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 #include "clang/Format/Format.h"
17 #include "AffectedRangeManager.h"
18 #include "ContinuationIndenter.h"
19 #include "FormatTokenLexer.h"
20 #include "SortJavaScriptImports.h"
21 #include "TokenAnalyzer.h"
22 #include "TokenAnnotator.h"
23 #include "UnwrappedLineFormatter.h"
24 #include "UnwrappedLineParser.h"
25 #include "WhitespaceManager.h"
26 #include "clang/Basic/Diagnostic.h"
27 #include "clang/Basic/DiagnosticOptions.h"
28 #include "clang/Basic/SourceManager.h"
29 #include "clang/Basic/VirtualFileSystem.h"
30 #include "clang/Lex/Lexer.h"
31 #include "llvm/ADT/STLExtras.h"
32 #include "llvm/Support/Allocator.h"
33 #include "llvm/Support/Debug.h"
34 #include "llvm/Support/Path.h"
35 #include "llvm/Support/Regex.h"
36 #include "llvm/Support/YAMLTraits.h"
37 #include <algorithm>
38 #include <memory>
39 #include <string>
40 
41 #define DEBUG_TYPE "format-formatter"
42 
43 using clang::format::FormatStyle;
44 
45 LLVM_YAML_IS_FLOW_SEQUENCE_VECTOR(std::string)
46 LLVM_YAML_IS_SEQUENCE_VECTOR(clang::format::FormatStyle::IncludeCategory)
47 
48 namespace llvm {
49 namespace yaml {
50 template <> struct ScalarEnumerationTraits<FormatStyle::LanguageKind> {
51   static void enumeration(IO &IO, FormatStyle::LanguageKind &Value) {
52     IO.enumCase(Value, "Cpp", FormatStyle::LK_Cpp);
53     IO.enumCase(Value, "Java", FormatStyle::LK_Java);
54     IO.enumCase(Value, "JavaScript", FormatStyle::LK_JavaScript);
55     IO.enumCase(Value, "ObjC", FormatStyle::LK_ObjC);
56     IO.enumCase(Value, "Proto", FormatStyle::LK_Proto);
57     IO.enumCase(Value, "TableGen", FormatStyle::LK_TableGen);
58   }
59 };
60 
61 template <> struct ScalarEnumerationTraits<FormatStyle::LanguageStandard> {
62   static void enumeration(IO &IO, FormatStyle::LanguageStandard &Value) {
63     IO.enumCase(Value, "Cpp03", FormatStyle::LS_Cpp03);
64     IO.enumCase(Value, "C++03", FormatStyle::LS_Cpp03);
65     IO.enumCase(Value, "Cpp11", FormatStyle::LS_Cpp11);
66     IO.enumCase(Value, "C++11", FormatStyle::LS_Cpp11);
67     IO.enumCase(Value, "Auto", FormatStyle::LS_Auto);
68   }
69 };
70 
71 template <> struct ScalarEnumerationTraits<FormatStyle::UseTabStyle> {
72   static void enumeration(IO &IO, FormatStyle::UseTabStyle &Value) {
73     IO.enumCase(Value, "Never", FormatStyle::UT_Never);
74     IO.enumCase(Value, "false", FormatStyle::UT_Never);
75     IO.enumCase(Value, "Always", FormatStyle::UT_Always);
76     IO.enumCase(Value, "true", FormatStyle::UT_Always);
77     IO.enumCase(Value, "ForIndentation", FormatStyle::UT_ForIndentation);
78     IO.enumCase(Value, "ForContinuationAndIndentation",
79                 FormatStyle::UT_ForContinuationAndIndentation);
80   }
81 };
82 
83 template <> struct ScalarEnumerationTraits<FormatStyle::JavaScriptQuoteStyle> {
84   static void enumeration(IO &IO, FormatStyle::JavaScriptQuoteStyle &Value) {
85     IO.enumCase(Value, "Leave", FormatStyle::JSQS_Leave);
86     IO.enumCase(Value, "Single", FormatStyle::JSQS_Single);
87     IO.enumCase(Value, "Double", FormatStyle::JSQS_Double);
88   }
89 };
90 
91 template <> struct ScalarEnumerationTraits<FormatStyle::ShortFunctionStyle> {
92   static void enumeration(IO &IO, FormatStyle::ShortFunctionStyle &Value) {
93     IO.enumCase(Value, "None", FormatStyle::SFS_None);
94     IO.enumCase(Value, "false", FormatStyle::SFS_None);
95     IO.enumCase(Value, "All", FormatStyle::SFS_All);
96     IO.enumCase(Value, "true", FormatStyle::SFS_All);
97     IO.enumCase(Value, "Inline", FormatStyle::SFS_Inline);
98     IO.enumCase(Value, "Empty", FormatStyle::SFS_Empty);
99   }
100 };
101 
102 template <> struct ScalarEnumerationTraits<FormatStyle::BinaryOperatorStyle> {
103   static void enumeration(IO &IO, FormatStyle::BinaryOperatorStyle &Value) {
104     IO.enumCase(Value, "All", FormatStyle::BOS_All);
105     IO.enumCase(Value, "true", FormatStyle::BOS_All);
106     IO.enumCase(Value, "None", FormatStyle::BOS_None);
107     IO.enumCase(Value, "false", FormatStyle::BOS_None);
108     IO.enumCase(Value, "NonAssignment", FormatStyle::BOS_NonAssignment);
109   }
110 };
111 
112 template <> struct ScalarEnumerationTraits<FormatStyle::BraceBreakingStyle> {
113   static void enumeration(IO &IO, FormatStyle::BraceBreakingStyle &Value) {
114     IO.enumCase(Value, "Attach", FormatStyle::BS_Attach);
115     IO.enumCase(Value, "Linux", FormatStyle::BS_Linux);
116     IO.enumCase(Value, "Mozilla", FormatStyle::BS_Mozilla);
117     IO.enumCase(Value, "Stroustrup", FormatStyle::BS_Stroustrup);
118     IO.enumCase(Value, "Allman", FormatStyle::BS_Allman);
119     IO.enumCase(Value, "GNU", FormatStyle::BS_GNU);
120     IO.enumCase(Value, "WebKit", FormatStyle::BS_WebKit);
121     IO.enumCase(Value, "Custom", FormatStyle::BS_Custom);
122   }
123 };
124 
125 template <>
126 struct ScalarEnumerationTraits<FormatStyle::ReturnTypeBreakingStyle> {
127   static void enumeration(IO &IO, FormatStyle::ReturnTypeBreakingStyle &Value) {
128     IO.enumCase(Value, "None", FormatStyle::RTBS_None);
129     IO.enumCase(Value, "All", FormatStyle::RTBS_All);
130     IO.enumCase(Value, "TopLevel", FormatStyle::RTBS_TopLevel);
131     IO.enumCase(Value, "TopLevelDefinitions",
132                 FormatStyle::RTBS_TopLevelDefinitions);
133     IO.enumCase(Value, "AllDefinitions", FormatStyle::RTBS_AllDefinitions);
134   }
135 };
136 
137 template <>
138 struct ScalarEnumerationTraits<FormatStyle::DefinitionReturnTypeBreakingStyle> {
139   static void
140   enumeration(IO &IO, FormatStyle::DefinitionReturnTypeBreakingStyle &Value) {
141     IO.enumCase(Value, "None", FormatStyle::DRTBS_None);
142     IO.enumCase(Value, "All", FormatStyle::DRTBS_All);
143     IO.enumCase(Value, "TopLevel", FormatStyle::DRTBS_TopLevel);
144 
145     // For backward compatibility.
146     IO.enumCase(Value, "false", FormatStyle::DRTBS_None);
147     IO.enumCase(Value, "true", FormatStyle::DRTBS_All);
148   }
149 };
150 
151 template <>
152 struct ScalarEnumerationTraits<FormatStyle::NamespaceIndentationKind> {
153   static void enumeration(IO &IO,
154                           FormatStyle::NamespaceIndentationKind &Value) {
155     IO.enumCase(Value, "None", FormatStyle::NI_None);
156     IO.enumCase(Value, "Inner", FormatStyle::NI_Inner);
157     IO.enumCase(Value, "All", FormatStyle::NI_All);
158   }
159 };
160 
161 template <> struct ScalarEnumerationTraits<FormatStyle::BracketAlignmentStyle> {
162   static void enumeration(IO &IO, FormatStyle::BracketAlignmentStyle &Value) {
163     IO.enumCase(Value, "Align", FormatStyle::BAS_Align);
164     IO.enumCase(Value, "DontAlign", FormatStyle::BAS_DontAlign);
165     IO.enumCase(Value, "AlwaysBreak", FormatStyle::BAS_AlwaysBreak);
166 
167     // For backward compatibility.
168     IO.enumCase(Value, "true", FormatStyle::BAS_Align);
169     IO.enumCase(Value, "false", FormatStyle::BAS_DontAlign);
170   }
171 };
172 
173 template <> struct ScalarEnumerationTraits<FormatStyle::PointerAlignmentStyle> {
174   static void enumeration(IO &IO, FormatStyle::PointerAlignmentStyle &Value) {
175     IO.enumCase(Value, "Middle", FormatStyle::PAS_Middle);
176     IO.enumCase(Value, "Left", FormatStyle::PAS_Left);
177     IO.enumCase(Value, "Right", FormatStyle::PAS_Right);
178 
179     // For backward compatibility.
180     IO.enumCase(Value, "true", FormatStyle::PAS_Left);
181     IO.enumCase(Value, "false", FormatStyle::PAS_Right);
182   }
183 };
184 
185 template <>
186 struct ScalarEnumerationTraits<FormatStyle::SpaceBeforeParensOptions> {
187   static void enumeration(IO &IO,
188                           FormatStyle::SpaceBeforeParensOptions &Value) {
189     IO.enumCase(Value, "Never", FormatStyle::SBPO_Never);
190     IO.enumCase(Value, "ControlStatements",
191                 FormatStyle::SBPO_ControlStatements);
192     IO.enumCase(Value, "Always", FormatStyle::SBPO_Always);
193 
194     // For backward compatibility.
195     IO.enumCase(Value, "false", FormatStyle::SBPO_Never);
196     IO.enumCase(Value, "true", FormatStyle::SBPO_ControlStatements);
197   }
198 };
199 
200 template <> struct MappingTraits<FormatStyle> {
201   static void mapping(IO &IO, FormatStyle &Style) {
202     // When reading, read the language first, we need it for getPredefinedStyle.
203     IO.mapOptional("Language", Style.Language);
204 
205     if (IO.outputting()) {
206       StringRef StylesArray[] = {"LLVM",    "Google", "Chromium",
207                                  "Mozilla", "WebKit", "GNU"};
208       ArrayRef<StringRef> Styles(StylesArray);
209       for (size_t i = 0, e = Styles.size(); i < e; ++i) {
210         StringRef StyleName(Styles[i]);
211         FormatStyle PredefinedStyle;
212         if (getPredefinedStyle(StyleName, Style.Language, &PredefinedStyle) &&
213             Style == PredefinedStyle) {
214           IO.mapOptional("# BasedOnStyle", StyleName);
215           break;
216         }
217       }
218     } else {
219       StringRef BasedOnStyle;
220       IO.mapOptional("BasedOnStyle", BasedOnStyle);
221       if (!BasedOnStyle.empty()) {
222         FormatStyle::LanguageKind OldLanguage = Style.Language;
223         FormatStyle::LanguageKind Language =
224             ((FormatStyle *)IO.getContext())->Language;
225         if (!getPredefinedStyle(BasedOnStyle, Language, &Style)) {
226           IO.setError(Twine("Unknown value for BasedOnStyle: ", BasedOnStyle));
227           return;
228         }
229         Style.Language = OldLanguage;
230       }
231     }
232 
233     // For backward compatibility.
