GoogleStyle.NamespaceIndentation = FormatStyle::NI_None;
214  GoogleStyle.ObjCSpaceBeforeProtocolList = false;
215  GoogleStyle.PointerBindsToType = true;
216  GoogleStyle.SpacesBeforeTrailingComments = 2;
217  GoogleStyle.Standard = FormatStyle::LS_Auto;
218  GoogleStyle.UseTab = false;
219
220  setDefaultPenalties(GoogleStyle);
221  GoogleStyle.PenaltyReturnTypeOnItsOwnLine = 200;
222
223  return GoogleStyle;
224}
225
226FormatStyle getChromiumStyle() {
227  FormatStyle ChromiumStyle = getGoogleStyle();
228  ChromiumStyle.AllowAllParametersOfDeclarationOnNextLine = false;
229  ChromiumStyle.AllowShortIfStatementsOnASingleLine = false;
230  ChromiumStyle.AllowShortLoopsOnASingleLine = false;
231  ChromiumStyle.BinPackParameters = false;
232  ChromiumStyle.DerivePointerBinding = false;
233  ChromiumStyle.Standard = FormatStyle::LS_Cpp03;
234  return ChromiumStyle;
235}
236
237FormatStyle getMozillaStyle() {
238  FormatStyle MozillaStyle = getLLVMStyle();
239  MozillaStyle.AllowAllParametersOfDeclarationOnNextLine = false;
240  MozillaStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = true;
241  MozillaStyle.DerivePointerBinding = true;
242  MozillaStyle.IndentCaseLabels = true;
243  MozillaStyle.ObjCSpaceBeforeProtocolList = false;
244  MozillaStyle.PenaltyReturnTypeOnItsOwnLine = 200;
245  MozillaStyle.PointerBindsToType = true;
246  return MozillaStyle;
247}
248
249FormatStyle getWebKitStyle() {
250  FormatStyle Style = getLLVMStyle();
251  Style.AccessModifierOffset = -4;
252  Style.AlignTrailingComments = false;
253  Style.BreakBeforeBinaryOperators = true;
254  Style.BreakBeforeBraces = FormatStyle::BS_Stroustrup;
255  Style.BreakConstructorInitializersBeforeComma = true;
256  Style.ColumnLimit = 0;
257  Style.IndentWidth = 4;
258  Style.NamespaceIndentation = FormatStyle::NI_Inner;
259  Style.PointerBindsToType = true;
260  return Style;
261}
262
263bool getPredefinedStyle(StringRef Name, FormatStyle *Style) {
264  if (Name.equals_lower("llvm"))
265    *Style = getLLVMStyle();
266  else if (Name.equals_lower("chromium"))
267    *Style = getChromiumStyle();
268  else if (Name.equals_lower("mozilla"))
269    *Style = getMozillaStyle();
270  else if (Name.equals_lower("google"))
271    *Style = getGoogleStyle();
272  else if (Name.equals_lower("webkit"))
273    *Style = getWebKitStyle();
274  else
275    return false;
276
277  return true;
278}
279
280llvm::error_code parseConfiguration(StringRef Text, FormatStyle *Style) {
281  if (Text.trim().empty())
282    return llvm::make_error_code(llvm::errc::invalid_argument);
283  llvm::yaml::Input Input(Text);
284  Input >> *Style;
285  return Input.error();
286}
287
288std::string configurationAsText(const FormatStyle &Style) {
289  std::string Text;
290  llvm::raw_string_ostream Stream(Text);
291  llvm::yaml::Output Output(Stream);
292  // We use the same mapping method for input and output, so we need a non-const
293  // reference here.
294  FormatStyle NonConstStyle = Style;
295  Output << NonConstStyle;
296  return Stream.str();
297}
298
299// Returns the length of everything up to the first possible line break after
300// the ), ], } or > matching \c Tok.
301static unsigned getLengthToMatchingParen(const FormatToken &Tok) {
302  if (Tok.MatchingParen == NULL)
303    return 0;
304  FormatToken *End = Tok.MatchingParen;
305  while (End->Next && !End->Next->CanBreakBefore) {
306    End = End->Next;
307  }
308  return End->TotalLength - Tok.TotalLength + 1;
309}
310
311namespace {
312
313class UnwrappedLineFormatter {
314public:
315  UnwrappedLineFormatter(const FormatStyle &Style, SourceManager &SourceMgr,
316                         const AnnotatedLine &Line, unsigned FirstIndent,
317                         const FormatToken *RootToken,
318                         WhitespaceManager &Whitespaces,
319                         encoding::Encoding Encoding,
320                         bool BinPackInconclusiveFunctions)
321      : Style(Style), SourceMgr(SourceMgr), Line(Line),
322        FirstIndent(FirstIndent), RootToken(RootToken),
323        Whitespaces(Whitespaces), Count(0), Encoding(Encoding),
324        BinPackInconclusiveFunctions(BinPackInconclusiveFunctions) {}
325
326  /// \brief Formats an \c UnwrappedLine.
327  void format(const AnnotatedLine *NextLine) {
328    // Initialize state dependent on indent.
329    LineState State;
330    State.Column = FirstIndent;
331    State.NextToken = RootToken;
332    State.Stack.push_back(ParenState(FirstIndent, FirstIndent,
333                                     /*AvoidBinPacking=*/false,
334                                     /*NoLineBreak=*/false));
335    State.LineContainsContinuedForLoopSection = false;
336    State.ParenLevel = 0;
337    State.StartOfStringLiteral = 0;
338    State.StartOfLineLevel = State.ParenLevel;
339    State.LowestLevelOnLine = State.ParenLevel;
340    State.IgnoreStackForComparison = false;
341
342    // The first token has already been indented and thus consumed.
343    moveStateToNextToken(State, /*DryRun=*/false, /*Newline=*/false);
344
345    if (Style.ColumnLimit == 0) {
346      formatWithoutColumnLimit(State);
347      return;
348    }
349
350    // If everything fits on a single line, just put it there.
351    unsigned ColumnLimit = Style.ColumnLimit;
352    if (NextLine && NextLine->InPPDirective &&
353        !NextLine->First->HasUnescapedNewline)
354      ColumnLimit = getColumnLimit();
355    if (Line.Last->TotalLength <= ColumnLimit - FirstIndent) {
356      while (State.NextToken != NULL) {
357        addTokenToState(false, false, State);
358      }
359    }
360
361    // If the ObjC method declaration does not fit on a line, we should format
362    // it with one arg per line.
363    if (Line.Type == LT_ObjCMethodDecl)
364      State.Stack.back().BreakBeforeParameter = true;
365
366    // Find best solution in solution space.
367    analyzeSolutionSpace(State);
368  }
369
370private:
371  void DebugTokenState(const FormatToken &FormatTok) {
372    const Token &Tok = FormatTok.Tok;
373    llvm::dbgs() << StringRef(SourceMgr.getCharacterData(Tok.getLocation()),
374                              Tok.getLength());
375    llvm::dbgs();
376  }
377
378  struct ParenState {
379    ParenState(unsigned Indent, unsigned LastSpace, bool AvoidBinPacking,
380               bool NoLineBreak)
381        : Indent(Indent), LastSpace(LastSpace), FirstLessLess(0),
382          BreakBeforeClosingBrace(false), QuestionColumn(0),
383          AvoidBinPacking(AvoidBinPacking), BreakBeforeParameter(false),
384          NoLineBreak(NoLineBreak), ColonPos(0), StartOfFunctionCall(0),
385          StartOfArraySubscripts(0), NestedNameSpecifierContinuation(0),
386          CallContinuation(0), VariablePos(0), ContainsLineBreak(false) {}
387
388    /// \brief The position to which a specific parenthesis level needs to be
389    /// indented.
390    unsigned Indent;
391
392    /// \brief The position of the last space on each level.
393    ///
394    /// Used e.g. to break like:
395    /// functionCall(Parameter, otherCall(
396    ///                             OtherParameter));
397    unsigned LastSpace;
398
399    /// \brief The position the first "<<" operator encountered on each level.
400    ///
401    /// Used to align "<<" operators. 0 if no such operator has been encountered
402    /// on a level.
403    unsigned FirstLessLess;
404
405    /// \brief Whether a newline needs to be inserted before the block's closing
406    /// brace.
407    ///
408    /// We only want to insert a newline before the closing brace if there also
409    /// was a newline after the beginning left brace.
410    bool BreakBeforeClosingBrace;
411
412    /// \brief The column of a \c ? in a conditional expression;
413    unsigned QuestionColumn;
414
415    /// \brief Avoid bin packing, i.e. multiple parameters/elements on multiple
416    /// lines, in this context.
417    bool AvoidBinPacking;
418
419    /// \brief Break after the next comma (or all the commas in this context if
420    /// \c AvoidBinPacking is \c true).
421    bool BreakBeforeParameter;
422
423    /// \brief Line breaking in this context would break a formatting rule.
424    bool NoLineBreak;
425
426    /// \brief The position of the colon in an ObjC method declaration/call.
427    unsigned ColonPos;
428
429    /// \brief The start of the most recent function in a builder-type call.
430    unsigned StartOfFunctionCall;
431
432    /// \brief Contains the start of array subscript expressions, so that they
433    /// can be aligned.
434    unsigned StartOfArraySubscripts;
435
436    /// \brief If a nested name specifier was broken over multiple lines, this
437    /// contains the start column of the second line. Otherwise 0.
438    unsigned NestedNameSpecifierContinuation;
439
440    /// \brief If a call expression was broken over multiple lines, this
441    /// contains the start column of the second line. Otherwise 0.
442    unsigned CallContinuation;
443
444    /// \brief The column of the first variable name in a variable declaration.
445    ///
446    /// Used to align further variables if necessary.
447    unsigned VariablePos;
448
449    /// \brief \c true if this \c ParenState already contains a line-break.
