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 
255 namespace Modularize {
256 
257 // Forwards.
258 class PreprocessorTrackerImpl;
259 
260 // Some handle types
261 typedef llvm::PooledStringPtr StringHandle;
262 
263 typedef int HeaderHandle;
264 const HeaderHandle HeaderHandleInvalid = -1;
265 
266 typedef int InclusionPathHandle;
267 const InclusionPathHandle InclusionPathHandleInvalid = -1;
268 
269 // Some utility functions.
270 
271 // Get a "file:line:column" source location string.
272 static std::string getSourceLocationString(clang::Preprocessor &PP,
273                                            clang::SourceLocation Loc) {
274   if (Loc.isInvalid())
275     return std::string("(none)");
276   else
277     return Loc.printToString(PP.getSourceManager());
278 }
279 
280 // Get just the file name from a source location.
281 static std::string getSourceLocationFile(clang::Preprocessor &PP,
282                                          clang::SourceLocation Loc) {
283   std::string Source(getSourceLocationString(PP, Loc));
284   size_t Offset = Source.find(':', 2);
285   if (Offset == std::string::npos)
286     return Source;
287   return Source.substr(0, Offset);
288 }
289 
290 // Get just the line and column from a source location.
291 static void getSourceLocationLineAndColumn(clang::Preprocessor &PP,
292                                            clang::SourceLocation Loc, int &Line,
293                                            int &Column) {
294   clang::PresumedLoc PLoc = PP.getSourceManager().getPresumedLoc(Loc);
295   if (PLoc.isInvalid()) {
296     Line = 0;
297     Column = 0;
298     return;
299   }
300   Line = PLoc.getLine();
301   Column = PLoc.getColumn();
302 }
303 
304 // Retrieve source snippet from file image.
305 std::string getSourceString(clang::Preprocessor &PP, clang::SourceRange Range) {
306   clang::SourceLocation BeginLoc = Range.getBegin();
307   clang::SourceLocation EndLoc = Range.getEnd();
308   const char *BeginPtr = PP.getSourceManager().getCharacterData(BeginLoc);
309   const char *EndPtr = PP.getSourceManager().getCharacterData(EndLoc);
310   size_t Length = EndPtr - BeginPtr;
311   return llvm::StringRef(BeginPtr, Length).trim().str();
312 }
313 
314 // Retrieve source line from file image given a location.
315 std::string getSourceLine(clang::Preprocessor &PP, 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 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 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 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   typedef clang::MacroInfo::tokens_iterator Iter;
409   for (Iter I = MI->tokens_begin(), E = MI->tokens_end(); I != E; ++I) {
410     clang::IdentifierInfo *II = I->getIdentifierInfo();
411     int ArgNo = (II && Args ? MI->getArgumentNum(II) : -1);
412     if (ArgNo == -1) {
413       // This isn't an argument, just add it.
414       if (II == nullptr)
415         Expanded += PP.getSpelling((*I)); // Not an identifier.
416       else {
417         // Token is for an identifier.
418         std::string Name = II->getName().str();
419         // Check for nexted macro references.
420         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
421         if (MacroInfo)
422           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
423         else
424           Expanded += Name;
425       }
426       continue;
427     }
428     // We get here if it's a function-style macro with arguments.
429     const clang::Token *ResultArgToks;
430     const clang::Token *ArgTok = Args->getUnexpArgument(ArgNo);
431     if (Args->ArgNeedsPreexpansion(ArgTok, PP))
432       ResultArgToks = &(const_cast<clang::MacroArgs *>(Args))
433           ->getPreExpArgument(ArgNo, MI, PP)[0];
434     else
435       ResultArgToks = ArgTok; // Use non-preexpanded Tokens.
436     // If the arg token didn't expand into anything, ignore it.
437     if (ResultArgToks->is(clang::tok::eof))
438       continue;
439     unsigned NumToks = clang::MacroArgs::getArgLength(ResultArgToks);
440     // Append the resulting argument expansions.
441     for (unsigned ArgumentIndex = 0; ArgumentIndex < NumToks; ++ArgumentIndex) {
442       const clang::Token &AT = ResultArgToks[ArgumentIndex];
443       clang::IdentifierInfo *II = AT.getIdentifierInfo();
444       if (II == nullptr)
445         Expanded += PP.getSpelling(AT); // Not an identifier.
446       else {
447         // It's an identifier.  Check for further expansion.