234     if (!IO.outputting()) {
235       IO.mapOptional("DerivePointerBinding", Style.DerivePointerAlignment);
236       IO.mapOptional("IndentFunctionDeclarationAfterType",
237                      Style.IndentWrappedFunctionNames);
238       IO.mapOptional("PointerBindsToType", Style.PointerAlignment);
239       IO.mapOptional("SpaceAfterControlStatementKeyword",
240                      Style.SpaceBeforeParens);
241     }
242 
243     IO.mapOptional("AccessModifierOffset", Style.AccessModifierOffset);
244     IO.mapOptional("AlignAfterOpenBracket", Style.AlignAfterOpenBracket);
245     IO.mapOptional("AlignConsecutiveAssignments",
246                    Style.AlignConsecutiveAssignments);
247     IO.mapOptional("AlignConsecutiveDeclarations",
248                    Style.AlignConsecutiveDeclarations);
249     IO.mapOptional("AlignEscapedNewlinesLeft", Style.AlignEscapedNewlinesLeft);
250     IO.mapOptional("AlignOperands", Style.AlignOperands);
251     IO.mapOptional("AlignTrailingComments", Style.AlignTrailingComments);
252     IO.mapOptional("AllowAllParametersOfDeclarationOnNextLine",
253                    Style.AllowAllParametersOfDeclarationOnNextLine);
254     IO.mapOptional("AllowShortBlocksOnASingleLine",
255                    Style.AllowShortBlocksOnASingleLine);
256     IO.mapOptional("AllowShortCaseLabelsOnASingleLine",
257                    Style.AllowShortCaseLabelsOnASingleLine);
258     IO.mapOptional("AllowShortFunctionsOnASingleLine",
259                    Style.AllowShortFunctionsOnASingleLine);
260     IO.mapOptional("AllowShortIfStatementsOnASingleLine",
261                    Style.AllowShortIfStatementsOnASingleLine);
262     IO.mapOptional("AllowShortLoopsOnASingleLine",
263                    Style.AllowShortLoopsOnASingleLine);
264     IO.mapOptional("AlwaysBreakAfterDefinitionReturnType",
265                    Style.AlwaysBreakAfterDefinitionReturnType);
266     IO.mapOptional("AlwaysBreakAfterReturnType",
267                    Style.AlwaysBreakAfterReturnType);
268     // If AlwaysBreakAfterDefinitionReturnType was specified but
269     // AlwaysBreakAfterReturnType was not, initialize the latter from the
270     // former for backwards compatibility.
271     if (Style.AlwaysBreakAfterDefinitionReturnType != FormatStyle::DRTBS_None &&
272         Style.AlwaysBreakAfterReturnType == FormatStyle::RTBS_None) {
273       if (Style.AlwaysBreakAfterDefinitionReturnType == FormatStyle::DRTBS_All)
274         Style.AlwaysBreakAfterReturnType = FormatStyle::RTBS_AllDefinitions;
275       else if (Style.AlwaysBreakAfterDefinitionReturnType ==
276                FormatStyle::DRTBS_TopLevel)
277         Style.AlwaysBreakAfterReturnType =
278             FormatStyle::RTBS_TopLevelDefinitions;
279     }
280 
281     IO.mapOptional("AlwaysBreakBeforeMultilineStrings",
282                    Style.AlwaysBreakBeforeMultilineStrings);
283     IO.mapOptional("AlwaysBreakTemplateDeclarations",
284                    Style.AlwaysBreakTemplateDeclarations);
285     IO.mapOptional("BinPackArguments", Style.BinPackArguments);
286     IO.mapOptional("BinPackParameters", Style.BinPackParameters);
287     IO.mapOptional("BraceWrapping", Style.BraceWrapping);
288     IO.mapOptional("BreakBeforeBinaryOperators",
289                    Style.BreakBeforeBinaryOperators);
290     IO.mapOptional("BreakBeforeBraces", Style.BreakBeforeBraces);
291     IO.mapOptional("BreakBeforeTernaryOperators",
292                    Style.BreakBeforeTernaryOperators);
293     IO.mapOptional("BreakConstructorInitializersBeforeComma",
294                    Style.BreakConstructorInitializersBeforeComma);
295     IO.mapOptional("BreakAfterJavaFieldAnnotations",
296                    Style.BreakAfterJavaFieldAnnotations);
297     IO.mapOptional("BreakStringLiterals", Style.BreakStringLiterals);
298     IO.mapOptional("ColumnLimit", Style.ColumnLimit);
299     IO.mapOptional("CommentPragmas", Style.CommentPragmas);
300     IO.mapOptional("ConstructorInitializerAllOnOneLineOrOnePerLine",
301                    Style.ConstructorInitializerAllOnOneLineOrOnePerLine);
302     IO.mapOptional("ConstructorInitializerIndentWidth",
303                    Style.ConstructorInitializerIndentWidth);
304     IO.mapOptional("ContinuationIndentWidth", Style.ContinuationIndentWidth);
305     IO.mapOptional("Cpp11BracedListStyle", Style.Cpp11BracedListStyle);
306     IO.mapOptional("DerivePointerAlignment", Style.DerivePointerAlignment);
307     IO.mapOptional("DisableFormat", Style.DisableFormat);
308     IO.mapOptional("ExperimentalAutoDetectBinPacking",
309                    Style.ExperimentalAutoDetectBinPacking);
310     IO.mapOptional("ForEachMacros", Style.ForEachMacros);
311     IO.mapOptional("IncludeCategories", Style.IncludeCategories);
312     IO.mapOptional("IncludeIsMainRegex", Style.IncludeIsMainRegex);
313     IO.mapOptional("IndentCaseLabels", Style.IndentCaseLabels);
314     IO.mapOptional("IndentWidth", Style.IndentWidth);
315     IO.mapOptional("IndentWrappedFunctionNames",
316                    Style.IndentWrappedFunctionNames);
317     IO.mapOptional("JavaScriptQuotes", Style.JavaScriptQuotes);
318     IO.mapOptional("JavaScriptWrapImports", Style.JavaScriptWrapImports);
319     IO.mapOptional("KeepEmptyLinesAtTheStartOfBlocks",
320                    Style.KeepEmptyLinesAtTheStartOfBlocks);
321     IO.mapOptional("MacroBlockBegin", Style.MacroBlockBegin);
322     IO.mapOptional("MacroBlockEnd", Style.MacroBlockEnd);
323     IO.mapOptional("MaxEmptyLinesToKeep", Style.MaxEmptyLinesToKeep);
324     IO.mapOptional("NamespaceIndentation", Style.NamespaceIndentation);
325     IO.mapOptional("ObjCBlockIndentWidth", Style.ObjCBlockIndentWidth);
326     IO.mapOptional("ObjCSpaceAfterProperty", Style.ObjCSpaceAfterProperty);
327     IO.mapOptional("ObjCSpaceBeforeProtocolList",
328                    Style.ObjCSpaceBeforeProtocolList);
329     IO.mapOptional("PenaltyBreakBeforeFirstCallParameter",
330                    Style.PenaltyBreakBeforeFirstCallParameter);
331     IO.mapOptional("PenaltyBreakComment", Style.PenaltyBreakComment);
332     IO.mapOptional("PenaltyBreakFirstLessLess",
333                    Style.PenaltyBreakFirstLessLess);
334     IO.mapOptional("PenaltyBreakString", Style.PenaltyBreakString);
335     IO.mapOptional("PenaltyExcessCharacter", Style.PenaltyExcessCharacter);
336     IO.mapOptional("PenaltyReturnTypeOnItsOwnLine",
337                    Style.PenaltyReturnTypeOnItsOwnLine);
338     IO.mapOptional("PointerAlignment", Style.PointerAlignment);
339     IO.mapOptional("ReflowComments", Style.ReflowComments);
340     IO.mapOptional("SortIncludes", Style.SortIncludes);
341     IO.mapOptional("SpaceAfterCStyleCast", Style.SpaceAfterCStyleCast);
342     IO.mapOptional("SpaceAfterTemplateKeyword", Style.SpaceAfterTemplateKeyword);
343     IO.mapOptional("SpaceBeforeAssignmentOperators",
344                    Style.SpaceBeforeAssignmentOperators);
345     IO.mapOptional("SpaceBeforeParens", Style.SpaceBeforeParens);
346     IO.mapOptional("SpaceInEmptyParentheses", Style.SpaceInEmptyParentheses);
347     IO.mapOptional("SpacesBeforeTrailingComments",
348                    Style.SpacesBeforeTrailingComments);
349     IO.mapOptional("SpacesInAngles", Style.SpacesInAngles);
350     IO.mapOptional("SpacesInContainerLiterals",
351                    Style.SpacesInContainerLiterals);
352     IO.mapOptional("SpacesInCStyleCastParentheses",
353                    Style.SpacesInCStyleCastParentheses);
354     IO.mapOptional("SpacesInParentheses", Style.SpacesInParentheses);
355     IO.mapOptional("SpacesInSquareBrackets", Style.SpacesInSquareBrackets);
356     IO.mapOptional("Standard", Style.Standard);
357     IO.mapOptional("TabWidth", Style.TabWidth);
358     IO.mapOptional("UseTab", Style.UseTab);
359   }
360 };
361 
362 template <> struct MappingTraits<FormatStyle::BraceWrappingFlags> {
363   static void mapping(IO &IO, FormatStyle::BraceWrappingFlags &Wrapping) {
364     IO.mapOptional("AfterClass", Wrapping.AfterClass);
365     IO.mapOptional("AfterControlStatement", Wrapping.AfterControlStatement);
366     IO.mapOptional("AfterEnum", Wrapping.AfterEnum);
367     IO.mapOptional("AfterFunction", Wrapping.AfterFunction);
368     IO.mapOptional("AfterNamespace", Wrapping.AfterNamespace);
369     IO.mapOptional("AfterObjCDeclaration", Wrapping.AfterObjCDeclaration);
370     IO.mapOptional("AfterStruct", Wrapping.AfterStruct);
371     IO.mapOptional("AfterUnion", Wrapping.AfterUnion);
372     IO.mapOptional("BeforeCatch", Wrapping.BeforeCatch);
373     IO.mapOptional("BeforeElse", Wrapping.BeforeElse);
374     IO.mapOptional("IndentBraces", Wrapping.IndentBraces);
375   }
376 };
377 
378 template <> struct MappingTraits<FormatStyle::IncludeCategory> {
379   static void mapping(IO &IO, FormatStyle::IncludeCategory &Category) {
380     IO.mapOptional("Regex", Category.Regex);
381     IO.mapOptional("Priority", Category.Priority);
382   }
383 };
384 
385 // Allows to read vector<FormatStyle> while keeping default values.
386 // IO.getContext() should contain a pointer to the FormatStyle structure, that
387 // will be used to get default values for missing keys.
388 // If the first element has no Language specified, it will be treated as the
389 // default one for the following elements.