450    ///
451    /// The first line break in a certain \c ParenState causes extra penalty so
452    /// that clang-format prefers similar breaks, i.e. breaks in the same
453    /// parenthesis.
454    bool ContainsLineBreak;
455
456    bool operator<(const ParenState &Other) const {
457      if (Indent != Other.Indent)
458        return Indent < Other.Indent;
459      if (LastSpace != Other.LastSpace)
460        return LastSpace < Other.LastSpace;
461      if (FirstLessLess != Other.FirstLessLess)
462        return FirstLessLess < Other.FirstLessLess;
463      if (BreakBeforeClosingBrace != Other.BreakBeforeClosingBrace)
464        return BreakBeforeClosingBrace;
465      if (QuestionColumn != Other.QuestionColumn)
466        return QuestionColumn < Other.QuestionColumn;
467      if (AvoidBinPacking != Other.AvoidBinPacking)
468        return AvoidBinPacking;
469      if (BreakBeforeParameter != Other.BreakBeforeParameter)
470        return BreakBeforeParameter;
471      if (NoLineBreak != Other.NoLineBreak)
472        return NoLineBreak;
473      if (ColonPos != Other.ColonPos)
474        return ColonPos < Other.ColonPos;
475      if (StartOfFunctionCall != Other.StartOfFunctionCall)
476        return StartOfFunctionCall < Other.StartOfFunctionCall;
477      if (StartOfArraySubscripts != Other.StartOfArraySubscripts)
478        return StartOfArraySubscripts < Other.StartOfArraySubscripts;
479      if (CallContinuation != Other.CallContinuation)
480        return CallContinuation < Other.CallContinuation;
481      if (VariablePos != Other.VariablePos)
482        return VariablePos < Other.VariablePos;
483      if (ContainsLineBreak != Other.ContainsLineBreak)
484        return ContainsLineBreak < Other.ContainsLineBreak;
485      return false;
486    }
487  };
488
489  /// \brief The current state when indenting a unwrapped line.
490  ///
491  /// As the indenting tries different combinations this is copied by value.
492  struct LineState {
493    /// \brief The number of used columns in the current line.
494    unsigned Column;
495
496    /// \brief The token that needs to be next formatted.
497    const FormatToken *NextToken;
498
499    /// \brief \c true if this line contains a continued for-loop section.
500    bool LineContainsContinuedForLoopSection;
501
502    /// \brief The level of nesting inside (), [], <> and {}.
503    unsigned ParenLevel;
504
505    /// \brief The \c ParenLevel at the start of this line.
506    unsigned StartOfLineLevel;
507
508    /// \brief The lowest \c ParenLevel on the current line.
509    unsigned LowestLevelOnLine;
510
511    /// \brief The start column of the string literal, if we're in a string
512    /// literal sequence, 0 otherwise.
513    unsigned StartOfStringLiteral;
514
515    /// \brief A stack keeping track of properties applying to parenthesis
516    /// levels.
517    std::vector<ParenState> Stack;
518
519    /// \brief Ignore the stack of \c ParenStates for state comparison.
520    ///
521    /// In long and deeply nested unwrapped lines, the current algorithm can
522    /// be insufficient for finding the best formatting with a reasonable amount
523    /// of time and memory. Setting this flag will effectively lead to the
524    /// algorithm not analyzing some combinations. However, these combinations
525    /// rarely contain the optimal solution: In short, accepting a higher
526    /// penalty early would need to lead to different values in the \c
527    /// ParenState stack (in an otherwise identical state) and these different
528    /// values would need to lead to a significant amount of avoided penalty
529    /// later.
530    ///
531    /// FIXME: Come up with a better algorithm instead.
532    bool IgnoreStackForComparison;
533
534    /// \brief Comparison operator to be able to used \c LineState in \c map.
535    bool operator<(const LineState &Other) const {
536      if (NextToken != Other.NextToken)
537        return NextToken < Other.NextToken;
538      if (Column != Other.Column)
539        return Column < Other.Column;
540      if (LineContainsContinuedForLoopSection !=
541          Other.LineContainsContinuedForLoopSection)
542        return LineContainsContinuedForLoopSection;
543      if (ParenLevel != Other.ParenLevel)
544        return ParenLevel < Other.ParenLevel;
545      if (StartOfLineLevel != Other.StartOfLineLevel)
546        return StartOfLineLevel < Other.StartOfLineLevel;
547      if (LowestLevelOnLine != Other.LowestLevelOnLine)
548        return LowestLevelOnLine < Other.LowestLevelOnLine;
549      if (StartOfStringLiteral != Other.StartOfStringLiteral)
550        return StartOfStringLiteral < Other.StartOfStringLiteral;
551      if (IgnoreStackForComparison || Other.IgnoreStackForComparison)
552        return false;
553      return Stack < Other.Stack;
554    }
555  };
556
557  /// \brief Formats the line starting at \p State, simply keeping all of the
558  /// input's line breaking decisions.
559  void formatWithoutColumnLimit(LineState &State) {
560    while (State.NextToken != NULL) {
561      bool Newline = mustBreak(State) ||
562                     (canBreak(State) && State.NextToken->NewlinesBefore > 0);
563      addTokenToState(Newline, /*DryRun=*/false, State);
564    }
565  }
566
567  /// \brief Appends the next token to \p State and updates information
568  /// necessary for indentation.
569  ///
570  /// Puts the token on the current line if \p Newline is \c false and adds a
571  /// line break and necessary indentation otherwise.
572  ///
573  /// If \p DryRun is \c false, also creates and stores the required
574  /// \c Replacement.
575  unsigned addTokenToState(bool Newline, bool DryRun, LineState &State) {
576    const FormatToken &Current = *State.NextToken;
577    const FormatToken &Previous = *State.NextToken->Previous;
578
579    // Extra penalty that needs to be added because of the way certain line
580    // breaks are chosen.
581    unsigned ExtraPenalty = 0;
582
583    if (State.Stack.size() == 0 || Current.Type == TT_ImplicitStringLiteral) {
584      // FIXME: Is this correct?
585      int WhitespaceLength = SourceMgr.getSpellingColumnNumber(
586                                 State.NextToken->WhitespaceRange.getEnd()) -
587                             SourceMgr.getSpellingColumnNumber(
588                                 State.NextToken->WhitespaceRange.getBegin());
589      State.Column += WhitespaceLength + State.NextToken->CodePointCount;
590      State.NextToken = State.NextToken->Next;
591      return 0;
592    }
593
594    // If we are continuing an expression, we want to indent an extra 4 spaces.
595    unsigned ContinuationIndent =
596        std::max(State.Stack.back().LastSpace, State.Stack.back().Indent) + 4;
597    if (Newline) {
598      State.Stack.back().ContainsLineBreak = true;
599      if (Current.is(tok::r_brace)) {
600        if (Current.BlockKind == BK_BracedInit)
601          State.Column = State.Stack[State.Stack.size() - 2].LastSpace;
602        else
603          State.Column = FirstIndent;
604      } else if (Current.is(tok::string_literal) &&
605                 State.StartOfStringLiteral != 0) {
606        State.Column = State.StartOfStringLiteral;
607        State.Stack.back().BreakBeforeParameter = true;
608      } else if (Current.is(tok::lessless) &&
609                 State.Stack.back().FirstLessLess != 0) {
610        State.Column = State.Stack.back().FirstLessLess;
611      } else if (Current.isOneOf(tok::period, tok::arrow) &&
612                 Current.Type != TT_DesignatedInitializerPeriod) {
613        if (State.Stack.back().CallContinuation == 0) {
614          State.Column = ContinuationIndent;
615          State.Stack.back().CallContinuation = State.Column;
616        } else {
617          State.Column = State.Stack.back().CallContinuation;
618        }
619      } else if (Current.Type == TT_ConditionalExpr) {
620        State.Column = State.Stack.back().QuestionColumn;
621      } else if (Previous.is(tok::comma) &&
622                 State.Stack.back().VariablePos != 0) {
623        State.Column = State.Stack.back().VariablePos;
624      } else if (Previous.ClosesTemplateDeclaration ||
625                 ((Current.Type == TT_StartOfName ||
626                   Current.is(tok::kw_operator)) &&
627                  State.ParenLevel == 0 &&
628                  (!Style.IndentFunctionDeclarationAfterType ||
629                   Line.StartsDefinition))) {
630        State.Column = State.Stack.back().Indent;
631      } else if (Current.Type == TT_ObjCSelectorName) {
632        if (State.Stack.back().ColonPos > Current.CodePointCount) {
633          State.Column = State.Stack.back().ColonPos - Current.CodePointCount;
634        } else {
635          State.Column = State.Stack.back().Indent;
636          State.Stack.back().ColonPos = State.Column + Current.CodePointCount;
637        }
638      } else if (Current.is(tok::l_square) &&
639                 Current.Type != TT_ObjCMethodExpr) {
640        if (State.Stack.back().StartOfArraySubscripts != 0)
641          State.Column = State.Stack.back().StartOfArraySubscripts;
642        else
643          State.Column = ContinuationIndent;
644      } else if (Current.Type == TT_StartOfName ||
645                 Previous.isOneOf(tok::coloncolon, tok::equal) ||
646                 Previous.Type == TT_ObjCMethodExpr) {
647        State.Column = ContinuationIndent;
648      } else {
649        State.Column = State.Stack.back().Indent;
650        // Ensure that we fall back to indenting 4 spaces instead of just
651        // flushing continuations left.
