1 //===--- PreprocessorTracker.cpp - Preprocessor tracking -*- C++ -*------===//
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 // The Basic Idea (Macro and Conditional Checking)
11 //
12 // Basically we install a PPCallbacks-derived object to track preprocessor
13 // activity, namely when a header file is entered/exited, when a macro
14 // is expanded, when "defined" is used, and when #if, #elif, #ifdef,
15 // and #ifndef are used.  We save the state of macro and "defined"
16 // expressions in a map, keyed on a name/file/line/column quadruple.
17 // The map entries store the different states (values) that a macro expansion,
18 // "defined" expression, or condition expression has in the course of
19 // processing for the one location in the one header containing it,
20 // plus a list of the nested include stacks for the states.  When a macro
21 // or "defined" expression evaluates to the same value, which is the
22 // desired case, only one state is stored.  Similarly, for conditional
23 // directives, we save the condition expression states in a separate map.
24 //
25 // This information is collected as modularize compiles all the headers
26 // given to it to process.  After all the compilations are performed,
27 // a check is performed for any entries in the maps that contain more
28 // than one different state, and for these an output message is generated.
29 //
30 // For example:
31 //
32 //   (...)/SubHeader.h:11:5:
33 //   #if SYMBOL == 1
34 //       ^
35 //   error: Macro instance 'SYMBOL' has different values in this header,
36 //          depending on how it was included.
37 //     'SYMBOL' expanded to: '1' with respect to these inclusion paths:
38 //       (...)/Header1.h
39 //         (...)/SubHeader.h
40 //   (...)/SubHeader.h:3:9:
41 //   #define SYMBOL 1
42 //             ^
43 //   Macro defined here.
44 //     'SYMBOL' expanded to: '2' with respect to these inclusion paths:
45 //       (...)/Header2.h
46 //           (...)/SubHeader.h
47 //   (...)/SubHeader.h:7:9:
48 //   #define SYMBOL 2
49 //             ^
50 //   Macro defined here.
51 //
52 // The Basic Idea ('Extern "C/C++" {}' Or 'namespace {}') With Nested
53 // '#include' Checking)
54 //
55 // To check for '#include' directives nested inside 'Extern "C/C++" {}'
56 // or 'namespace {}' blocks, we keep track of the '#include' directives
57 // while running the preprocessor, and later during a walk of the AST
58 // we call a function to check for any '#include' directies inside
59 // an 'Extern "C/C++" {}' or 'namespace {}' block, given its source
60 // range.
61 //
62 // Design and Implementation Details (Macro and Conditional Checking)
63 //
64 // A PreprocessorTrackerImpl class implements the PreprocessorTracker
65 // interface. It uses a PreprocessorCallbacks class derived from PPCallbacks
66 // to track preprocessor activity, namely entering/exiting a header, macro
67 // expansions, use of "defined" expressions, and #if, #elif, #ifdef, and
68 // #ifndef conditional directives. PreprocessorTrackerImpl stores a map
69 // of MacroExpansionTracker objects keyed on a name/file/line/column
70 // value represented by a light-weight PPItemKey value object. This
71 // is the key top-level data structure tracking the values of macro
72 // expansion instances.  Similarly, it stores a map of ConditionalTracker
73 // objects with the same kind of key, for tracking preprocessor conditional
74 // directives.
75 //
76 // The MacroExpansionTracker object represents one macro reference or use
77 // of a "defined" expression in a header file. It stores a handle to a
78 // string representing the unexpanded macro instance, a handle to a string
79 // representing the unpreprocessed source line containing the unexpanded
80 // macro instance, and a vector of one or more MacroExpansionInstance
81 // objects.
82 //
83 // The MacroExpansionInstance object represents one or more expansions
84 // of a macro reference, for the case where the macro expands to the same
85 // value. MacroExpansionInstance stores a handle to a string representing
86 // the expanded macro value, a PPItemKey representing the file/line/column
87 // where the macro was defined, a handle to a string representing the source
88 // line containing the macro definition, and a vector of InclusionPathHandle
89 // values that represents the hierarchies of include files for each case
90 // where the particular header containing the macro reference was referenced
91 // or included.
92 
93 // In the normal case where a macro instance always expands to the same
94 // value, the MacroExpansionTracker object will only contain one
95 // MacroExpansionInstance representing all the macro expansion instances.
96 // If a case was encountered where a macro instance expands to a value
97 // that is different from that seen before, or the macro was defined in
98 // a different place, a new MacroExpansionInstance object representing
99 // that case will be added to the vector in MacroExpansionTracker. If a
100 // macro instance expands to a value already seen before, the
101 // InclusionPathHandle representing that case's include file hierarchy
102 // will be added to the existing MacroExpansionInstance object.
103 
104 // For checking conditional directives, the ConditionalTracker class
105 // functions similarly to MacroExpansionTracker, but tracks an #if,
106 // #elif, #ifdef, or #ifndef directive in a header file.  It stores
107 // a vector of one or two ConditionalExpansionInstance objects,
108 // representing the cases where the conditional expression evaluates
109 // to true or false.  This latter object stores the evaluated value
110 // of the condition expression (a bool) and a vector of
111 // InclusionPathHandles.
112 //
113 // To reduce the instances of string and object copying, the
114 // PreprocessorTrackerImpl class uses a StringPool to save all stored
115 // strings, and defines a StringHandle type to abstract the references
116 // to the strings.
117 //
118 // PreprocessorTrackerImpl also maintains a list representing the unique
119 // headers, which is just a vector of StringHandle's for the header file
120 // paths. A HeaderHandle abstracts a reference to a header, and is simply
121 // the index of the stored header file path.
122 //
123 // A HeaderInclusionPath class abstracts a unique hierarchy of header file
124 // inclusions. It simply stores a vector of HeaderHandles ordered from the
125 // top-most header (the one from the header list passed to modularize) down
126 // to the header containing the macro reference. PreprocessorTrackerImpl
127 // stores a vector of these objects. An InclusionPathHandle typedef
128 // abstracts a reference to one of the HeaderInclusionPath objects, and is
129 // simply the index of the stored HeaderInclusionPath object. The
130 // MacroExpansionInstance object stores a vector of these handles so that
131 // the reporting function can display the include hierarchies for the macro
132 // expansion instances represented by that object, to help the user
133 // understand how the header was included. (A future enhancement might
134 // be to associate a line number for the #include directives, but I
135 // think not doing so is good enough for the present.)
136 //
137 // A key reason for using these opaque handles was to try to keep all the
138 // internal objects light-weight value objects, in order to reduce string
139 // and object copying overhead, and to abstract this implementation detail.
140 //
141 // The key data structures are built up while modularize runs the headers
142 // through the compilation. A PreprocessorTracker instance is created and
143 // passed down to the AST action and consumer objects in modularize. For
144 // each new compilation instance, the consumer calls the
145 // PreprocessorTracker's handleNewPreprocessorEntry function, which sets
146 // up a PreprocessorCallbacks object for the preprocessor. At the end of
147 // the compilation instance, the PreprocessorTracker's
148 // handleNewPreprocessorExit function handles cleaning up with respect
149 // to the preprocessing instance.
150 //
151 // The PreprocessorCallbacks object uses an overidden FileChanged callback
152 // to determine when a header is entered and exited (including exiting the
153 // header during #include directives). It calls PreprocessorTracker's
154 // handleHeaderEntry and handleHeaderExit functions upon entering and
155 // exiting a header. These functions manage a stack of header handles
156 // representing by a vector, pushing and popping header handles as headers
157 // are entered and exited. When a HeaderInclusionPath object is created,
158 // it simply copies this stack.
159 //
160 // The PreprocessorCallbacks object uses an overridden MacroExpands callback
161 // to track when a macro expansion is performed. It calls a couple of helper
162 // functions to get the unexpanded and expanded macro values as strings, but
163 // then calls PreprocessorTrackerImpl's addMacroExpansionInstance function to
164 // do the rest of the work. The getMacroExpandedString function uses the
165 // preprocessor's getSpelling to convert tokens to strings using the
166 // information passed to the MacroExpands callback, and simply concatenates
167 // them. It makes recursive calls to itself to handle nested macro
168 // definitions, and also handles function-style macros.