390 template <> struct DocumentListTraits<std::vector<FormatStyle>> {
391   static size_t size(IO &IO, std::vector<FormatStyle> &Seq) {
392     return Seq.size();
393   }
394   static FormatStyle &element(IO &IO, std::vector<FormatStyle> &Seq,
395                               size_t Index) {
396     if (Index >= Seq.size()) {
397       assert(Index == Seq.size());
398       FormatStyle Template;
399       if (Seq.size() > 0 && Seq[0].Language == FormatStyle::LK_None) {
400         Template = Seq[0];
401       } else {
402         Template = *((const FormatStyle *)IO.getContext());
403         Template.Language = FormatStyle::LK_None;
404       }
405       Seq.resize(Index + 1, Template);
406     }
407     return Seq[Index];
408   }
409 };
410 } // namespace yaml
411 } // namespace llvm
412 
413 namespace clang {
414 namespace format {
415 
416 const std::error_category &getParseCategory() {
417   static ParseErrorCategory C;
418   return C;
419 }
420 std::error_code make_error_code(ParseError e) {
421   return std::error_code(static_cast<int>(e), getParseCategory());
422 }
423 
424 const char *ParseErrorCategory::name() const noexcept {
425   return "clang-format.parse_error";
426 }
427 
428 std::string ParseErrorCategory::message(int EV) const {
429   switch (static_cast<ParseError>(EV)) {
430   case ParseError::Success:
431     return "Success";
432   case ParseError::Error:
433     return "Invalid argument";
434   case ParseError::Unsuitable:
435     return "Unsuitable";
436   }
437   llvm_unreachable("unexpected parse error");
438 }
439 
440 static FormatStyle expandPresets(const FormatStyle &Style) {
441   if (Style.BreakBeforeBraces == FormatStyle::BS_Custom)
442     return Style;
443   FormatStyle Expanded = Style;
444   Expanded.BraceWrapping = {false, false, false, false, false, false,
445                             false, false, false, false, false};
446   switch (Style.BreakBeforeBraces) {
447   case FormatStyle::BS_Linux:
448     Expanded.BraceWrapping.AfterClass = true;
449     Expanded.BraceWrapping.AfterFunction = true;
450     Expanded.BraceWrapping.AfterNamespace = true;
451     break;
452   case FormatStyle::BS_Mozilla:
453     Expanded.BraceWrapping.AfterClass = true;
454     Expanded.BraceWrapping.AfterEnum = true;
455     Expanded.BraceWrapping.AfterFunction = true;
456     Expanded.BraceWrapping.AfterStruct = true;
457     Expanded.BraceWrapping.AfterUnion = true;
458     break;
459   case FormatStyle::BS_Stroustrup:
460     Expanded.BraceWrapping.AfterFunction = true;
461     Expanded.BraceWrapping.BeforeCatch = true;
462     Expanded.BraceWrapping.BeforeElse = true;
463     break;
464   case FormatStyle::BS_Allman:
465     Expanded.BraceWrapping.AfterClass = true;
466     Expanded.BraceWrapping.AfterControlStatement = true;
467     Expanded.BraceWrapping.AfterEnum = true;
468     Expanded.BraceWrapping.AfterFunction = true;
469     Expanded.BraceWrapping.AfterNamespace = true;
470     Expanded.BraceWrapping.AfterObjCDeclaration = true;
471     Expanded.BraceWrapping.AfterStruct = true;
472     Expanded.BraceWrapping.BeforeCatch = true;
473     Expanded.BraceWrapping.BeforeElse = true;
474     break;
475   case FormatStyle::BS_GNU:
476     Expanded.BraceWrapping = {true, true, true, true, true, true,
477                               true, true, true, true, true};
478     break;
479   case FormatStyle::BS_WebKit:
480     Expanded.BraceWrapping.AfterFunction = true;
481     break;
482   default:
483     break;
484   }
485   return Expanded;
486 }
487 
488 FormatStyle getLLVMStyle() {
489   FormatStyle LLVMStyle;
490   LLVMStyle.Language = FormatStyle::LK_Cpp;
491   LLVMStyle.AccessModifierOffset = -2;
492   LLVMStyle.AlignEscapedNewlinesLeft = false;
493   LLVMStyle.AlignAfterOpenBracket = FormatStyle::BAS_Align;
494   LLVMStyle.AlignOperands = true;
495   LLVMStyle.AlignTrailingComments = true;
496   LLVMStyle.AlignConsecutiveAssignments = false;
497   LLVMStyle.AlignConsecutiveDeclarations = false;
498   LLVMStyle.AllowAllParametersOfDeclarationOnNextLine = true;
499   LLVMStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_All;
500   LLVMStyle.AllowShortBlocksOnASingleLine = false;
501   LLVMStyle.AllowShortCaseLabelsOnASingleLine = false;
502   LLVMStyle.AllowShortIfStatementsOnASingleLine = false;
503   LLVMStyle.AllowShortLoopsOnASingleLine = false;
504   LLVMStyle.AlwaysBreakAfterReturnType = FormatStyle::RTBS_None;
505   LLVMStyle.AlwaysBreakAfterDefinitionReturnType = FormatStyle::DRTBS_None;
506   LLVMStyle.AlwaysBreakBeforeMultilineStrings = false;
507   LLVMStyle.AlwaysBreakTemplateDeclarations = false;
508   LLVMStyle.BinPackParameters = true;
509   LLVMStyle.BinPackArguments = true;
510   LLVMStyle.BreakBeforeBinaryOperators = FormatStyle::BOS_None;
511   LLVMStyle.BreakBeforeTernaryOperators = true;
512   LLVMStyle.BreakBeforeBraces = FormatStyle::BS_Attach;
513   LLVMStyle.BraceWrapping = {false, false, false, false, false, false,
514                              false, false, false, false, false};
515   LLVMStyle.BreakAfterJavaFieldAnnotations = false;
516   LLVMStyle.BreakConstructorInitializersBeforeComma = false;
517   LLVMStyle.BreakStringLiterals = true;
518   LLVMStyle.ColumnLimit = 80;
519   LLVMStyle.CommentPragmas = "^ IWYU pragma:";
520   LLVMStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = false;
521   LLVMStyle.ConstructorInitializerIndentWidth = 4;
522   LLVMStyle.ContinuationIndentWidth = 4;
523   LLVMStyle.Cpp11BracedListStyle = true;
524   LLVMStyle.DerivePointerAlignment = false;
525   LLVMStyle.ExperimentalAutoDetectBinPacking = false;
526   LLVMStyle.ForEachMacros.push_back("foreach");
527   LLVMStyle.ForEachMacros.push_back("Q_FOREACH");
528   LLVMStyle.ForEachMacros.push_back("BOOST_FOREACH");
529   LLVMStyle.IncludeCategories = {{"^\"(llvm|llvm-c|clang|clang-c)/", 2},
530                                  {"^(<|\"(gtest|isl|json)/)", 3},
531                                  {".*", 1}};
532   LLVMStyle.IncludeIsMainRegex = "$";
533   LLVMStyle.IndentCaseLabels = false;
534   LLVMStyle.IndentWrappedFunctionNames = false;
535   LLVMStyle.IndentWidth = 2;
536   LLVMStyle.JavaScriptQuotes = FormatStyle::JSQS_Leave;
537   LLVMStyle.JavaScriptWrapImports = true;
538   LLVMStyle.TabWidth = 8;
539   LLVMStyle.MaxEmptyLinesToKeep = 1;
540   LLVMStyle.KeepEmptyLinesAtTheStartOfBlocks = true;
541   LLVMStyle.NamespaceIndentation = FormatStyle::NI_None;
542   LLVMStyle.ObjCBlockIndentWidth = 2;
543   LLVMStyle.ObjCSpaceAfterProperty = false;
544   LLVMStyle.ObjCSpaceBeforeProtocolList = true;
545   LLVMStyle.PointerAlignment = FormatStyle::PAS_Right;
546   LLVMStyle.SpacesBeforeTrailingComments = 1;
547   LLVMStyle.Standard = FormatStyle::LS_Cpp11;
548   LLVMStyle.UseTab = FormatStyle::UT_Never;
549   LLVMStyle.JavaScriptQuotes = FormatStyle::JSQS_Leave;
550   LLVMStyle.ReflowComments = true;
551   LLVMStyle.SpacesInParentheses = false;
552   LLVMStyle.SpacesInSquareBrackets = false;
553   LLVMStyle.SpaceInEmptyParentheses = false;
554   LLVMStyle.SpacesInContainerLiterals = true;
555   LLVMStyle.SpacesInCStyleCastParentheses = false;
556   LLVMStyle.SpaceAfterCStyleCast = false;
557   LLVMStyle.SpaceAfterTemplateKeyword = true;
558   LLVMStyle.SpaceBeforeParens = FormatStyle::SBPO_ControlStatements;
559   LLVMStyle.SpaceBeforeAssignmentOperators = true;
560   LLVMStyle.SpacesInAngles = false;
561 
562   LLVMStyle.PenaltyBreakComment = 300;
563   LLVMStyle.PenaltyBreakFirstLessLess = 120;
564   LLVMStyle.PenaltyBreakString = 1000;
565   LLVMStyle.PenaltyExcessCharacter = 1000000;
566   LLVMStyle.PenaltyReturnTypeOnItsOwnLine = 60;
567   LLVMStyle.PenaltyBreakBeforeFirstCallParameter = 19;
568 
569   LLVMStyle.DisableFormat = false;
570   LLVMStyle.SortIncludes = true;
571 
572   return LLVMStyle;
573 }
574 
575 FormatStyle getGoogleStyle(FormatStyle::LanguageKind Language) {
576   FormatStyle GoogleStyle = getLLVMStyle();
577   GoogleStyle.Language = Language;
578 
579   GoogleStyle.AccessModifierOffset = -1;
580   GoogleStyle.AlignEscapedNewlinesLeft = true;
581   GoogleStyle.AllowShortIfStatementsOnASingleLine = true;
582   GoogleStyle.AllowShortLoopsOnASingleLine = true;
583   GoogleStyle.AlwaysBreakBeforeMultilineStrings = true;
584   GoogleStyle.AlwaysBreakTemplateDeclarations = true;
585   GoogleStyle.ConstructorInitializerAllOnOneLineOrOnePerLine = true;
586   GoogleStyle.DerivePointerAlignment = true;
587   GoogleStyle.IncludeCategories = {{"^<.*\\.h>", 1}, {"^<.*", 2}, {".*", 3}};
588   GoogleStyle.IncludeIsMainRegex = "([-_](test|unittest))?$";
589   GoogleStyle.IndentCaseLabels = true;
590   GoogleStyle.KeepEmptyLinesAtTheStartOfBlocks = false;
591   GoogleStyle.ObjCSpaceAfterProperty = false;
592   GoogleStyle.ObjCSpaceBeforeProtocolList = false;
593   GoogleStyle.PointerAlignment = FormatStyle::PAS_Left;
594   GoogleStyle.SpacesBeforeTrailingComments = 2;
595   GoogleStyle.Standard = FormatStyle::LS_Auto;
596 
597   GoogleStyle.PenaltyReturnTypeOnItsOwnLine = 200;
598   GoogleStyle.PenaltyBreakBeforeFirstCallParameter = 1;
599 
600   if (Language == FormatStyle::LK_Java) {
601     GoogleStyle.AlignAfterOpenBracket = FormatStyle::BAS_DontAlign;
602     GoogleStyle.AlignOperands = false;
603     GoogleStyle.AlignTrailingComments = false;
604     GoogleStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_Empty;
605     GoogleStyle.AllowShortIfStatementsOnASingleLine = false;
606     GoogleStyle.AlwaysBreakBeforeMultilineStrings = false;
607     GoogleStyle.BreakBeforeBinaryOperators = FormatStyle::BOS_NonAssignment;
608     GoogleStyle.ColumnLimit = 100;
609     GoogleStyle.SpaceAfterCStyleCast = true;
610     GoogleStyle.SpacesBeforeTrailingComments = 1;
611   } else if (Language == FormatStyle::LK_JavaScript) {
612     GoogleStyle.AlignAfterOpenBracket = FormatStyle::BAS_AlwaysBreak;
613     GoogleStyle.AlignOperands = false;
614     GoogleStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_Empty;
615     GoogleStyle.AlwaysBreakBeforeMultilineStrings = false;
616     GoogleStyle.BreakBeforeTernaryOperators = false;
617     GoogleStyle.CommentPragmas =
618         "(taze:|@(export|requirecss|return|returns|see|visibility)) ";
619     GoogleStyle.MaxEmptyLinesToKeep = 3;
620     GoogleStyle.NamespaceIndentation = FormatStyle::NI_All;
621     GoogleStyle.SpacesInContainerLiterals = false;
622     GoogleStyle.JavaScriptQuotes = FormatStyle::JSQS_Single;
623     GoogleStyle.JavaScriptWrapImports = false;
624   } else if (Language == FormatStyle::LK_Proto) {