652        if (State.Column == FirstIndent)
653          State.Column += 4;
654      }
655
656      if (Current.is(tok::question))
657        State.Stack.back().BreakBeforeParameter = true;
658      if ((Previous.isOneOf(tok::comma, tok::semi) &&
659           !State.Stack.back().AvoidBinPacking) ||
660          Previous.Type == TT_BinaryOperator)
661        State.Stack.back().BreakBeforeParameter = false;
662      if (Previous.Type == TT_TemplateCloser && State.ParenLevel == 0)
663        State.Stack.back().BreakBeforeParameter = false;
664
665      if (!DryRun) {
666        unsigned NewLines = 1;
667        if (Current.is(tok::comment))
668          NewLines = std::max(
669              NewLines,
670              std::min(Current.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1));
671        Whitespaces.replaceWhitespace(Current, NewLines, State.Column,
672                                      State.Column, Line.InPPDirective);
673      }
674
675      if (!Current.isTrailingComment())
676        State.Stack.back().LastSpace = State.Column;
677      if (Current.isOneOf(tok::arrow, tok::period) &&
678          Current.Type != TT_DesignatedInitializerPeriod)
679        State.Stack.back().LastSpace += Current.CodePointCount;
680      State.StartOfLineLevel = State.ParenLevel;
681      State.LowestLevelOnLine = State.ParenLevel;
682
683      // Any break on this level means that the parent level has been broken
684      // and we need to avoid bin packing there.
685      for (unsigned i = 0, e = State.Stack.size() - 1; i != e; ++i) {
686        State.Stack[i].BreakBeforeParameter = true;
687      }
688      const FormatToken *TokenBefore = Current.getPreviousNonComment();
689      if (TokenBefore && !TokenBefore->isOneOf(tok::comma, tok::semi) &&
690          TokenBefore->Type != TT_TemplateCloser &&
691          TokenBefore->Type != TT_BinaryOperator && !TokenBefore->opensScope())
692        State.Stack.back().BreakBeforeParameter = true;
693
694      // If we break after {, we should also break before the corresponding }.
695      if (Previous.is(tok::l_brace))
696        State.Stack.back().BreakBeforeClosingBrace = true;
697
698      if (State.Stack.back().AvoidBinPacking) {
699        // If we are breaking after '(', '{', '<', this is not bin packing
700        // unless AllowAllParametersOfDeclarationOnNextLine is false.
701        if (!(Previous.isOneOf(tok::l_paren, tok::l_brace) ||
702              Previous.Type == TT_BinaryOperator) ||
703            (!Style.AllowAllParametersOfDeclarationOnNextLine &&
704             Line.MustBeDeclaration))
705          State.Stack.back().BreakBeforeParameter = true;
706      }
707
708      // Breaking before the first "<<" is generally not desirable.
709      if (Current.is(tok::lessless) && State.Stack.back().FirstLessLess == 0)
710        ExtraPenalty += Style.PenaltyBreakFirstLessLess;
711
712    } else {
713      if (Current.is(tok::equal) &&
714          (RootToken->is(tok::kw_for) || State.ParenLevel == 0) &&
715          State.Stack.back().VariablePos == 0) {
716        State.Stack.back().VariablePos = State.Column;
717        // Move over * and & if they are bound to the variable name.
718        const FormatToken *Tok = &Previous;
719        while (Tok && State.Stack.back().VariablePos >= Tok->CodePointCount) {
720          State.Stack.back().VariablePos -= Tok->CodePointCount;
721          if (Tok->SpacesRequiredBefore != 0)
722            break;
723          Tok = Tok->Previous;
724        }
725        if (Previous.PartOfMultiVariableDeclStmt)
726          State.Stack.back().LastSpace = State.Stack.back().VariablePos;
727      }
728
729      unsigned Spaces = State.NextToken->SpacesRequiredBefore;
730
731      if (!DryRun)
732        Whitespaces.replaceWhitespace(Current, 0, Spaces,
733                                      State.Column + Spaces);
734
735      if (Current.Type == TT_ObjCSelectorName &&
736          State.Stack.back().ColonPos == 0) {
737        if (State.Stack.back().Indent + Current.LongestObjCSelectorName >
738            State.Column + Spaces + Current.CodePointCount)
739          State.Stack.back().ColonPos =
740              State.Stack.back().Indent + Current.LongestObjCSelectorName;
741        else
742          State.Stack.back().ColonPos =
743              State.Column + Spaces + Current.CodePointCount;
744      }
745
746      if (Previous.opensScope() && Previous.Type != TT_ObjCMethodExpr &&
747          Current.Type != TT_LineComment)
748        State.Stack.back().Indent = State.Column + Spaces;
749      if (Previous.is(tok::comma) && !Current.isTrailingComment() &&
750          State.Stack.back().AvoidBinPacking)
751        State.Stack.back().NoLineBreak = true;
752
753      State.Column += Spaces;
754      if (Current.is(tok::l_paren) && Previous.isOneOf(tok::kw_if, tok::kw_for))
755        // Treat the condition inside an if as if it was a second function
756        // parameter, i.e. let nested calls have an indent of 4.
757        State.Stack.back().LastSpace = State.Column + 1; // 1 is length of "(".
758      else if (Previous.is(tok::comma))
759        State.Stack.back().LastSpace = State.Column;
760      else if ((Previous.Type == TT_BinaryOperator ||
761                Previous.Type == TT_ConditionalExpr ||
762                Previous.Type == TT_CtorInitializerColon) &&
763               !(Previous.getPrecedence() == prec::Assignment &&
764                 Current.FakeLParens.empty()))
765        // Always indent relative to the RHS of the expression unless this is a
766        // simple assignment without binary expression on the RHS.
767        State.Stack.back().LastSpace = State.Column;
768      else if (Previous.Type == TT_InheritanceColon)
769        State.Stack.back().Indent = State.Column;
770      else if (Previous.opensScope()) {
771        // If a function has multiple parameters (including a single parameter
772        // that is a binary expression) or a trailing call, indent all
773        // parameters from the opening parenthesis. This avoids confusing
774        // indents like:
775        //   OuterFunction(InnerFunctionCall(
776        //       ParameterToInnerFunction),
777        //                 SecondParameterToOuterFunction);
778        bool HasMultipleParameters = !Current.FakeLParens.empty();
779        bool HasTrailingCall = false;
780        if (Previous.MatchingParen) {
781          const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
782          if (Next && Next->isOneOf(tok::period, tok::arrow))
783            HasTrailingCall = true;
784        }
785        if (HasMultipleParameters || HasTrailingCall)
786          State.Stack.back().LastSpace = State.Column;
787      }
788    }
789
790    return moveStateToNextToken(State, DryRun, Newline) + ExtraPenalty;
791  }
792
793  /// \brief Mark the next token as consumed in \p State and modify its stacks
794  /// accordingly.
795  unsigned moveStateToNextToken(LineState &State, bool DryRun, bool Newline) {
796    const FormatToken &Current = *State.NextToken;
797    assert(State.Stack.size());
798
799    if (Current.Type == TT_InheritanceColon)
800      State.Stack.back().AvoidBinPacking = true;
801    if (Current.is(tok::lessless) && State.Stack.back().FirstLessLess == 0)
802      State.Stack.back().FirstLessLess = State.Column;
803    if (Current.is(tok::l_square) &&
804        State.Stack.back().StartOfArraySubscripts == 0)
805      State.Stack.back().StartOfArraySubscripts = State.Column;
806    if (Current.is(tok::question))
807      State.Stack.back().QuestionColumn = State.Column;
808    if (!Current.opensScope() && !Current.closesScope())
809      State.LowestLevelOnLine =
810          std::min(State.LowestLevelOnLine, State.ParenLevel);
811    if (Current.isOneOf(tok::period, tok::arrow) &&
812        Line.Type == LT_BuilderTypeCall && State.ParenLevel == 0)
813      State.Stack.back().StartOfFunctionCall =
814          Current.LastInChainOfCalls ? 0
815                                     : State.Column + Current.CodePointCount;
816    if (Current.Type == TT_CtorInitializerColon) {
817      // Indent 2 from the column, so:
818      // SomeClass::SomeClass()
819      //     : First(...), ...
820      //       Next(...)
821      //       ^ line up here.
822      if (!Style.BreakConstructorInitializersBeforeComma)
823        State.Stack.back().Indent = State.Column + 2;
824      if (Style.ConstructorInitializerAllOnOneLineOrOnePerLine)
825        State.Stack.back().AvoidBinPacking = true;
826      State.Stack.back().BreakBeforeParameter = false;
827    }
828
829    // If return returns a binary expression, align after it.
830    if (Current.is(tok::kw_return) && !Current.FakeLParens.empty())
831      State.Stack.back().LastSpace = State.Column + 7;
832
833    // In ObjC method declaration we align on the ":" of parameters, but we need
834    // to ensure that we indent parameters on subsequent lines by at least 4.
835    if (Current.Type == TT_ObjCMethodSpecifier)
836      State.Stack.back().Indent += 4;
837
838    // Insert scopes created by fake parenthesis.
839    const FormatToken *Previous = Current.getPreviousNonComment();
840    // Don't add extra indentation for the first fake parenthesis after
841    // 'return', assignements or opening <({[. The indentation for these cases
842    // is special cased.
843    bool SkipFirstExtraIndent =
844        Current.is(tok::kw_return) ||
845        (Previous && (Previous->opensScope() ||
846                      Previous->getPrecedence() == prec::Assignment));
847    for (SmallVectorImpl<prec::Level>::const_reverse_iterator
848             I = Current.FakeLParens.rbegin(),
849             E = Current.FakeLParens.rend();
850         I != E; ++I) {
851      ParenState NewParenState = State.Stack.back();
852      NewParenState.ContainsLineBreak = false;
853      NewParenState.Indent =
854          std::max(std::max(State.Column, NewParenState.Indent),
855                   State.Stack.back().LastSpace);
856
857      // Always indent conditional expressions. Never indent expression where
858      // the 'operator' is ',', ';' or an assignment (i.e. *I <=
859      // prec::Assignment) as those have different indentation rules. Indent
860      // other expression, unless the indentation needs to be skipped.