169 //
170 // PreprocessorTrackerImpl's addMacroExpansionInstance function looks for
171 // an existing MacroExpansionTracker entry in its map of MacroExampleTracker
172 // objects. If none exists, it adds one with one MacroExpansionInstance and
173 // returns. If a MacroExpansionTracker object already exists, it looks for
174 // an existing MacroExpansionInstance object stored in the
175 // MacroExpansionTracker object, one that matches the macro expanded value
176 // and the macro definition location. If a matching MacroExpansionInstance
177 // object is found, it just adds the current HeaderInclusionPath object to
178 // it. If not found, it creates and stores a new MacroExpantionInstance
179 // object. The addMacroExpansionInstance function calls a couple of helper
180 // functions to get the pre-formatted location and source line strings for
181 // the macro reference and the macro definition stored as string handles.
182 // These helper functions use the current source manager from the
183 // preprocessor. This is done in advance at this point in time because the
184 // source manager doesn't exist at the time of the reporting.
185 //
186 // For conditional check, the PreprocessorCallbacks class overrides the
187 // PPCallbacks handlers for #if, #elif, #ifdef, and #ifndef.  These handlers
188 // call the addConditionalExpansionInstance method of
189 // PreprocessorTrackerImpl.  The process is similar to that of macros, but
190 // with some different data and error messages.  A lookup is performed for
191 // the conditional, and if a ConditionalTracker object doesn't yet exist for
192 // the conditional, a new one is added, including adding a
193 // ConditionalExpansionInstance object to it to represent the condition
194 // expression state.  If a ConditionalTracker for the conditional does
195 // exist, a lookup is made for a ConditionalExpansionInstance object
196 // matching the condition expression state.  If one exists, a
197 // HeaderInclusionPath is added to it.  Otherwise a new
198 // ConditionalExpansionInstance  entry is made.  If a ConditionalTracker
199 // has two ConditionalExpansionInstance objects, it means there was a
200 // conflict, meaning the conditional expression evaluated differently in
201 // one or more cases.
202 //
203 // After modularize has performed all the compilations, it enters a phase
204 // of error reporting. This new feature adds to this reporting phase calls
205 // to the PreprocessorTracker's reportInconsistentMacros and
206 // reportInconsistentConditionals functions. These functions walk the maps
207 // of MacroExpansionTracker's and ConditionalTracker's respectively. If
208 // any of these objects have more than one MacroExpansionInstance or
209 // ConditionalExpansionInstance objects, it formats and outputs an error
210 // message like the example shown previously, using the stored data.
211 //
212 // A potential issue is that there is some overlap between the #if/#elif
213 // conditional and macro reporting.  I could disable the #if and #elif,
214 // leaving just the #ifdef and #ifndef, since these don't overlap.  Or,
215 // to make clearer the separate reporting phases, I could add an output
216 // message marking the phases.
217 //
218 // Design and Implementation Details ('Extern "C/C++" {}' Or
219 // 'namespace {}') With Nested '#include' Checking)
220 //
221 // We override the InclusionDirective in PPCallbacks to record information
222 // about each '#include' directive encountered during preprocessing.
223 // We co-opt the PPItemKey class to store the information about each
224 // '#include' directive, including the source file name containing the
225 // directive, the name of the file being included, and the source line
226 // and column of the directive.  We store these object in a vector,
227 // after first check to see if an entry already exists.
228 //
229 // Later, while the AST is being walked for other checks, we provide
230 // visit handlers for 'extern "C/C++" {}' and 'namespace (name) {}'
231 // blocks, checking to see if any '#include' directives occurred
232 // within the blocks, reporting errors if any found.
233 //
234 // Future Directions
235 //
236 // We probably should add options to disable any of the checks, in case
237 // there is some problem with them, or the messages get too verbose.
238 //
239 // With the map of all the macro and conditional expansion instances,
240 // it might be possible to add to the existing modularize error messages
241 // (the second part referring to definitions being different), attempting
242 // to tie them to the last macro conflict encountered with respect to the
243 // order of the code encountered.
244 //
245 //===--------------------------------------------------------------------===//
246 
247 #include "clang/Lex/LexDiagnostic.h"
248 #include "PreprocessorTracker.h"
249 #include "clang/Lex/MacroArgs.h"
250 #include "clang/Lex/PPCallbacks.h"
251 #include "llvm/ADT/SmallSet.h"
252 #include "llvm/Support/StringPool.h"
253 #include "llvm/Support/raw_ostream.h"
254 #include "ModularizeUtilities.h"
255 
256 namespace Modularize {
257 
258 // Some handle types
259 typedef llvm::PooledStringPtr StringHandle;
260 
261 typedef int HeaderHandle;
262 const HeaderHandle HeaderHandleInvalid = -1;
263 
264 typedef int InclusionPathHandle;
265 const InclusionPathHandle InclusionPathHandleInvalid = -1;
266 
267 // Some utility functions.
268 
269 // Get a "file:line:column" source location string.
270 static std::string getSourceLocationString(clang::Preprocessor &PP,
271                                            clang::SourceLocation Loc) {
272   if (Loc.isInvalid())
273     return std::string("(none)");
274   else
275     return Loc.printToString(PP.getSourceManager());
276 }
277 
278 // Get just the file name from a source location.
279 static std::string getSourceLocationFile(clang::Preprocessor &PP,
280                                          clang::SourceLocation Loc) {
281   std::string Source(getSourceLocationString(PP, Loc));
282   size_t Offset = Source.find(':', 2);
283   if (Offset == std::string::npos)
284     return Source;
285   return Source.substr(0, Offset);
286 }
287 
288 // Get just the line and column from a source location.
289 static void getSourceLocationLineAndColumn(clang::Preprocessor &PP,
290                                            clang::SourceLocation Loc, int &Line,
291                                            int &Column) {
292   clang::PresumedLoc PLoc = PP.getSourceManager().getPresumedLoc(Loc);
293   if (PLoc.isInvalid()) {
294     Line = 0;
295     Column = 0;
296     return;
297   }
298   Line = PLoc.getLine();
299   Column = PLoc.getColumn();
300 }
301 
302 // Retrieve source snippet from file image.
303 static std::string getSourceString(clang::Preprocessor &PP,
304                                    clang::SourceRange Range) {
305   clang::SourceLocation BeginLoc = Range.getBegin();
306   clang::SourceLocation EndLoc = Range.getEnd();
307   const char *BeginPtr = PP.getSourceManager().getCharacterData(BeginLoc);
308   const char *EndPtr = PP.getSourceManager().getCharacterData(EndLoc);
309   size_t Length = EndPtr - BeginPtr;
310   return llvm::StringRef(BeginPtr, Length).trim().str();
311 }
312 
313 // Retrieve source line from file image given a location.
314 static std::string getSourceLine(clang::Preprocessor &PP,
315                                  clang::SourceLocation Loc) {
316   const llvm::MemoryBuffer *MemBuffer =
317       PP.getSourceManager().getBuffer(PP.getSourceManager().getFileID(Loc));
318   const char *Buffer = MemBuffer->getBufferStart();
319   const char *BufferEnd = MemBuffer->getBufferEnd();
320   const char *BeginPtr = PP.getSourceManager().getCharacterData(Loc);
321   const char *EndPtr = BeginPtr;
322   while (BeginPtr > Buffer) {
323     if (*BeginPtr == '\n') {
324       BeginPtr++;
325       break;
326     }
327     BeginPtr--;
328   }
329   while (EndPtr < BufferEnd) {
330     if (*EndPtr == '\n') {
331       break;
332     }
333     EndPtr++;
334   }
335   size_t Length = EndPtr - BeginPtr;
336   return llvm::StringRef(BeginPtr, Length).str();
337 }
338 
339 // Retrieve source line from file image given a file ID and line number.