625     GoogleStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_None;
626     GoogleStyle.SpacesInContainerLiterals = false;
627   } else if (Language == FormatStyle::LK_ObjC) {
628     GoogleStyle.ColumnLimit = 100;
629   }
630 
631   return GoogleStyle;
632 }
633 
634 FormatStyle getChromiumStyle(FormatStyle::LanguageKind Language) {
635   FormatStyle ChromiumStyle = getGoogleStyle(Language);
636   if (Language == FormatStyle::LK_Java) {
637     ChromiumStyle.AllowShortIfStatementsOnASingleLine = true;
638     ChromiumStyle.BreakAfterJavaFieldAnnotations = true;
639     ChromiumStyle.ContinuationIndentWidth = 8;
640     ChromiumStyle.IndentWidth = 4;
641   } else {
642     ChromiumStyle.AllowAllParametersOfDeclarationOnNextLine = false;
643     ChromiumStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_Inline;
644     ChromiumStyle.AllowShortIfStatementsOnASingleLine = false;
645     ChromiumStyle.AllowShortLoopsOnASingleLine = false;
646     ChromiumStyle.BinPackParameters = false;
647     ChromiumStyle.DerivePointerAlignment = false;
648   }
649   ChromiumStyle.SortIncludes = false;
650   return ChromiumStyle;
651 }
652 
653 FormatStyle getMozillaStyle() {
654   FormatStyle MozillaStyle = getLLVMStyle();
655   MozillaStyle.AllowAllParametersOfDeclarationOnNextLine = false;
656   MozillaStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_Inline;
657   MozillaStyle.AlwaysBreakAfterReturnType =
658       FormatStyle::RTBS_TopLevel;
659   MozillaStyle.AlwaysBreakAfterDefinitionReturnType =
660       FormatStyle::DRTBS_TopLevel;
661   MozillaStyle.AlwaysBreakTemplateDeclarations = true;
662   MozillaStyle.BinPackParameters = false;
663   MozillaStyle.BinPackArguments = false;
664   MozillaStyle.BreakBeforeBraces = FormatStyle::BS_Mozilla;
665   MozillaStyle.BreakConstructorInitializersBeforeComma = true;
666   MozillaStyle.ConstructorInitializerIndentWidth = 2;
667   MozillaStyle.ContinuationIndentWidth = 2;
668   MozillaStyle.Cpp11BracedListStyle = false;
669   MozillaStyle.IndentCaseLabels = true;
670   MozillaStyle.ObjCSpaceAfterProperty = true;
671   MozillaStyle.ObjCSpaceBeforeProtocolList = false;
672   MozillaStyle.PenaltyReturnTypeOnItsOwnLine = 200;
673   MozillaStyle.PointerAlignment = FormatStyle::PAS_Left;
674   MozillaStyle.SpaceAfterTemplateKeyword = false;
675   return MozillaStyle;
676 }
677 
678 FormatStyle getWebKitStyle() {
679   FormatStyle Style = getLLVMStyle();
680   Style.AccessModifierOffset = -4;
681   Style.AlignAfterOpenBracket = FormatStyle::BAS_DontAlign;
682   Style.AlignOperands = false;
683   Style.AlignTrailingComments = false;
684   Style.BreakBeforeBinaryOperators = FormatStyle::BOS_All;
685   Style.BreakBeforeBraces = FormatStyle::BS_WebKit;
686   Style.BreakConstructorInitializersBeforeComma = true;
687   Style.Cpp11BracedListStyle = false;
688   Style.ColumnLimit = 0;
689   Style.IndentWidth = 4;
690   Style.NamespaceIndentation = FormatStyle::NI_Inner;
691   Style.ObjCBlockIndentWidth = 4;
692   Style.ObjCSpaceAfterProperty = true;
693   Style.PointerAlignment = FormatStyle::PAS_Left;
694   return Style;
695 }
696 
697 FormatStyle getGNUStyle() {
698   FormatStyle Style = getLLVMStyle();
699   Style.AlwaysBreakAfterDefinitionReturnType = FormatStyle::DRTBS_All;
700   Style.AlwaysBreakAfterReturnType = FormatStyle::RTBS_AllDefinitions;
701   Style.BreakBeforeBinaryOperators = FormatStyle::BOS_All;
702   Style.BreakBeforeBraces = FormatStyle::BS_GNU;
703   Style.BreakBeforeTernaryOperators = true;
704   Style.Cpp11BracedListStyle = false;
705   Style.ColumnLimit = 79;
706   Style.SpaceBeforeParens = FormatStyle::SBPO_Always;
707   Style.Standard = FormatStyle::LS_Cpp03;
708   return Style;
709 }
710 
711 FormatStyle getNoStyle() {
712   FormatStyle NoStyle = getLLVMStyle();
713   NoStyle.DisableFormat = true;
714   NoStyle.SortIncludes = false;
715   return NoStyle;
716 }
717 
718 bool getPredefinedStyle(StringRef Name, FormatStyle::LanguageKind Language,
719                         FormatStyle *Style) {
720   if (Name.equals_lower("llvm")) {
721     *Style = getLLVMStyle();
722   } else if (Name.equals_lower("chromium")) {
723     *Style = getChromiumStyle(Language);
724   } else if (Name.equals_lower("mozilla")) {
725     *Style = getMozillaStyle();
726   } else if (Name.equals_lower("google")) {
727     *Style = getGoogleStyle(Language);
728   } else if (Name.equals_lower("webkit")) {
729     *Style = getWebKitStyle();
730   } else if (Name.equals_lower("gnu")) {
731     *Style = getGNUStyle();
732   } else if (Name.equals_lower("none")) {
733     *Style = getNoStyle();
734   } else {
735     return false;
736   }
737 
738   Style->Language = Language;
739   return true;
740 }
741 
742 std::error_code parseConfiguration(StringRef Text, FormatStyle *Style) {
743   assert(Style);
744   FormatStyle::LanguageKind Language = Style->Language;
745   assert(Language != FormatStyle::LK_None);
746   if (Text.trim().empty())
747     return make_error_code(ParseError::Error);
748 
749   std::vector<FormatStyle> Styles;
750   llvm::yaml::Input Input(Text);
751   // DocumentListTraits<vector<FormatStyle>> uses the context to get default
752   // values for the fields, keys for which are missing from the configuration.
753   // Mapping also uses the context to get the language to find the correct
754   // base style.
755   Input.setContext(Style);
756   Input >> Styles;
757   if (Input.error())
758     return Input.error();
759 
760   for (unsigned i = 0; i < Styles.size(); ++i) {
761     // Ensures that only the first configuration can skip the Language option.
762     if (Styles[i].Language == FormatStyle::LK_None && i != 0)
763       return make_error_code(ParseError::Error);
764     // Ensure that each language is configured at most once.
765     for (unsigned j = 0; j < i; ++j) {
766       if (Styles[i].Language == Styles[j].Language) {
767         DEBUG(llvm::dbgs()
768               << "Duplicate languages in the config file on positions " << j
769               << " and " << i << "\n");
770         return make_error_code(ParseError::Error);
771       }
772     }
773   }
774   // Look for a suitable configuration starting from the end, so we can
775   // find the configuration for the specific language first, and the default
776   // configuration (which can only be at slot 0) after it.
777   for (int i = Styles.size() - 1; i >= 0; --i) {
778     if (Styles[i].Language == Language ||
779         Styles[i].Language == FormatStyle::LK_None) {
780       *Style = Styles[i];
781       Style->Language = Language;
782       return make_error_code(ParseError::Success);
783     }
784   }
785   return make_error_code(ParseError::Unsuitable);
786 }
787 
788 std::string configurationAsText(const FormatStyle &Style) {
789   std::string Text;
790   llvm::raw_string_ostream Stream(Text);
791   llvm::yaml::Output Output(Stream);
792   // We use the same mapping method for input and output, so we need a non-const
793   // reference here.
794   FormatStyle NonConstStyle = expandPresets(Style);
795   Output << NonConstStyle;
796   return Stream.str();
797 }
798 
799 namespace {
800 
801 class JavaScriptRequoter : public TokenAnalyzer {
802 public:
803   JavaScriptRequoter(const Environment &Env, const FormatStyle &Style)
804       : TokenAnalyzer(Env, Style) {}
805 
806   tooling::Replacements
807   analyze(TokenAnnotator &Annotator,
808           SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
809           FormatTokenLexer &Tokens) override {
810     AffectedRangeMgr.computeAffectedLines(AnnotatedLines.begin(),
811                                           AnnotatedLines.end());
812     tooling::Replacements Result;
813     requoteJSStringLiteral(AnnotatedLines, Result);
814     return Result;
815   }
816 
817 private:
818   // Replaces double/single-quoted string literal as appropriate, re-escaping
819   // the contents in the process.
820   void requoteJSStringLiteral(SmallVectorImpl<AnnotatedLine *> &Lines,
821                               tooling::Replacements &Result) {
822     for (AnnotatedLine *Line : Lines) {
823       requoteJSStringLiteral(Line->Children, Result);
824       if (!Line->Affected)
825         continue;
826       for (FormatToken *FormatTok = Line->First; FormatTok;
827            FormatTok = FormatTok->Next) {
828         StringRef Input = FormatTok->TokenText;
829         if (FormatTok->Finalized || !FormatTok->isStringLiteral() ||
830             // NB: testing for not starting with a double quote to avoid
831             // breaking `template strings`.
832             (Style.JavaScriptQuotes == FormatStyle::JSQS_Single &&
833              !Input.startswith("\"")) ||
834             (Style.JavaScriptQuotes == FormatStyle::JSQS_Double &&
835              !Input.startswith("\'")))
836           continue;
837 
838         // Change start and end quote.
839         bool IsSingle = Style.JavaScriptQuotes == FormatStyle::JSQS_Single;
840         SourceLocation Start = FormatTok->Tok.getLocation();
841         auto Replace = [&](SourceLocation Start, unsigned Length,
842                            StringRef ReplacementText) {
843           auto Err = Result.add(tooling::Replacement(
844               Env.getSourceManager(), Start, Length, ReplacementText));
845           // FIXME: handle error. For now, print error message and skip the
846           // replacement for release version.
847           if (Err)
848             llvm::errs() << llvm::toString(std::move(Err)) << "\n";
849           assert(!Err);
850         };
851         Replace(Start, 1, IsSingle ? "'" : "\"");
852         Replace(FormatTok->Tok.getEndLoc().getLocWithOffset(-1), 1,
853                 IsSingle ? "'" : "\"");
854 
855         // Escape internal quotes.
856         bool Escaped = false;
857         for (size_t i = 1; i < Input.size() - 1; i++) {
858           switch (Input[i]) {
859           case '\\':
860             if (!Escaped && i + 1 < Input.size() &&
861                 ((IsSingle && Input[i + 1] == '"') ||
862                  (!IsSingle && Input[i + 1] == '\''))) {
863               // Remove this \, it's escaping a " or ' that no longer needs
864               // escaping
865               Replace(Start.getLocWithOffset(i), 1, "");
866               continue;
867             }
868             Escaped = !Escaped;
869             break;
870           case '\"':
871           case '\'':
872             if (!Escaped && IsSingle == (Input[i] == '\'')) {
873               // Escape the quote.
