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 "ContinuationIndenter.h"
19 #include "TokenAnnotator.h"
20 #include "UnwrappedLineParser.h"
21 #include "WhitespaceManager.h"
22 #include "clang/Basic/Diagnostic.h"
23 #include "clang/Basic/SourceManager.h"
24 #include "clang/Format/Format.h"
25 #include "clang/Lex/Lexer.h"
26 #include "llvm/ADT/STLExtras.h"
27 #include "llvm/Support/Allocator.h"
28 #include "llvm/Support/Debug.h"
29 #include "llvm/Support/YAMLTraits.h"
30 #include "llvm/Support/Path.h"
31 #include <queue>
32 #include <string>
33 
34 using clang::format::FormatStyle;
35 
36 namespace llvm {
37 namespace yaml {
38 template <> struct ScalarEnumerationTraits<FormatStyle::LanguageKind> {
39   static void enumeration(IO &IO, FormatStyle::LanguageKind &Value) {
40     IO.enumCase(Value, "Cpp", FormatStyle::LK_Cpp);
41     IO.enumCase(Value, "JavaScript", FormatStyle::LK_JavaScript);
42   }
43 };
44 
45 template <> struct ScalarEnumerationTraits<FormatStyle::LanguageStandard> {
46   static void enumeration(IO &IO, FormatStyle::LanguageStandard &Value) {
47     IO.enumCase(Value, "Cpp03", FormatStyle::LS_Cpp03);
48     IO.enumCase(Value, "C++03", FormatStyle::LS_Cpp03);
49     IO.enumCase(Value, "Cpp11", FormatStyle::LS_Cpp11);
50     IO.enumCase(Value, "C++11", FormatStyle::LS_Cpp11);
51     IO.enumCase(Value, "Auto", FormatStyle::LS_Auto);
52   }
53 };
54 
55 template <> struct ScalarEnumerationTraits<FormatStyle::UseTabStyle> {
56   static void enumeration(IO &IO, FormatStyle::UseTabStyle &Value) {
57     IO.enumCase(Value, "Never", FormatStyle::UT_Never);
58     IO.enumCase(Value, "false", FormatStyle::UT_Never);
59     IO.enumCase(Value, "Always", FormatStyle::UT_Always);
60     IO.enumCase(Value, "true", FormatStyle::UT_Always);
61     IO.enumCase(Value, "ForIndentation", FormatStyle::UT_ForIndentation);
62   }
63 };
64 
65 template <> struct ScalarEnumerationTraits<FormatStyle::BraceBreakingStyle> {
66   static void enumeration(IO &IO, FormatStyle::BraceBreakingStyle &Value) {
67     IO.enumCase(Value, "Attach", FormatStyle::BS_Attach);
68     IO.enumCase(Value, "Linux", FormatStyle::BS_Linux);
69     IO.enumCase(Value, "Stroustrup", FormatStyle::BS_Stroustrup);
70     IO.enumCase(Value, "Allman", FormatStyle::BS_Allman);
71     IO.enumCase(Value, "GNU", FormatStyle::BS_GNU);
72   }
73 };
74 
75 template <>
76 struct ScalarEnumerationTraits<FormatStyle::NamespaceIndentationKind> {
77   static void enumeration(IO &IO,
78                           FormatStyle::NamespaceIndentationKind &Value) {
79     IO.enumCase(Value, "None", FormatStyle::NI_None);
80     IO.enumCase(Value, "Inner", FormatStyle::NI_Inner);
81     IO.enumCase(Value, "All", FormatStyle::NI_All);
82   }
83 };
84 
85 template <>
86 struct ScalarEnumerationTraits<FormatStyle::SpaceBeforeParensOptions> {
87   static void enumeration(IO &IO,
88                           FormatStyle::SpaceBeforeParensOptions &Value) {
89     IO.enumCase(Value, "Never", FormatStyle::SBPO_Never);
90     IO.enumCase(Value, "ControlStatements",
91                 FormatStyle::SBPO_ControlStatements);
92     IO.enumCase(Value, "Always", FormatStyle::SBPO_Always);
93 
94     // For backward compatibility.
95     IO.enumCase(Value, "false", FormatStyle::SBPO_Never);
96     IO.enumCase(Value, "true", FormatStyle::SBPO_ControlStatements);
97   }
98 };
99 
100 template <> struct MappingTraits<FormatStyle> {
101   static void mapping(IO &IO, FormatStyle &Style) {
102     // When reading, read the language first, we need it for getPredefinedStyle.
103     IO.mapOptional("Language", Style.Language);
104 
105     if (IO.outputting()) {
106       StringRef StylesArray[] = { "LLVM",    "Google", "Chromium",
107                                   "Mozilla", "WebKit", "GNU" };
108       ArrayRef<StringRef> Styles(StylesArray);
109       for (size_t i = 0, e = Styles.size(); i < e; ++i) {
110         StringRef StyleName(Styles[i]);
111         FormatStyle PredefinedStyle;
112         if (getPredefinedStyle(StyleName, Style.Language, &PredefinedStyle) &&
113             Style == PredefinedStyle) {
114           IO.mapOptional("# BasedOnStyle", StyleName);
115           break;
116         }
117       }
118     } else {
119       StringRef BasedOnStyle;
120       IO.mapOptional("BasedOnStyle", BasedOnStyle);
121       if (!BasedOnStyle.empty()) {
122         FormatStyle::LanguageKind OldLanguage = Style.Language;
123         FormatStyle::LanguageKind Language =
124             ((FormatStyle *)IO.getContext())->Language;
125         if (!getPredefinedStyle(BasedOnStyle, Language, &Style)) {
126           IO.setError(Twine("Unknown value for BasedOnStyle: ", BasedOnStyle));
127           return;
128         }
129         Style.Language = OldLanguage;
130       }
131     }
132 
133     IO.mapOptional("AccessModifierOffset", Style.AccessModifierOffset);
134     IO.mapOptional("ConstructorInitializerIndentWidth",
135                    Style.ConstructorInitializerIndentWidth);
136     IO.mapOptional("AlignEscapedNewlinesLeft", Style.AlignEscapedNewlinesLeft);
137     IO.mapOptional("AlignTrailingComments", Style.AlignTrailingComments);
138     IO.mapOptional("AllowAllParametersOfDeclarationOnNextLine",
139                    Style.AllowAllParametersOfDeclarationOnNextLine);
140     IO.mapOptional("AllowShortIfStatementsOnASingleLine",
141                    Style.AllowShortIfStatementsOnASingleLine);
142     IO.mapOptional("AllowShortLoopsOnASingleLine",
143                    Style.AllowShortLoopsOnASingleLine);
144     IO.mapOptional("AllowShortFunctionsOnASingleLine",
145                    Style.AllowShortFunctionsOnASingleLine);
146     IO.mapOptional("AlwaysBreakTemplateDeclarations",
147                    Style.AlwaysBreakTemplateDeclarations);
148     IO.mapOptional("AlwaysBreakBeforeMultilineStrings",
149                    Style.AlwaysBreakBeforeMultilineStrings);
150     IO.mapOptional("BreakBeforeBinaryOperators",
151                    Style.BreakBeforeBinaryOperators);
152     IO.mapOptional("BreakBeforeTernaryOperators",
153                    Style.BreakBeforeTernaryOperators);
154     IO.mapOptional("BreakConstructorInitializersBeforeComma",
155                    Style.BreakConstructorInitializersBeforeComma);
156     IO.mapOptional("BinPackParameters", Style.BinPackParameters);
157     IO.mapOptional("ColumnLimit", Style.ColumnLimit);
158     IO.mapOptional("ConstructorInitializerAllOnOneLineOrOnePerLine",
159                    Style.ConstructorInitializerAllOnOneLineOrOnePerLine);
160     IO.mapOptional("DerivePointerBinding", Style.DerivePointerBinding);
161     IO.mapOptional("ExperimentalAutoDetectBinPacking",
162                    Style.ExperimentalAutoDetectBinPacking);
163     IO.mapOptional("IndentCaseLabels", Style.IndentCaseLabels);
164     IO.mapOptional("MaxEmptyLinesToKeep", Style.MaxEmptyLinesToKeep);
165     IO.mapOptional("NamespaceIndentation", Style.NamespaceIndentation);
166     IO.mapOptional("ObjCSpaceBeforeProtocolList",
167                    Style.ObjCSpaceBeforeProtocolList);
168     IO.mapOptional("PenaltyBreakBeforeFirstCallParameter",
169                    Style.PenaltyBreakBeforeFirstCallParameter);
170     IO.mapOptional("PenaltyBreakComment", Style.PenaltyBreakComment);
171     IO.mapOptional("PenaltyBreakString", Style.PenaltyBreakString);
172     IO.mapOptional("PenaltyBreakFirstLessLess",
173                    Style.PenaltyBreakFirstLessLess);
174     IO.mapOptional("PenaltyExcessCharacter", Style.PenaltyExcessCharacter);
175     IO.mapOptional("PenaltyReturnTypeOnItsOwnLine",
176                    Style.PenaltyReturnTypeOnItsOwnLine);
177     IO.mapOptional("PointerBindsToType", Style.PointerBindsToType);
178     IO.mapOptional("SpacesBeforeTrailingComments",
179                    Style.SpacesBeforeTrailingComments);
180     IO.mapOptional("Cpp11BracedListStyle", Style.Cpp11BracedListStyle);
181     IO.mapOptional("Standard", Style.Standard);
182     IO.mapOptional("IndentWidth", Style.IndentWidth);
183     IO.mapOptional("TabWidth", Style.TabWidth);
184     IO.mapOptional("UseTab", Style.UseTab);
185     IO.mapOptional("BreakBeforeBraces", Style.BreakBeforeBraces);
186     IO.mapOptional("IndentFunctionDeclarationAfterType",
187                    Style.IndentFunctionDeclarationAfterType);
188     IO.mapOptional("SpacesInParentheses", Style.SpacesInParentheses);
189     IO.mapOptional("SpacesInAngles", Style.SpacesInAngles);
190     IO.mapOptional("SpaceInEmptyParentheses", Style.SpaceInEmptyParentheses);
191     IO.mapOptional("SpacesInCStyleCastParentheses",
192                    Style.SpacesInCStyleCastParentheses);
193     IO.mapOptional("SpaceBeforeAssignmentOperators",
194                    Style.SpaceBeforeAssignmentOperators);
195     IO.mapOptional("ContinuationIndentWidth", Style.ContinuationIndentWidth);
196 
197     // For backward compatibility.