448         std::string Name = II->getName().str();
449         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
450         if (MacroInfo)
451           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
452         else
453           Expanded += Name;
454       }
455     }
456   }
457   return Expanded;
458 }
459 
460 // Get the string representing a vector of Tokens.
461 std::string
462 getTokensSpellingString(clang::Preprocessor &PP,
463                         llvm::SmallVectorImpl<clang::Token> &Tokens) {
464   std::string Expanded;
465   // Walk over the macro Tokens.
466   typedef llvm::SmallVectorImpl<clang::Token>::iterator Iter;
467   for (Iter I = Tokens.begin(), E = Tokens.end(); I != E; ++I)
468     Expanded += PP.getSpelling(*I); // Not an identifier.
469   return llvm::StringRef(Expanded).trim().str();
470 }
471 
472 // Get the expansion for a macro instance, given the information
473 // provided by PPCallbacks.
474 std::string getExpandedString(clang::Preprocessor &PP,
475                               llvm::StringRef MacroName,
476                               const clang::MacroInfo *MI,
477                               const clang::MacroArgs *Args) {
478   std::string Expanded;
479   // Walk over the macro Tokens.
480   typedef clang::MacroInfo::tokens_iterator Iter;
481   for (Iter I = MI->tokens_begin(), E = MI->tokens_end(); I != E; ++I) {
482     clang::IdentifierInfo *II = I->getIdentifierInfo();
483     int ArgNo = (II && Args ? MI->getArgumentNum(II) : -1);
484     if (ArgNo == -1) {
485       // This isn't an argument, just add it.
486       if (II == nullptr)
487         Expanded += PP.getSpelling((*I)); // Not an identifier.
488       else {
489         // Token is for an identifier.
490         std::string Name = II->getName().str();
491         // Check for nexted macro references.
492         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
493         if (MacroInfo)
494           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
495         else
496           Expanded += Name;
497       }
498       continue;
499     }
500     // We get here if it's a function-style macro with arguments.
501     const clang::Token *ResultArgToks;
502     const clang::Token *ArgTok = Args->getUnexpArgument(ArgNo);
503     if (Args->ArgNeedsPreexpansion(ArgTok, PP))
504       ResultArgToks = &(const_cast<clang::MacroArgs *>(Args))
505           ->getPreExpArgument(ArgNo, MI, PP)[0];
506     else
507       ResultArgToks = ArgTok; // Use non-preexpanded Tokens.
508     // If the arg token didn't expand into anything, ignore it.
509     if (ResultArgToks->is(clang::tok::eof))
510       continue;
511     unsigned NumToks = clang::MacroArgs::getArgLength(ResultArgToks);
512     // Append the resulting argument expansions.
513     for (unsigned ArgumentIndex = 0; ArgumentIndex < NumToks; ++ArgumentIndex) {
514       const clang::Token &AT = ResultArgToks[ArgumentIndex];
515       clang::IdentifierInfo *II = AT.getIdentifierInfo();
516       if (II == nullptr)
517         Expanded += PP.getSpelling(AT); // Not an identifier.
518       else {
519         // It's an identifier.  Check for further expansion.
520         std::string Name = II->getName().str();
521         clang::MacroInfo *MacroInfo = PP.getMacroInfo(II);
522         if (MacroInfo)
523           Expanded += getMacroExpandedString(PP, Name, MacroInfo, nullptr);
524         else
525           Expanded += Name;
526       }
527     }
528   }
529   return Expanded;
530 }
531 
532 // ConditionValueKind strings.
533 const char *
534 ConditionValueKindStrings[] = {
535   "(not evaluated)", "false", "true"
536 };
537 
538 bool operator<(const StringHandle &H1, const StringHandle &H2) {
539   const char *S1 = (H1 ? *H1 : "");
540   const char *S2 = (H2 ? *H2 : "");
541   int Diff = strcmp(S1, S2);
542   return Diff < 0;
543 }
544 bool operator>(const StringHandle &H1, const StringHandle &H2) {
545   const char *S1 = (H1 ? *H1 : "");
546   const char *S2 = (H2 ? *H2 : "");
547   int Diff = strcmp(S1, S2);
548   return Diff > 0;
549 }
550 
551 // Preprocessor item key.