861      if (*I == prec::Conditional ||
862          (!SkipFirstExtraIndent && *I > prec::Assignment &&
863           !Style.BreakBeforeBinaryOperators))
864        NewParenState.Indent += 4;
865      if (Previous && !Previous->opensScope())
866        NewParenState.BreakBeforeParameter = false;
867      State.Stack.push_back(NewParenState);
868      SkipFirstExtraIndent = false;
869    }
870
871    // If we encounter an opening (, [, { or <, we add a level to our stacks to
872    // prepare for the following tokens.
873    if (Current.opensScope()) {
874      unsigned NewIndent;
875      unsigned LastSpace = State.Stack.back().LastSpace;
876      bool AvoidBinPacking;
877      if (Current.is(tok::l_brace)) {
878        NewIndent =
879            LastSpace + (Style.Cpp11BracedListStyle ? 4 : Style.IndentWidth);
880        const FormatToken *NextNoComment = Current.getNextNonComment();
881        AvoidBinPacking = NextNoComment &&
882                          NextNoComment->Type == TT_DesignatedInitializerPeriod;
883      } else {
884        NewIndent =
885            4 + std::max(LastSpace, State.Stack.back().StartOfFunctionCall);
886        AvoidBinPacking = !Style.BinPackParameters ||
887                          (Style.ExperimentalAutoDetectBinPacking &&
888                           (Current.PackingKind == PPK_OnePerLine ||
889                            (!BinPackInconclusiveFunctions &&
890                             Current.PackingKind == PPK_Inconclusive)));
891      }
892
893      State.Stack.push_back(ParenState(NewIndent, LastSpace, AvoidBinPacking,
894                                       State.Stack.back().NoLineBreak));
895      ++State.ParenLevel;
896    }
897
898    // If this '[' opens an ObjC call, determine whether all parameters fit into
899    // one line and put one per line if they don't.
900    if (Current.is(tok::l_square) && Current.Type == TT_ObjCMethodExpr &&
901        Current.MatchingParen != NULL) {
902      if (getLengthToMatchingParen(Current) + State.Column > getColumnLimit())
903        State.Stack.back().BreakBeforeParameter = true;
904    }
905
906    // If we encounter a closing ), ], } or >, we can remove a level from our
907    // stacks.
908    if (Current.isOneOf(tok::r_paren, tok::r_square) ||
909        (Current.is(tok::r_brace) && State.NextToken != RootToken) ||
910        State.NextToken->Type == TT_TemplateCloser) {
911      State.Stack.pop_back();
912      --State.ParenLevel;
913    }
914    if (Current.is(tok::r_square)) {
915      // If this ends the array subscript expr, reset the corresponding value.
916      const FormatToken *NextNonComment = Current.getNextNonComment();
917      if (NextNonComment && NextNonComment->isNot(tok::l_square))
918        State.Stack.back().StartOfArraySubscripts = 0;
919    }
920
921    // Remove scopes created by fake parenthesis.
922    for (unsigned i = 0, e = Current.FakeRParens; i != e; ++i) {
923      unsigned VariablePos = State.Stack.back().VariablePos;
924      State.Stack.pop_back();
925      State.Stack.back().VariablePos = VariablePos;
926    }
927
928    if (Current.is(tok::string_literal) && State.StartOfStringLiteral == 0) {
929      State.StartOfStringLiteral = State.Column;
930    } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash,
931                                tok::string_literal)) {
932      State.StartOfStringLiteral = 0;
933    }
934
935    State.Column += Current.CodePointCount;
936
937    State.NextToken = State.NextToken->Next;
938
939    if (!Newline && Style.AlwaysBreakBeforeMultilineStrings &&
940        Current.is(tok::string_literal) && Current.CanBreakBefore)
941      return 0;
942
943    return breakProtrudingToken(Current, State, DryRun);
944  }
945
946  /// \brief If the current token sticks out over the end of the line, break
947  /// it if possible.
948  ///
949  /// \returns An extra penalty if a token was broken, otherwise 0.
950  ///
951  /// The returned penalty will cover the cost of the additional line breaks and
952  /// column limit violation in all lines except for the last one. The penalty
953  /// for the column limit violation in the last line (and in single line
954  /// tokens) is handled in \c addNextStateToQueue.
955  unsigned breakProtrudingToken(const FormatToken &Current, LineState &State,
956                                bool DryRun) {
957    llvm::OwningPtr<BreakableToken> Token;
958    unsigned StartColumn = State.Column - Current.CodePointCount;
959    unsigned OriginalStartColumn =
960        SourceMgr.getSpellingColumnNumber(Current.getStartOfNonWhitespace()) -
961        1;
962
963    if (Current.is(tok::string_literal) &&
964        Current.Type != TT_ImplicitStringLiteral) {
965      // Only break up default narrow strings.
966      if (!Current.TokenText.startswith("\""))
967        return 0;
968      // Don't break string literals with escaped newlines. As clang-format must
969      // not change the string's content, it is unlikely that we'll end up with
970      // a better format.
971      if (Current.TokenText.find("\\\n") != StringRef::npos)
972        return 0;
973      // Exempts unterminated string literals from line breaking. The user will
974      // likely want to terminate the string before any line breaking is done.
975      if (Current.IsUnterminatedLiteral)
976         return 0;
977
978      Token.reset(new BreakableStringLiteral(Current, StartColumn,
979                                             Line.InPPDirective, Encoding));
980    } else if (Current.Type == TT_BlockComment && Current.isTrailingComment()) {
981      Token.reset(new BreakableBlockComment(
982          Style, Current, StartColumn, OriginalStartColumn, !Current.Previous,
983          Line.InPPDirective, Encoding));
984    } else if (Current.Type == TT_LineComment &&
985               (Current.Previous == NULL ||
986                Current.Previous->Type != TT_ImplicitStringLiteral)) {
987      // Don't break line comments with escaped newlines. These look like
988      // separate line comments, but in fact contain a single line comment with
989      // multiple lines including leading whitespace and the '//' markers.
990      //
991      // FIXME: If we want to handle them correctly, we'll need to adjust
992      // leading whitespace in consecutive lines when changing indentation of
993      // the first line similar to what we do with block comments.
994      StringRef::size_type EscapedNewlinePos = Current.TokenText.find("\\\n");
995      if (EscapedNewlinePos != StringRef::npos) {
996        State.Column =
997            StartColumn +
998            encoding::getCodePointCount(
999                Current.TokenText.substr(0, EscapedNewlinePos), Encoding) +
1000            1;
1001        return 0;
1002      }
1003
1004      Token.reset(new BreakableLineComment(Current, StartColumn,
1005                                           Line.InPPDirective, Encoding));
1006    } else {
1007      return 0;
1008    }
1009    if (Current.UnbreakableTailLength >= getColumnLimit())
1010      return 0;
1011
1012    unsigned RemainingSpace = getColumnLimit() - Current.UnbreakableTailLength;
1013    bool BreakInserted = false;
1014    unsigned Penalty = 0;
1015    unsigned RemainingTokenColumns = 0;
1016    for (unsigned LineIndex = 0, EndIndex = Token->getLineCount();
1017         LineIndex != EndIndex; ++LineIndex) {
1018      if (!DryRun)
1019        Token->replaceWhitespaceBefore(LineIndex, Whitespaces);
1020      unsigned TailOffset = 0;
1021      RemainingTokenColumns = Token->getLineLengthAfterSplit(
1022          LineIndex, TailOffset, StringRef::npos);
1023      while (RemainingTokenColumns > RemainingSpace) {
1024        BreakableToken::Split Split =
1025            Token->getSplit(LineIndex, TailOffset, getColumnLimit());
1026        if (Split.first == StringRef::npos) {
1027          // The last line's penalty is handled in addNextStateToQueue().
1028          if (LineIndex < EndIndex - 1)
1029            Penalty += Style.PenaltyExcessCharacter *
1030                       (RemainingTokenColumns - RemainingSpace);
1031          break;
1032        }
1033        assert(Split.first != 0);
1034        unsigned NewRemainingTokenColumns = Token->getLineLengthAfterSplit(
1035            LineIndex, TailOffset + Split.first + Split.second,
1036            StringRef::npos);
1037        assert(NewRemainingTokenColumns < RemainingTokenColumns);
1038        if (!DryRun)
1039          Token->insertBreak(LineIndex, TailOffset, Split, Whitespaces);
1040        Penalty += Current.is(tok::string_literal) ? Style.PenaltyBreakString
1041                                                   : Style.PenaltyBreakComment;
1042        unsigned ColumnsUsed =
1043            Token->getLineLengthAfterSplit(LineIndex, TailOffset, Split.first);
1044        if (ColumnsUsed > getColumnLimit()) {
1045          Penalty +=
1046              Style.PenaltyExcessCharacter * (ColumnsUsed - getColumnLimit());
1047        }
1048        TailOffset += Split.first + Split.second;
1049        RemainingTokenColumns = NewRemainingTokenColumns;
1050        BreakInserted = true;
1051      }
1052    }
1053
1054    State.Column = RemainingTokenColumns;
1055
1056    if (BreakInserted) {
1057      // If we break the token inside a parameter list, we need to break before
1058      // the next parameter on all levels, so that the next parameter is clearly
1059      // visible. Line comments already introduce a break.
1060      if (Current.Type != TT_LineComment) {
1061        for (unsigned i = 0, e = State.Stack.size(); i != e; ++i)
1062          State.Stack[i].BreakBeforeParameter = true;
1063      }
1064
1065      State.Stack.back().LastSpace = StartColumn;
1066    }
1067    return Penalty;
1068  }
1069
1070  unsigned getColumnLimit() {
1071    // In preprocessor directives reserve two chars for trailing " \"
1072    return Style.ColumnLimit - (Line.InPPDirective ? 2 : 0);
1073  }
1074
1075  /// \brief An edge in the solution space from \c Previous->State to \c State,
1076  /// inserting a newline dependent on the \c NewLine.