198     if (!IO.outputting()) {
199       IO.mapOptional("SpaceAfterControlStatementKeyword",
200                      Style.SpaceBeforeParens);
201     }
202     IO.mapOptional("SpaceBeforeParens", Style.SpaceBeforeParens);
203   }
204 };
205 
206 // Allows to read vector<FormatStyle> while keeping default values.
207 // IO.getContext() should contain a pointer to the FormatStyle structure, that
208 // will be used to get default values for missing keys.
209 // If the first element has no Language specified, it will be treated as the
210 // default one for the following elements.
211 template <> struct DocumentListTraits<std::vector<FormatStyle> > {
212   static size_t size(IO &IO, std::vector<FormatStyle> &Seq) {
213     return Seq.size();
214   }
215   static FormatStyle &element(IO &IO, std::vector<FormatStyle> &Seq,
216                               size_t Index) {
217     if (Index >= Seq.size()) {
218       assert(Index == Seq.size());
219       FormatStyle Template;
220       if (Seq.size() > 0 && Seq[0].Language == FormatStyle::LK_None) {
221         Template = Seq[0];
222       } else {
223         Template = *((const FormatStyle*)IO.getContext());
224         Template.Language = FormatStyle::LK_None;
225       }
226       Seq.resize(Index + 1, Template);
227     }
228     return Seq[Index];
229   }
230 };
231 }
232 }
233 
234 namespace clang {
235 namespace format {
236 
237 FormatStyle getLLVMStyle() {
238   FormatStyle LLVMStyle;
239   LLVMStyle.Language = FormatStyle::LK_Cpp;
240   LLVMStyle.AccessModifierOffset = -2;
241   LLVMStyle.AlignEscapedNewlinesLeft = false;
242   LLVMStyle.AlignTrailingComments = true;
243   LLVMStyle.AllowAllParametersOfDeclarationOnNextLine = true;
244   LLVMStyle.AllowShortFunctionsOnASingleLine = true;
245   LLVMStyle.AllowShortIfStatementsOnASingleLine = false;
246   LLVMStyle.AllowShortLoopsOnASingleLine = false;
247   LLVMStyle.AlwaysBreakBeforeMultilineStrings = false;
248   LLVMStyle.AlwaysBreakTemplateDeclarations = false;
249   LLVMStyle.BinPackParameters = true;
250   LLVMStyle.BreakBeforeBinaryOperators = false;
251   LLVMStyle.BreakBeforeTernaryOperators = true;
252   LLVMStyle.BreakBeforeBraces = FormatStyle::BS_Attach;
253   LLVMStyle.BreakConstructorInitializersBeforeComma = false;
254   LLVMStyle.ColumnLimit = 80;
255   LLVMStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = false;
256   LLVMStyle.ConstructorInitializerIndentWidth = 4;
257   LLVMStyle.Cpp11BracedListStyle = false;
258   LLVMStyle.DerivePointerBinding = false;
259   LLVMStyle.ExperimentalAutoDetectBinPacking = false;
260   LLVMStyle.IndentCaseLabels = false;
261   LLVMStyle.IndentFunctionDeclarationAfterType = false;
262   LLVMStyle.IndentWidth = 2;
263   LLVMStyle.TabWidth = 8;
264   LLVMStyle.MaxEmptyLinesToKeep = 1;
265   LLVMStyle.NamespaceIndentation = FormatStyle::NI_None;
266   LLVMStyle.ObjCSpaceBeforeProtocolList = true;
267   LLVMStyle.PointerBindsToType = false;
268   LLVMStyle.SpacesBeforeTrailingComments = 1;
269   LLVMStyle.Standard = FormatStyle::LS_Cpp03;
270   LLVMStyle.UseTab = FormatStyle::UT_Never;
271   LLVMStyle.SpacesInParentheses = false;
272   LLVMStyle.SpaceInEmptyParentheses = false;
273   LLVMStyle.SpacesInCStyleCastParentheses = false;
274   LLVMStyle.SpaceBeforeParens = FormatStyle::SBPO_ControlStatements;
275   LLVMStyle.SpaceBeforeAssignmentOperators = true;
276   LLVMStyle.ContinuationIndentWidth = 4;
277   LLVMStyle.SpacesInAngles = false;
278 
279   LLVMStyle.PenaltyBreakComment = 300;
280   LLVMStyle.PenaltyBreakFirstLessLess = 120;
281   LLVMStyle.PenaltyBreakString = 1000;
282   LLVMStyle.PenaltyExcessCharacter = 1000000;
283   LLVMStyle.PenaltyReturnTypeOnItsOwnLine = 60;
284   LLVMStyle.PenaltyBreakBeforeFirstCallParameter = 19;
285 
286   return LLVMStyle;
287 }
288 
289 FormatStyle getGoogleStyle() {
290   FormatStyle GoogleStyle = getLLVMStyle();
291   GoogleStyle.AccessModifierOffset = -1;
292   GoogleStyle.AlignEscapedNewlinesLeft = true;
293   GoogleStyle.AllowShortIfStatementsOnASingleLine = true;
294   GoogleStyle.AllowShortLoopsOnASingleLine = true;
295   GoogleStyle.AlwaysBreakBeforeMultilineStrings = true;
296   GoogleStyle.AlwaysBreakTemplateDeclarations = true;
297   GoogleStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = true;
298   GoogleStyle.Cpp11BracedListStyle = true;
299   GoogleStyle.DerivePointerBinding = true;
300   GoogleStyle.IndentCaseLabels = true;
301   GoogleStyle.IndentFunctionDeclarationAfterType = true;
302   GoogleStyle.ObjCSpaceBeforeProtocolList = false;
303   GoogleStyle.PointerBindsToType = true;
304   GoogleStyle.SpacesBeforeTrailingComments = 2;
305   GoogleStyle.Standard = FormatStyle::LS_Auto;
306 
307   GoogleStyle.PenaltyReturnTypeOnItsOwnLine = 200;
308   GoogleStyle.PenaltyBreakBeforeFirstCallParameter = 1;
309 
310   return GoogleStyle;
311 }
312 
313 FormatStyle getGoogleJSStyle() {
314   FormatStyle GoogleJSStyle = getGoogleStyle();
315   GoogleJSStyle.Language = FormatStyle::LK_JavaScript;
316   GoogleJSStyle.BreakBeforeTernaryOperators = false;
317   // FIXME: Currently unimplemented:
318   // var arr = [1, 2, 3];  // No space after [ or before ].
319   // var obj = {a: 1, b: 2, c: 3};  // No space after ':'.
320   return GoogleJSStyle;
321 }
322 
323 FormatStyle getChromiumStyle() {
324   FormatStyle ChromiumStyle = getGoogleStyle();
325   ChromiumStyle.AllowAllParametersOfDeclarationOnNextLine = false;
326   ChromiumStyle.AllowShortIfStatementsOnASingleLine = false;
327   ChromiumStyle.AllowShortLoopsOnASingleLine = false;
328   ChromiumStyle.BinPackParameters = false;
329   ChromiumStyle.DerivePointerBinding = false;
330   ChromiumStyle.Standard = FormatStyle::LS_Cpp03;
331   return ChromiumStyle;
332 }
333 
334 FormatStyle getMozillaStyle() {
335   FormatStyle MozillaStyle = getLLVMStyle();
336   MozillaStyle.AllowAllParametersOfDeclarationOnNextLine = false;
337   MozillaStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = true;
338   MozillaStyle.DerivePointerBinding = true;
339   MozillaStyle.IndentCaseLabels = true;
340   MozillaStyle.ObjCSpaceBeforeProtocolList = false;
341   MozillaStyle.PenaltyReturnTypeOnItsOwnLine = 200;
342   MozillaStyle.PointerBindsToType = true;
343   return MozillaStyle;
344 }
345 
346 FormatStyle getWebKitStyle() {
347   FormatStyle Style = getLLVMStyle();
348   Style.AccessModifierOffset = -4;
349   Style.AlignTrailingComments = false;
350   Style.BreakBeforeBinaryOperators = true;
351   Style.BreakBeforeBraces = FormatStyle::BS_Stroustrup;
352   Style.BreakConstructorInitializersBeforeComma = true;
353   Style.ColumnLimit = 0;
354   Style.IndentWidth = 4;
355   Style.NamespaceIndentation = FormatStyle::NI_Inner;
356   Style.PointerBindsToType = true;
357   return Style;
358 }
359 
360 FormatStyle getGNUStyle() {
361   FormatStyle Style = getLLVMStyle();
362   Style.BreakBeforeBinaryOperators = true;
363   Style.BreakBeforeBraces = FormatStyle::BS_GNU;
364   Style.BreakBeforeTernaryOperators = true;
365   Style.ColumnLimit = 79;
366   Style.SpaceBeforeParens = FormatStyle::SBPO_Always;
367   return Style;
368 }
369 
370 bool getPredefinedStyle(StringRef Name, FormatStyle::LanguageKind Language,
371                         FormatStyle *Style) {
372   if (Name.equals_lower("llvm")) {
373     *Style = getLLVMStyle();
374   } else if (Name.equals_lower("chromium")) {
375     *Style = getChromiumStyle();
376   } else if (Name.equals_lower("mozilla")) {
377     *Style = getMozillaStyle();
378   } else if (Name.equals_lower("google")) {
379     *Style = Language == FormatStyle::LK_JavaScript ? getGoogleJSStyle()
380                                                     : getGoogleStyle();
381   } else if (Name.equals_lower("webkit")) {
382     *Style = getWebKitStyle();
383   } else if (Name.equals_lower("gnu")) {
384     *Style = getGNUStyle();
385   } else {
386     return false;
387   }
388 
389   Style->Language = Language;
390   return true;
391 }
392 
393 llvm::error_code parseConfiguration(StringRef Text, FormatStyle *Style) {
394   assert(Style);
395   FormatStyle::LanguageKind Language = Style->Language;
396   assert(Language != FormatStyle::LK_None);
397   if (Text.trim().empty())
398     return llvm::make_error_code(llvm::errc::invalid_argument);
399 
400   std::vector<FormatStyle> Styles;
401   llvm::yaml::Input Input(Text);
402   // DocumentListTraits<vector<FormatStyle>> uses the context to get default
403   // values for the fields, keys for which are missing from the configuration.