340 static std::string getSourceLine(clang::Preprocessor &PP, clang::FileID FileID,
341                                  int Line) {
342   const llvm::MemoryBuffer *MemBuffer = PP.getSourceManager().getBuffer(FileID);
343   const char *Buffer = MemBuffer->getBufferStart();
344   const char *BufferEnd = MemBuffer->getBufferEnd();
345   const char *BeginPtr = Buffer;
346   const char *EndPtr = BufferEnd;
347   int LineCounter = 1;
348   if (Line == 1)
349     BeginPtr = Buffer;
350   else {
351     while (Buffer < BufferEnd) {
352       if (*Buffer == '\n') {
353         if (++LineCounter == Line) {
354           BeginPtr = Buffer++ + 1;
355           break;
356         }
357       }
358       Buffer++;
359     }
360   }
361   while (Buffer < BufferEnd) {
362     if (*Buffer == '\n') {
363       EndPtr = Buffer;
364       break;
365     }
366     Buffer++;
367   }
368   size_t Length = EndPtr - BeginPtr;
369   return llvm::StringRef(BeginPtr, Length).str();
370 }
371 
372 // Get the string for the Unexpanded macro instance.
373 // The soureRange is expected to end at the last token
374 // for the macro instance, which in the case of a function-style
375 // macro will be a ')', but for an object-style macro, it
376 // will be the macro name itself.
377 static std::string getMacroUnexpandedString(clang::SourceRange Range,
378                                             clang::Preprocessor &PP,
379                                             llvm::StringRef MacroName,
380                                             const clang::MacroInfo *MI) {
381   clang::SourceLocation BeginLoc(Range.getBegin());
382   const char *BeginPtr = PP.getSourceManager().getCharacterData(BeginLoc);
383   size_t Length;
384   std::string Unexpanded;
385   if (MI->isFunctionLike()) {
386     clang::SourceLocation EndLoc(Range.getEnd());
387     const char *EndPtr = PP.getSourceManager().getCharacterData(EndLoc) + 1;
388     Length = (EndPtr - BeginPtr) + 1; // +1 is ')' width.
389   } else
390     Length = MacroName.size();
391   return llvm::StringRef(BeginPtr, Length).trim().str();
392 }
393 
394 // Get the expansion for a macro instance, given the information
395 // provided by PPCallbacks.
396 // FIXME: This doesn't support function-style macro instances
397 // passed as arguments to another function-style macro. However,
398 // since it still expands the inner arguments, it still
399 // allows modularize to effectively work with respect to macro
400 // consistency checking, although it displays the incorrect
401 // expansion in error messages.
402 static std::string getMacroExpandedString(clang::Preprocessor &PP,
403                                           llvm::StringRef MacroName,
404                                           const clang::MacroInfo *MI,
405                                           const clang::MacroArgs *Args) {
406   std::string Expanded;
407   // Walk over the macro Tokens.
408   for (const auto &T : MI->tokens()) {
409     clang::IdentifierInfo *II = T.getIdentifierInfo();
410     int ArgNo = (II && Args ? MI->getArgumentNum(II) : -1);
411     if (ArgNo == -1) {
412       // This isn't an argument, just add it.
413       if (II == nullptr)
414         Expanded += PP.getSpelling(T); // Not an identifier.
415       else {
416         // Token is for an identifier.
417         std::string Name = II->getName().str();
418         // Check for nexted macro references.
419         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
420         if (MacroInfo && (Name != MacroName))
421           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
422         else
423           Expanded += Name;
424       }
425       continue;
426     }
427     // We get here if it's a function-style macro with arguments.
428     const clang::Token *ResultArgToks;
429     const clang::Token *ArgTok = Args->getUnexpArgument(ArgNo);
430     if (Args->ArgNeedsPreexpansion(ArgTok, PP))
431       ResultArgToks = &(const_cast<clang::MacroArgs *>(Args))
432           ->getPreExpArgument(ArgNo, MI, PP)[0];
433     else
434       ResultArgToks = ArgTok; // Use non-preexpanded Tokens.
435     // If the arg token didn't expand into anything, ignore it.
436     if (ResultArgToks->is(clang::tok::eof))
437       continue;
438     unsigned NumToks = clang::MacroArgs::getArgLength(ResultArgToks);
439     // Append the resulting argument expansions.
440     for (unsigned ArgumentIndex = 0; ArgumentIndex < NumToks; ++ArgumentIndex) {
441       const clang::Token &AT = ResultArgToks[ArgumentIndex];
442       clang::IdentifierInfo *II = AT.getIdentifierInfo();
443       if (II == nullptr)
444         Expanded += PP.getSpelling(AT); // Not an identifier.
445       else {
446         // It's an identifier.  Check for further expansion.
447         std::string Name = II->getName().str();
448         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
449         if (MacroInfo)
450           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
451         else
452           Expanded += Name;
453       }
454     }
455   }
456   return Expanded;
457 }
458 
459 namespace {
460 
461 // ConditionValueKind strings.
462 const char *
463 ConditionValueKindStrings[] = {
464   "(not evaluated)", "false", "true"
465 };
466 
467 bool operator<(const StringHandle &H1, const StringHandle &H2) {
468   const char *S1 = (H1 ? *H1 : "");
469   const char *S2 = (H2 ? *H2 : "");
470   int Diff = strcmp(S1, S2);
471   return Diff < 0;
472 }
473 bool operator>(const StringHandle &H1, const StringHandle &H2) {
474   const char *S1 = (H1 ? *H1 : "");
475   const char *S2 = (H2 ? *H2 : "");
476   int Diff = strcmp(S1, S2);
477   return Diff > 0;
478 }
479 
480 // Preprocessor item key.
481 //
482 // This class represents a location in a source file, for use
483 // as a key representing a unique name/file/line/column quadruplet,
484 // which in this case is used to identify a macro expansion instance,
485 // but could be used for other things as well.
486 // The file is a header file handle, the line is a line number,
487 // and the column is a column number.
488 class PPItemKey {
489 public:
490   PPItemKey(clang::Preprocessor &PP, StringHandle Name, HeaderHandle File,
491             clang::SourceLocation Loc)
492       : Name(Name), File(File) {
493     getSourceLocationLineAndColumn(PP, Loc, Line, Column);
494   }
495   PPItemKey(StringHandle Name, HeaderHandle File, int Line, int Column)
496       : Name(Name), File(File), Line(Line), Column(Column) {}
497   PPItemKey(const PPItemKey &Other)
498       : Name(Other.Name), File(Other.File), Line(Other.Line),
499         Column(Other.Column) {}
500   PPItemKey() : File(HeaderHandleInvalid), Line(0), Column(0) {}
501   bool operator==(const PPItemKey &Other) const {
502     if (Name != Other.Name)
503       return false;
504     if (File != Other.File)
505       return false;
506     if (Line != Other.Line)
507       return false;
508     return Column == Other.Column;
509   }
510   bool operator<(const PPItemKey &Other) const {
511     if (Name < Other.Name)
512       return true;
513     else if (Name > Other.Name)
514       return false;
515     if (File < Other.File)
516       return true;
517     else if (File > Other.File)
518       return false;
519     if (Line < Other.Line)
520       return true;
521     else if (Line > Other.Line)
522       return false;
523     return Column < Other.Column;
524   }
525   StringHandle Name;
526   HeaderHandle File;
527   int Line;
528   int Column;
529 };
530 
531 // Header inclusion path.
532 class HeaderInclusionPath {
533 public:
534   HeaderInclusionPath(std::vector<HeaderHandle> HeaderInclusionPath)
535       : Path(HeaderInclusionPath) {}
536   HeaderInclusionPath(const HeaderInclusionPath &Other) : Path(Other.Path) {}
537   HeaderInclusionPath() {}
538   std::vector<HeaderHandle> Path;
539 };
540 
541 // Macro expansion instance.
542 //
543 // This class represents an instance of a macro expansion with a
544 // unique value.  It also stores the unique header inclusion paths
545 // for use in telling the user the nested include path to the header.