552 //
553 // This class represents a location in a source file, for use
554 // as a key representing a unique name/file/line/column quadruplet,
555 // which in this case is used to identify a macro expansion instance,
556 // but could be used for other things as well.
557 // The file is a header file handle, the line is a line number,
558 // and the column is a column number.
559 class PPItemKey {
560 public:
561   PPItemKey(clang::Preprocessor &PP, StringHandle Name, HeaderHandle File,
562             clang::SourceLocation Loc)
563       : Name(Name), File(File) {
564     getSourceLocationLineAndColumn(PP, Loc, Line, Column);
565   }
566   PPItemKey(StringHandle Name, HeaderHandle File, int Line, int Column)
567       : Name(Name), File(File), Line(Line), Column(Column) {}
568   PPItemKey(const PPItemKey &Other)
569       : Name(Other.Name), File(Other.File), Line(Other.Line),
570         Column(Other.Column) {}
571   PPItemKey() : File(HeaderHandleInvalid), Line(0), Column(0) {}
572   bool operator==(const PPItemKey &Other) const {
573     if (Name != Other.Name)
574       return false;
575     if (File != Other.File)
576       return false;
577     if (Line != Other.Line)
578       return false;
579     return Column == Other.Column;
580   }
581   bool operator<(const PPItemKey &Other) const {
582     if (Name < Other.Name)
583       return true;
584     else if (Name > Other.Name)
585       return false;
586     if (File < Other.File)
587       return true;
588     else if (File > Other.File)
589       return false;
590     if (Line < Other.Line)
591       return true;
592     else if (Line > Other.Line)
593       return false;
594     return Column < Other.Column;
595   }
596   StringHandle Name;
597   HeaderHandle File;
598   int Line;
599   int Column;
600 };
601 
602 // Header inclusion path.
603 class HeaderInclusionPath {
604 public:
605   HeaderInclusionPath(std::vector<HeaderHandle> HeaderInclusionPath)
606       : Path(HeaderInclusionPath) {}
607   HeaderInclusionPath(const HeaderInclusionPath &Other) : Path(Other.Path) {}
608   HeaderInclusionPath() {}
609   std::vector<HeaderHandle> Path;
610 };
611 
612 // Macro expansion instance.
613 //
614 // This class represents an instance of a macro expansion with a
615 // unique value.  It also stores the unique header inclusion paths
616 // for use in telling the user the nested include path to the header.
617 class MacroExpansionInstance {
618 public:
619   MacroExpansionInstance(StringHandle MacroExpanded,
620                          PPItemKey &DefinitionLocation,
621                          StringHandle DefinitionSourceLine,
622                          InclusionPathHandle H)
623       : MacroExpanded(MacroExpanded), DefinitionLocation(DefinitionLocation),
624         DefinitionSourceLine(DefinitionSourceLine) {
625     InclusionPathHandles.push_back(H);
626   }
627   MacroExpansionInstance() {}
628 
629   // Check for the presence of a header inclusion path handle entry.
630   // Return false if not found.
631   bool haveInclusionPathHandle(InclusionPathHandle H) {
632     for (std::vector<InclusionPathHandle>::iterator
633              I = InclusionPathHandles.begin(),
634              E = InclusionPathHandles.end();
635          I != E; ++I) {
636       if (*I == H)
637         return true;
638     }
639     return InclusionPathHandleInvalid;
640   }
641   // Add a new header inclusion path entry, if not already present.
642   void addInclusionPathHandle(InclusionPathHandle H) {
643     if (!haveInclusionPathHandle(H))
644       InclusionPathHandles.push_back(H);
645   }
646 
647   // A string representing the macro instance after preprocessing.
648   StringHandle MacroExpanded;
649   // A file/line/column triplet representing the macro definition location.
650   PPItemKey DefinitionLocation;
651   // A place to save the macro definition line string.
652   StringHandle DefinitionSourceLine;
653   // The header inclusion path handles for all the instances.