404   // Mapping also uses the context to get the language to find the correct
405   // base style.
406   Input.setContext(Style);
407   Input >> Styles;
408   if (Input.error())
409     return Input.error();
410 
411   for (unsigned i = 0; i < Styles.size(); ++i) {
412     // Ensures that only the first configuration can skip the Language option.
413     if (Styles[i].Language == FormatStyle::LK_None && i != 0)
414       return llvm::make_error_code(llvm::errc::invalid_argument);
415     // Ensure that each language is configured at most once.
416     for (unsigned j = 0; j < i; ++j) {
417       if (Styles[i].Language == Styles[j].Language) {
418         DEBUG(llvm::dbgs()
419               << "Duplicate languages in the config file on positions " << j
420               << " and " << i << "\n");
421         return llvm::make_error_code(llvm::errc::invalid_argument);
422       }
423     }
424   }
425   // Look for a suitable configuration starting from the end, so we can
426   // find the configuration for the specific language first, and the default
427   // configuration (which can only be at slot 0) after it.
428   for (int i = Styles.size() - 1; i >= 0; --i) {
429     if (Styles[i].Language == Language ||
430         Styles[i].Language == FormatStyle::LK_None) {
431       *Style = Styles[i];
432       Style->Language = Language;
433       return llvm::make_error_code(llvm::errc::success);
434     }
435   }
436   return llvm::make_error_code(llvm::errc::not_supported);
437 }
438 
439 std::string configurationAsText(const FormatStyle &Style) {
440   std::string Text;
441   llvm::raw_string_ostream Stream(Text);
442   llvm::yaml::Output Output(Stream);
443   // We use the same mapping method for input and output, so we need a non-const
444   // reference here.
445   FormatStyle NonConstStyle = Style;
446   Output << NonConstStyle;
447   return Stream.str();
448 }
449 
450 namespace {
451 
452 class NoColumnLimitFormatter {
453 public:
454   NoColumnLimitFormatter(ContinuationIndenter *Indenter) : Indenter(Indenter) {}
455 
456   /// \brief Formats the line starting at \p State, simply keeping all of the
457   /// input's line breaking decisions.
458   void format(unsigned FirstIndent, const AnnotatedLine *Line) {
459     LineState State =
460         Indenter->getInitialState(FirstIndent, Line, /*DryRun=*/false);
461     while (State.NextToken != NULL) {
462       bool Newline =
463           Indenter->mustBreak(State) ||
464           (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
465       Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
466     }
467   }
468 
469 private:
470   ContinuationIndenter *Indenter;
471 };
472 
473 class LineJoiner {
474 public:
475   LineJoiner(const FormatStyle &Style) : Style(Style) {}
476 
477   /// \brief Calculates how many lines can be merged into 1 starting at \p I.
478   unsigned
479   tryFitMultipleLinesInOne(unsigned Indent,
480                            SmallVectorImpl<AnnotatedLine *>::const_iterator I,
481                            SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
482     // We can never merge stuff if there are trailing line comments.
483     AnnotatedLine *TheLine = *I;
484     if (TheLine->Last->Type == TT_LineComment)
485       return 0;
486 
487     if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
488       return 0;
489 
490     unsigned Limit =
491         Style.ColumnLimit == 0 ? UINT_MAX : Style.ColumnLimit - Indent;
492     // If we already exceed the column limit, we set 'Limit' to 0. The different
493     // tryMerge..() functions can then decide whether to still do merging.
494     Limit = TheLine->Last->TotalLength > Limit
495                 ? 0
496                 : Limit - TheLine->Last->TotalLength;
497 
498     if (I + 1 == E || I[1]->Type == LT_Invalid)
499       return 0;
500 
501     if (TheLine->Last->Type == TT_FunctionLBrace) {
502       return Style.AllowShortFunctionsOnASingleLine
503                  ? tryMergeSimpleBlock(I, E, Limit)
504                  : 0;
505     }
506     if (TheLine->Last->is(tok::l_brace)) {
507       return Style.BreakBeforeBraces == FormatStyle::BS_Attach
508                  ? tryMergeSimpleBlock(I, E, Limit)
509                  : 0;
510     }
511     if (I[1]->First->Type == TT_FunctionLBrace &&
512         Style.BreakBeforeBraces != FormatStyle::BS_Attach) {
513       // Reduce the column limit by the number of spaces we need to insert
514       // around braces.
515       Limit = Limit > 3 ? Limit - 3 : 0;
516       unsigned MergedLines = 0;
517       if (Style.AllowShortFunctionsOnASingleLine) {
518         MergedLines = tryMergeSimpleBlock(I + 1, E, Limit);
519         // If we managed to merge the block, count the function header, which is
520         // on a separate line.
521         if (MergedLines > 0)
522           ++MergedLines;
523       }
524       return MergedLines;
525     }
526     if (TheLine->First->is(tok::kw_if)) {
527       return Style.AllowShortIfStatementsOnASingleLine
528                  ? tryMergeSimpleControlStatement(I, E, Limit)
529                  : 0;
530     }
531     if (TheLine->First->isOneOf(tok::kw_for, tok::kw_while)) {
532       return Style.AllowShortLoopsOnASingleLine
533                  ? tryMergeSimpleControlStatement(I, E, Limit)
534                  : 0;
535     }
536     if (TheLine->InPPDirective &&
537         (TheLine->First->HasUnescapedNewline || TheLine->First->IsFirst)) {
538       return tryMergeSimplePPDirective(I, E, Limit);
539     }
540     return 0;
541   }
542 
543 private:
544   unsigned
545   tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
546                             SmallVectorImpl<AnnotatedLine *>::const_iterator E,
547                             unsigned Limit) {
548     if (Limit == 0)
549       return 0;
550     if (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline)
551       return 0;
552     if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
553       return 0;
554     if (1 + I[1]->Last->TotalLength > Limit)
555       return 0;
556     return 1;
557   }
558 
559   unsigned tryMergeSimpleControlStatement(
560       SmallVectorImpl<AnnotatedLine *>::const_iterator I,
561       SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
562     if (Limit == 0)
563       return 0;
564     if ((Style.BreakBeforeBraces == FormatStyle::BS_Allman ||
565          Style.BreakBeforeBraces == FormatStyle::BS_GNU) &&
566         I[1]->First->is(tok::l_brace))
567       return 0;
568     if (I[1]->InPPDirective != (*I)->InPPDirective ||
569         (I[1]->InPPDirective && I[1]->First->HasUnescapedNewline))
570       return 0;
571     AnnotatedLine &Line = **I;
572     if (Line.Last->isNot(tok::r_paren))
573       return 0;
574     if (1 + I[1]->Last->TotalLength > Limit)
575       return 0;
576     if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for,
577                              tok::kw_while) ||
578         I[1]->First->Type == TT_LineComment)
579       return 0;
580     // Only inline simple if's (no nested if or else).
581     if (I + 2 != E && Line.First->is(tok::kw_if) &&
582         I[2]->First->is(tok::kw_else))
583       return 0;
584     return 1;
585   }
586 
587   unsigned
588   tryMergeSimpleBlock(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
589                       SmallVectorImpl<AnnotatedLine *>::const_iterator E,
590                       unsigned Limit) {
591     // First, check that the current line allows merging. This is the case if
592     // we're not in a control flow statement and the last token is an opening
593     // brace.
594     AnnotatedLine &Line = **I;
595     if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::r_brace,
596                             tok::kw_else, tok::kw_try, tok::kw_catch,
597                             tok::kw_for,
598                             // This gets rid of all ObjC @ keywords and methods.
599                             tok::at, tok::minus, tok::plus))
600       return 0;
601 
602     FormatToken *Tok = I[1]->First;
603     if (Tok->is(tok::r_brace) && !Tok->MustBreakBefore &&
604         (Tok->getNextNonComment() == NULL ||
605          Tok->getNextNonComment()->is(tok::semi))) {
606       // We merge empty blocks even if the line exceeds the column limit.
607       Tok->SpacesRequiredBefore = 0;
608       Tok->CanBreakBefore = true;
609       return 1;
610     } else if (Limit != 0 && Line.First->isNot(tok::kw_namespace)) {
611       // Check that we still have three lines and they fit into the limit.
612       if (I + 2 == E || I[2]->Type == LT_Invalid)
613         return 0;
614 
615       if (!nextTwoLinesFitInto(I, Limit))
616         return 0;
617 
618       // Second, check that the next line does not contain any braces - if it
619       // does, readability declines when putting it into a single line.
620       if (I[1]->Last->Type == TT_LineComment || Tok->MustBreakBefore)
621         return 0;
622       do {
623         if (Tok->isOneOf(tok::l_brace, tok::r_brace))
624           return 0;
625         Tok = Tok->Next;
626       } while (Tok != NULL);
627 
628       // Last, check that the third line contains a single closing brace.