546 class MacroExpansionInstance {
547 public:
548   MacroExpansionInstance(StringHandle MacroExpanded,
549                          PPItemKey &DefinitionLocation,
550                          StringHandle DefinitionSourceLine,
551                          InclusionPathHandle H)
552       : MacroExpanded(MacroExpanded), DefinitionLocation(DefinitionLocation),
553         DefinitionSourceLine(DefinitionSourceLine) {
554     InclusionPathHandles.push_back(H);
555   }
556   MacroExpansionInstance() {}
557 
558   // Check for the presence of a header inclusion path handle entry.
559   // Return false if not found.
560   bool haveInclusionPathHandle(InclusionPathHandle H) {
561     for (std::vector<InclusionPathHandle>::iterator
562              I = InclusionPathHandles.begin(),
563              E = InclusionPathHandles.end();
564          I != E; ++I) {
565       if (*I == H)
566         return true;
567     }
568     return InclusionPathHandleInvalid;
569   }
570   // Add a new header inclusion path entry, if not already present.
571   void addInclusionPathHandle(InclusionPathHandle H) {
572     if (!haveInclusionPathHandle(H))
573       InclusionPathHandles.push_back(H);
574   }
575 
576   // A string representing the macro instance after preprocessing.
577   StringHandle MacroExpanded;
578   // A file/line/column triplet representing the macro definition location.
579   PPItemKey DefinitionLocation;
580   // A place to save the macro definition line string.
581   StringHandle DefinitionSourceLine;
582   // The header inclusion path handles for all the instances.
583   std::vector<InclusionPathHandle> InclusionPathHandles;
584 };
585 
586 // Macro expansion instance tracker.
587 //
588 // This class represents one macro expansion, keyed by a PPItemKey.
589 // It stores a string representing the macro reference in the source,
590 // and a list of ConditionalExpansionInstances objects representing
591 // the unique values the condition expands to in instances of the header.
592 class MacroExpansionTracker {
593 public:
594   MacroExpansionTracker(StringHandle MacroUnexpanded,
595                         StringHandle MacroExpanded,
596                         StringHandle InstanceSourceLine,
597                         PPItemKey &DefinitionLocation,
598                         StringHandle DefinitionSourceLine,
599                         InclusionPathHandle InclusionPathHandle)
600       : MacroUnexpanded(MacroUnexpanded),
601         InstanceSourceLine(InstanceSourceLine) {
602     addMacroExpansionInstance(MacroExpanded, DefinitionLocation,
603                               DefinitionSourceLine, InclusionPathHandle);
604   }
605   MacroExpansionTracker() {}
606 
607   // Find a matching macro expansion instance.
608   MacroExpansionInstance *
609   findMacroExpansionInstance(StringHandle MacroExpanded,
610                              PPItemKey &DefinitionLocation) {
611     for (std::vector<MacroExpansionInstance>::iterator
612              I = MacroExpansionInstances.begin(),
613              E = MacroExpansionInstances.end();
614          I != E; ++I) {
615       if ((I->MacroExpanded == MacroExpanded) &&
616           (I->DefinitionLocation == DefinitionLocation)) {
617         return &*I; // Found.
618       }
619     }
620     return nullptr; // Not found.
621   }
622 
623   // Add a macro expansion instance.
624   void addMacroExpansionInstance(StringHandle MacroExpanded,
625                                  PPItemKey &DefinitionLocation,
626                                  StringHandle DefinitionSourceLine,
627                                  InclusionPathHandle InclusionPathHandle) {
628     MacroExpansionInstances.push_back(
629         MacroExpansionInstance(MacroExpanded, DefinitionLocation,
630                                DefinitionSourceLine, InclusionPathHandle));
631   }
632 
633   // Return true if there is a mismatch.
634   bool hasMismatch() { return MacroExpansionInstances.size() > 1; }
635 
636   // A string representing the macro instance without expansion.
637   StringHandle MacroUnexpanded;
638   // A place to save the macro instance source line string.
639   StringHandle InstanceSourceLine;
640   // The macro expansion instances.
641   // If all instances of the macro expansion expand to the same value,
642   // This vector will only have one instance.
643   std::vector<MacroExpansionInstance> MacroExpansionInstances;
644 };
645 
646 // Conditional expansion instance.
647 //
648 // This class represents an instance of a condition exoression result
649 // with a unique value.  It also stores the unique header inclusion paths
650 // for use in telling the user the nested include path to the header.
651 class ConditionalExpansionInstance {
652 public:
653   ConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue, InclusionPathHandle H)
654       : ConditionValue(ConditionValue) {
655     InclusionPathHandles.push_back(H);
656   }
657   ConditionalExpansionInstance() {}
658 
659   // Check for the presence of a header inclusion path handle entry.
660   // Return false if not found.
661   bool haveInclusionPathHandle(InclusionPathHandle H) {
662     for (std::vector<InclusionPathHandle>::iterator
663              I = InclusionPathHandles.begin(),
664              E = InclusionPathHandles.end();
665          I != E; ++I) {
666       if (*I == H)
667         return true;
668     }
669     return InclusionPathHandleInvalid;
670   }
671   // Add a new header inclusion path entry, if not already present.
672   void addInclusionPathHandle(InclusionPathHandle H) {
673     if (!haveInclusionPathHandle(H))
674       InclusionPathHandles.push_back(H);
675   }
676 
677   // A flag representing the evaluated condition value.
678   clang::PPCallbacks::ConditionValueKind ConditionValue;
679   // The header inclusion path handles for all the instances.
680   std::vector<InclusionPathHandle> InclusionPathHandles;
681 };
682 
683 // Conditional directive instance tracker.
684 //
685 // This class represents one conditional directive, keyed by a PPItemKey.
686 // It stores a string representing the macro reference in the source,
687 // and a list of ConditionExpansionInstance objects representing
688 // the unique value the condition expression expands to in instances of
689 // the header.
690 class ConditionalTracker {
691 public:
692   ConditionalTracker(clang::tok::PPKeywordKind DirectiveKind,
693                      clang::PPCallbacks::ConditionValueKind ConditionValue,
694                      StringHandle ConditionUnexpanded,
695                      InclusionPathHandle InclusionPathHandle)
696       : DirectiveKind(DirectiveKind), ConditionUnexpanded(ConditionUnexpanded) {
697     addConditionalExpansionInstance(ConditionValue, InclusionPathHandle);
698   }
699   ConditionalTracker() {}
700 
701   // Find a matching condition expansion instance.
702   ConditionalExpansionInstance *
703   findConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue) {
704     for (std::vector<ConditionalExpansionInstance>::iterator
705              I = ConditionalExpansionInstances.begin(),
706              E = ConditionalExpansionInstances.end();
707          I != E; ++I) {
708       if (I->ConditionValue == ConditionValue) {
709         return &*I; // Found.
710       }
711     }
712     return nullptr; // Not found.
713   }
714 
715   // Add a conditional expansion instance.
716   void
717   addConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue,
718                                   InclusionPathHandle InclusionPathHandle) {
719     ConditionalExpansionInstances.push_back(
720         ConditionalExpansionInstance(ConditionValue, InclusionPathHandle));
721   }
722 
723   // Return true if there is a mismatch.
724   bool hasMismatch() { return ConditionalExpansionInstances.size() > 1; }
725 
726   // The kind of directive.
727   clang::tok::PPKeywordKind DirectiveKind;
728   // A string representing the macro instance without expansion.
729   StringHandle ConditionUnexpanded;
730   // The condition expansion instances.
731   // If all instances of the conditional expression expand to the same value,
732   // This vector will only have one instance.
733   std::vector<ConditionalExpansionInstance> ConditionalExpansionInstances;
734 };
735 
736 class PreprocessorTrackerImpl;
737 
738 // Preprocessor callbacks for modularize.
739 //
740 // This class derives from the Clang PPCallbacks class to track preprocessor
741 // actions, such as changing files and handling preprocessor directives and
742 // macro expansions.  It has to figure out when a new header file is entered
743 // and left, as the provided handler is not particularly clear about it.
744 class PreprocessorCallbacks : public clang::PPCallbacks {
745 public:
746   PreprocessorCallbacks(PreprocessorTrackerImpl &ppTracker,
747                         clang::Preprocessor &PP, llvm::StringRef rootHeaderFile)
748       : PPTracker(ppTracker), PP(PP), RootHeaderFile(rootHeaderFile) {}
749   ~PreprocessorCallbacks() override {}
750 
751   // Overridden handlers.