1077  struct StateNode {
1078    StateNode(const LineState &State, bool NewLine, StateNode *Previous)
1079        : State(State), NewLine(NewLine), Previous(Previous) {}
1080    LineState State;
1081    bool NewLine;
1082    StateNode *Previous;
1083  };
1084
1085  /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
1086  ///
1087  /// In case of equal penalties, we want to prefer states that were inserted
1088  /// first. During state generation we make sure that we insert states first
1089  /// that break the line as late as possible.
1090  typedef std::pair<unsigned, unsigned> OrderedPenalty;
1091
1092  /// \brief An item in the prioritized BFS search queue. The \c StateNode's
1093  /// \c State has the given \c OrderedPenalty.
1094  typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
1095
1096  /// \brief The BFS queue type.
1097  typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
1098                              std::greater<QueueItem> > QueueType;
1099
1100  /// \brief Analyze the entire solution space starting from \p InitialState.
1101  ///
1102  /// This implements a variant of Dijkstra's algorithm on the graph that spans
1103  /// the solution space (\c LineStates are the nodes). The algorithm tries to
1104  /// find the shortest path (the one with lowest penalty) from \p InitialState
1105  /// to a state where all tokens are placed.
1106  void analyzeSolutionSpace(LineState &InitialState) {
1107    std::set<LineState> Seen;
1108
1109    // Insert start element into queue.
1110    StateNode *Node =
1111        new (Allocator.Allocate()) StateNode(InitialState, false, NULL);
1112    Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
1113    ++Count;
1114
1115    // While not empty, take first element and follow edges.
1116    while (!Queue.empty()) {
1117      unsigned Penalty = Queue.top().first.first;
1118      StateNode *Node = Queue.top().second;
1119      if (Node->State.NextToken == NULL) {
1120        DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
1121        break;
1122      }
1123      Queue.pop();
1124
1125      // Cut off the analysis of certain solutions if the analysis gets too
1126      // complex. See description of IgnoreStackForComparison.
1127      if (Count > 10000)
1128        Node->State.IgnoreStackForComparison = true;
1129
1130      if (!Seen.insert(Node->State).second)
1131        // State already examined with lower penalty.
1132        continue;
1133
1134      addNextStateToQueue(Penalty, Node, /*NewLine=*/false);
1135      addNextStateToQueue(Penalty, Node, /*NewLine=*/true);
1136    }
1137
1138    if (Queue.empty())
1139      // We were unable to find a solution, do nothing.
1140      // FIXME: Add diagnostic?
1141      return;
1142
1143    // Reconstruct the solution.
1144    reconstructPath(InitialState, Queue.top().second);
1145    DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
1146    DEBUG(llvm::dbgs() << "---\n");
1147  }
1148
1149  void reconstructPath(LineState &State, StateNode *Current) {
1150    std::deque<StateNode *> Path;
1151    // We do not need a break before the initial token.
1152    while (Current->Previous) {
1153      Path.push_front(Current);
1154      Current = Current->Previous;
1155    }
1156    for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
1157         I != E; ++I) {
1158      DEBUG({
1159        if ((*I)->NewLine) {
1160          llvm::dbgs() << "Penalty for splitting before "
1161                       << (*I)->Previous->State.NextToken->Tok.getName() << ": "
1162                       << (*I)->Previous->State.NextToken->SplitPenalty << "\n";
1163        }
1164      });
1165      addTokenToState((*I)->NewLine, false, State);
1166    }
1167  }
1168
1169  /// \brief Add the following state to the analysis queue \c Queue.
1170  ///
1171  /// Assume the current state is \p PreviousNode and has been reached with a
1172  /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
1173  void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
1174                           bool NewLine) {
1175    if (NewLine && !canBreak(PreviousNode->State))
1176      return;
1177    if (!NewLine && mustBreak(PreviousNode->State))
1178      return;
1179    if (NewLine) {
1180      if (!PreviousNode->State.Stack.back().ContainsLineBreak)
1181        Penalty += 15;
1182      Penalty += PreviousNode->State.NextToken->SplitPenalty;
1183    }
1184
1185    StateNode *Node = new (Allocator.Allocate())
1186        StateNode(PreviousNode->State, NewLine, PreviousNode);
1187    Penalty += addTokenToState(NewLine, true, Node->State);
1188    if (Node->State.Column > getColumnLimit()) {
1189      unsigned ExcessCharacters = Node->State.Column - getColumnLimit();
1190      Penalty += Style.PenaltyExcessCharacter * ExcessCharacters;
1191    }
1192
1193    Queue.push(QueueItem(OrderedPenalty(Penalty, Count), Node));
1194    ++Count;
1195  }
1196
1197  /// \brief Returns \c true, if a line break after \p State is allowed.
1198  bool canBreak(const LineState &State) {
1199    const FormatToken &Current = *State.NextToken;
1200    const FormatToken &Previous = *Current.Previous;
1201    assert(&Previous == Current.Previous);
1202    if (!Current.CanBreakBefore &&
1203        !(Current.is(tok::r_brace) &&
1204          State.Stack.back().BreakBeforeClosingBrace))
1205      return false;
1206    // The opening "{" of a braced list has to be on the same line as the first
1207    // element if it is nested in another braced init list or function call.
1208    if (!Current.MustBreakBefore && Previous.is(tok::l_brace) &&
1209        Previous.Previous &&
1210        Previous.Previous->isOneOf(tok::l_brace, tok::l_paren, tok::comma))
1211      return false;
1212    // This prevents breaks like:
1213    //   ...
1214    //   SomeParameter, OtherParameter).DoSomething(
1215    //   ...
1216    // As they hide "DoSomething" and are generally bad for readability.
1217    if (Previous.opensScope() &&
1218        State.LowestLevelOnLine < State.StartOfLineLevel)
1219      return false;
1220    return !State.Stack.back().NoLineBreak;
1221  }
1222
1223  /// \brief Returns \c true, if a line break after \p State is mandatory.
1224  bool mustBreak(const LineState &State) {
1225    const FormatToken &Current = *State.NextToken;
1226    const FormatToken &Previous = *Current.Previous;
1227    if (Current.MustBreakBefore || Current.Type == TT_InlineASMColon)
1228      return true;
1229    if (!Style.Cpp11BracedListStyle && Current.is(tok::r_brace) &&
1230        State.Stack.back().BreakBeforeClosingBrace)
1231       return true;
1232    if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection)
1233      return true;
1234    if (Style.BreakConstructorInitializersBeforeComma) {
1235      if (Previous.Type == TT_CtorInitializerComma)
1236        return false;
1237      if (Current.Type == TT_CtorInitializerComma)
1238        return true;
1239    }
1240    if ((Previous.isOneOf(tok::comma, tok::semi) || Current.is(tok::question) ||
1241         (Current.Type == TT_ConditionalExpr &&
1242          !(Current.is(tok::colon) && Previous.is(tok::question)))) &&
1243        State.Stack.back().BreakBeforeParameter &&
1244        !Current.isTrailingComment() &&
1245        !Current.isOneOf(tok::r_paren, tok::r_brace))
1246      return true;
1247    if (Style.AlwaysBreakBeforeMultilineStrings &&
1248        State.Column > State.Stack.back().Indent &&
1249        Current.is(tok::string_literal) && Previous.isNot(tok::lessless) &&
1250        Previous.Type != TT_InlineASMColon &&
1251        ((Current.getNextNonComment() &&
1252          Current.getNextNonComment()->is(tok::string_literal)) ||
1253         (Current.TokenText.find("\\\n") != StringRef::npos)))
1254      return true;
1255
1256    if (!Style.BreakBeforeBinaryOperators) {
1257      // If we need to break somewhere inside the LHS of a binary expression, we
1258      // should also break after the operator. Otherwise, the formatting would
1259      // hide the operator precedence, e.g. in:
1260      //   if (aaaaaaaaaaaaaa ==
1261      //           bbbbbbbbbbbbbb && c) {..
1262      // For comparisons, we only apply this rule, if the LHS is a binary
1263      // expression itself as otherwise, the line breaks seem superfluous.
1264      // We need special cases for ">>" which we have split into two ">" while
1265      // lexing in order to make template parsing easier.
1266      //
1267      // FIXME: We'll need something similar for styles that break before binary
1268      // operators.
1269      bool IsComparison = (Previous.getPrecedence() == prec::Relational ||
1270                           Previous.getPrecedence() == prec::Equality) &&
1271                          Previous.Previous && Previous.Previous->Type !=
1272                                                   TT_BinaryOperator; // For >>.
1273      bool LHSIsBinaryExpr =
1274          Previous.Previous && Previous.Previous->FakeRParens > 0;
1275      if (Previous.Type == TT_BinaryOperator &&
1276          (!IsComparison || LHSIsBinaryExpr) &&
1277          Current.Type != TT_BinaryOperator && // For >>.
1278          !Current.isTrailingComment() &&
1279          !Previous.isOneOf(tok::lessless, tok::question) &&
1280          Previous.getPrecedence() != prec::Assignment &&
1281          State.Stack.back().BreakBeforeParameter)
1282        return true;
1283    }
1284
1285    // Same as above, but for the first "<<" operator.
1286    if (Current.is(tok::lessless) && State.Stack.back().BreakBeforeParameter &&
1287        State.Stack.back().FirstLessLess == 0)
1288      return true;
1289
1290    // FIXME: Comparing LongestObjCSelectorName to 0 is a hacky way of finding
1291    // out whether it is the first parameter. Clean this up.
1292    if (Current.Type == TT_ObjCSelectorName &&
1293        Current.LongestObjCSelectorName == 0 &&
1294        State.Stack.back().BreakBeforeParameter)
1295      return true;
1296    if ((Current.Type == TT_CtorInitializerColon ||
1297         (Previous.ClosesTemplateDeclaration && State.ParenLevel == 0)))
1298      return true;
1299
1300    if ((Current.Type == TT_StartOfName || Current.is(tok::kw_operator)) &&
1301        Line.MightBeFunctionDecl && State.Stack.back().BreakBeforeParameter &&
1302        State.ParenLevel == 0)
1303      return true;
1304    return false;
1305  }
1306
1307  // Returns the total number of columns required for the remaining tokens.