629       Tok = I[2]->First;
630       if (Tok->getNextNonComment() != NULL || Tok->isNot(tok::r_brace) ||
631           Tok->MustBreakBefore)
632         return 0;
633 
634       return 2;
635     }
636     return 0;
637   }
638 
639   bool nextTwoLinesFitInto(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
640                            unsigned Limit) {
641     return 1 + I[1]->Last->TotalLength + 1 + I[2]->Last->TotalLength <= Limit;
642   }
643 
644   const FormatStyle &Style;
645 };
646 
647 class UnwrappedLineFormatter {
648 public:
649   UnwrappedLineFormatter(ContinuationIndenter *Indenter,
650                          WhitespaceManager *Whitespaces,
651                          const FormatStyle &Style)
652       : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
653         Joiner(Style) {}
654 
655   unsigned format(const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
656                   int AdditionalIndent = 0, bool FixBadIndentation = false) {
657     assert(!Lines.empty());
658     unsigned Penalty = 0;
659     std::vector<int> IndentForLevel;
660     for (unsigned i = 0, e = Lines[0]->Level; i != e; ++i)
661       IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
662     const AnnotatedLine *PreviousLine = NULL;
663     for (SmallVectorImpl<AnnotatedLine *>::const_iterator I = Lines.begin(),
664                                                           E = Lines.end();
665          I != E; ++I) {
666       const AnnotatedLine &TheLine = **I;
667       const FormatToken *FirstTok = TheLine.First;
668       int Offset = getIndentOffset(*FirstTok);
669 
670       // Determine indent and try to merge multiple unwrapped lines.
671       unsigned Indent;
672       if (TheLine.InPPDirective) {
673         Indent = TheLine.Level * Style.IndentWidth;
674       } else {
675         while (IndentForLevel.size() <= TheLine.Level)
676           IndentForLevel.push_back(-1);
677         IndentForLevel.resize(TheLine.Level + 1);
678         Indent = getIndent(IndentForLevel, TheLine.Level);
679       }
680       unsigned LevelIndent = Indent;
681       if (static_cast<int>(Indent) + Offset >= 0)
682         Indent += Offset;
683 
684       // Merge multiple lines if possible.
685       unsigned MergedLines = Joiner.tryFitMultipleLinesInOne(Indent, I, E);
686       if (MergedLines > 0 && Style.ColumnLimit == 0) {
687         // Disallow line merging if there is a break at the start of one of the
688         // input lines.
689         for (unsigned i = 0; i < MergedLines; ++i) {
690           if (I[i + 1]->First->NewlinesBefore > 0)
691             MergedLines = 0;
692         }
693       }
694       if (!DryRun) {
695         for (unsigned i = 0; i < MergedLines; ++i) {
696           join(*I[i], *I[i + 1]);
697         }
698       }
699       I += MergedLines;
700 
701       bool FixIndentation =
702           FixBadIndentation && (LevelIndent != FirstTok->OriginalColumn);
703       if (TheLine.First->is(tok::eof)) {
704         if (PreviousLine && PreviousLine->Affected && !DryRun) {
705           // Remove the file's trailing whitespace.
706           unsigned Newlines = std::min(FirstTok->NewlinesBefore, 1u);
707           Whitespaces->replaceWhitespace(*TheLine.First, Newlines,
708                                          /*IndentLevel=*/0, /*Spaces=*/0,
709                                          /*TargetColumn=*/0);
710         }
711       } else if (TheLine.Type != LT_Invalid &&
712                  (TheLine.Affected || FixIndentation)) {
713         if (FirstTok->WhitespaceRange.isValid()) {
714           if (!DryRun)
715             formatFirstToken(*TheLine.First, PreviousLine, TheLine.Level,
716                              Indent, TheLine.InPPDirective);
717         } else {
718           Indent = LevelIndent = FirstTok->OriginalColumn;
719         }
720 
721         // If everything fits on a single line, just put it there.
722         unsigned ColumnLimit = Style.ColumnLimit;
723         if (I + 1 != E) {
724           AnnotatedLine *NextLine = I[1];
725           if (NextLine->InPPDirective && !NextLine->First->HasUnescapedNewline)
726             ColumnLimit = getColumnLimit(TheLine.InPPDirective);
727         }
728 
729         if (TheLine.Last->TotalLength + Indent <= ColumnLimit) {
730           LineState State = Indenter->getInitialState(Indent, &TheLine, DryRun);
731           while (State.NextToken != NULL)
732             Indenter->addTokenToState(State, /*Newline=*/false, DryRun);
733         } else if (Style.ColumnLimit == 0) {
734           // FIXME: Implement nested blocks for ColumnLimit = 0.
735           NoColumnLimitFormatter Formatter(Indenter);
736           if (!DryRun)
737             Formatter.format(Indent, &TheLine);
738         } else {
739           Penalty += format(TheLine, Indent, DryRun);
740         }
741 
742         if (!TheLine.InPPDirective)
743           IndentForLevel[TheLine.Level] = LevelIndent;
744       } else if (TheLine.ChildrenAffected) {
745         format(TheLine.Children, DryRun);
746       } else {
747         // Format the first token if necessary, and notify the WhitespaceManager
748         // about the unchanged whitespace.
749         for (FormatToken *Tok = TheLine.First; Tok != NULL; Tok = Tok->Next) {
750           if (Tok == TheLine.First &&
751               (Tok->NewlinesBefore > 0 || Tok->IsFirst)) {
752             unsigned LevelIndent = Tok->OriginalColumn;
753             if (!DryRun) {
754               // Remove trailing whitespace of the previous line.
755               if ((PreviousLine && PreviousLine->Affected) ||
756                   TheLine.LeadingEmptyLinesAffected) {
757                 formatFirstToken(*Tok, PreviousLine, TheLine.Level, LevelIndent,
758                                  TheLine.InPPDirective);
759               } else {
760                 Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
761               }
762             }
763 
764             if (static_cast<int>(LevelIndent) - Offset >= 0)
765               LevelIndent -= Offset;
766             if (Tok->isNot(tok::comment) && !TheLine.InPPDirective)
767               IndentForLevel[TheLine.Level] = LevelIndent;
768           } else if (!DryRun) {
769             Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
770           }
771         }
772       }
773       if (!DryRun) {
774         for (FormatToken *Tok = TheLine.First; Tok != NULL; Tok = Tok->Next) {
775           Tok->Finalized = true;
776         }
777       }
778       PreviousLine = *I;
779     }
780     return Penalty;
781   }
782 
783 private:
784   /// \brief Formats an \c AnnotatedLine and returns the penalty.
785   ///
786   /// If \p DryRun is \c false, directly applies the changes.
787   unsigned format(const AnnotatedLine &Line, unsigned FirstIndent,
788                   bool DryRun) {
789     LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
790 
791     // If the ObjC method declaration does not fit on a line, we should format
792     // it with one arg per line.
793     if (State.Line->Type == LT_ObjCMethodDecl)
794       State.Stack.back().BreakBeforeParameter = true;
795 
796     // Find best solution in solution space.
797     return analyzeSolutionSpace(State, DryRun);
798   }
799 
800   /// \brief An edge in the solution space from \c Previous->State to \c State,
801   /// inserting a newline dependent on the \c NewLine.
802   struct StateNode {
803     StateNode(const LineState &State, bool NewLine, StateNode *Previous)
804         : State(State), NewLine(NewLine), Previous(Previous) {}
805     LineState State;
806     bool NewLine;
807     StateNode *Previous;
808   };
809 
810   /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
811   ///
812   /// In case of equal penalties, we want to prefer states that were inserted
813   /// first. During state generation we make sure that we insert states first
814   /// that break the line as late as possible.
815   typedef std::pair<unsigned, unsigned> OrderedPenalty;
816 
817   /// \brief An item in the prioritized BFS search queue. The \c StateNode's
818   /// \c State has the given \c OrderedPenalty.
819   typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
820 
821   /// \brief The BFS queue type.
822   typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
823                               std::greater<QueueItem> > QueueType;
824 
825   /// \brief Get the offset of the line relatively to the level.
826   ///
827   /// For example, 'public:' labels in classes are offset by 1 or 2
828   /// characters to the left from their level.
829   int getIndentOffset(const FormatToken &RootToken) {
830     if (RootToken.isAccessSpecifier(false) || RootToken.isObjCAccessSpecifier())
831       return Style.AccessModifierOffset;
832     return 0;
833   }
834 
835   /// \brief Add a new line and the required indent before the first Token
836   /// of the \c UnwrappedLine if there was no structural parsing error.
837   void formatFirstToken(FormatToken &RootToken,
838                         const AnnotatedLine *PreviousLine, unsigned IndentLevel,
839                         unsigned Indent, bool InPPDirective) {
840     unsigned Newlines =
841         std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
842     // Remove empty lines before "}" where applicable.
843     if (RootToken.is(tok::r_brace) &&
844         (!RootToken.Next ||
845          (RootToken.Next->is(tok::semi) && !RootToken.Next->Next)))
846       Newlines = std::min(Newlines, 1u);
847     if (Newlines == 0 && !RootToken.IsFirst)
848       Newlines = 1;
849 
850     // Insert extra new line before access specifiers.
851     if (PreviousLine && PreviousLine->Last->isOneOf(tok::semi, tok::r_brace) &&
852         RootToken.isAccessSpecifier() && RootToken.NewlinesBefore == 1)
853       ++Newlines;
854 
855     // Remove empty lines after access specifiers.
856     if (PreviousLine && PreviousLine->First->isAccessSpecifier())
857       Newlines = std::min(1u, Newlines);
858 
859     Whitespaces->replaceWhitespace(RootToken, Newlines, IndentLevel, Indent,
860                                    Indent, InPPDirective &&
861                                                !RootToken.HasUnescapedNewline);
862   }
863 
864   /// \brief Get the indent of \p Level from \p IndentForLevel.