874               Replace(Start.getLocWithOffset(i), 0, "\\");
875             }
876             Escaped = false;
877             break;
878           default:
879             Escaped = false;
880             break;
881           }
882         }
883       }
884     }
885   }
886 };
887 
888 class Formatter : public TokenAnalyzer {
889 public:
890   Formatter(const Environment &Env, const FormatStyle &Style,
891             bool *IncompleteFormat)
892       : TokenAnalyzer(Env, Style), IncompleteFormat(IncompleteFormat) {}
893 
894   tooling::Replacements
895   analyze(TokenAnnotator &Annotator,
896           SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
897           FormatTokenLexer &Tokens) override {
898     tooling::Replacements Result;
899     deriveLocalStyle(AnnotatedLines);
900     AffectedRangeMgr.computeAffectedLines(AnnotatedLines.begin(),
901                                           AnnotatedLines.end());
902     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
903       Annotator.calculateFormattingInformation(*AnnotatedLines[i]);
904     }
905     Annotator.setCommentLineLevels(AnnotatedLines);
906 
907     WhitespaceManager Whitespaces(
908         Env.getSourceManager(), Style,
909         inputUsesCRLF(Env.getSourceManager().getBufferData(Env.getFileID())));
910     ContinuationIndenter Indenter(Style, Tokens.getKeywords(),
911                                   Env.getSourceManager(), Whitespaces, Encoding,
912                                   BinPackInconclusiveFunctions);
913     UnwrappedLineFormatter(&Indenter, &Whitespaces, Style, Tokens.getKeywords(),
914                            IncompleteFormat)
915         .format(AnnotatedLines);
916     for (const auto &R : Whitespaces.generateReplacements())
917       if (Result.add(R))
918         return Result;
919     return Result;
920   }
921 
922 private:
923 
924   static bool inputUsesCRLF(StringRef Text) {
925     return Text.count('\r') * 2 > Text.count('\n');
926   }
927 
928   bool
929   hasCpp03IncompatibleFormat(const SmallVectorImpl<AnnotatedLine *> &Lines) {
930     for (const AnnotatedLine *Line : Lines) {
931       if (hasCpp03IncompatibleFormat(Line->Children))
932         return true;
933       for (FormatToken *Tok = Line->First->Next; Tok; Tok = Tok->Next) {
934         if (Tok->WhitespaceRange.getBegin() == Tok->WhitespaceRange.getEnd()) {
935           if (Tok->is(tok::coloncolon) && Tok->Previous->is(TT_TemplateOpener))
936             return true;
937           if (Tok->is(TT_TemplateCloser) &&
938               Tok->Previous->is(TT_TemplateCloser))
939             return true;
940         }
941       }
942     }
943     return false;
944   }
945 
946   int countVariableAlignments(const SmallVectorImpl<AnnotatedLine *> &Lines) {
947     int AlignmentDiff = 0;
948     for (const AnnotatedLine *Line : Lines) {
949       AlignmentDiff += countVariableAlignments(Line->Children);
950       for (FormatToken *Tok = Line->First; Tok && Tok->Next; Tok = Tok->Next) {
951         if (!Tok->is(TT_PointerOrReference))
952           continue;
953         bool SpaceBefore =
954             Tok->WhitespaceRange.getBegin() != Tok->WhitespaceRange.getEnd();
955         bool SpaceAfter = Tok->Next->WhitespaceRange.getBegin() !=
956                           Tok->Next->WhitespaceRange.getEnd();
957         if (SpaceBefore && !SpaceAfter)
958           ++AlignmentDiff;
959         if (!SpaceBefore && SpaceAfter)
960           --AlignmentDiff;
961       }
962     }
963     return AlignmentDiff;
964   }
965 
966   void
967   deriveLocalStyle(const SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
968     bool HasBinPackedFunction = false;
969     bool HasOnePerLineFunction = false;
970     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
971       if (!AnnotatedLines[i]->First->Next)
972         continue;
973       FormatToken *Tok = AnnotatedLines[i]->First->Next;
974       while (Tok->Next) {
975         if (Tok->PackingKind == PPK_BinPacked)
976           HasBinPackedFunction = true;
977         if (Tok->PackingKind == PPK_OnePerLine)
978           HasOnePerLineFunction = true;
979 
980         Tok = Tok->Next;
981       }
982     }
983     if (Style.DerivePointerAlignment)
984       Style.PointerAlignment = countVariableAlignments(AnnotatedLines) <= 0
985                                    ? FormatStyle::PAS_Left
986                                    : FormatStyle::PAS_Right;
987     if (Style.Standard == FormatStyle::LS_Auto)
988       Style.Standard = hasCpp03IncompatibleFormat(AnnotatedLines)
989                            ? FormatStyle::LS_Cpp11
990                            : FormatStyle::LS_Cpp03;
991     BinPackInconclusiveFunctions =
992         HasBinPackedFunction || !HasOnePerLineFunction;
993   }
994 
995   bool BinPackInconclusiveFunctions;
996   bool *IncompleteFormat;
997 };
998 
999 // This class clean up the erroneous/redundant code around the given ranges in
1000 // file.
1001 class Cleaner : public TokenAnalyzer {
1002 public:
1003   Cleaner(const Environment &Env, const FormatStyle &Style)
1004       : TokenAnalyzer(Env, Style),
1005         DeletedTokens(FormatTokenLess(Env.getSourceManager())) {}
1006 
1007   // FIXME: eliminate unused parameters.
1008   tooling::Replacements
1009   analyze(TokenAnnotator &Annotator,
1010           SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
1011           FormatTokenLexer &Tokens) override {
1012     // FIXME: in the current implementation the granularity of affected range
1013     // is an annotated line. However, this is not sufficient. Furthermore,
1014     // redundant code introduced by replacements does not necessarily
1015     // intercept with ranges of replacements that result in the redundancy.
1016     // To determine if some redundant code is actually introduced by
1017     // replacements(e.g. deletions), we need to come up with a more
1018     // sophisticated way of computing affected ranges.
1019     AffectedRangeMgr.computeAffectedLines(AnnotatedLines.begin(),
1020                                           AnnotatedLines.end());
1021 
1022     checkEmptyNamespace(AnnotatedLines);
1023 
1024     for (auto &Line : AnnotatedLines) {
1025       if (Line->Affected) {
1026         cleanupRight(Line->First, tok::comma, tok::comma);
1027         cleanupRight(Line->First, TT_CtorInitializerColon, tok::comma);
1028         cleanupRight(Line->First, tok::l_paren, tok::comma);
1029         cleanupLeft(Line->First, tok::comma, tok::r_paren);
1030         cleanupLeft(Line->First, TT_CtorInitializerComma, tok::l_brace);
1031         cleanupLeft(Line->First, TT_CtorInitializerColon, tok::l_brace);
1032         cleanupLeft(Line->First, TT_CtorInitializerColon, tok::equal);
1033       }
1034     }
1035 
1036     return generateFixes();
1037   }
1038 
1039 private:
1040   bool containsOnlyComments(const AnnotatedLine &Line) {
1041     for (FormatToken *Tok = Line.First; Tok != nullptr; Tok = Tok->Next) {
1042       if (Tok->isNot(tok::comment))
1043         return false;
1044     }
1045     return true;
1046   }
1047 
1048   // Iterate through all lines and remove any empty (nested) namespaces.
1049   void checkEmptyNamespace(SmallVectorImpl<AnnotatedLine *> &AnnotatedLines) {
1050     std::set<unsigned> DeletedLines;
1051     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
1052       auto &Line = *AnnotatedLines[i];
1053       if (Line.startsWith(tok::kw_namespace) ||
1054           Line.startsWith(tok::kw_inline, tok::kw_namespace)) {
1055         checkEmptyNamespace(AnnotatedLines, i, i, DeletedLines);
1056       }
1057     }
1058 
1059     for (auto Line : DeletedLines) {
1060       FormatToken *Tok = AnnotatedLines[Line]->First;
1061       while (Tok) {
1062         deleteToken(Tok);
1063         Tok = Tok->Next;
1064       }
1065     }
1066   }
1067 
1068   // The function checks if the namespace, which starts from \p CurrentLine, and
1069   // its nested namespaces are empty and delete them if they are empty. It also
1070   // sets \p NewLine to the last line checked.
1071   // Returns true if the current namespace is empty.
1072   bool checkEmptyNamespace(SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
1073                            unsigned CurrentLine, unsigned &NewLine,
1074                            std::set<unsigned> &DeletedLines) {
1075     unsigned InitLine = CurrentLine, End = AnnotatedLines.size();
1076     if (Style.BraceWrapping.AfterNamespace) {
1077       // If the left brace is in a new line, we should consume it first so that
1078       // it does not make the namespace non-empty.
1079       // FIXME: error handling if there is no left brace.
1080       if (!AnnotatedLines[++CurrentLine]->startsWith(tok::l_brace)) {
1081         NewLine = CurrentLine;
1082         return false;
1083       }
1084     } else if (!AnnotatedLines[CurrentLine]->endsWith(tok::l_brace)) {
1085       return false;
1086     }
1087     while (++CurrentLine < End) {
1088       if (AnnotatedLines[CurrentLine]->startsWith(tok::r_brace))
1089         break;
1090 
1091       if (AnnotatedLines[CurrentLine]->startsWith(tok::kw_namespace) ||
1092           AnnotatedLines[CurrentLine]->startsWith(tok::kw_inline,
1093                                                   tok::kw_namespace)) {
1094         if (!checkEmptyNamespace(AnnotatedLines, CurrentLine, NewLine,
1095                                  DeletedLines))
1096           return false;
1097         CurrentLine = NewLine;
1098         continue;
1099       }
1100 
1101       if (containsOnlyComments(*AnnotatedLines[CurrentLine]))
1102         continue;
1103 
1104       // If there is anything other than comments or nested namespaces in the
1105       // current namespace, the namespace cannot be empty.
1106       NewLine = CurrentLine;
1107       return false;
1108     }
1109 
1110     NewLine = CurrentLine;
1111     if (CurrentLine >= End)
1112       return false;
1113 
1114     // Check if the empty namespace is actually affected by changed ranges.
1115     if (!AffectedRangeMgr.affectsCharSourceRange(CharSourceRange::getCharRange(
1116             AnnotatedLines[InitLine]->First->Tok.getLocation(),
1117             AnnotatedLines[CurrentLine]->Last->Tok.getEndLoc())))
1118       return false;
1119 
1120     for (unsigned i = InitLine; i <= CurrentLine; ++i) {
1121       DeletedLines.insert(i);
1122     }
1123 
1124     return true;
1125   }
1126 
1127   // Checks pairs {start, start->next},..., {end->previous, end} and deletes one
1128   // of the token in the pair if the left token has \p LK token kind and the
1129   // right token has \p RK token kind. If \p DeleteLeft is true, the left token
1130   // is deleted on match; otherwise, the right token is deleted.
1131   template <typename LeftKind, typename RightKind>
1132   void cleanupPair(FormatToken *Start, LeftKind LK, RightKind RK,
1133                    bool DeleteLeft) {
1134     auto NextNotDeleted = [this](const FormatToken &Tok) -> FormatToken * {
1135       for (auto *Res = Tok.Next; Res; Res = Res->Next)
1136         if (!Res->is(tok::comment) &&
1137             DeletedTokens.find(Res) == DeletedTokens.end())
1138           return Res;
1139       return nullptr;
1140     };
1141     for (auto *Left = Start; Left;) {
1142       auto *Right = NextNotDeleted(*Left);
1143       if (!Right)
1144         break;
1145       if (Left->is(LK) && Right->is(RK)) {
1146         deleteToken(DeleteLeft ? Left : Right);
1147         for (auto *Tok = Left->Next; Tok && Tok != Right; Tok = Tok->Next)
1148           deleteToken(Tok);
1149         // If the right token is deleted, we should keep the left token
1150         // unchanged and pair it with the new right token.
1151         if (!DeleteLeft)
1152           continue;
1153       }
1154       Left = Right;
1155     }
1156   }
1157 
1158   template <typename LeftKind, typename RightKind>
1159   void cleanupLeft(FormatToken *Start, LeftKind LK, RightKind RK) {
1160     cleanupPair(Start, LK, RK, /*DeleteLeft=*/true);
1161   }
1162 
1163   template <typename LeftKind, typename RightKind>
1164   void cleanupRight(FormatToken *Start, LeftKind LK, RightKind RK) {
1165     cleanupPair(Start, LK, RK, /*DeleteLeft=*/false);
1166   }
1167 
1168   // Delete the given token.
1169   inline void deleteToken(FormatToken *Tok) {
1170     if (Tok)
1171       DeletedTokens.insert(Tok);
1172   }
1173 
1174   tooling::Replacements generateFixes() {
1175     tooling::Replacements Fixes;
1176     std::vector<FormatToken *> Tokens;
1177     std::copy(DeletedTokens.begin(), DeletedTokens.end(),
1178               std::back_inserter(Tokens));
1179 
1180     // Merge multiple continuous token deletions into one big deletion so that
1181     // the number of replacements can be reduced. This makes computing affected
1182     // ranges more efficient when we run reformat on the changed code.
1183     unsigned Idx = 0;
1184     while (Idx < Tokens.size()) {
1185       unsigned St = Idx, End = Idx;
1186       while ((End + 1) < Tokens.size() &&
1187              Tokens[End]->Next == Tokens[End + 1]) {
1188         End++;
1189       }
1190       auto SR = CharSourceRange::getCharRange(Tokens[St]->Tok.getLocation(),
1191                                               Tokens[End]->Tok.getEndLoc());
1192       auto Err =
1193           Fixes.add(tooling::Replacement(Env.getSourceManager(), SR, ""));
1194       // FIXME: better error handling. for now just print error message and skip
1195       // for the release version.