654   std::vector<InclusionPathHandle> InclusionPathHandles;
655 };
656 
657 // Macro expansion instance tracker.
658 //
659 // This class represents one macro expansion, keyed by a PPItemKey.
660 // It stores a string representing the macro reference in the source,
661 // and a list of ConditionalExpansionInstances objects representing
662 // the unique values the condition expands to in instances of the header.
663 class MacroExpansionTracker {
664 public:
665   MacroExpansionTracker(StringHandle MacroUnexpanded,
666                         StringHandle MacroExpanded,
667                         StringHandle InstanceSourceLine,
668                         PPItemKey &DefinitionLocation,
669                         StringHandle DefinitionSourceLine,
670                         InclusionPathHandle InclusionPathHandle)
671       : MacroUnexpanded(MacroUnexpanded),
672         InstanceSourceLine(InstanceSourceLine) {
673     addMacroExpansionInstance(MacroExpanded, DefinitionLocation,
674                               DefinitionSourceLine, InclusionPathHandle);
675   }
676   MacroExpansionTracker() {}
677 
678   // Find a matching macro expansion instance.
679   MacroExpansionInstance *
680   findMacroExpansionInstance(StringHandle MacroExpanded,
681                              PPItemKey &DefinitionLocation) {
682     for (std::vector<MacroExpansionInstance>::iterator
683              I = MacroExpansionInstances.begin(),
684              E = MacroExpansionInstances.end();
685          I != E; ++I) {
686       if ((I->MacroExpanded == MacroExpanded) &&
687           (I->DefinitionLocation == DefinitionLocation)) {
688         return &*I; // Found.
689       }
690     }
691     return nullptr; // Not found.
692   }
693 
694   // Add a macro expansion instance.
695   void addMacroExpansionInstance(StringHandle MacroExpanded,
696                                  PPItemKey &DefinitionLocation,
697                                  StringHandle DefinitionSourceLine,
698                                  InclusionPathHandle InclusionPathHandle) {
699     MacroExpansionInstances.push_back(
700         MacroExpansionInstance(MacroExpanded, DefinitionLocation,
701                                DefinitionSourceLine, InclusionPathHandle));
702   }
703 
704   // Return true if there is a mismatch.
705   bool hasMismatch() { return MacroExpansionInstances.size() > 1; }
706 
707   // A string representing the macro instance without expansion.
708   StringHandle MacroUnexpanded;
709   // A place to save the macro instance source line string.
710   StringHandle InstanceSourceLine;
711   // The macro expansion instances.
712   // If all instances of the macro expansion expand to the same value,
713   // This vector will only have one instance.
714   std::vector<MacroExpansionInstance> MacroExpansionInstances;
715 };
716 
717 // Conditional expansion instance.
718 //
719 // This class represents an instance of a condition exoression result
720 // with a unique value.  It also stores the unique header inclusion paths
721 // for use in telling the user the nested include path to the header.
722 class ConditionalExpansionInstance {
723 public:
724   ConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue, InclusionPathHandle H)
725       : ConditionValue(ConditionValue) {
726     InclusionPathHandles.push_back(H);
727   }
728   ConditionalExpansionInstance() {}
729 
730   // Check for the presence of a header inclusion path handle entry.
731   // Return false if not found.
732   bool haveInclusionPathHandle(InclusionPathHandle H) {
733     for (std::vector<InclusionPathHandle>::iterator
734              I = InclusionPathHandles.begin(),
735              E = InclusionPathHandles.end();
736          I != E; ++I) {
737       if (*I == H)
738         return true;
739     }
740     return InclusionPathHandleInvalid;
741   }
742   // Add a new header inclusion path entry, if not already present.
743   void addInclusionPathHandle(InclusionPathHandle H) {
744     if (!haveInclusionPathHandle(H))
745       InclusionPathHandles.push_back(H);
746   }
747 
748   // A flag representing the evaluated condition value.
749   clang::PPCallbacks::ConditionValueKind ConditionValue;
750   // The header inclusion path handles for all the instances.
751   std::vector<InclusionPathHandle> InclusionPathHandles;
752 };
753 
754 // Conditional directive instance tracker.
755 //
756 // This class represents one conditional directive, keyed by a PPItemKey.