865   ///
866   /// \p IndentForLevel must contain the indent for the level \c l
867   /// at \p IndentForLevel[l], or a value < 0 if the indent for
868   /// that level is unknown.
869   unsigned getIndent(const std::vector<int> IndentForLevel, unsigned Level) {
870     if (IndentForLevel[Level] != -1)
871       return IndentForLevel[Level];
872     if (Level == 0)
873       return 0;
874     return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
875   }
876 
877   void join(AnnotatedLine &A, const AnnotatedLine &B) {
878     assert(!A.Last->Next);
879     assert(!B.First->Previous);
880     if (B.Affected)
881       A.Affected = true;
882     A.Last->Next = B.First;
883     B.First->Previous = A.Last;
884     B.First->CanBreakBefore = true;
885     unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
886     for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
887       Tok->TotalLength += LengthA;
888       A.Last = Tok;
889     }
890   }
891 
892   unsigned getColumnLimit(bool InPPDirective) const {
893     // In preprocessor directives reserve two chars for trailing " \"
894     return Style.ColumnLimit - (InPPDirective ? 2 : 0);
895   }
896 
897   /// \brief Analyze the entire solution space starting from \p InitialState.
898   ///
899   /// This implements a variant of Dijkstra's algorithm on the graph that spans
900   /// the solution space (\c LineStates are the nodes). The algorithm tries to
901   /// find the shortest path (the one with lowest penalty) from \p InitialState
902   /// to a state where all tokens are placed. Returns the penalty.
903   ///
904   /// If \p DryRun is \c false, directly applies the changes.
905   unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun = false) {
906     std::set<LineState> Seen;
907 
908     // Increasing count of \c StateNode items we have created. This is used to
909     // create a deterministic order independent of the container.
910     unsigned Count = 0;
911     QueueType Queue;
912 
913     // Insert start element into queue.
914     StateNode *Node =
915         new (Allocator.Allocate()) StateNode(InitialState, false, NULL);
916     Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
917     ++Count;
918 
919     unsigned Penalty = 0;
920 
921     // While not empty, take first element and follow edges.
922     while (!Queue.empty()) {
923       Penalty = Queue.top().first.first;
924       StateNode *Node = Queue.top().second;
925       if (Node->State.NextToken == NULL) {
926         DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
927         break;
928       }
929       Queue.pop();
930 
931       // Cut off the analysis of certain solutions if the analysis gets too
932       // complex. See description of IgnoreStackForComparison.
933       if (Count > 10000)
934         Node->State.IgnoreStackForComparison = true;
935 
936       if (!Seen.insert(Node->State).second)
937         // State already examined with lower penalty.
938         continue;
939 
940       FormatDecision LastFormat = Node->State.NextToken->Decision;
941       if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
942         addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
943       if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
944         addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
945     }
946 
947     if (Queue.empty()) {
948       // We were unable to find a solution, do nothing.
949       // FIXME: Add diagnostic?
950       DEBUG(llvm::dbgs() << "Could not find a solution.\n");
951       return 0;
952     }
953 
954     // Reconstruct the solution.
955     if (!DryRun)
956       reconstructPath(InitialState, Queue.top().second);
957 
958     DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
959     DEBUG(llvm::dbgs() << "---\n");
960 
961     return Penalty;
962   }
963 
964   void reconstructPath(LineState &State, StateNode *Current) {
965     std::deque<StateNode *> Path;
966     // We do not need a break before the initial token.
967     while (Current->Previous) {
968       Path.push_front(Current);
969       Current = Current->Previous;
970     }
971     for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
972          I != E; ++I) {
973       unsigned Penalty = 0;
974       formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
975       Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
976 
977       DEBUG({
978         if ((*I)->NewLine) {
979           llvm::dbgs() << "Penalty for placing "
980                        << (*I)->Previous->State.NextToken->Tok.getName() << ": "
981                        << Penalty << "\n";
982         }
983       });
984     }
985   }
986 
987   /// \brief Add the following state to the analysis queue \c Queue.
988   ///
989   /// Assume the current state is \p PreviousNode and has been reached with a
990   /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
991   void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
992                            bool NewLine, unsigned *Count, QueueType *Queue) {
993     if (NewLine && !Indenter->canBreak(PreviousNode->State))
994       return;
995     if (!NewLine && Indenter->mustBreak(PreviousNode->State))
996       return;
997 
998     StateNode *Node = new (Allocator.Allocate())
999         StateNode(PreviousNode->State, NewLine, PreviousNode);
1000     if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
1001       return;
1002 
1003     Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
1004 
1005     Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
1006     ++(*Count);
1007   }
1008 
1009   /// \brief If the \p State's next token is an r_brace closing a nested block,
1010   /// format the nested block before it.
1011   ///
1012   /// Returns \c true if all children could be placed successfully and adapts
1013   /// \p Penalty as well as \p State. If \p DryRun is false, also directly
1014   /// creates changes using \c Whitespaces.
1015   ///
1016   /// The crucial idea here is that children always get formatted upon
1017   /// encountering the closing brace right after the nested block. Now, if we
1018   /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
1019   /// \c false), the entire block has to be kept on the same line (which is only
1020   /// possible if it fits on the line, only contains a single statement, etc.
1021   ///
1022   /// If \p NewLine is true, we format the nested block on separate lines, i.e.
1023   /// break after the "{", format all lines with correct indentation and the put
1024   /// the closing "}" on yet another new line.
1025   ///
1026   /// This enables us to keep the simple structure of the
1027   /// \c UnwrappedLineFormatter, where we only have two options for each token:
1028   /// break or don't break.
1029   bool formatChildren(LineState &State, bool NewLine, bool DryRun,
1030                       unsigned &Penalty) {
1031     FormatToken &Previous = *State.NextToken->Previous;
1032     const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
1033     if (!LBrace || LBrace->isNot(tok::l_brace) ||
1034         LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
1035       // The previous token does not open a block. Nothing to do. We don't
1036       // assert so that we can simply call this function for all tokens.
1037       return true;
1038 
1039     if (NewLine) {
1040       int AdditionalIndent = State.Stack.back().Indent -
1041                              Previous.Children[0]->Level * Style.IndentWidth;
1042       Penalty += format(Previous.Children, DryRun, AdditionalIndent,
1043                         /*FixBadIndentation=*/true);
1044       return true;
1045     }
1046 
1047     // Cannot merge multiple statements into a single line.
1048     if (Previous.Children.size() > 1)
1049       return false;
1050 
1051     // We can't put the closing "}" on a line with a trailing comment.
1052     if (Previous.Children[0]->Last->isTrailingComment())
1053       return false;
1054 
1055     if (!DryRun) {
1056       Whitespaces->replaceWhitespace(
1057           *Previous.Children[0]->First,
1058           /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1,
1059           /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
1060     }
1061     Penalty += format(*Previous.Children[0], State.Column + 1, DryRun);
1062 
1063     State.Column += 1 + Previous.Children[0]->Last->TotalLength;
1064     return true;
1065   }
1066 
1067   ContinuationIndenter *Indenter;
1068   WhitespaceManager *Whitespaces;
1069   FormatStyle Style;
1070   LineJoiner Joiner;
1071 
1072   llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
1073 };
1074 
1075 class FormatTokenLexer {
1076 public:
1077   FormatTokenLexer(Lexer &Lex, SourceManager &SourceMgr, FormatStyle &Style,
1078                    encoding::Encoding Encoding)
1079       : FormatTok(NULL), IsFirstToken(true), GreaterStashed(false), Column(0),
1080         TrailingWhitespace(0), Lex(Lex), SourceMgr(SourceMgr), Style(Style),
1081         IdentTable(getFormattingLangOpts()), Encoding(Encoding) {
1082     Lex.SetKeepWhitespaceMode(true);
1083   }
1084 
1085   ArrayRef<FormatToken *> lex() {
1086     assert(Tokens.empty());
1087     do {
1088       Tokens.push_back(getNextToken());
1089       tryMergePreviousTokens();
1090     } while (Tokens.back()->Tok.isNot(tok::eof));
1091     return Tokens;
1092   }
1093 
1094   IdentifierTable &getIdentTable() { return IdentTable; }
1095 
1096 private:
1097   void tryMergePreviousTokens() {
1098     if (tryMerge_TMacro())
1099       return;
1100 
1101     if (Style.Language == FormatStyle::LK_JavaScript) {
1102       static tok::TokenKind JSIdentity[] = { tok::equalequal, tok::equal };
1103       static tok::TokenKind JSNotIdentity[] = { tok::exclaimequal, tok::equal };
1104       static tok::TokenKind JSShiftEqual[] = { tok::greater, tok::greater,
1105                                                tok::greaterequal };
1106       // FIXME: We probably need to change token type to mimic operator with the
1107       // correct priority.