1196       if (Err)
1197         llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1198       assert(!Err && "Fixes must not conflict!");
1199       Idx = End + 1;
1200     }
1201 
1202     return Fixes;
1203   }
1204 
1205   // Class for less-than inequality comparason for the set `RedundantTokens`.
1206   // We store tokens in the order they appear in the translation unit so that
1207   // we do not need to sort them in `generateFixes()`.
1208   struct FormatTokenLess {
1209     FormatTokenLess(const SourceManager &SM) : SM(SM) {}
1210 
1211     bool operator()(const FormatToken *LHS, const FormatToken *RHS) const {
1212       return SM.isBeforeInTranslationUnit(LHS->Tok.getLocation(),
1213                                           RHS->Tok.getLocation());
1214     }
1215     const SourceManager &SM;
1216   };
1217 
1218   // Tokens to be deleted.
1219   std::set<FormatToken *, FormatTokenLess> DeletedTokens;
1220 };
1221 
1222 struct IncludeDirective {
1223   StringRef Filename;
1224   StringRef Text;
1225   unsigned Offset;
1226   int Category;
1227 };
1228 
1229 } // end anonymous namespace
1230 
1231 // Determines whether 'Ranges' intersects with ('Start', 'End').
1232 static bool affectsRange(ArrayRef<tooling::Range> Ranges, unsigned Start,
1233                          unsigned End) {
1234   for (auto Range : Ranges) {
1235     if (Range.getOffset() < End &&
1236         Range.getOffset() + Range.getLength() > Start)
1237       return true;
1238   }
1239   return false;
1240 }
1241 
1242 // Returns a pair (Index, OffsetToEOL) describing the position of the cursor
1243 // before sorting/deduplicating. Index is the index of the include under the
1244 // cursor in the original set of includes. If this include has duplicates, it is
1245 // the index of the first of the duplicates as the others are going to be
1246 // removed. OffsetToEOL describes the cursor's position relative to the end of
1247 // its current line.
1248 // If `Cursor` is not on any #include, `Index` will be UINT_MAX.
1249 static std::pair<unsigned, unsigned>
1250 FindCursorIndex(const SmallVectorImpl<IncludeDirective> &Includes,
1251                 const SmallVectorImpl<unsigned> &Indices, unsigned Cursor) {
1252   unsigned CursorIndex = UINT_MAX;
1253   unsigned OffsetToEOL = 0;
1254   for (int i = 0, e = Includes.size(); i != e; ++i) {
1255     unsigned Start = Includes[Indices[i]].Offset;
1256     unsigned End = Start + Includes[Indices[i]].Text.size();
1257     if (!(Cursor >= Start && Cursor < End))
1258       continue;
1259     CursorIndex = Indices[i];
1260     OffsetToEOL = End - Cursor;
1261     // Put the cursor on the only remaining #include among the duplicate
1262     // #includes.
1263     while (--i >= 0 && Includes[CursorIndex].Text == Includes[Indices[i]].Text)
1264       CursorIndex = i;
1265     break;
1266   }
1267   return std::make_pair(CursorIndex, OffsetToEOL);
1268 }
1269 
1270 // Sorts and deduplicate a block of includes given by 'Includes' alphabetically
1271 // adding the necessary replacement to 'Replaces'. 'Includes' must be in strict
1272 // source order.
1273 // #include directives with the same text will be deduplicated, and only the
1274 // first #include in the duplicate #includes remains. If the `Cursor` is
1275 // provided and put on a deleted #include, it will be moved to the remaining
1276 // #include in the duplicate #includes.
1277 static void sortCppIncludes(const FormatStyle &Style,
1278                             const SmallVectorImpl<IncludeDirective> &Includes,
1279                             ArrayRef<tooling::Range> Ranges, StringRef FileName,
1280                             tooling::Replacements &Replaces, unsigned *Cursor) {
1281   unsigned IncludesBeginOffset = Includes.front().Offset;
1282   unsigned IncludesEndOffset =
1283       Includes.back().Offset + Includes.back().Text.size();
1284   unsigned IncludesBlockSize = IncludesEndOffset - IncludesBeginOffset;
1285   if (!affectsRange(Ranges, IncludesBeginOffset, IncludesEndOffset))
1286     return;
1287   SmallVector<unsigned, 16> Indices;
1288   for (unsigned i = 0, e = Includes.size(); i != e; ++i)
1289     Indices.push_back(i);
1290   std::stable_sort(
1291       Indices.begin(), Indices.end(), [&](unsigned LHSI, unsigned RHSI) {
1292         return std::tie(Includes[LHSI].Category, Includes[LHSI].Filename) <
1293                std::tie(Includes[RHSI].Category, Includes[RHSI].Filename);
1294       });
1295   // The index of the include on which the cursor will be put after
1296   // sorting/deduplicating.
1297   unsigned CursorIndex;
1298   // The offset from cursor to the end of line.
1299   unsigned CursorToEOLOffset;
1300   if (Cursor)
1301     std::tie(CursorIndex, CursorToEOLOffset) =
1302         FindCursorIndex(Includes, Indices, *Cursor);
1303 
1304   // Deduplicate #includes.
1305   Indices.erase(std::unique(Indices.begin(), Indices.end(),
1306                             [&](unsigned LHSI, unsigned RHSI) {
1307                               return Includes[LHSI].Text == Includes[RHSI].Text;
1308                             }),
1309                 Indices.end());
1310 
1311   // If the #includes are out of order, we generate a single replacement fixing
1312   // the entire block. Otherwise, no replacement is generated.
1313   if (Indices.size() == Includes.size() &&
1314       std::is_sorted(Indices.begin(), Indices.end()))
1315     return;
1316 
1317   std::string result;
1318   for (unsigned Index : Indices) {
1319     if (!result.empty())
1320       result += "\n";
1321     result += Includes[Index].Text;
1322     if (Cursor && CursorIndex == Index)
1323       *Cursor = IncludesBeginOffset + result.size() - CursorToEOLOffset;
1324   }
1325 
1326   auto Err = Replaces.add(tooling::Replacement(
1327       FileName, Includes.front().Offset, IncludesBlockSize, result));
1328   // FIXME: better error handling. For now, just skip the replacement for the
1329   // release version.
1330   if (Err)
1331     llvm::errs() << llvm::toString(std::move(Err)) << "\n";
1332   assert(!Err);
1333 }
1334 
1335 namespace {
1336 
1337 // This class manages priorities of #include categories and calculates
1338 // priorities for headers.
1339 class IncludeCategoryManager {
1340 public:
1341   IncludeCategoryManager(const FormatStyle &Style, StringRef FileName)
1342       : Style(Style), FileName(FileName) {
1343     FileStem = llvm::sys::path::stem(FileName);
1344     for (const auto &Category : Style.IncludeCategories)
1345       CategoryRegexs.emplace_back(Category.Regex);
1346     IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") ||
1347                  FileName.endswith(".cpp") || FileName.endswith(".c++") ||
1348                  FileName.endswith(".cxx") || FileName.endswith(".m") ||
1349                  FileName.endswith(".mm");
1350   }
1351 
1352   // Returns the priority of the category which \p IncludeName belongs to.
1353   // If \p CheckMainHeader is true and \p IncludeName is a main header, returns
1354   // 0. Otherwise, returns the priority of the matching category or INT_MAX.
1355   int getIncludePriority(StringRef IncludeName, bool CheckMainHeader) {
1356     int Ret = INT_MAX;
1357     for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i)
1358       if (CategoryRegexs[i].match(IncludeName)) {
1359         Ret = Style.IncludeCategories[i].Priority;
1360         break;
1361       }
1362     if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName))
1363       Ret = 0;
1364     return Ret;
1365   }
1366 
1367 private:
1368   bool isMainHeader(StringRef IncludeName) const {
1369     if (!IncludeName.startswith("\""))
1370       return false;
1371     StringRef HeaderStem =
1372         llvm::sys::path::stem(IncludeName.drop_front(1).drop_back(1));
1373     if (FileStem.startswith(HeaderStem)) {
1374       llvm::Regex MainIncludeRegex(
1375           (HeaderStem + Style.IncludeIsMainRegex).str());
1376       if (MainIncludeRegex.match(FileStem))
1377         return true;
1378     }
1379     return false;
1380   }
1381 
1382   const FormatStyle &Style;
1383   bool IsMainFile;
1384   StringRef FileName;
1385   StringRef FileStem;
1386   SmallVector<llvm::Regex, 4> CategoryRegexs;
1387 };
1388 
1389 const char IncludeRegexPattern[] =
1390     R"(^[\t\ ]*#[\t\ ]*(import|include)[^"<]*(["<][^">]*[">]))";
1391 
1392 } // anonymous namespace
1393 
1394 tooling::Replacements sortCppIncludes(const FormatStyle &Style, StringRef Code,
1395                                       ArrayRef<tooling::Range> Ranges,
1396                                       StringRef FileName,
1397                                       tooling::Replacements &Replaces,
1398                                       unsigned *Cursor) {
1399   unsigned Prev = 0;
1400   unsigned SearchFrom = 0;
1401   llvm::Regex IncludeRegex(IncludeRegexPattern);
1402   SmallVector<StringRef, 4> Matches;
1403   SmallVector<IncludeDirective, 16> IncludesInBlock;
1404 
1405   // In compiled files, consider the first #include to be the main #include of
1406   // the file if it is not a system #include. This ensures that the header
1407   // doesn't have hidden dependencies
1408   // (http://llvm.org/docs/CodingStandards.html#include-style).
1409   //
1410   // FIXME: Do some sanity checking, e.g. edit distance of the base name, to fix
1411   // cases where the first #include is unlikely to be the main header.
1412   IncludeCategoryManager Categories(Style, FileName);
1413   bool FirstIncludeBlock = true;
1414   bool MainIncludeFound = false;
1415   bool FormattingOff = false;
1416 
1417   for (;;) {
1418     auto Pos = Code.find('\n', SearchFrom);
1419     StringRef Line =
1420         Code.substr(Prev, (Pos != StringRef::npos ? Pos : Code.size()) - Prev);
1421 
1422     StringRef Trimmed = Line.trim();
1423     if (Trimmed == "// clang-format off")
1424       FormattingOff = true;
1425     else if (Trimmed == "// clang-format on")
1426       FormattingOff = false;
1427 
1428     if (!FormattingOff && !Line.endswith("\\")) {
1429       if (IncludeRegex.match(Line, &Matches)) {
1430         StringRef IncludeName = Matches[2];
1431         int Category = Categories.getIncludePriority(
1432             IncludeName,
1433             /*CheckMainHeader=*/!MainIncludeFound && FirstIncludeBlock);
1434         if (Category == 0)
1435           MainIncludeFound = true;
1436         IncludesInBlock.push_back({IncludeName, Line, Prev, Category});
1437       } else if (!IncludesInBlock.empty()) {
1438         sortCppIncludes(Style, IncludesInBlock, Ranges, FileName, Replaces,
1439                         Cursor);
1440         IncludesInBlock.clear();
1441         FirstIncludeBlock = false;
1442       }
1443       Prev = Pos + 1;
1444     }
1445     if (Pos == StringRef::npos || Pos + 1 == Code.size())
1446       break;
1447     SearchFrom = Pos + 1;
1448   }
1449   if (!IncludesInBlock.empty())
1450     sortCppIncludes(Style, IncludesInBlock, Ranges, FileName, Replaces, Cursor);
1451   return Replaces;
1452 }
1453 
1454 tooling::Replacements sortIncludes(const FormatStyle &Style, StringRef Code,
1455                                    ArrayRef<tooling::Range> Ranges,
1456                                    StringRef FileName, unsigned *Cursor) {
1457   tooling::Replacements Replaces;
1458   if (!Style.SortIncludes)
1459     return Replaces;
1460   if (Style.Language == FormatStyle::LanguageKind::LK_JavaScript)
1461     return sortJavaScriptImports(Style, Code, Ranges, FileName);
1462   sortCppIncludes(Style, Code, Ranges, FileName, Replaces, Cursor);
1463   return Replaces;
1464 }
1465 
1466 template <typename T>
1467 static llvm::Expected<tooling::Replacements>
1468 processReplacements(T ProcessFunc, StringRef Code,
1469                     const tooling::Replacements &Replaces,
1470                     const FormatStyle &Style) {
1471   if (Replaces.empty())
1472     return tooling::Replacements();
1473 
1474   auto NewCode = applyAllReplacements(Code, Replaces);
1475   if (!NewCode)
1476     return NewCode.takeError();
1477   std::vector<tooling::Range> ChangedRanges = Replaces.getAffectedRanges();
1478   StringRef FileName = Replaces.begin()->getFilePath();
1479 
1480   tooling::Replacements FormatReplaces =
1481       ProcessFunc(Style, *NewCode, ChangedRanges, FileName);
1482 
1483   return Replaces.merge(FormatReplaces);
1484 }
1485 
1486 llvm::Expected<tooling::Replacements>
1487 formatReplacements(StringRef Code, const tooling::Replacements &Replaces,
1488                    const FormatStyle &Style) {
1489   // We need to use lambda function here since there are two versions of
1490   // `sortIncludes`.