752   void InclusionDirective(clang::SourceLocation HashLoc,
753                           const clang::Token &IncludeTok,
754                           llvm::StringRef FileName, bool IsAngled,
755                           clang::CharSourceRange FilenameRange,
756                           const clang::FileEntry *File,
757                           llvm::StringRef SearchPath,
758                           llvm::StringRef RelativePath,
759                           const clang::Module *Imported) override;
760   void FileChanged(clang::SourceLocation Loc,
761                    clang::PPCallbacks::FileChangeReason Reason,
762                    clang::SrcMgr::CharacteristicKind FileType,
763                    clang::FileID PrevFID = clang::FileID()) override;
764   void MacroExpands(const clang::Token &MacroNameTok,
765                     const clang::MacroDefinition &MD, clang::SourceRange Range,
766                     const clang::MacroArgs *Args) override;
767   void Defined(const clang::Token &MacroNameTok,
768                const clang::MacroDefinition &MD,
769                clang::SourceRange Range) override;
770   void If(clang::SourceLocation Loc, clang::SourceRange ConditionRange,
771           clang::PPCallbacks::ConditionValueKind ConditionResult) override;
772   void Elif(clang::SourceLocation Loc, clang::SourceRange ConditionRange,
773             clang::PPCallbacks::ConditionValueKind ConditionResult,
774             clang::SourceLocation IfLoc) override;
775   void Ifdef(clang::SourceLocation Loc, const clang::Token &MacroNameTok,
776              const clang::MacroDefinition &MD) override;
777   void Ifndef(clang::SourceLocation Loc, const clang::Token &MacroNameTok,
778               const clang::MacroDefinition &MD) override;
779 
780 private:
781   PreprocessorTrackerImpl &PPTracker;
782   clang::Preprocessor &PP;
783   std::string RootHeaderFile;
784 };
785 
786 // Preprocessor macro expansion item map types.
787 typedef std::map<PPItemKey, MacroExpansionTracker> MacroExpansionMap;
788 typedef std::map<PPItemKey, MacroExpansionTracker>::iterator
789 MacroExpansionMapIter;
790 
791 // Preprocessor conditional expansion item map types.
792 typedef std::map<PPItemKey, ConditionalTracker> ConditionalExpansionMap;
793 typedef std::map<PPItemKey, ConditionalTracker>::iterator
794 ConditionalExpansionMapIter;
795 
796 // Preprocessor tracker for modularize.
797 //
798 // This class stores information about all the headers processed in the
799 // course of running modularize.
800 class PreprocessorTrackerImpl : public PreprocessorTracker {
801 public:
802   PreprocessorTrackerImpl(llvm::SmallVector<std::string, 32> &Headers,
803         bool DoBlockCheckHeaderListOnly)
804       : BlockCheckHeaderListOnly(DoBlockCheckHeaderListOnly),
805         CurrentInclusionPathHandle(InclusionPathHandleInvalid),
806         InNestedHeader(false) {
807     // Use canonical header path representation.
808     for (llvm::ArrayRef<std::string>::iterator I = Headers.begin(),
809       E = Headers.end();
810       I != E; ++I) {
811       HeaderList.push_back(getCanonicalPath(*I));
812     }
813   }
814 
815   ~PreprocessorTrackerImpl() override {}
816 
817   // Handle entering a preprocessing session.
818   void handlePreprocessorEntry(clang::Preprocessor &PP,
819                                llvm::StringRef rootHeaderFile) override {
820     HeadersInThisCompile.clear();
821     assert((HeaderStack.size() == 0) && "Header stack should be empty.");
822     pushHeaderHandle(addHeader(rootHeaderFile));
823     PP.addPPCallbacks(llvm::make_unique<PreprocessorCallbacks>(*this, PP,
824                                                                rootHeaderFile));
825   }
826   // Handle exiting a preprocessing session.
827   void handlePreprocessorExit() override { HeaderStack.clear(); }
828 
829   // Handle include directive.
830   // This function is called every time an include directive is seen by the
831   // preprocessor, for the purpose of later checking for 'extern "" {}' or
832   // "namespace {}" blocks containing #include directives.
833   void handleIncludeDirective(llvm::StringRef DirectivePath, int DirectiveLine,
834                               int DirectiveColumn,
835                               llvm::StringRef TargetPath) override {
836     // If it's not a header in the header list, ignore it with respect to
837     // the check.
838     if (BlockCheckHeaderListOnly && !isHeaderListHeader(TargetPath))
839       return;
840     HeaderHandle CurrentHeaderHandle = findHeaderHandle(DirectivePath);
841     StringHandle IncludeHeaderHandle = addString(TargetPath);
842     for (std::vector<PPItemKey>::const_iterator I = IncludeDirectives.begin(),
843                                                 E = IncludeDirectives.end();
844          I != E; ++I) {
845       // If we already have an entry for this directive, return now.
846       if ((I->File == CurrentHeaderHandle) && (I->Line == DirectiveLine))
847         return;
848     }
849     PPItemKey IncludeDirectiveItem(IncludeHeaderHandle, CurrentHeaderHandle,
850                                    DirectiveLine, DirectiveColumn);
851     IncludeDirectives.push_back(IncludeDirectiveItem);
852   }
853 
854   // Check for include directives within the given source line range.
855   // Report errors if any found.  Returns true if no include directives
856   // found in block.
857   bool checkForIncludesInBlock(clang::Preprocessor &PP,
858                                clang::SourceRange BlockSourceRange,
859                                const char *BlockIdentifierMessage,
860                                llvm::raw_ostream &OS) override {
861     clang::SourceLocation BlockStartLoc = BlockSourceRange.getBegin();
862     clang::SourceLocation BlockEndLoc = BlockSourceRange.getEnd();
863     // Use block location to get FileID of both the include directive
864     // and block statement.
865     clang::FileID FileID = PP.getSourceManager().getFileID(BlockStartLoc);
866     std::string SourcePath = getSourceLocationFile(PP, BlockStartLoc);
867     SourcePath = ModularizeUtilities::getCanonicalPath(SourcePath);
868     HeaderHandle SourceHandle = findHeaderHandle(SourcePath);
869     if (SourceHandle == -1)
870       return true;
871     int BlockStartLine, BlockStartColumn, BlockEndLine, BlockEndColumn;
872     bool returnValue = true;
873     getSourceLocationLineAndColumn(PP, BlockStartLoc, BlockStartLine,
874                                    BlockStartColumn);
875     getSourceLocationLineAndColumn(PP, BlockEndLoc, BlockEndLine,
876                                    BlockEndColumn);
877     for (std::vector<PPItemKey>::const_iterator I = IncludeDirectives.begin(),
878                                                 E = IncludeDirectives.end();
879          I != E; ++I) {
880       // If we find an entry within the block, report an error.
881       if ((I->File == SourceHandle) && (I->Line >= BlockStartLine) &&
882           (I->Line < BlockEndLine)) {
883         returnValue = false;
884         OS << SourcePath << ":" << I->Line << ":" << I->Column << ":\n";
885         OS << getSourceLine(PP, FileID, I->Line) << "\n";
886         if (I->Column > 0)
887           OS << std::string(I->Column - 1, ' ') << "^\n";
888         OS << "error: Include directive within " << BlockIdentifierMessage
889            << ".\n";
890         OS << SourcePath << ":" << BlockStartLine << ":" << BlockStartColumn
891            << ":\n";
892         OS << getSourceLine(PP, BlockStartLoc) << "\n";
893         if (BlockStartColumn > 0)
894           OS << std::string(BlockStartColumn - 1, ' ') << "^\n";
895         OS << "The \"" << BlockIdentifierMessage << "\" block is here.\n";
896       }
897     }
898     return returnValue;
899   }
900 
901   // Handle entering a header source file.
902   void handleHeaderEntry(clang::Preprocessor &PP, llvm::StringRef HeaderPath) {
903     // Ignore <built-in> and <command-line> to reduce message clutter.