1108       if (tryMergeTokens(JSIdentity))
1109         return;
1110       if (tryMergeTokens(JSNotIdentity))
1111         return;
1112       if (tryMergeTokens(JSShiftEqual))
1113         return;
1114     }
1115   }
1116 
1117   bool tryMergeTokens(ArrayRef<tok::TokenKind> Kinds) {
1118     if (Tokens.size() < Kinds.size())
1119       return false;
1120 
1121     SmallVectorImpl<FormatToken *>::const_iterator First =
1122         Tokens.end() - Kinds.size();
1123     if (!First[0]->is(Kinds[0]))
1124       return false;
1125     unsigned AddLength = 0;
1126     for (unsigned i = 1; i < Kinds.size(); ++i) {
1127       if (!First[i]->is(Kinds[i]) || First[i]->WhitespaceRange.getBegin() !=
1128                                          First[i]->WhitespaceRange.getEnd())
1129         return false;
1130       AddLength += First[i]->TokenText.size();
1131     }
1132     Tokens.resize(Tokens.size() - Kinds.size() + 1);
1133     First[0]->TokenText = StringRef(First[0]->TokenText.data(),
1134                                     First[0]->TokenText.size() + AddLength);
1135     First[0]->ColumnWidth += AddLength;
1136     return true;
1137   }
1138 
1139   bool tryMerge_TMacro() {
1140     if (Tokens.size() < 4)
1141       return false;
1142     FormatToken *Last = Tokens.back();
1143     if (!Last->is(tok::r_paren))
1144       return false;
1145 
1146     FormatToken *String = Tokens[Tokens.size() - 2];
1147     if (!String->is(tok::string_literal) || String->IsMultiline)
1148       return false;
1149 
1150     if (!Tokens[Tokens.size() - 3]->is(tok::l_paren))
1151       return false;
1152 
1153     FormatToken *Macro = Tokens[Tokens.size() - 4];
1154     if (Macro->TokenText != "_T")
1155       return false;
1156 
1157     const char *Start = Macro->TokenText.data();
1158     const char *End = Last->TokenText.data() + Last->TokenText.size();
1159     String->TokenText = StringRef(Start, End - Start);
1160     String->IsFirst = Macro->IsFirst;
1161     String->LastNewlineOffset = Macro->LastNewlineOffset;
1162     String->WhitespaceRange = Macro->WhitespaceRange;
1163     String->OriginalColumn = Macro->OriginalColumn;
1164     String->ColumnWidth = encoding::columnWidthWithTabs(
1165         String->TokenText, String->OriginalColumn, Style.TabWidth, Encoding);
1166 
1167     Tokens.pop_back();
1168     Tokens.pop_back();
1169     Tokens.pop_back();
1170     Tokens.back() = String;
1171     return true;
1172   }
1173 
1174   FormatToken *getNextToken() {
1175     if (GreaterStashed) {
1176       // Create a synthesized second '>' token.
1177       // FIXME: Increment Column and set OriginalColumn.
1178       Token Greater = FormatTok->Tok;
1179       FormatTok = new (Allocator.Allocate()) FormatToken;
1180       FormatTok->Tok = Greater;
1181       SourceLocation GreaterLocation =
1182           FormatTok->Tok.getLocation().getLocWithOffset(1);
1183       FormatTok->WhitespaceRange =
1184           SourceRange(GreaterLocation, GreaterLocation);
1185       FormatTok->TokenText = ">";
1186       FormatTok->ColumnWidth = 1;
1187       GreaterStashed = false;
1188       return FormatTok;
1189     }
1190 
1191     FormatTok = new (Allocator.Allocate()) FormatToken;
1192     readRawToken(*FormatTok);
1193     SourceLocation WhitespaceStart =
1194         FormatTok->Tok.getLocation().getLocWithOffset(-TrailingWhitespace);
1195     FormatTok->IsFirst = IsFirstToken;
1196     IsFirstToken = false;
1197 
1198     // Consume and record whitespace until we find a significant token.
1199     unsigned WhitespaceLength = TrailingWhitespace;
1200     while (FormatTok->Tok.is(tok::unknown)) {
1201       for (int i = 0, e = FormatTok->TokenText.size(); i != e; ++i) {
1202         switch (FormatTok->TokenText[i]) {
1203         case '\n':
1204           ++FormatTok->NewlinesBefore;
1205           // FIXME: This is technically incorrect, as it could also
1206           // be a literal backslash at the end of the line.
1207           if (i == 0 || (FormatTok->TokenText[i - 1] != '\\' &&
1208                          (FormatTok->TokenText[i - 1] != '\r' || i == 1 ||
1209                           FormatTok->TokenText[i - 2] != '\\')))
1210             FormatTok->HasUnescapedNewline = true;
1211           FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
1212           Column = 0;
1213           break;
1214         case '\r':
1215         case '\f':
1216         case '\v':
1217           Column = 0;
1218           break;
1219         case ' ':
1220           ++Column;
1221           break;
1222         case '\t':
1223           Column += Style.TabWidth - Column % Style.TabWidth;
1224           break;
1225         case '\\':
1226           ++Column;
1227           if (i + 1 == e || (FormatTok->TokenText[i + 1] != '\r' &&
1228                              FormatTok->TokenText[i + 1] != '\n'))
1229             FormatTok->Type = TT_ImplicitStringLiteral;
1230           break;
1231         default:
1232           FormatTok->Type = TT_ImplicitStringLiteral;
1233           ++Column;
1234           break;
1235         }
1236       }
1237 
1238       if (FormatTok->Type == TT_ImplicitStringLiteral)
1239         break;
1240       WhitespaceLength += FormatTok->Tok.getLength();
1241 
1242       readRawToken(*FormatTok);
1243     }
1244 
1245     // In case the token starts with escaped newlines, we want to
1246     // take them into account as whitespace - this pattern is quite frequent
1247     // in macro definitions.
1248     // FIXME: Add a more explicit test.
1249     while (FormatTok->TokenText.size() > 1 && FormatTok->TokenText[0] == '\\' &&
1250            FormatTok->TokenText[1] == '\n') {
1251       // FIXME: ++FormatTok->NewlinesBefore is missing...
1252       WhitespaceLength += 2;
1253       Column = 0;
1254       FormatTok->TokenText = FormatTok->TokenText.substr(2);
1255     }
1256 
1257     FormatTok->WhitespaceRange = SourceRange(
1258         WhitespaceStart, WhitespaceStart.getLocWithOffset(WhitespaceLength));
1259 
1260     FormatTok->OriginalColumn = Column;
1261 
1262     TrailingWhitespace = 0;
1263     if (FormatTok->Tok.is(tok::comment)) {
1264       // FIXME: Add the trimmed whitespace to Column.
1265       StringRef UntrimmedText = FormatTok->TokenText;
1266       FormatTok->TokenText = FormatTok->TokenText.rtrim(" \t\v\f");
1267       TrailingWhitespace = UntrimmedText.size() - FormatTok->TokenText.size();
1268     } else if (FormatTok->Tok.is(tok::raw_identifier)) {
1269       IdentifierInfo &Info = IdentTable.get(FormatTok->TokenText);
1270       FormatTok->Tok.setIdentifierInfo(&Info);
1271       FormatTok->Tok.setKind(Info.getTokenID());
1272     } else if (FormatTok->Tok.is(tok::greatergreater)) {
1273       FormatTok->Tok.setKind(tok::greater);
1274       FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
1275       GreaterStashed = true;
1276     }
1277 
1278     // Now FormatTok is the next non-whitespace token.
1279 
1280     StringRef Text = FormatTok->TokenText;
1281     size_t FirstNewlinePos = Text.find('\n');
1282     if (FirstNewlinePos == StringRef::npos) {
1283       // FIXME: ColumnWidth actually depends on the start column, we need to
1284       // take this into account when the token is moved.
1285       FormatTok->ColumnWidth =
1286           encoding::columnWidthWithTabs(Text, Column, Style.TabWidth, Encoding);
1287       Column += FormatTok->ColumnWidth;
1288     } else {
1289       FormatTok->IsMultiline = true;
1290       // FIXME: ColumnWidth actually depends on the start column, we need to
1291       // take this into account when the token is moved.
1292       FormatTok->ColumnWidth = encoding::columnWidthWithTabs(
1293           Text.substr(0, FirstNewlinePos), Column, Style.TabWidth, Encoding);
1294 
1295       // The last line of the token always starts in column 0.
1296       // Thus, the length can be precomputed even in the presence of tabs.
1297       FormatTok->LastLineColumnWidth = encoding::columnWidthWithTabs(
1298           Text.substr(Text.find_last_of('\n') + 1), 0, Style.TabWidth,
1299           Encoding);
1300       Column = FormatTok->LastLineColumnWidth;
1301     }
1302 
1303     return FormatTok;
1304   }
1305 
1306   FormatToken *FormatTok;
1307   bool IsFirstToken;
1308   bool GreaterStashed;
1309   unsigned Column;
1310   unsigned TrailingWhitespace;
1311   Lexer &Lex;
1312   SourceManager &SourceMgr;
1313   FormatStyle &Style;
1314   IdentifierTable IdentTable;
1315   encoding::Encoding Encoding;
1316   llvm::SpecificBumpPtrAllocator<FormatToken> Allocator;
1317   SmallVector<FormatToken *, 16> Tokens;
1318 
1319   void readRawToken(FormatToken &Tok) {
1320     Lex.LexFromRawLexer(Tok.Tok);
1321     Tok.TokenText = StringRef(SourceMgr.getCharacterData(Tok.Tok.getLocation()),
1322                               Tok.Tok.getLength());
1323     // For formatting, treat unterminated string literals like normal string
1324     // literals.