1308  unsigned getRemainingLength(const LineState &State) {
1309    if (State.NextToken && State.NextToken->Previous)
1310      return Line.Last->TotalLength - State.NextToken->Previous->TotalLength;
1311    return 0;
1312  }
1313
1314  FormatStyle Style;
1315  SourceManager &SourceMgr;
1316  const AnnotatedLine &Line;
1317  const unsigned FirstIndent;
1318  const FormatToken *RootToken;
1319  WhitespaceManager &Whitespaces;
1320
1321  llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1322  QueueType Queue;
1323  // Increasing count of \c StateNode items we have created. This is used
1324  // to create a deterministic order independent of the container.
1325  unsigned Count;
1326  encoding::Encoding Encoding;
1327  bool BinPackInconclusiveFunctions;
1328};
1329
1330class FormatTokenLexer {
1331public:
1332  FormatTokenLexer(Lexer &Lex, SourceManager &SourceMgr,
1333                   encoding::Encoding Encoding)
1334      : FormatTok(NULL), GreaterStashed(false), TrailingWhitespace(0), Lex(Lex),
1335        SourceMgr(SourceMgr), IdentTable(getFormattingLangOpts()),
1336        Encoding(Encoding) {
1337    Lex.SetKeepWhitespaceMode(true);
1338  }
1339
1340  ArrayRef<FormatToken *> lex() {
1341    assert(Tokens.empty());
1342    do {
1343      Tokens.push_back(getNextToken());
1344    } while (Tokens.back()->Tok.isNot(tok::eof));
1345    return Tokens;
1346  }
1347
1348  IdentifierTable &getIdentTable() { return IdentTable; }
1349
1350private:
1351  FormatToken *getNextToken() {
1352    if (GreaterStashed) {
1353      // Create a synthesized second '>' token.
1354      Token Greater = FormatTok->Tok;
1355      FormatTok = new (Allocator.Allocate()) FormatToken;
1356      FormatTok->Tok = Greater;
1357      SourceLocation GreaterLocation =
1358          FormatTok->Tok.getLocation().getLocWithOffset(1);
1359      FormatTok->WhitespaceRange =
1360          SourceRange(GreaterLocation, GreaterLocation);
1361      FormatTok->TokenText = ">";
1362      FormatTok->CodePointCount = 1;
1363      GreaterStashed = false;
1364      return FormatTok;
1365    }
1366
1367    FormatTok = new (Allocator.Allocate()) FormatToken;
1368    readRawToken(*FormatTok);
1369    SourceLocation WhitespaceStart =
1370        FormatTok->Tok.getLocation().getLocWithOffset(-TrailingWhitespace);
1371    if (SourceMgr.getFileOffset(WhitespaceStart) == 0)
1372      FormatTok->IsFirst = true;
1373
1374    // Consume and record whitespace until we find a significant token.
1375    unsigned WhitespaceLength = TrailingWhitespace;
1376    while (FormatTok->Tok.is(tok::unknown)) {
1377      unsigned Newlines = FormatTok->TokenText.count('\n');
1378      if (Newlines > 0)
1379        FormatTok->LastNewlineOffset =
1380            WhitespaceLength + FormatTok->TokenText.rfind('\n') + 1;
1381      FormatTok->NewlinesBefore += Newlines;
1382      unsigned EscapedNewlines = FormatTok->TokenText.count("\\\n");
1383      FormatTok->HasUnescapedNewline |= EscapedNewlines != Newlines;
1384      WhitespaceLength += FormatTok->Tok.getLength();
1385
1386      readRawToken(*FormatTok);
1387    }
1388
1389    // In case the token starts with escaped newlines, we want to
1390    // take them into account as whitespace - this pattern is quite frequent
1391    // in macro definitions.
1392    // FIXME: What do we want to do with other escaped spaces, and escaped
1393    // spaces or newlines in the middle of tokens?
1394    // FIXME: Add a more explicit test.
1395    while (FormatTok->TokenText.size() > 1 && FormatTok->TokenText[0] == '\\' &&
1396           FormatTok->TokenText[1] == '\n') {
1397      // FIXME: ++FormatTok->NewlinesBefore is missing...
1398      WhitespaceLength += 2;
1399      FormatTok->TokenText = FormatTok->TokenText.substr(2);
1400    }
1401
1402    TrailingWhitespace = 0;
1403    if (FormatTok->Tok.is(tok::comment)) {
1404      StringRef UntrimmedText = FormatTok->TokenText;
1405      FormatTok->TokenText = FormatTok->TokenText.rtrim();
1406      TrailingWhitespace = UntrimmedText.size() - FormatTok->TokenText.size();
1407    } else if (FormatTok->Tok.is(tok::raw_identifier)) {
1408      IdentifierInfo &Info = IdentTable.get(FormatTok->TokenText);
1409      FormatTok->Tok.setIdentifierInfo(&Info);
1410      FormatTok->Tok.setKind(Info.getTokenID());
1411    } else if (FormatTok->Tok.is(tok::greatergreater)) {
1412      FormatTok->Tok.setKind(tok::greater);
1413      FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
1414      GreaterStashed = true;
1415    }
1416
1417    // Now FormatTok is the next non-whitespace token.
1418    FormatTok->CodePointCount =
1419        encoding::getCodePointCount(FormatTok->TokenText, Encoding);
1420
1421    FormatTok->WhitespaceRange = SourceRange(
1422        WhitespaceStart, WhitespaceStart.getLocWithOffset(WhitespaceLength));
1423    return FormatTok;
1424  }
1425
1426  FormatToken *FormatTok;
1427  bool GreaterStashed;
1428  unsigned TrailingWhitespace;
1429  Lexer &Lex;
1430  SourceManager &SourceMgr;
1431  IdentifierTable IdentTable;
1432  encoding::Encoding Encoding;
1433  llvm::SpecificBumpPtrAllocator<FormatToken> Allocator;
1434  SmallVector<FormatToken *, 16> Tokens;
1435
1436  void readRawToken(FormatToken &Tok) {
1437    Lex.LexFromRawLexer(Tok.Tok);
1438    Tok.TokenText = StringRef(SourceMgr.getCharacterData(Tok.Tok.getLocation()),
1439                              Tok.Tok.getLength());
1440
1441    // For formatting, treat unterminated string literals like normal string
1442    // literals.
1443    if (Tok.is(tok::unknown) && !Tok.TokenText.empty() &&
1444        Tok.TokenText[0] == '"') {
1445      Tok.Tok.setKind(tok::string_literal);
1446      Tok.IsUnterminatedLiteral = true;
1447    }
1448  }
1449};
1450
1451class Formatter : public UnwrappedLineConsumer {
1452public:
1453  Formatter(const FormatStyle &Style, Lexer &Lex, SourceManager &SourceMgr,
1454            const std::vector<CharSourceRange> &Ranges)
1455      : Style(Style), Lex(Lex), SourceMgr(SourceMgr),
1456        Whitespaces(SourceMgr, Style), Ranges(Ranges),
1457        Encoding(encoding::detectEncoding(Lex.getBuffer())) {
1458    DEBUG(llvm::dbgs() << "File encoding: "
1459                       << (Encoding == encoding::Encoding_UTF8 ? "UTF8"
1460                                                               : "unknown")
1461                       << "\n");
1462  }
1463
1464  virtual ~Formatter() {}
1465
1466  tooling::Replacements format() {
1467    FormatTokenLexer Tokens(Lex, SourceMgr, Encoding);
1468
1469    UnwrappedLineParser Parser(Style, Tokens.lex(), *this);
1470    bool StructuralError = Parser.parse();
1471    TokenAnnotator Annotator(Style, Tokens.getIdentTable().get("in"));
1472    for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1473      Annotator.annotate(AnnotatedLines[i]);
1474    }
1475    deriveLocalStyle();
1476    for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1477      Annotator.calculateFormattingInformation(AnnotatedLines[i]);
1478    }
1479
1480    // Adapt level to the next line if this is a comment.
1481    // FIXME: Can/should this be done in the UnwrappedLineParser?
1482    const AnnotatedLine *NextNonCommentLine = NULL;
1483    for (unsigned i = AnnotatedLines.size() - 1; i > 0; --i) {
1484      if (NextNonCommentLine && AnnotatedLines[i].First->is(tok::comment) &&
1485          !AnnotatedLines[i].First->Next)
1486        AnnotatedLines[i].Level = NextNonCommentLine->Level;
1487      else
1488        NextNonCommentLine = AnnotatedLines[i].First->isNot(tok::r_brace)
1489                                 ? &AnnotatedLines[i]
1490                                 : NULL;
1491    }
1492
1493    std::vector<int> IndentForLevel;
1494    bool PreviousLineWasTouched = false;
1495    const FormatToken *PreviousLineLastToken = 0;
1496    bool FormatPPDirective = false;
1497    for (std::vector<AnnotatedLine>::iterator I = AnnotatedLines.begin(),
1498                                              E = AnnotatedLines.end();
1499         I != E; ++I) {
1500      const AnnotatedLine &TheLine = *I;
1501      const FormatToken *FirstTok = TheLine.First;
1502      int Offset = getIndentOffset(*TheLine.First);
1503
1504      // Check whether this line is part of a formatted preprocessor directive.
1505      if (FirstTok->HasUnescapedNewline)
1506        FormatPPDirective = false;
1507      if (!FormatPPDirective && TheLine.InPPDirective &&
1508          (touchesLine(TheLine) || touchesPPDirective(I + 1, E)))
1509        FormatPPDirective = true;
1510
1511      // Determine indent and try to merge multiple unwrapped lines.