1491   auto SortIncludes = [](const FormatStyle &Style, StringRef Code,
1492                          std::vector<tooling::Range> Ranges,
1493                          StringRef FileName) -> tooling::Replacements {
1494     return sortIncludes(Style, Code, Ranges, FileName);
1495   };
1496   auto SortedReplaces =
1497       processReplacements(SortIncludes, Code, Replaces, Style);
1498   if (!SortedReplaces)
1499     return SortedReplaces.takeError();
1500 
1501   // We need to use lambda function here since there are two versions of
1502   // `reformat`.
1503   auto Reformat = [](const FormatStyle &Style, StringRef Code,
1504                      std::vector<tooling::Range> Ranges,
1505                      StringRef FileName) -> tooling::Replacements {
1506     return reformat(Style, Code, Ranges, FileName);
1507   };
1508   return processReplacements(Reformat, Code, *SortedReplaces, Style);
1509 }
1510 
1511 namespace {
1512 
1513 inline bool isHeaderInsertion(const tooling::Replacement &Replace) {
1514   return Replace.getOffset() == UINT_MAX && Replace.getLength() == 0 &&
1515          llvm::Regex(IncludeRegexPattern).match(Replace.getReplacementText());
1516 }
1517 
1518 inline bool isHeaderDeletion(const tooling::Replacement &Replace) {
1519   return Replace.getOffset() == UINT_MAX && Replace.getLength() == 1;
1520 }
1521 
1522 // Returns the offset after skipping a sequence of tokens, matched by \p
1523 // GetOffsetAfterSequence, from the start of the code.
1524 // \p GetOffsetAfterSequence should be a function that matches a sequence of
1525 // tokens and returns an offset after the sequence.
1526 unsigned getOffsetAfterTokenSequence(
1527     StringRef FileName, StringRef Code, const FormatStyle &Style,
1528     std::function<unsigned(const SourceManager &, Lexer &, Token &)>
1529         GetOffsetAfterSequense) {
1530   std::unique_ptr<Environment> Env =
1531       Environment::CreateVirtualEnvironment(Code, FileName, /*Ranges=*/{});
1532   const SourceManager &SourceMgr = Env->getSourceManager();
1533   Lexer Lex(Env->getFileID(), SourceMgr.getBuffer(Env->getFileID()), SourceMgr,
1534             getFormattingLangOpts(Style));
1535   Token Tok;
1536   // Get the first token.
1537   Lex.LexFromRawLexer(Tok);
1538   return GetOffsetAfterSequense(SourceMgr, Lex, Tok);
1539 }
1540 
1541 // Check if a sequence of tokens is like "#<Name> <raw_identifier>". If it is,
1542 // \p Tok will be the token after this directive; otherwise, it can be any token
1543 // after the given \p Tok (including \p Tok).
1544 bool checkAndConsumeDirectiveWithName(Lexer &Lex, StringRef Name, Token &Tok) {
1545   bool Matched = Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
1546                  Tok.is(tok::raw_identifier) &&
1547                  Tok.getRawIdentifier() == Name && !Lex.LexFromRawLexer(Tok) &&
1548                  tok::raw_identifier;
1549   if (Matched)
1550     Lex.LexFromRawLexer(Tok);
1551   return Matched;
1552 }
1553 
1554 void skipComments(Lexer &Lex, Token &Tok) {
1555   while (Tok.is(tok::comment))
1556     if (Lex.LexFromRawLexer(Tok))
1557       return;
1558 }
1559 
1560 // Returns the offset after header guard directives and any comments
1561 // before/after header guards. If no header guard presents in the code, this
1562 // will returns the offset after skipping all comments from the start of the
1563 // code.
1564 unsigned getOffsetAfterHeaderGuardsAndComments(StringRef FileName,
1565                                                StringRef Code,
1566                                                const FormatStyle &Style) {
1567   return getOffsetAfterTokenSequence(
1568       FileName, Code, Style,
1569       [](const SourceManager &SM, Lexer &Lex, Token Tok) {
1570         skipComments(Lex, Tok);
1571         unsigned InitialOffset = SM.getFileOffset(Tok.getLocation());
1572         if (checkAndConsumeDirectiveWithName(Lex, "ifndef", Tok)) {
1573           skipComments(Lex, Tok);
1574           if (checkAndConsumeDirectiveWithName(Lex, "define", Tok))
1575             return SM.getFileOffset(Tok.getLocation());
1576         }
1577         return InitialOffset;
1578       });
1579 }
1580 
1581 // Check if a sequence of tokens is like
1582 //    "#include ("header.h" | <header.h>)".
1583 // If it is, \p Tok will be the token after this directive; otherwise, it can be
1584 // any token after the given \p Tok (including \p Tok).
1585 bool checkAndConsumeInclusiveDirective(Lexer &Lex, Token &Tok) {
1586   auto Matched = [&]() {
1587     Lex.LexFromRawLexer(Tok);
1588     return true;
1589   };
1590   if (Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
1591       Tok.is(tok::raw_identifier) && Tok.getRawIdentifier() == "include") {
1592     if (Lex.LexFromRawLexer(Tok))
1593       return false;
1594     if (Tok.is(tok::string_literal))
1595       return Matched();
1596     if (Tok.is(tok::less)) {
1597       while (!Lex.LexFromRawLexer(Tok) && Tok.isNot(tok::greater)) {
1598       }
1599       if (Tok.is(tok::greater))
1600         return Matched();
1601     }
1602   }
1603   return false;
1604 }
1605 
1606 // Returns the offset of the last #include directive after which a new
1607 // #include can be inserted. This ignores #include's after the #include block(s)
1608 // in the beginning of a file to avoid inserting headers into code sections
1609 // where new #include's should not be added by default.
1610 // These code sections include:
1611 //      - raw string literals (containing #include).
1612 //      - #if blocks.
1613 //      - Special #include's among declarations (e.g. functions).
1614 //
1615 // If no #include after which a new #include can be inserted, this returns the
1616 // offset after skipping all comments from the start of the code.
1617 // Inserting after an #include is not allowed if it comes after code that is not
1618 // #include (e.g. pre-processing directive that is not #include, declarations).
1619 unsigned getMaxHeaderInsertionOffset(StringRef FileName, StringRef Code,
1620                                      const FormatStyle &Style) {
1621   return getOffsetAfterTokenSequence(
1622       FileName, Code, Style,
1623       [](const SourceManager &SM, Lexer &Lex, Token Tok) {
1624         skipComments(Lex, Tok);
1625         unsigned MaxOffset = SM.getFileOffset(Tok.getLocation());
1626         while (checkAndConsumeInclusiveDirective(Lex, Tok))
1627           MaxOffset = SM.getFileOffset(Tok.getLocation());
1628         return MaxOffset;
1629       });
1630 }
1631 
1632 bool isDeletedHeader(llvm::StringRef HeaderName,
1633                      const std::set<llvm::StringRef> &HeadersToDelete) {
1634   return HeadersToDelete.count(HeaderName) ||
1635          HeadersToDelete.count(HeaderName.trim("\"<>"));
1636 }
1637 
1638 // FIXME: insert empty lines between newly created blocks.
1639 tooling::Replacements
1640 fixCppIncludeInsertions(StringRef Code, const tooling::Replacements &Replaces,
1641                         const FormatStyle &Style) {
1642   if (Style.Language != FormatStyle::LanguageKind::LK_Cpp)
1643     return Replaces;
1644 
1645   tooling::Replacements HeaderInsertions;
1646   std::set<llvm::StringRef> HeadersToDelete;
1647   tooling::Replacements Result;
1648   for (const auto &R : Replaces) {
1649     if (isHeaderInsertion(R)) {
1650       // Replacements from \p Replaces must be conflict-free already, so we can
1651       // simply consume the error.
1652       llvm::consumeError(HeaderInsertions.add(R));
1653     } else if (isHeaderDeletion(R)) {
1654       HeadersToDelete.insert(R.getReplacementText());
1655     } else if (R.getOffset() == UINT_MAX) {
1656       llvm::errs() << "Insertions other than header #include insertion are "
1657                       "not supported! "
1658                    << R.getReplacementText() << "\n";
1659     } else {
1660       llvm::consumeError(Result.add(R));
1661     }
1662   }
1663   if (HeaderInsertions.empty() && HeadersToDelete.empty())
1664     return Replaces;
1665 
1666   llvm::Regex IncludeRegex(IncludeRegexPattern);
1667   llvm::Regex DefineRegex(R"(^[\t\ ]*#[\t\ ]*define[\t\ ]*[^\\]*$)");
1668   SmallVector<StringRef, 4> Matches;
1669 
1670   StringRef FileName = Replaces.begin()->getFilePath();
1671   IncludeCategoryManager Categories(Style, FileName);
1672 
1673   // Record the offset of the end of the last include in each category.
1674   std::map<int, int> CategoryEndOffsets;
1675   // All possible priorities.
1676   // Add 0 for main header and INT_MAX for headers that are not in any category.
1677   std::set<int> Priorities = {0, INT_MAX};
1678   for (const auto &Category : Style.IncludeCategories)
1679     Priorities.insert(Category.Priority);
1680   int FirstIncludeOffset = -1;
1681   // All new headers should be inserted after this offset.
1682   unsigned MinInsertOffset =
1683       getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style);
1684   StringRef TrimmedCode = Code.drop_front(MinInsertOffset);
1685   // Max insertion offset in the original code.
1686   unsigned MaxInsertOffset =
1687       MinInsertOffset +
1688       getMaxHeaderInsertionOffset(FileName, TrimmedCode, Style);
1689   SmallVector<StringRef, 32> Lines;
1690   TrimmedCode.split(Lines, '\n');
1691   unsigned Offset = MinInsertOffset;
1692   unsigned NextLineOffset;
1693   std::set<StringRef> ExistingIncludes;
1694   for (auto Line : Lines) {
1695     NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1);
1696     if (IncludeRegex.match(Line, &Matches)) {
1697       // The header name with quotes or angle brackets.
1698       StringRef IncludeName = Matches[2];
1699       ExistingIncludes.insert(IncludeName);
1700       // Only record the offset of current #include if we can insert after it.
1701       if (Offset <= MaxInsertOffset) {
1702         int Category = Categories.getIncludePriority(
1703             IncludeName, /*CheckMainHeader=*/FirstIncludeOffset < 0);
1704         CategoryEndOffsets[Category] = NextLineOffset;
1705         if (FirstIncludeOffset < 0)
1706           FirstIncludeOffset = Offset;
1707       }
1708       if (isDeletedHeader(IncludeName, HeadersToDelete)) {
1709         // If this is the last line without trailing newline, we need to make
1710         // sure we don't delete across the file boundary.
1711         unsigned Length = std::min(Line.size() + 1, Code.size() - Offset);
1712         llvm::Error Err =
1713             Result.add(tooling::Replacement(FileName, Offset, Length, ""));
1714         if (Err) {
1715           // Ignore the deletion on conflict.