904     if (HeaderPath.startswith("<"))
905       return;
906     HeaderHandle H = addHeader(HeaderPath);
907     if (H != getCurrentHeaderHandle())
908       pushHeaderHandle(H);
909     // Check for nested header.
910     if (!InNestedHeader)
911       InNestedHeader = !HeadersInThisCompile.insert(H).second;
912   }
913 
914   // Handle exiting a header source file.
915   void handleHeaderExit(llvm::StringRef HeaderPath) {
916     // Ignore <built-in> and <command-line> to reduce message clutter.
917     if (HeaderPath.startswith("<"))
918       return;
919     HeaderHandle H = findHeaderHandle(HeaderPath);
920     HeaderHandle TH;
921     if (isHeaderHandleInStack(H)) {
922       do {
923         TH = getCurrentHeaderHandle();
924         popHeaderHandle();
925       } while ((TH != H) && (HeaderStack.size() != 0));
926     }
927     InNestedHeader = false;
928   }
929 
930   // Lookup/add string.
931   StringHandle addString(llvm::StringRef Str) { return Strings.intern(Str); }
932 
933   // Convert to a canonical path.
934   std::string getCanonicalPath(llvm::StringRef path) const {
935     std::string CanonicalPath(path);
936     std::replace(CanonicalPath.begin(), CanonicalPath.end(), '\\', '/');
937     return CanonicalPath;
938   }
939 
940   // Return true if the given header is in the header list.
941   bool isHeaderListHeader(llvm::StringRef HeaderPath) const {
942     std::string CanonicalPath = getCanonicalPath(HeaderPath);
943     for (llvm::ArrayRef<std::string>::iterator I = HeaderList.begin(),
944         E = HeaderList.end();
945         I != E; ++I) {
946       if (*I == CanonicalPath)
947         return true;
948     }
949     return false;
950   }
951 
952   // Get the handle of a header file entry.
953   // Return HeaderHandleInvalid if not found.
954   HeaderHandle findHeaderHandle(llvm::StringRef HeaderPath) const {
955     std::string CanonicalPath = getCanonicalPath(HeaderPath);
956     HeaderHandle H = 0;
957     for (std::vector<StringHandle>::const_iterator I = HeaderPaths.begin(),
958                                                    E = HeaderPaths.end();
959          I != E; ++I, ++H) {
960       if (**I == CanonicalPath)
961         return H;
962     }
963     return HeaderHandleInvalid;
964   }
965 
966   // Add a new header file entry, or return existing handle.
967   // Return the header handle.
968   HeaderHandle addHeader(llvm::StringRef HeaderPath) {
969     std::string CanonicalPath = getCanonicalPath(HeaderPath);
970     HeaderHandle H = findHeaderHandle(CanonicalPath);
971     if (H == HeaderHandleInvalid) {
972       H = HeaderPaths.size();
973       HeaderPaths.push_back(addString(CanonicalPath));
974     }
975     return H;
976   }
977 
978   // Return a header file path string given its handle.
979   StringHandle getHeaderFilePath(HeaderHandle H) const {
980     if ((H >= 0) && (H < (HeaderHandle)HeaderPaths.size()))
981       return HeaderPaths[H];
982     return StringHandle();
983   }
984 
985   // Returns a handle to the inclusion path.
986   InclusionPathHandle pushHeaderHandle(HeaderHandle H) {
987     HeaderStack.push_back(H);
988     return CurrentInclusionPathHandle = addInclusionPathHandle(HeaderStack);
989   }
990   // Pops the last header handle from the stack;
991   void popHeaderHandle() {
992     // assert((HeaderStack.size() != 0) && "Header stack already empty.");
993     if (HeaderStack.size() != 0) {
994       HeaderStack.pop_back();
995       CurrentInclusionPathHandle = addInclusionPathHandle(HeaderStack);
996     }
997   }
998   // Get the top handle on the header stack.
999   HeaderHandle getCurrentHeaderHandle() const {
1000     if (HeaderStack.size() != 0)
1001       return HeaderStack.back();
1002     return HeaderHandleInvalid;
1003   }
1004 
1005   // Check for presence of header handle in the header stack.
1006   bool isHeaderHandleInStack(HeaderHandle H) const {
1007     for (std::vector<HeaderHandle>::const_iterator I = HeaderStack.begin(),
1008                                                    E = HeaderStack.end();
1009          I != E; ++I) {
1010       if (*I == H)
1011         return true;
1012     }
1013     return false;
1014   }
1015 
1016   // Get the handle of a header inclusion path entry.
1017   // Return InclusionPathHandleInvalid if not found.
1018   InclusionPathHandle
1019   findInclusionPathHandle(const std::vector<HeaderHandle> &Path) const {
1020     InclusionPathHandle H = 0;
1021     for (std::vector<HeaderInclusionPath>::const_iterator
1022              I = InclusionPaths.begin(),
1023              E = InclusionPaths.end();
1024          I != E; ++I, ++H) {
1025       if (I->Path == Path)
1026         return H;
1027     }
1028     return HeaderHandleInvalid;
1029   }
1030   // Add a new header inclusion path entry, or return existing handle.
1031   // Return the header inclusion path entry handle.
1032   InclusionPathHandle
1033   addInclusionPathHandle(const std::vector<HeaderHandle> &Path) {
1034     InclusionPathHandle H = findInclusionPathHandle(Path);
1035     if (H == HeaderHandleInvalid) {
1036       H = InclusionPaths.size();
1037       InclusionPaths.push_back(HeaderInclusionPath(Path));
1038     }
1039     return H;
1040   }
1041   // Return the current inclusion path handle.
1042   InclusionPathHandle getCurrentInclusionPathHandle() const {
1043     return CurrentInclusionPathHandle;
1044   }
1045 
1046   // Return an inclusion path given its handle.
1047   const std::vector<HeaderHandle> &
1048   getInclusionPath(InclusionPathHandle H) const {
1049     if ((H >= 0) && (H <= (InclusionPathHandle)InclusionPaths.size()))
1050       return InclusionPaths[H].Path;
1051     static std::vector<HeaderHandle> Empty;
1052     return Empty;
1053   }
1054 
1055   // Add a macro expansion instance.
1056   void addMacroExpansionInstance(clang::Preprocessor &PP, HeaderHandle H,
1057                                  clang::SourceLocation InstanceLoc,
1058                                  clang::SourceLocation DefinitionLoc,
1059                                  clang::IdentifierInfo *II,
1060                                  llvm::StringRef MacroUnexpanded,
1061                                  llvm::StringRef MacroExpanded,
1062                                  InclusionPathHandle InclusionPathHandle) {
1063     if (InNestedHeader)
1064       return;
1065     StringHandle MacroName = addString(II->getName());
1066     PPItemKey InstanceKey(PP, MacroName, H, InstanceLoc);
1067     PPItemKey DefinitionKey(PP, MacroName, H, DefinitionLoc);
1068     MacroExpansionMapIter I = MacroExpansions.find(InstanceKey);
1069     // If existing instance of expansion not found, add one.
1070     if (I == MacroExpansions.end()) {
1071       std::string InstanceSourceLine =
1072           getSourceLocationString(PP, InstanceLoc) + ":\n" +
1073           getSourceLine(PP, InstanceLoc) + "\n";
1074       std::string DefinitionSourceLine =
1075           getSourceLocationString(PP, DefinitionLoc) + ":\n" +
1076           getSourceLine(PP, DefinitionLoc) + "\n";
1077       MacroExpansions[InstanceKey] = MacroExpansionTracker(
1078           addString(MacroUnexpanded), addString(MacroExpanded),
1079           addString(InstanceSourceLine), DefinitionKey,
1080           addString(DefinitionSourceLine), InclusionPathHandle);
1081     } else {
1082       // We've seen the macro before.  Get its tracker.
1083       MacroExpansionTracker &CondTracker = I->second;
1084       // Look up an existing instance value for the macro.
1085       MacroExpansionInstance *MacroInfo =
1086           CondTracker.findMacroExpansionInstance(addString(MacroExpanded),
1087                                                  DefinitionKey);
1088       // If found, just add the inclusion path to the instance.