1512      while (IndentForLevel.size() <= TheLine.Level)
1513        IndentForLevel.push_back(-1);
1514      IndentForLevel.resize(TheLine.Level + 1);
1515      unsigned Indent = getIndent(IndentForLevel, TheLine.Level);
1516      if (static_cast<int>(Indent) + Offset >= 0)
1517        Indent += Offset;
1518      tryFitMultipleLinesInOne(Indent, I, E);
1519
1520      bool WasMoved = PreviousLineWasTouched && FirstTok->NewlinesBefore == 0;
1521      if (TheLine.First->is(tok::eof)) {
1522        if (PreviousLineWasTouched) {
1523          unsigned NewLines = std::min(FirstTok->NewlinesBefore, 1u);
1524          Whitespaces.replaceWhitespace(*TheLine.First, NewLines, /*Indent*/ 0,
1525                                        /*TargetColumn*/ 0);
1526        }
1527      } else if (TheLine.Type != LT_Invalid &&
1528                 (WasMoved || FormatPPDirective || touchesLine(TheLine))) {
1529        unsigned LevelIndent = getIndent(IndentForLevel, TheLine.Level);
1530        if (FirstTok->WhitespaceRange.isValid() &&
1531            // Insert a break even if there is a structural error in case where
1532            // we break apart a line consisting of multiple unwrapped lines.
1533            (FirstTok->NewlinesBefore == 0 || !StructuralError)) {
1534          formatFirstToken(*TheLine.First, PreviousLineLastToken, Indent,
1535                           TheLine.InPPDirective);
1536        } else {
1537          Indent = LevelIndent =
1538              SourceMgr.getSpellingColumnNumber(FirstTok->Tok.getLocation()) -
1539              1;
1540        }
1541        UnwrappedLineFormatter Formatter(Style, SourceMgr, TheLine, Indent,
1542                                         TheLine.First, Whitespaces, Encoding,
1543                                         BinPackInconclusiveFunctions);
1544        Formatter.format(I + 1 != E ? &*(I + 1) : NULL);
1545        IndentForLevel[TheLine.Level] = LevelIndent;
1546        PreviousLineWasTouched = true;
1547      } else {
1548        // Format the first token if necessary, and notify the WhitespaceManager
1549        // about the unchanged whitespace.
1550        for (const FormatToken *Tok = TheLine.First; Tok != NULL;
1551             Tok = Tok->Next) {
1552          if (Tok == TheLine.First &&
1553              (Tok->NewlinesBefore > 0 || Tok->IsFirst)) {
1554            unsigned LevelIndent =
1555                SourceMgr.getSpellingColumnNumber(Tok->Tok.getLocation()) - 1;
1556            // Remove trailing whitespace of the previous line if it was
1557            // touched.
1558            if (PreviousLineWasTouched || touchesEmptyLineBefore(TheLine)) {
1559              formatFirstToken(*Tok, PreviousLineLastToken, LevelIndent,
1560                               TheLine.InPPDirective);
1561            } else {
1562              Whitespaces.addUntouchableToken(*Tok, TheLine.InPPDirective);
1563            }
1564
1565            if (static_cast<int>(LevelIndent) - Offset >= 0)
1566              LevelIndent -= Offset;
1567            if (Tok->isNot(tok::comment))
1568              IndentForLevel[TheLine.Level] = LevelIndent;
1569          } else {
1570            Whitespaces.addUntouchableToken(*Tok, TheLine.InPPDirective);
1571          }
1572        }
1573        // If we did not reformat this unwrapped line, the column at the end of
1574        // the last token is unchanged - thus, we can calculate the end of the
1575        // last token.
1576        PreviousLineWasTouched = false;
1577      }
1578      PreviousLineLastToken = I->Last;
1579    }
1580    return Whitespaces.generateReplacements();
1581  }
1582
1583private:
1584  void deriveLocalStyle() {
1585    unsigned CountBoundToVariable = 0;
1586    unsigned CountBoundToType = 0;
1587    bool HasCpp03IncompatibleFormat = false;
1588    bool HasBinPackedFunction = false;
1589    bool HasOnePerLineFunction = false;
1590    for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1591      if (!AnnotatedLines[i].First->Next)
1592        continue;
1593      FormatToken *Tok = AnnotatedLines[i].First->Next;
1594      while (Tok->Next) {
1595        if (Tok->Type == TT_PointerOrReference) {
1596          bool SpacesBefore =
1597              Tok->WhitespaceRange.getBegin() != Tok->WhitespaceRange.getEnd();
1598          bool SpacesAfter = Tok->Next->WhitespaceRange.getBegin() !=
1599                             Tok->Next->WhitespaceRange.getEnd();
1600          if (SpacesBefore && !SpacesAfter)
1601            ++CountBoundToVariable;
1602          else if (!SpacesBefore && SpacesAfter)
1603            ++CountBoundToType;
1604        }
1605
1606        if (Tok->Type == TT_TemplateCloser &&
1607            Tok->Previous->Type == TT_TemplateCloser &&
1608            Tok->WhitespaceRange.getBegin() == Tok->WhitespaceRange.getEnd())
1609          HasCpp03IncompatibleFormat = true;
1610
1611        if (Tok->PackingKind == PPK_BinPacked)
1612          HasBinPackedFunction = true;
1613        if (Tok->PackingKind == PPK_OnePerLine)
1614          HasOnePerLineFunction = true;
1615
1616        Tok = Tok->Next;
1617      }
1618    }
1619    if (Style.DerivePointerBinding) {
1620      if (CountBoundToType > CountBoundToVariable)
1621        Style.PointerBindsToType = true;
1622      else if (CountBoundToType < CountBoundToVariable)
1623        Style.PointerBindsToType = false;
1624    }
1625    if (Style.Standard == FormatStyle::LS_Auto) {
1626      Style.Standard = HasCpp03IncompatibleFormat ? FormatStyle::LS_Cpp11
1627                                                  : FormatStyle::LS_Cpp03;
1628    }
1629    BinPackInconclusiveFunctions =
1630        HasBinPackedFunction || !HasOnePerLineFunction;
1631  }
1632
1633  /// \brief Get the indent of \p Level from \p IndentForLevel.
1634  ///
1635  /// \p IndentForLevel must contain the indent for the level \c l
1636  /// at \p IndentForLevel[l], or a value < 0 if the indent for
1637  /// that level is unknown.
1638  unsigned getIndent(const std::vector<int> IndentForLevel, unsigned Level) {
1639    if (IndentForLevel[Level] != -1)
1640      return IndentForLevel[Level];
1641    if (Level == 0)
1642      return 0;
1643    return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
1644  }
1645
1646  /// \brief Get the offset of the line relatively to the level.
1647  ///
1648  /// For example, 'public:' labels in classes are offset by 1 or 2
1649  /// characters to the left from their level.
1650  int getIndentOffset(const FormatToken &RootToken) {
1651    if (RootToken.isAccessSpecifier(false) || RootToken.isObjCAccessSpecifier())
1652      return Style.AccessModifierOffset;
1653    return 0;
1654  }
1655
1656  /// \brief Tries to merge lines into one.
1657  ///
1658  /// This will change \c Line and \c AnnotatedLine to contain the merged line,
1659  /// if possible; note that \c I will be incremented when lines are merged.
1660  void tryFitMultipleLinesInOne(unsigned Indent,
1661                                std::vector<AnnotatedLine>::iterator &I,
1662                                std::vector<AnnotatedLine>::iterator E) {
1663    // We can never merge stuff if there are trailing line comments.
1664    if (I->Last->Type == TT_LineComment)
1665      return;
1666
1667    if (Indent > Style.ColumnLimit)
1668      return;
1669
1670    unsigned Limit = Style.ColumnLimit - Indent;
1671    // If we already exceed the column limit, we set 'Limit' to 0. The different
1672    // tryMerge..() functions can then decide whether to still do merging.
1673    Limit = I->Last->TotalLength > Limit ? 0 : Limit - I->Last->TotalLength;
1674
1675    if (I + 1 == E || (I + 1)->Type == LT_Invalid)
1676      return;
1677
1678    if (I->Last->is(tok::l_brace)) {
1679      tryMergeSimpleBlock(I, E, Limit);
1680    } else if (Style.AllowShortIfStatementsOnASingleLine &&
1681               I->First->is(tok::kw_if)) {
1682      tryMergeSimpleControlStatement(I, E, Limit);
1683    } else if (Style.AllowShortLoopsOnASingleLine &&
1684               I->First->isOneOf(tok::kw_for, tok::kw_while)) {
1685      tryMergeSimpleControlStatement(I, E, Limit);
1686    } else if (I->InPPDirective &&
1687               (I->First->HasUnescapedNewline || I->First->IsFirst)) {
1688      tryMergeSimplePPDirective(I, E, Limit);
1689    }
1690  }
1691
1692  void tryMergeSimplePPDirective(std::vector<AnnotatedLine>::iterator &I,
1693                                 std::vector<AnnotatedLine>::iterator E,
1694                                 unsigned Limit) {
1695    if (Limit == 0)
1696      return;
1697    AnnotatedLine &Line = *I;
1698    if (!(I + 1)->InPPDirective || (I + 1)->First->HasUnescapedNewline)
1699      return;
1700    if (I + 2 != E && (I + 2)->InPPDirective &&
1701        !(I + 2)->First->HasUnescapedNewline)
1702      return;
1703    if (1 + (I + 1)->Last->TotalLength > Limit)
1704      return;
1705    join(Line, *(++I));
1706  }
1707
1708  void tryMergeSimpleControlStatement(std::vector<AnnotatedLine>::iterator &I,
1709                                      std::vector<AnnotatedLine>::iterator E,
1710                                      unsigned Limit) {
1711    if (Limit == 0)
1712      return;
1713    if (Style.BreakBeforeBraces == FormatStyle::BS_Allman &&
1714        (I + 1)->First->is(tok::l_brace))
1715      return;
1716    if ((I + 1)->InPPDirective != I->InPPDirective ||
1717        ((I + 1)->InPPDirective && (I + 1)->First->HasUnescapedNewline))
1718      return;
1719    AnnotatedLine &Line = *I;
1720    if (Line.Last->isNot(tok::r_paren))
1721      return;
1722    if (1 + (I + 1)->Last->TotalLength > Limit)
1723      return;
1724    if ((I + 1)->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for,
1725                                tok::kw_while) ||
1726        (I + 1)->First->Type == TT_LineComment)
1727      return;
1728    // Only inline simple if's (no nested if or else).