1716           llvm::errs() << "Failed to add header deletion replacement for "
1717                        << IncludeName << ": " << llvm::toString(std::move(Err))
1718                        << "\n";
1719         }
1720       }
1721     }
1722     Offset = NextLineOffset;
1723   }
1724 
1725   // Populate CategoryEndOfssets:
1726   // - Ensure that CategoryEndOffset[Highest] is always populated.
1727   // - If CategoryEndOffset[Priority] isn't set, use the next higher value that
1728   //   is set, up to CategoryEndOffset[Highest].
1729   auto Highest = Priorities.begin();
1730   if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) {
1731     if (FirstIncludeOffset >= 0)
1732       CategoryEndOffsets[*Highest] = FirstIncludeOffset;
1733     else
1734       CategoryEndOffsets[*Highest] = MinInsertOffset;
1735   }
1736   // By this point, CategoryEndOffset[Highest] is always set appropriately:
1737   //  - to an appropriate location before/after existing #includes, or
1738   //  - to right after the header guard, or
1739   //  - to the beginning of the file.
1740   for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; ++I)
1741     if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end())
1742       CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)];
1743 
1744   bool NeedNewLineAtEnd = !Code.empty() && Code.back() != '\n';
1745   for (const auto &R : HeaderInsertions) {
1746     auto IncludeDirective = R.getReplacementText();
1747     bool Matched = IncludeRegex.match(IncludeDirective, &Matches);
1748     assert(Matched && "Header insertion replacement must have replacement text "
1749                       "'#include ...'");
1750     (void)Matched;
1751     auto IncludeName = Matches[2];
1752     if (ExistingIncludes.find(IncludeName) != ExistingIncludes.end()) {
1753       DEBUG(llvm::dbgs() << "Skip adding existing include : " << IncludeName
1754                          << "\n");
1755       continue;
1756     }
1757     int Category =
1758         Categories.getIncludePriority(IncludeName, /*CheckMainHeader=*/true);
1759     Offset = CategoryEndOffsets[Category];
1760     std::string NewInclude = !IncludeDirective.endswith("\n")
1761                                  ? (IncludeDirective + "\n").str()
1762                                  : IncludeDirective.str();
1763     // When inserting headers at end of the code, also append '\n' to the code
1764     // if it does not end with '\n'.
1765     if (NeedNewLineAtEnd && Offset == Code.size()) {
1766       NewInclude = "\n" + NewInclude;
1767       NeedNewLineAtEnd = false;
1768     }
1769     auto NewReplace = tooling::Replacement(FileName, Offset, 0, NewInclude);
1770     auto Err = Result.add(NewReplace);
1771     if (Err) {
1772       llvm::consumeError(std::move(Err));
1773       unsigned NewOffset = Result.getShiftedCodePosition(Offset);
1774       NewReplace = tooling::Replacement(FileName, NewOffset, 0, NewInclude);
1775       Result = Result.merge(tooling::Replacements(NewReplace));
1776     }
1777   }
1778   return Result;
1779 }
1780 
1781 } // anonymous namespace
1782 
1783 llvm::Expected<tooling::Replacements>
1784 cleanupAroundReplacements(StringRef Code, const tooling::Replacements &Replaces,
1785                           const FormatStyle &Style) {
1786   // We need to use lambda function here since there are two versions of
1787   // `cleanup`.
1788   auto Cleanup = [](const FormatStyle &Style, StringRef Code,
1789                     std::vector<tooling::Range> Ranges,
1790                     StringRef FileName) -> tooling::Replacements {
1791     return cleanup(Style, Code, Ranges, FileName);
1792   };
1793   // Make header insertion replacements insert new headers into correct blocks.
1794   tooling::Replacements NewReplaces =
1795       fixCppIncludeInsertions(Code, Replaces, Style);
1796   return processReplacements(Cleanup, Code, NewReplaces, Style);
1797 }
1798 
1799 tooling::Replacements reformat(const FormatStyle &Style, StringRef Code,
1800                                ArrayRef<tooling::Range> Ranges,
1801                                StringRef FileName, bool *IncompleteFormat) {
1802   FormatStyle Expanded = expandPresets(Style);
1803   if (Expanded.DisableFormat)
1804     return tooling::Replacements();
1805 
1806   auto Env = Environment::CreateVirtualEnvironment(Code, FileName, Ranges);
1807 
1808   if (Style.Language == FormatStyle::LK_JavaScript &&
1809       Style.JavaScriptQuotes != FormatStyle::JSQS_Leave) {
1810     JavaScriptRequoter Requoter(*Env, Expanded);
1811     tooling::Replacements Requotes = Requoter.process();
1812     if (!Requotes.empty()) {
1813       auto NewCode = applyAllReplacements(Code, Requotes);
1814       if (NewCode) {
1815         auto NewEnv = Environment::CreateVirtualEnvironment(
1816             *NewCode, FileName,
1817             tooling::calculateRangesAfterReplacements(Requotes, Ranges));
1818         Formatter Format(*NewEnv, Expanded, IncompleteFormat);
1819         return Requotes.merge(Format.process());
1820       }
1821     }
1822   }
1823 
1824   Formatter Format(*Env, Expanded, IncompleteFormat);
1825   return Format.process();
1826 }
1827 
1828 tooling::Replacements cleanup(const FormatStyle &Style, StringRef Code,
1829                               ArrayRef<tooling::Range> Ranges,
1830                               StringRef FileName) {
1831   std::unique_ptr<Environment> Env =
1832       Environment::CreateVirtualEnvironment(Code, FileName, Ranges);
1833   Cleaner Clean(*Env, Style);
1834   return Clean.process();
1835 }
1836 
1837 LangOptions getFormattingLangOpts(const FormatStyle &Style) {
1838   LangOptions LangOpts;
1839   LangOpts.CPlusPlus = 1;
1840   LangOpts.CPlusPlus11 = Style.Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
1841   LangOpts.CPlusPlus14 = Style.Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
1842   LangOpts.LineComment = 1;
1843   bool AlternativeOperators = Style.Language == FormatStyle::LK_Cpp;
1844   LangOpts.CXXOperatorNames = AlternativeOperators ? 1 : 0;
1845   LangOpts.Bool = 1;
1846   LangOpts.ObjC1 = 1;
1847   LangOpts.ObjC2 = 1;
1848   LangOpts.MicrosoftExt = 1;    // To get kw___try, kw___finally.
1849   LangOpts.DeclSpecKeyword = 1; // To get __declspec.
1850   return LangOpts;
1851 }
1852 
1853 const char *StyleOptionHelpDescription =
1854     "Coding style, currently supports:\n"
1855     "  LLVM, Google, Chromium, Mozilla, WebKit.\n"
1856     "Use -style=file to load style configuration from\n"
1857     ".clang-format file located in one of the parent\n"
1858     "directories of the source file (or current\n"
1859     "directory for stdin).\n"
1860     "Use -style=\"{key: value, ...}\" to set specific\n"
1861     "parameters, e.g.:\n"
1862     "  -style=\"{BasedOnStyle: llvm, IndentWidth: 8}\"";
1863 
1864 static FormatStyle::LanguageKind getLanguageByFileName(StringRef FileName) {
1865   if (FileName.endswith(".java"))
1866     return FormatStyle::LK_Java;
1867   if (FileName.endswith_lower(".js") || FileName.endswith_lower(".ts"))
1868     return FormatStyle::LK_JavaScript; // JavaScript or TypeScript.
1869   if (FileName.endswith(".m") || FileName.endswith(".mm"))
1870     return FormatStyle::LK_ObjC;
1871   if (FileName.endswith_lower(".proto") ||
1872       FileName.endswith_lower(".protodevel"))
1873     return FormatStyle::LK_Proto;
1874   if (FileName.endswith_lower(".td"))
1875     return FormatStyle::LK_TableGen;
1876   return FormatStyle::LK_Cpp;
1877 }
1878 
1879 FormatStyle getStyle(StringRef StyleName, StringRef FileName,
1880                      StringRef FallbackStyle, StringRef Code,
1881                      vfs::FileSystem *FS) {
1882   if (!FS) {
1883     FS = vfs::getRealFileSystem().get();
1884   }
1885   FormatStyle Style = getLLVMStyle();
1886   Style.Language = getLanguageByFileName(FileName);
1887 
1888   // This is a very crude detection of whether a header contains ObjC code that
1889   // should be improved over time and probably be done on tokens, not one the
1890   // bare content of the file.
1891   if (Style.Language == FormatStyle::LK_Cpp && FileName.endswith(".h") &&
1892       (Code.contains("\n- (") || Code.contains("\n+ (")))
1893     Style.Language = FormatStyle::LK_ObjC;
1894 
1895   if (!getPredefinedStyle(FallbackStyle, Style.Language, &Style)) {
1896     llvm::errs() << "Invalid fallback style \"" << FallbackStyle
1897                  << "\" using LLVM style\n";
1898     return Style;
1899   }
1900 
1901   if (StyleName.startswith("{")) {
1902     // Parse YAML/JSON style from the command line.
1903     if (std::error_code ec = parseConfiguration(StyleName, &Style)) {
1904       llvm::errs() << "Error parsing -style: " << ec.message() << ", using "
1905                    << FallbackStyle << " style\n";
1906     }
1907     return Style;
1908   }
1909 
1910   if (!StyleName.equals_lower("file")) {
1911     if (!getPredefinedStyle(StyleName, Style.Language, &Style))
1912       llvm::errs() << "Invalid value for -style, using " << FallbackStyle
1913                    << " style\n";
1914     return Style;
1915   }
1916 
1917   // Look for .clang-format/_clang-format file in the file's parent directories.
1918   SmallString<128> UnsuitableConfigFiles;
1919   SmallString<128> Path(FileName);
1920   llvm::sys::fs::make_absolute(Path);
1921   for (StringRef Directory = Path; !Directory.empty();
1922        Directory = llvm::sys::path::parent_path(Directory)) {
1923 
1924     auto Status = FS->status(Directory);
1925     if (!Status ||
1926         Status->getType() != llvm::sys::fs::file_type::directory_file) {
1927       continue;
1928     }
1929 
1930     SmallString<128> ConfigFile(Directory);
1931 
1932     llvm::sys::path::append(ConfigFile, ".clang-format");
1933     DEBUG(llvm::dbgs() << "Trying " << ConfigFile << "...\n");
1934 
1935     Status = FS->status(ConfigFile.str());
1936     bool IsFile =
1937         Status && (Status->getType() == llvm::sys::fs::file_type::regular_file);
1938     if (!IsFile) {
1939       // Try _clang-format too, since dotfiles are not commonly used on Windows.
1940       ConfigFile = Directory;
1941       llvm::sys::path::append(ConfigFile, "_clang-format");
1942       DEBUG(llvm::dbgs() << "Trying " << ConfigFile << "...\n");
1943       Status = FS->status(ConfigFile.str());
1944       IsFile = Status &&
1945                (Status->getType() == llvm::sys::fs::file_type::regular_file);
1946     }
1947 
1948     if (IsFile) {
1949       llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> Text =
1950           FS->getBufferForFile(ConfigFile.str());
1951       if (std::error_code EC = Text.getError()) {
1952         llvm::errs() << EC.message() << "\n";
1953         break;
1954       }
1955       if (std::error_code ec =
1956               parseConfiguration(Text.get()->getBuffer(), &Style)) {
1957         if (ec == ParseError::Unsuitable) {
1958           if (!UnsuitableConfigFiles.empty())
1959             UnsuitableConfigFiles.append(", ");
1960           UnsuitableConfigFiles.append(ConfigFile);
1961           continue;
1962         }
1963         llvm::errs() << "Error reading " << ConfigFile << ": " << ec.message()
1964                      << "\n";
1965         break;
1966       }
1967       DEBUG(llvm::dbgs() << "Using configuration file " << ConfigFile << "\n");
1968       return Style;
1969     }
1970   }
1971   if (!UnsuitableConfigFiles.empty()) {
1972     llvm::errs() << "Configuration file(s) do(es) not support "
1973                  << getLanguageName(Style.Language) << ": "
1974                  << UnsuitableConfigFiles << "\n";
1975   }
1976   return Style;
1977 }
1978 
1979 } // namespace format
1980 } // namespace clang
1981