757 // It stores a string representing the macro reference in the source,
758 // and a list of ConditionExpansionInstance objects representing
759 // the unique value the condition expression expands to in instances of
760 // the header.
761 class ConditionalTracker {
762 public:
763   ConditionalTracker(clang::tok::PPKeywordKind DirectiveKind,
764                      clang::PPCallbacks::ConditionValueKind ConditionValue,
765                      StringHandle ConditionUnexpanded,
766                      InclusionPathHandle InclusionPathHandle)
767       : DirectiveKind(DirectiveKind), ConditionUnexpanded(ConditionUnexpanded) {
768     addConditionalExpansionInstance(ConditionValue, InclusionPathHandle);
769   }
770   ConditionalTracker() {}
771 
772   // Find a matching condition expansion instance.
773   ConditionalExpansionInstance *
774   findConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue) {
775     for (std::vector<ConditionalExpansionInstance>::iterator
776              I = ConditionalExpansionInstances.begin(),
777              E = ConditionalExpansionInstances.end();
778          I != E; ++I) {
779       if (I->ConditionValue == ConditionValue) {
780         return &*I; // Found.
781       }
782     }
783     return nullptr; // Not found.
784   }
785 
786   // Add a conditional expansion instance.
787   void
788   addConditionalExpansionInstance(clang::PPCallbacks::ConditionValueKind ConditionValue,
789                                   InclusionPathHandle InclusionPathHandle) {
790     ConditionalExpansionInstances.push_back(
791         ConditionalExpansionInstance(ConditionValue, InclusionPathHandle));
792   }
793 
794   // Return true if there is a mismatch.
795   bool hasMismatch() { return ConditionalExpansionInstances.size() > 1; }
796 
797   // The kind of directive.
798   clang::tok::PPKeywordKind DirectiveKind;
799   // A string representing the macro instance without expansion.
800   StringHandle ConditionUnexpanded;
801   // The condition expansion instances.
802   // If all instances of the conditional expression expand to the same value,
803   // This vector will only have one instance.
804   std::vector<ConditionalExpansionInstance> ConditionalExpansionInstances;
805 };
806 
807 // Preprocessor callbacks for modularize.
808 //
809 // This class derives from the Clang PPCallbacks class to track preprocessor
810 // actions, such as changing files and handling preprocessor directives and
811 // macro expansions.  It has to figure out when a new header file is entered
812 // and left, as the provided handler is not particularly clear about it.
813 class PreprocessorCallbacks : public clang::PPCallbacks {
814 public:
815   PreprocessorCallbacks(PreprocessorTrackerImpl &ppTracker,
816                         clang::Preprocessor &PP, llvm::StringRef rootHeaderFile)
817       : PPTracker(ppTracker), PP(PP), RootHeaderFile(rootHeaderFile) {}
818   ~PreprocessorCallbacks() {}
819 
820   // Overridden handlers.
821   void InclusionDirective(clang::SourceLocation HashLoc,
822                           const clang::Token &IncludeTok,
823                           llvm::StringRef FileName, bool IsAngled,
824                           clang::CharSourceRange FilenameRange,
825                           const clang::FileEntry *File,
826                           llvm::StringRef SearchPath,
827                           llvm::StringRef RelativePath,
828                           const clang::Module *Imported);
829   void FileChanged(clang::SourceLocation Loc,
830                    clang::PPCallbacks::FileChangeReason Reason,
831                    clang::SrcMgr::CharacteristicKind FileType,
832                    clang::FileID PrevFID = clang::FileID());
833   void MacroExpands(const clang::Token &MacroNameTok,
834                     const clang::MacroDirective *MD, clang::SourceRange Range,
835                     const clang::MacroArgs *Args);
836   void Defined(const clang::Token &MacroNameTok,
837                const clang::MacroDirective *MD, clang::SourceRange Range);
838   void If(clang::SourceLocation Loc, clang::SourceRange ConditionRange,
839           clang::PPCallbacks::ConditionValueKind ConditionResult);
840   void Elif(clang::SourceLocation Loc, clang::SourceRange ConditionRange,
841             clang::PPCallbacks::ConditionValueKind ConditionResult, clang::SourceLocation IfLoc);
842   void Ifdef(clang::SourceLocation Loc, const clang::Token &MacroNameTok,
843              const clang::MacroDirective *MD);
844   void Ifndef(clang::SourceLocation Loc, const clang::Token &MacroNameTok,
845               const clang::MacroDirective *MD);
846 
847 private:
848   PreprocessorTrackerImpl &PPTracker;
849   clang::Preprocessor &PP;
850   std::string RootHeaderFile;
851 };
852 
853 // Preprocessor macro expansion item map types.