1325     if (Tok.is(tok::unknown) && !Tok.TokenText.empty() &&
1326         Tok.TokenText[0] == '"') {
1327       Tok.Tok.setKind(tok::string_literal);
1328       Tok.IsUnterminatedLiteral = true;
1329     }
1330   }
1331 };
1332 
1333 static StringRef getLanguageName(FormatStyle::LanguageKind Language) {
1334   switch (Language) {
1335   case FormatStyle::LK_Cpp:
1336     return "C++";
1337   case FormatStyle::LK_JavaScript:
1338     return "JavaScript";
1339   default:
1340     return "Unknown";
1341   }
1342 }
1343 
1344 class Formatter : public UnwrappedLineConsumer {
1345 public:
1346   Formatter(const FormatStyle &Style, Lexer &Lex, SourceManager &SourceMgr,
1347             const std::vector<CharSourceRange> &Ranges)
1348       : Style(Style), Lex(Lex), SourceMgr(SourceMgr),
1349         Whitespaces(SourceMgr, Style, inputUsesCRLF(Lex.getBuffer())),
1350         Ranges(Ranges.begin(), Ranges.end()), UnwrappedLines(1),
1351         Encoding(encoding::detectEncoding(Lex.getBuffer())) {
1352     DEBUG(llvm::dbgs() << "File encoding: "
1353                        << (Encoding == encoding::Encoding_UTF8 ? "UTF8"
1354                                                                : "unknown")
1355                        << "\n");
1356     DEBUG(llvm::dbgs() << "Language: " << getLanguageName(Style.Language)
1357                        << "\n");
1358   }
1359 
1360   tooling::Replacements format() {
1361     tooling::Replacements Result;
1362     FormatTokenLexer Tokens(Lex, SourceMgr, Style, Encoding);
1363 
1364     UnwrappedLineParser Parser(Style, Tokens.lex(), *this);
1365     bool StructuralError = Parser.parse();
1366     assert(UnwrappedLines.rbegin()->empty());
1367     for (unsigned Run = 0, RunE = UnwrappedLines.size(); Run + 1 != RunE;
1368          ++Run) {
1369       DEBUG(llvm::dbgs() << "Run " << Run << "...\n");
1370       SmallVector<AnnotatedLine *, 16> AnnotatedLines;
1371       for (unsigned i = 0, e = UnwrappedLines[Run].size(); i != e; ++i) {
1372         AnnotatedLines.push_back(new AnnotatedLine(UnwrappedLines[Run][i]));
1373       }
1374       tooling::Replacements RunResult =
1375           format(AnnotatedLines, StructuralError, Tokens);
1376       DEBUG({
1377         llvm::dbgs() << "Replacements for run " << Run << ":\n";
1378         for (tooling::Replacements::iterator I = RunResult.begin(),
1379                                              E = RunResult.end();
1380              I != E; ++I) {
1381           llvm::dbgs() << I->toString() << "\n";
1382         }
1383       });
1384       for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1385         delete AnnotatedLines[i];
1386       }
1387       Result.insert(RunResult.begin(), RunResult.end());
1388       Whitespaces.reset();
1389     }
1390     return Result;
1391   }
1392 
1393   tooling::Replacements format(SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
1394                                bool StructuralError, FormatTokenLexer &Tokens) {
1395     TokenAnnotator Annotator(Style, Tokens.getIdentTable().get("in"));
1396     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1397       Annotator.annotate(*AnnotatedLines[i]);
1398     }
1399     deriveLocalStyle(AnnotatedLines);
1400     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1401       Annotator.calculateFormattingInformation(*AnnotatedLines[i]);
1402     }
1403     computeAffectedLines(AnnotatedLines.begin(), AnnotatedLines.end());
1404 
1405     Annotator.setCommentLineLevels(AnnotatedLines);
1406     ContinuationIndenter Indenter(Style, SourceMgr, Whitespaces, Encoding,
1407                                   BinPackInconclusiveFunctions);
1408     UnwrappedLineFormatter Formatter(&Indenter, &Whitespaces, Style);
1409     Formatter.format(AnnotatedLines, /*DryRun=*/false);
1410     return Whitespaces.generateReplacements();
1411   }
1412 
1413 private:
1414   // Determines which lines are affected by the SourceRanges given as input.
1415   // Returns \c true if at least one line between I and E or one of their
1416   // children is affected.
1417   bool computeAffectedLines(SmallVectorImpl<AnnotatedLine *>::iterator I,
1418                             SmallVectorImpl<AnnotatedLine *>::iterator E) {
1419     bool SomeLineAffected = false;
1420     const AnnotatedLine *PreviousLine = NULL;
1421     while (I != E) {
1422       AnnotatedLine *Line = *I;
1423       Line->LeadingEmptyLinesAffected = affectsLeadingEmptyLines(*Line->First);
1424 
1425       // If a line is part of a preprocessor directive, it needs to be formatted
1426       // if any token within the directive is affected.
1427       if (Line->InPPDirective) {
1428         FormatToken *Last = Line->Last;
1429         SmallVectorImpl<AnnotatedLine *>::iterator PPEnd = I + 1;
1430         while (PPEnd != E && !(*PPEnd)->First->HasUnescapedNewline) {
1431           Last = (*PPEnd)->Last;
1432           ++PPEnd;
1433         }
1434 
1435         if (affectsTokenRange(*Line->First, *Last,
1436                               /*IncludeLeadingNewlines=*/false)) {
1437           SomeLineAffected = true;
1438           markAllAsAffected(I, PPEnd);
1439         }
1440         I = PPEnd;
1441         continue;
1442       }
1443 
1444       if (nonPPLineAffected(Line, PreviousLine))
1445         SomeLineAffected = true;
1446 
1447       PreviousLine = Line;
1448       ++I;
1449     }
1450     return SomeLineAffected;
1451   }
1452 
1453   // Determines whether 'Line' is affected by the SourceRanges given as input.
1454   // Returns \c true if line or one if its children is affected.
1455   bool nonPPLineAffected(AnnotatedLine *Line,
1456                          const AnnotatedLine *PreviousLine) {
1457     bool SomeLineAffected = false;
1458     Line->ChildrenAffected =
1459         computeAffectedLines(Line->Children.begin(), Line->Children.end());
1460     if (Line->ChildrenAffected)
1461       SomeLineAffected = true;
1462 
1463     // Stores whether one of the line's tokens is directly affected.
1464     bool SomeTokenAffected = false;
1465     // Stores whether we need to look at the leading newlines of the next token
1466     // in order to determine whether it was affected.
1467     bool IncludeLeadingNewlines = false;
1468 
1469     // Stores whether the first child line of any of this line's tokens is
1470     // affected.
1471     bool SomeFirstChildAffected = false;
1472 
1473     for (FormatToken *Tok = Line->First; Tok; Tok = Tok->Next) {
1474       // Determine whether 'Tok' was affected.
1475       if (affectsTokenRange(*Tok, *Tok, IncludeLeadingNewlines))
1476         SomeTokenAffected = true;
1477 
1478       // Determine whether the first child of 'Tok' was affected.
1479       if (!Tok->Children.empty() && Tok->Children.front()->Affected)
1480         SomeFirstChildAffected = true;
1481 
1482       IncludeLeadingNewlines = Tok->Children.empty();
1483     }
1484 
1485     // Was this line moved, i.e. has it previously been on the same line as an
1486     // affected line?
1487     bool LineMoved = PreviousLine && PreviousLine->Affected &&
1488                      Line->First->NewlinesBefore == 0;
1489 
1490     bool IsContinuedComment = Line->First->is(tok::comment) &&
1491                               Line->First->Next == NULL &&
1492                               Line->First->NewlinesBefore < 2 && PreviousLine &&
1493                               PreviousLine->Affected &&
1494                               PreviousLine->Last->is(tok::comment);
1495 
1496     if (SomeTokenAffected || SomeFirstChildAffected || LineMoved ||
1497         IsContinuedComment) {
1498       Line->Affected = true;
1499       SomeLineAffected = true;
1500     }
1501     return SomeLineAffected;
1502   }
1503 
1504   // Marks all lines between I and E as well as all their children as affected.
1505   void markAllAsAffected(SmallVectorImpl<AnnotatedLine *>::iterator I,
1506                          SmallVectorImpl<AnnotatedLine *>::iterator E) {
1507     while (I != E) {
1508       (*I)->Affected = true;
1509       markAllAsAffected((*I)->Children.begin(), (*I)->Children.end());
1510       ++I;
1511     }
1512   }
1513 
1514   // Returns true if the range from 'First' to 'Last' intersects with one of the
1515   // input ranges.
1516   bool affectsTokenRange(const FormatToken &First, const FormatToken &Last,
1517                          bool IncludeLeadingNewlines) {
1518     SourceLocation Start = First.WhitespaceRange.getBegin();
1519     if (!IncludeLeadingNewlines)
1520       Start = Start.getLocWithOffset(First.LastNewlineOffset);
1521     SourceLocation End = Last.getStartOfNonWhitespace();
1522     if (Last.TokenText.size() > 0)
1523       End = End.getLocWithOffset(Last.TokenText.size() - 1);
1524     CharSourceRange Range = CharSourceRange::getCharRange(Start, End);
1525     return affectsCharSourceRange(Range);
1526   }
1527 
1528   // Returns true if one of the input ranges intersect the leading empty lines
1529   // before 'Tok'.
1530   bool affectsLeadingEmptyLines(const FormatToken &Tok) {
1531     CharSourceRange EmptyLineRange = CharSourceRange::getCharRange(
1532         Tok.WhitespaceRange.getBegin(),
1533         Tok.WhitespaceRange.getBegin().getLocWithOffset(Tok.LastNewlineOffset));
1534     return affectsCharSourceRange(EmptyLineRange);
1535   }
1536 
1537   // Returns true if 'Range' intersects with one of the input ranges.