1729    if (I + 2 != E && Line.First->is(tok::kw_if) &&
1730        (I + 2)->First->is(tok::kw_else))
1731      return;
1732    join(Line, *(++I));
1733  }
1734
1735  void tryMergeSimpleBlock(std::vector<AnnotatedLine>::iterator &I,
1736                           std::vector<AnnotatedLine>::iterator E,
1737                           unsigned Limit) {
1738    // No merging if the brace already is on the next line.
1739    if (Style.BreakBeforeBraces != FormatStyle::BS_Attach)
1740      return;
1741
1742    // First, check that the current line allows merging. This is the case if
1743    // we're not in a control flow statement and the last token is an opening
1744    // brace.
1745    AnnotatedLine &Line = *I;
1746    if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::r_brace,
1747                            tok::kw_else, tok::kw_try, tok::kw_catch,
1748                            tok::kw_for,
1749                            // This gets rid of all ObjC @ keywords and methods.
1750                            tok::at, tok::minus, tok::plus))
1751      return;
1752
1753    FormatToken *Tok = (I + 1)->First;
1754    if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
1755        (Tok->getNextNonComment() == NULL ||
1756         Tok->getNextNonComment()->is(tok::semi))) {
1757      // We merge empty blocks even if the line exceeds the column limit.
1758      Tok->SpacesRequiredBefore = 0;
1759      Tok->CanBreakBefore = true;
1760      join(Line, *(I + 1));
1761      I += 1;
1762    } else if (Limit != 0 && Line.First->isNot(tok::kw_namespace)) {
1763      // Check that we still have three lines and they fit into the limit.
1764      if (I + 2 == E || (I + 2)->Type == LT_Invalid ||
1765          !nextTwoLinesFitInto(I, Limit))
1766        return;
1767
1768      // Second, check that the next line does not contain any braces - if it
1769      // does, readability declines when putting it into a single line.
1770      if ((I + 1)->Last->Type == TT_LineComment || Tok->MustBreakBefore)
1771        return;
1772      do {
1773        if (Tok->isOneOf(tok::l_brace, tok::r_brace))
1774          return;
1775        Tok = Tok->Next;
1776      } while (Tok != NULL);
1777
1778      // Last, check that the third line contains a single closing brace.
1779      Tok = (I + 2)->First;
1780      if (Tok->getNextNonComment() != NULL || Tok->isNot(tok::r_brace) ||
1781          Tok->MustBreakBefore)
1782        return;
1783
1784      join(Line, *(I + 1));
1785      join(Line, *(I + 2));
1786      I += 2;
1787    }
1788  }
1789
1790  bool nextTwoLinesFitInto(std::vector<AnnotatedLine>::iterator I,
1791                           unsigned Limit) {
1792    return 1 + (I + 1)->Last->TotalLength + 1 + (I + 2)->Last->TotalLength <=
1793           Limit;
1794  }
1795
1796  void join(AnnotatedLine &A, const AnnotatedLine &B) {
1797    assert(!A.Last->Next);
1798    assert(!B.First->Previous);
1799    A.Last->Next = B.First;
1800    B.First->Previous = A.Last;
1801    unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
1802    for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
1803      Tok->TotalLength += LengthA;
1804      A.Last = Tok;
1805    }
1806  }
1807
1808  bool touchesRanges(const CharSourceRange &Range) {
1809    for (unsigned i = 0, e = Ranges.size(); i != e; ++i) {
1810      if (!SourceMgr.isBeforeInTranslationUnit(Range.getEnd(),
1811                                               Ranges[i].getBegin()) &&
1812          !SourceMgr.isBeforeInTranslationUnit(Ranges[i].getEnd(),
1813                                               Range.getBegin()))
1814        return true;
1815    }
1816    return false;
1817  }
1818
1819  bool touchesLine(const AnnotatedLine &TheLine) {
1820    const FormatToken *First = TheLine.First;
1821    const FormatToken *Last = TheLine.Last;
1822    CharSourceRange LineRange = CharSourceRange::getCharRange(
1823        First->WhitespaceRange.getBegin().getLocWithOffset(
1824            First->LastNewlineOffset),
1825        Last->Tok.getLocation().getLocWithOffset(Last->TokenText.size() - 1));
1826    return touchesRanges(LineRange);
1827  }
1828
1829  bool touchesPPDirective(std::vector<AnnotatedLine>::iterator I,
1830                          std::vector<AnnotatedLine>::iterator E) {
1831    for (; I != E; ++I) {
1832      if (I->First->HasUnescapedNewline)
1833        return false;
1834      if (touchesLine(*I))
1835        return true;
1836    }
1837    return false;
1838  }
1839
1840  bool touchesEmptyLineBefore(const AnnotatedLine &TheLine) {
1841    const FormatToken *First = TheLine.First;
1842    CharSourceRange LineRange = CharSourceRange::getCharRange(
1843        First->WhitespaceRange.getBegin(),
1844        First->WhitespaceRange.getBegin().getLocWithOffset(
1845            First->LastNewlineOffset));
1846    return touchesRanges(LineRange);
1847  }
1848
1849  virtual void consumeUnwrappedLine(const UnwrappedLine &TheLine) {
1850    AnnotatedLines.push_back(AnnotatedLine(TheLine));
1851  }
1852
1853  /// \brief Add a new line and the required indent before the first Token
1854  /// of the \c UnwrappedLine if there was no structural parsing error.
1855  /// Returns the indent level of the \c UnwrappedLine.
1856  void formatFirstToken(const FormatToken &RootToken,
1857                        const FormatToken *PreviousToken, unsigned Indent,
1858                        bool InPPDirective) {
1859    unsigned Newlines =
1860        std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
1861    // Remove empty lines before "}" where applicable.
1862    if (RootToken.is(tok::r_brace) &&
1863        (!RootToken.Next ||
1864         (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)))
1865      Newlines = std::min(Newlines, 1u);
1866    if (Newlines == 0 && !RootToken.IsFirst)
1867      Newlines = 1;
1868
1869    // Insert extra new line before access specifiers.
1870    if (PreviousToken && PreviousToken->isOneOf(tok::semi, tok::r_brace) &&
1871        RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
1872      ++Newlines;
1873
1874    Whitespaces.replaceWhitespace(
1875        RootToken, Newlines, Indent, Indent,
1876        InPPDirective && !RootToken.HasUnescapedNewline);
1877  }
1878
1879  FormatStyle Style;
1880  Lexer &Lex;
1881  SourceManager &SourceMgr;
1882  WhitespaceManager Whitespaces;
1883  std::vector<CharSourceRange> Ranges;
1884  std::vector<AnnotatedLine> AnnotatedLines;
1885
1886  encoding::Encoding Encoding;
1887  bool BinPackInconclusiveFunctions;
1888};
1889
1890} // end anonymous namespace
1891
1892tooling::Replacements reformat(const FormatStyle &Style, Lexer &Lex,
1893                               SourceManager &SourceMgr,
1894                               std::vector<CharSourceRange> Ranges) {
1895  Formatter formatter(Style, Lex, SourceMgr, Ranges);
1896  return formatter.format();
1897}
1898
1899tooling::Replacements reformat(const FormatStyle &Style, StringRef Code,
1900                               std::vector<tooling::Range> Ranges,
1901                               StringRef FileName) {
1902  FileManager Files((FileSystemOptions()));
1903  DiagnosticsEngine Diagnostics(
1904      IntrusiveRefCntPtr<DiagnosticIDs>(new DiagnosticIDs),
1905      new DiagnosticOptions);
1906  SourceManager SourceMgr(Diagnostics, Files);
1907  llvm::MemoryBuffer *Buf = llvm::MemoryBuffer::getMemBuffer(Code, FileName);
1908  const clang::FileEntry *Entry =
1909      Files.getVirtualFile(FileName, Buf->getBufferSize(), 0);
1910  SourceMgr.overrideFileContents(Entry, Buf);
1911  FileID ID =
1912      SourceMgr.createFileID(Entry, SourceLocation(), clang::SrcMgr::C_User);
1913  Lexer Lex(ID, SourceMgr.getBuffer(ID), SourceMgr,
1914            getFormattingLangOpts(Style.Standard));
1915  SourceLocation StartOfFile = SourceMgr.getLocForStartOfFile(ID);
1916  std::vector<CharSourceRange> CharRanges;
1917  for (unsigned i = 0, e = Ranges.size(); i != e; ++i) {
1918    SourceLocation Start = StartOfFile.getLocWithOffset(Ranges[i].getOffset());
1919    SourceLocation End = Start.getLocWithOffset(Ranges[i].getLength());
1920    CharRanges.push_back(CharSourceRange::getCharRange(Start, End));
1921  }
1922  return reformat(Style, Lex, SourceMgr, CharRanges);
1923}
1924
1925LangOptions getFormattingLangOpts(FormatStyle::LanguageStandard Standard) {
1926  LangOptions LangOpts;
1927  LangOpts.CPlusPlus = 1;
1928  LangOpts.CPlusPlus11 = Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
1929  LangOpts.LineComment = 1;
1930  LangOpts.Bool = 1;
1931  LangOpts.ObjC1 = 1;
1932  LangOpts.ObjC2 = 1;
1933  return LangOpts;
1934}
1935
1936} // namespace format
1937} // namespace clang
1938