1089       if (MacroInfo)
1090         MacroInfo->addInclusionPathHandle(InclusionPathHandle);
1091       else {
1092         // Otherwise add a new instance with the unique value.
1093         std::string DefinitionSourceLine =
1094             getSourceLocationString(PP, DefinitionLoc) + ":\n" +
1095             getSourceLine(PP, DefinitionLoc) + "\n";
1096         CondTracker.addMacroExpansionInstance(
1097             addString(MacroExpanded), DefinitionKey,
1098             addString(DefinitionSourceLine), InclusionPathHandle);
1099       }
1100     }
1101   }
1102 
1103   // Add a conditional expansion instance.
1104   void
1105   addConditionalExpansionInstance(clang::Preprocessor &PP, HeaderHandle H,
1106                                   clang::SourceLocation InstanceLoc,
1107                                   clang::tok::PPKeywordKind DirectiveKind,
1108                                   clang::PPCallbacks::ConditionValueKind ConditionValue,
1109                                   llvm::StringRef ConditionUnexpanded,
1110                                   InclusionPathHandle InclusionPathHandle) {
1111     // Ignore header guards, assuming the header guard is the only conditional.
1112     if (InNestedHeader)
1113       return;
1114     StringHandle ConditionUnexpandedHandle(addString(ConditionUnexpanded));
1115     PPItemKey InstanceKey(PP, ConditionUnexpandedHandle, H, InstanceLoc);
1116     ConditionalExpansionMapIter I = ConditionalExpansions.find(InstanceKey);
1117     // If existing instance of condition not found, add one.
1118     if (I == ConditionalExpansions.end()) {
1119       std::string InstanceSourceLine =
1120           getSourceLocationString(PP, InstanceLoc) + ":\n" +
1121           getSourceLine(PP, InstanceLoc) + "\n";
1122       ConditionalExpansions[InstanceKey] =
1123           ConditionalTracker(DirectiveKind, ConditionValue,
1124                              ConditionUnexpandedHandle, InclusionPathHandle);
1125     } else {
1126       // We've seen the conditional before.  Get its tracker.
1127       ConditionalTracker &CondTracker = I->second;
1128       // Look up an existing instance value for the condition.
1129       ConditionalExpansionInstance *MacroInfo =
1130           CondTracker.findConditionalExpansionInstance(ConditionValue);
1131       // If found, just add the inclusion path to the instance.
1132       if (MacroInfo)
1133         MacroInfo->addInclusionPathHandle(InclusionPathHandle);
1134       else {
1135         // Otherwise add a new instance with the unique value.
1136         CondTracker.addConditionalExpansionInstance(ConditionValue,
1137                                                     InclusionPathHandle);
1138       }
1139     }
1140   }
1141 
1142   // Report on inconsistent macro instances.
1143   // Returns true if any mismatches.
1144   bool reportInconsistentMacros(llvm::raw_ostream &OS) override {
1145     bool ReturnValue = false;
1146     // Walk all the macro expansion trackers in the map.
1147     for (MacroExpansionMapIter I = MacroExpansions.begin(),
1148                                E = MacroExpansions.end();
1149          I != E; ++I) {
1150       const PPItemKey &ItemKey = I->first;
1151       MacroExpansionTracker &MacroExpTracker = I->second;
1152       // If no mismatch (only one instance value) continue.
1153       if (!MacroExpTracker.hasMismatch())
1154         continue;
1155       // Tell caller we found one or more errors.
1156       ReturnValue = true;
1157       // Start the error message.
1158       OS << *MacroExpTracker.InstanceSourceLine;
1159       if (ItemKey.Column > 0)
1160         OS << std::string(ItemKey.Column - 1, ' ') << "^\n";
1161       OS << "error: Macro instance '" << *MacroExpTracker.MacroUnexpanded
1162          << "' has different values in this header, depending on how it was "
1163             "included.\n";
1164       // Walk all the instances.
1165       for (std::vector<MacroExpansionInstance>::iterator
1166                IMT = MacroExpTracker.MacroExpansionInstances.begin(),
1167                EMT = MacroExpTracker.MacroExpansionInstances.end();
1168            IMT != EMT; ++IMT) {
1169         MacroExpansionInstance &MacroInfo = *IMT;
1170         OS << "  '" << *MacroExpTracker.MacroUnexpanded << "' expanded to: '"
1171            << *MacroInfo.MacroExpanded
1172            << "' with respect to these inclusion paths:\n";
1173         // Walk all the inclusion path hierarchies.
1174         for (std::vector<InclusionPathHandle>::iterator
1175                  IIP = MacroInfo.InclusionPathHandles.begin(),
1176                  EIP = MacroInfo.InclusionPathHandles.end();
1177              IIP != EIP; ++IIP) {
1178           const std::vector<HeaderHandle> &ip = getInclusionPath(*IIP);
1179           int Count = (int)ip.size();
1180           for (int Index = 0; Index < Count; ++Index) {
1181             HeaderHandle H = ip[Index];
1182             OS << std::string((Index * 2) + 4, ' ') << *getHeaderFilePath(H)
1183                << "\n";
1184           }
1185         }
1186         // For a macro that wasn't defined, we flag it by using the
1187         // instance location.
1188         // If there is a definition...
1189         if (MacroInfo.DefinitionLocation.Line != ItemKey.Line) {
1190           OS << *MacroInfo.DefinitionSourceLine;
1191           if (MacroInfo.DefinitionLocation.Column > 0)
1192             OS << std::string(MacroInfo.DefinitionLocation.Column - 1, ' ')
1193                << "^\n";
1194           OS << "Macro defined here.\n";
1195         } else
1196           OS << "(no macro definition)"
1197              << "\n";
1198       }
1199     }
1200     return ReturnValue;
1201   }
1202 
1203   // Report on inconsistent conditional instances.
1204   // Returns true if any mismatches.
1205   bool reportInconsistentConditionals(llvm::raw_ostream &OS) override {
1206     bool ReturnValue = false;
1207     // Walk all the conditional trackers in the map.
1208     for (ConditionalExpansionMapIter I = ConditionalExpansions.begin(),
1209                                      E = ConditionalExpansions.end();
1210          I != E; ++I) {
1211       const PPItemKey &ItemKey = I->first;
1212       ConditionalTracker &CondTracker = I->second;
1213       if (!CondTracker.hasMismatch())
1214         continue;
1215       // Tell caller we found one or more errors.
1216       ReturnValue = true;
1217       // Start the error message.
1218       OS << *HeaderPaths[ItemKey.File] << ":" << ItemKey.Line << ":"
1219          << ItemKey.Column << "\n";
1220       OS << "#" << getDirectiveSpelling(CondTracker.DirectiveKind) << " "
1221          << *CondTracker.ConditionUnexpanded << "\n";
1222       OS << "^\n";
1223       OS << "error: Conditional expression instance '"
1224          << *CondTracker.ConditionUnexpanded
1225          << "' has different values in this header, depending on how it was "
1226             "included.\n";
1227       // Walk all the instances.
1228       for (std::vector<ConditionalExpansionInstance>::iterator
1229                IMT = CondTracker.ConditionalExpansionInstances.begin(),
1230                EMT = CondTracker.ConditionalExpansionInstances.end();
1231            IMT != EMT; ++IMT) {
1232         ConditionalExpansionInstance &MacroInfo = *IMT;
1233         OS << "  '" << *CondTracker.ConditionUnexpanded << "' expanded to: '"
1234            << ConditionValueKindStrings[MacroInfo.ConditionValue]
1235            << "' with respect to these inclusion paths:\n";
1236         // Walk all the inclusion path hierarchies.
1237         for (std::vector<InclusionPathHandle>::iterator
1238                  IIP = MacroInfo.InclusionPathHandles.begin(),
1239                  EIP = MacroInfo.InclusionPathHandles.end();
1240              IIP != EIP; ++IIP) {
1241           const std::vector<HeaderHandle> &ip = getInclusionPath(*IIP);
1242           int Count = (int)ip.size();
1243           for (int Index = 0; Index < Count; ++Index) {
1244             HeaderHandle H = ip[Index];
1245             OS << std::string((Index * 2) + 4, ' ') << *getHeaderFilePath(H)
1246                << "\n";
1247           }
1248         }
1249       }
1250     }
1251     return ReturnValue;
1252   }
1253 
1254   // Get directive spelling.