854 typedef std::map<PPItemKey, MacroExpansionTracker> MacroExpansionMap;
855 typedef std::map<PPItemKey, MacroExpansionTracker>::iterator
856 MacroExpansionMapIter;
857 
858 // Preprocessor conditional expansion item map types.
859 typedef std::map<PPItemKey, ConditionalTracker> ConditionalExpansionMap;
860 typedef std::map<PPItemKey, ConditionalTracker>::iterator
861 ConditionalExpansionMapIter;
862 
863 // Preprocessor tracker for modularize.
864 //
865 // This class stores information about all the headers processed in the
866 // course of running modularize.
867 class PreprocessorTrackerImpl : public PreprocessorTracker {
868 public:
869   PreprocessorTrackerImpl(llvm::SmallVector<std::string, 32> &Headers,
870         bool DoBlockCheckHeaderListOnly)
871       : BlockCheckHeaderListOnly(DoBlockCheckHeaderListOnly),
872         CurrentInclusionPathHandle(InclusionPathHandleInvalid),
873         InNestedHeader(false) {
874     // Use canonical header path representation.
875     for (llvm::ArrayRef<std::string>::iterator I = Headers.begin(),
876       E = Headers.end();
877       I != E; ++I) {
878       HeaderList.push_back(getCanonicalPath(*I));
879     }
880   }
881 
882   ~PreprocessorTrackerImpl() {}
883 
884   // Handle entering a preprocessing session.
885   void handlePreprocessorEntry(clang::Preprocessor &PP,
886                                llvm::StringRef rootHeaderFile) {
887     HeadersInThisCompile.clear();
888     assert((HeaderStack.size() == 0) && "Header stack should be empty.");
889     pushHeaderHandle(addHeader(rootHeaderFile));
890     PP.addPPCallbacks(llvm::make_unique<PreprocessorCallbacks>(*this, PP,
891                                                                rootHeaderFile));
892   }
893   // Handle exiting a preprocessing session.
894   void handlePreprocessorExit() { HeaderStack.clear(); }
895 
896   // Handle include directive.
897   // This function is called every time an include directive is seen by the
898   // preprocessor, for the purpose of later checking for 'extern "" {}' or
899   // "namespace {}" blocks containing #include directives.
900   void handleIncludeDirective(llvm::StringRef DirectivePath, int DirectiveLine,
901                               int DirectiveColumn, llvm::StringRef TargetPath) {
902     // If it's not a header in the header list, ignore it with respect to
903     // the check.
904     if (BlockCheckHeaderListOnly && !isHeaderListHeader(TargetPath))
905       return;
906     HeaderHandle CurrentHeaderHandle = findHeaderHandle(DirectivePath);
907     StringHandle IncludeHeaderHandle = addString(TargetPath);
908     for (std::vector<PPItemKey>::const_iterator I = IncludeDirectives.begin(),
909                                                 E = IncludeDirectives.end();
910          I != E; ++I) {
911       // If we already have an entry for this directive, return now.
912       if ((I->File == CurrentHeaderHandle) && (I->Line == DirectiveLine))
913         return;
914     }
915     PPItemKey IncludeDirectiveItem(IncludeHeaderHandle, CurrentHeaderHandle,
916                                    DirectiveLine, DirectiveColumn);
917     IncludeDirectives.push_back(IncludeDirectiveItem);
918   }
919 
920   // Check for include directives within the given source line range.
921   // Report errors if any found.  Returns true if no include directives
922   // found in block.
923   bool checkForIncludesInBlock(clang::Preprocessor &PP,
924                                clang::SourceRange BlockSourceRange,
925                                const char *BlockIdentifierMessage,
926                                llvm::raw_ostream &OS) {
927     clang::SourceLocation BlockStartLoc = BlockSourceRange.getBegin();
928     clang::SourceLocation BlockEndLoc = BlockSourceRange.getEnd();
929     // Use block location to get FileID of both the include directive
930     // and block statement.