1538   bool affectsCharSourceRange(const CharSourceRange &Range) {
1539     for (SmallVectorImpl<CharSourceRange>::const_iterator I = Ranges.begin(),
1540                                                           E = Ranges.end();
1541          I != E; ++I) {
1542       if (!SourceMgr.isBeforeInTranslationUnit(Range.getEnd(), I->getBegin()) &&
1543           !SourceMgr.isBeforeInTranslationUnit(I->getEnd(), Range.getBegin()))
1544         return true;
1545     }
1546     return false;
1547   }
1548 
1549   static bool inputUsesCRLF(StringRef Text) {
1550     return Text.count('\r') * 2 > Text.count('\n');
1551   }
1552 
1553   void
1554   deriveLocalStyle(const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
1555     unsigned CountBoundToVariable = 0;
1556     unsigned CountBoundToType = 0;
1557     bool HasCpp03IncompatibleFormat = false;
1558     bool HasBinPackedFunction = false;
1559     bool HasOnePerLineFunction = false;
1560     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1561       if (!AnnotatedLines[i]->First->Next)
1562         continue;
1563       FormatToken *Tok = AnnotatedLines[i]->First->Next;
1564       while (Tok->Next) {
1565         if (Tok->Type == TT_PointerOrReference) {
1566           bool SpacesBefore =
1567               Tok->WhitespaceRange.getBegin() != Tok->WhitespaceRange.getEnd();
1568           bool SpacesAfter = Tok->Next->WhitespaceRange.getBegin() !=
1569                              Tok->Next->WhitespaceRange.getEnd();
1570           if (SpacesBefore && !SpacesAfter)
1571             ++CountBoundToVariable;
1572           else if (!SpacesBefore && SpacesAfter)
1573             ++CountBoundToType;
1574         }
1575 
1576         if (Tok->WhitespaceRange.getBegin() == Tok->WhitespaceRange.getEnd()) {
1577           if (Tok->is(tok::coloncolon) &&
1578               Tok->Previous->Type == TT_TemplateOpener)
1579             HasCpp03IncompatibleFormat = true;
1580           if (Tok->Type == TT_TemplateCloser &&
1581               Tok->Previous->Type == TT_TemplateCloser)
1582             HasCpp03IncompatibleFormat = true;
1583         }
1584 
1585         if (Tok->PackingKind == PPK_BinPacked)
1586           HasBinPackedFunction = true;
1587         if (Tok->PackingKind == PPK_OnePerLine)
1588           HasOnePerLineFunction = true;
1589 
1590         Tok = Tok->Next;
1591       }
1592     }
1593     if (Style.DerivePointerBinding) {
1594       if (CountBoundToType > CountBoundToVariable)
1595         Style.PointerBindsToType = true;
1596       else if (CountBoundToType < CountBoundToVariable)
1597         Style.PointerBindsToType = false;
1598     }
1599     if (Style.Standard == FormatStyle::LS_Auto) {
1600       Style.Standard = HasCpp03IncompatibleFormat ? FormatStyle::LS_Cpp11
1601                                                   : FormatStyle::LS_Cpp03;
1602     }
1603     BinPackInconclusiveFunctions =
1604         HasBinPackedFunction || !HasOnePerLineFunction;
1605   }
1606 
1607   virtual void consumeUnwrappedLine(const UnwrappedLine &TheLine) {
1608     assert(!UnwrappedLines.empty());
1609     UnwrappedLines.back().push_back(TheLine);
1610   }
1611 
1612   virtual void finishRun() {
1613     UnwrappedLines.push_back(SmallVector<UnwrappedLine, 16>());
1614   }
1615 
1616   FormatStyle Style;
1617   Lexer &Lex;
1618   SourceManager &SourceMgr;
1619   WhitespaceManager Whitespaces;
1620   SmallVector<CharSourceRange, 8> Ranges;
1621   SmallVector<SmallVector<UnwrappedLine, 16>, 2> UnwrappedLines;
1622 
1623   encoding::Encoding Encoding;
1624   bool BinPackInconclusiveFunctions;
1625 };
1626 
1627 } // end anonymous namespace
1628 
1629 tooling::Replacements reformat(const FormatStyle &Style, Lexer &Lex,
1630                                SourceManager &SourceMgr,
1631                                std::vector<CharSourceRange> Ranges) {
1632   Formatter formatter(Style, Lex, SourceMgr, Ranges);
1633   return formatter.format();
1634 }
1635 
1636 tooling::Replacements reformat(const FormatStyle &Style, StringRef Code,
1637                                std::vector<tooling::Range> Ranges,
1638                                StringRef FileName) {
1639   FileManager Files((FileSystemOptions()));
1640   DiagnosticsEngine Diagnostics(
1641       IntrusiveRefCntPtr<DiagnosticIDs>(new DiagnosticIDs),
1642       new DiagnosticOptions);
1643   SourceManager SourceMgr(Diagnostics, Files);
1644   llvm::MemoryBuffer *Buf = llvm::MemoryBuffer::getMemBuffer(Code, FileName);
1645   const clang::FileEntry *Entry =
1646       Files.getVirtualFile(FileName, Buf->getBufferSize(), 0);
1647   SourceMgr.overrideFileContents(Entry, Buf);
1648   FileID ID =
1649       SourceMgr.createFileID(Entry, SourceLocation(), clang::SrcMgr::C_User);
1650   Lexer Lex(ID, SourceMgr.getBuffer(ID), SourceMgr,
1651             getFormattingLangOpts(Style.Standard));
1652   SourceLocation StartOfFile = SourceMgr.getLocForStartOfFile(ID);
1653   std::vector<CharSourceRange> CharRanges;
1654   for (unsigned i = 0, e = Ranges.size(); i != e; ++i) {
1655     SourceLocation Start = StartOfFile.getLocWithOffset(Ranges[i].getOffset());
1656     SourceLocation End = Start.getLocWithOffset(Ranges[i].getLength());
1657     CharRanges.push_back(CharSourceRange::getCharRange(Start, End));
1658   }
1659   return reformat(Style, Lex, SourceMgr, CharRanges);
1660 }
1661 
1662 LangOptions getFormattingLangOpts(FormatStyle::LanguageStandard Standard) {
1663   LangOptions LangOpts;
1664   LangOpts.CPlusPlus = 1;
1665   LangOpts.CPlusPlus11 = Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
1666   LangOpts.LineComment = 1;
1667   LangOpts.Bool = 1;
1668   LangOpts.ObjC1 = 1;
1669   LangOpts.ObjC2 = 1;
1670   return LangOpts;
1671 }
1672 
1673 const char *StyleOptionHelpDescription =
1674     "Coding style, currently supports:\n"
1675     "  LLVM, Google, Chromium, Mozilla, WebKit.\n"
1676     "Use -style=file to load style configuration from\n"
1677     ".clang-format file located in one of the parent\n"
1678     "directories of the source file (or current\n"
1679     "directory for stdin).\n"
1680     "Use -style=\"{key: value, ...}\" to set specific\n"
1681     "parameters, e.g.:\n"
1682     "  -style=\"{BasedOnStyle: llvm, IndentWidth: 8}\"";
1683 
1684 static FormatStyle::LanguageKind getLanguageByFileName(StringRef FileName) {
1685   if (FileName.endswith_lower(".js")) {
1686     return FormatStyle::LK_JavaScript;
1687   }
1688   return FormatStyle::LK_Cpp;
1689 }
1690 
1691 FormatStyle getStyle(StringRef StyleName, StringRef FileName,
1692                      StringRef FallbackStyle) {
1693   FormatStyle Style = getLLVMStyle();
1694   Style.Language = getLanguageByFileName(FileName);
1695   if (!getPredefinedStyle(FallbackStyle, Style.Language, &Style)) {
1696     llvm::errs() << "Invalid fallback style \"" << FallbackStyle
1697                  << "\" using LLVM style\n";
1698     return Style;
1699   }
1700 
1701   if (StyleName.startswith("{")) {
1702     // Parse YAML/JSON style from the command line.
1703     if (llvm::error_code ec = parseConfiguration(StyleName, &Style)) {
1704       llvm::errs() << "Error parsing -style: " << ec.message() << ", using "
1705                    << FallbackStyle << " style\n";
1706     }
1707     return Style;
1708   }
1709 
1710   if (!StyleName.equals_lower("file")) {
1711     if (!getPredefinedStyle(StyleName, Style.Language, &Style))
1712       llvm::errs() << "Invalid value for -style, using " << FallbackStyle
1713                    << " style\n";
1714     return Style;
1715   }
1716 
1717   // Look for .clang-format/_clang-format file in the file's parent directories.
1718   SmallString<128> UnsuitableConfigFiles;
1719   SmallString<128> Path(FileName);
1720   llvm::sys::fs::make_absolute(Path);
1721   for (StringRef Directory = Path; !Directory.empty();
1722        Directory = llvm::sys::path::parent_path(Directory)) {
1723     if (!llvm::sys::fs::is_directory(Directory))
1724       continue;
1725     SmallString<128> ConfigFile(Directory);
1726 
1727     llvm::sys::path::append(ConfigFile, ".clang-format");
1728     DEBUG(llvm::dbgs() << "Trying " << ConfigFile << "...\n");
1729     bool IsFile = false;
1730     // Ignore errors from is_regular_file: we only need to know if we can read
1731     // the file or not.
1732     llvm::sys::fs::is_regular_file(Twine(ConfigFile), IsFile);
1733 
1734     if (!IsFile) {
1735       // Try _clang-format too, since dotfiles are not commonly used on Windows.
1736       ConfigFile = Directory;
1737       llvm::sys::path::append(ConfigFile, "_clang-format");
1738       DEBUG(llvm::dbgs() << "Trying " << ConfigFile << "...\n");
1739       llvm::sys::fs::is_regular_file(Twine(ConfigFile), IsFile);
1740     }
1741 
1742     if (IsFile) {
1743       OwningPtr<llvm::MemoryBuffer> Text;
1744       if (llvm::error_code ec =
1745               llvm::MemoryBuffer::getFile(ConfigFile.c_str(), Text)) {
1746         llvm::errs() << ec.message() << "\n";
1747         break;
1748       }
1749       if (llvm::error_code ec = parseConfiguration(Text->getBuffer(), &Style)) {
1750         if (ec == llvm::errc::not_supported) {
1751           if (!UnsuitableConfigFiles.empty())
1752             UnsuitableConfigFiles.append(", ");
1753           UnsuitableConfigFiles.append(ConfigFile);
1754           continue;
1755         }
1756         llvm::errs() << "Error reading " << ConfigFile << ": " << ec.message()
1757                      << "\n";
1758         break;
1759       }
1760       DEBUG(llvm::dbgs() << "Using configuration file " << ConfigFile << "\n");
1761       return Style;
1762     }
1763   }
1764   llvm::errs() << "Can't find usable .clang-format, using " << FallbackStyle
1765                << " style\n";
1766   if (!UnsuitableConfigFiles.empty()) {
1767     llvm::errs() << "Configuration file(s) do(es) not support "
1768                  << getLanguageName(Style.Language) << ": "
1769                  << UnsuitableConfigFiles << "\n";
1770   }
1771   return Style;
1772 }
1773 
1774 } // namespace format
1775 } // namespace clang
1776