1255   static const char *getDirectiveSpelling(clang::tok::PPKeywordKind kind) {
1256     switch (kind) {
1257     case clang::tok::pp_if:
1258       return "if";
1259     case clang::tok::pp_elif:
1260       return "elif";
1261     case clang::tok::pp_ifdef:
1262       return "ifdef";
1263     case clang::tok::pp_ifndef:
1264       return "ifndef";
1265     default:
1266       return "(unknown)";
1267     }
1268   }
1269 
1270 private:
1271   llvm::SmallVector<std::string, 32> HeaderList;
1272   // Only do extern, namespace check for headers in HeaderList.
1273   bool BlockCheckHeaderListOnly;
1274   llvm::StringPool Strings;
1275   std::vector<StringHandle> HeaderPaths;
1276   std::vector<HeaderHandle> HeaderStack;
1277   std::vector<HeaderInclusionPath> InclusionPaths;
1278   InclusionPathHandle CurrentInclusionPathHandle;
1279   llvm::SmallSet<HeaderHandle, 128> HeadersInThisCompile;
1280   std::vector<PPItemKey> IncludeDirectives;
1281   MacroExpansionMap MacroExpansions;
1282   ConditionalExpansionMap ConditionalExpansions;
1283   bool InNestedHeader;
1284 };
1285 
1286 } // namespace
1287 
1288 // PreprocessorTracker functions.
1289 
1290 // PreprocessorTracker desctructor.
1291 PreprocessorTracker::~PreprocessorTracker() {}
1292 
1293 // Create instance of PreprocessorTracker.
1294 PreprocessorTracker *PreprocessorTracker::create(
1295     llvm::SmallVector<std::string, 32> &Headers,
1296     bool DoBlockCheckHeaderListOnly) {
1297   return new PreprocessorTrackerImpl(Headers, DoBlockCheckHeaderListOnly);
1298 }
1299 
1300 // Preprocessor callbacks for modularize.
1301 
1302 // Handle include directive.
1303 void PreprocessorCallbacks::InclusionDirective(
1304     clang::SourceLocation HashLoc, const clang::Token &IncludeTok,
1305     llvm::StringRef FileName, bool IsAngled,
1306     clang::CharSourceRange FilenameRange, const clang::FileEntry *File,
1307     llvm::StringRef SearchPath, llvm::StringRef RelativePath,
1308     const clang::Module *Imported) {
1309   int DirectiveLine, DirectiveColumn;
1310   std::string HeaderPath = getSourceLocationFile(PP, HashLoc);
1311   getSourceLocationLineAndColumn(PP, HashLoc, DirectiveLine, DirectiveColumn);
1312   PPTracker.handleIncludeDirective(HeaderPath, DirectiveLine, DirectiveColumn,
1313                                    FileName);
1314 }
1315 
1316 // Handle file entry/exit.
1317 void PreprocessorCallbacks::FileChanged(
1318     clang::SourceLocation Loc, clang::PPCallbacks::FileChangeReason Reason,
1319     clang::SrcMgr::CharacteristicKind FileType, clang::FileID PrevFID) {
1320   switch (Reason) {
1321   case EnterFile:
1322     PPTracker.handleHeaderEntry(PP, getSourceLocationFile(PP, Loc));
1323     break;
1324   case ExitFile: {
1325     const clang::FileEntry *F =
1326         PP.getSourceManager().getFileEntryForID(PrevFID);
1327     if (F)
1328       PPTracker.handleHeaderExit(F->getName());
1329   } break;
1330   case SystemHeaderPragma:
1331   case RenameFile:
1332     break;
1333   }
1334 }
1335 
1336 // Handle macro expansion.
1337 void PreprocessorCallbacks::MacroExpands(const clang::Token &MacroNameTok,
1338                                          const clang::MacroDefinition &MD,
1339                                          clang::SourceRange Range,
1340                                          const clang::MacroArgs *Args) {
1341   clang::SourceLocation Loc = Range.getBegin();
1342   // Ignore macro argument expansions.
1343   if (!Loc.isFileID())
1344     return;
1345   clang::IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
1346   const clang::MacroInfo *MI = MD.getMacroInfo();
1347   std::string MacroName = II->getName().str();
1348   std::string Unexpanded(getMacroUnexpandedString(Range, PP, MacroName, MI));
1349   std::string Expanded(getMacroExpandedString(PP, MacroName, MI, Args));
1350   PPTracker.addMacroExpansionInstance(
1351       PP, PPTracker.getCurrentHeaderHandle(), Loc, MI->getDefinitionLoc(), II,
1352       Unexpanded, Expanded, PPTracker.getCurrentInclusionPathHandle());
1353 }
1354 
1355 void PreprocessorCallbacks::Defined(const clang::Token &MacroNameTok,
1356                                     const clang::MacroDefinition &MD,
1357                                     clang::SourceRange Range) {
1358   clang::SourceLocation Loc(Range.getBegin());
1359   clang::IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
1360   const clang::MacroInfo *MI = MD.getMacroInfo();
1361   std::string MacroName = II->getName().str();
1362   std::string Unexpanded(getSourceString(PP, Range));
1363   PPTracker.addMacroExpansionInstance(
1364       PP, PPTracker.getCurrentHeaderHandle(), Loc,
1365       (MI ? MI->getDefinitionLoc() : Loc), II, Unexpanded,
1366       (MI ? "true" : "false"), PPTracker.getCurrentInclusionPathHandle());
1367 }
1368 
1369 void PreprocessorCallbacks::If(clang::SourceLocation Loc,
1370                                clang::SourceRange ConditionRange,
1371                                clang::PPCallbacks::ConditionValueKind ConditionResult) {
1372   std::string Unexpanded(getSourceString(PP, ConditionRange));
1373   PPTracker.addConditionalExpansionInstance(
1374       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_if,
1375       ConditionResult, Unexpanded, PPTracker.getCurrentInclusionPathHandle());
1376 }
1377 
1378 void PreprocessorCallbacks::Elif(clang::SourceLocation Loc,
1379                                  clang::SourceRange ConditionRange,
1380                                  clang::PPCallbacks::ConditionValueKind ConditionResult,
1381                                  clang::SourceLocation IfLoc) {
1382   std::string Unexpanded(getSourceString(PP, ConditionRange));
1383   PPTracker.addConditionalExpansionInstance(
1384       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_elif,
1385       ConditionResult, Unexpanded, PPTracker.getCurrentInclusionPathHandle());
1386 }
1387 
1388 void PreprocessorCallbacks::Ifdef(clang::SourceLocation Loc,
1389                                   const clang::Token &MacroNameTok,
1390                                   const clang::MacroDefinition &MD) {
1391   clang::PPCallbacks::ConditionValueKind IsDefined =
1392     (MD ? clang::PPCallbacks::CVK_True : clang::PPCallbacks::CVK_False );
1393   PPTracker.addConditionalExpansionInstance(
1394       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_ifdef,
1395       IsDefined, PP.getSpelling(MacroNameTok),
1396       PPTracker.getCurrentInclusionPathHandle());
1397 }
1398 
1399 void PreprocessorCallbacks::Ifndef(clang::SourceLocation Loc,
1400                                    const clang::Token &MacroNameTok,
1401                                    const clang::MacroDefinition &MD) {
1402   clang::PPCallbacks::ConditionValueKind IsNotDefined =
1403     (!MD ? clang::PPCallbacks::CVK_True : clang::PPCallbacks::CVK_False );
1404   PPTracker.addConditionalExpansionInstance(
1405       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_ifndef,
1406       IsNotDefined, PP.getSpelling(MacroNameTok),
1407       PPTracker.getCurrentInclusionPathHandle());
1408 }
1409 } // end namespace Modularize
1410