931     clang::FileID FileID = PP.getSourceManager().getFileID(BlockStartLoc);
932     std::string SourcePath = getSourceLocationFile(PP, BlockStartLoc);
933     HeaderHandle SourceHandle = findHeaderHandle(SourcePath);
934     // FIXME: Go back and fix getSourceLocation to use a canonical form.
935     if (SourceHandle == -1)
936       return true;
937     int BlockStartLine, BlockStartColumn, BlockEndLine, BlockEndColumn;
938     bool returnValue = true;
939     getSourceLocationLineAndColumn(PP, BlockStartLoc, BlockStartLine,
940                                    BlockStartColumn);
941     getSourceLocationLineAndColumn(PP, BlockEndLoc, BlockEndLine,
942                                    BlockEndColumn);
943     for (std::vector<PPItemKey>::const_iterator I = IncludeDirectives.begin(),
944                                                 E = IncludeDirectives.end();
945          I != E; ++I) {
946       // If we find an entry within the block, report an error.
947       if ((I->File == SourceHandle) && (I->Line >= BlockStartLine) &&
948           (I->Line < BlockEndLine)) {
949         returnValue = false;
950         OS << SourcePath << ":" << I->Line << ":" << I->Column << ":\n";
951         OS << getSourceLine(PP, FileID, I->Line) << "\n";
952         if (I->Column > 0)
953           OS << std::string(I->Column - 1, ' ') << "^\n";
954         OS << "error: Include directive within " << BlockIdentifierMessage
955            << ".\n";
956         OS << SourcePath << ":" << BlockStartLine << ":" << BlockStartColumn
957            << ":\n";
958         OS << getSourceLine(PP, BlockStartLoc) << "\n";
959         if (BlockStartColumn > 0)
960           OS << std::string(BlockStartColumn - 1, ' ') << "^\n";
961         OS << "The \"" << BlockIdentifierMessage << "\" block is here.\n";
962       }
963     }
964     return returnValue;
965   }
966 
967   // Handle entering a header source file.
968   void handleHeaderEntry(clang::Preprocessor &PP, llvm::StringRef HeaderPath) {
969     // Ignore <built-in> and <command-line> to reduce message clutter.
970     if (HeaderPath.startswith("<"))
971       return;
972     HeaderHandle H = addHeader(HeaderPath);
973     if (H != getCurrentHeaderHandle())
974       pushHeaderHandle(H);
975     // Check for nested header.
976     if (!InNestedHeader)
977       InNestedHeader = !HeadersInThisCompile.insert(H).second;
978   }
979 
980   // Handle exiting a header source file.
981   void handleHeaderExit(llvm::StringRef HeaderPath) {
982     // Ignore <built-in> and <command-line> to reduce message clutter.
983     if (HeaderPath.startswith("<"))
984       return;
985     HeaderHandle H = findHeaderHandle(HeaderPath);
986     HeaderHandle TH;
987     if (isHeaderHandleInStack(H)) {
988       do {
989         TH = getCurrentHeaderHandle();
990         popHeaderHandle();
991       } while ((TH != H) && (HeaderStack.size() != 0));
992     }
993     InNestedHeader = false;
994   }
995 
996   // Lookup/add string.
997   StringHandle addString(llvm::StringRef Str) { return Strings.intern(Str); }
998 
999   // Convert to a canonical path.
1000   std::string getCanonicalPath(llvm::StringRef path) const {
1001     std::string CanonicalPath(path);
1002     std::replace(CanonicalPath.begin(), CanonicalPath.end(), '\\', '/');
1003     return CanonicalPath;
1004   }
1005 
1006   // Return true if the given header is in the header list.
1007   bool isHeaderListHeader(llvm::StringRef HeaderPath) const {
1008     std::string CanonicalPath = getCanonicalPath(HeaderPath);
1009     for (llvm::ArrayRef<std::string>::iterator I = HeaderList.begin(),
1010         E = HeaderList.end();
1011         I != E; ++I) {
1012       if (*I == CanonicalPath)
1013         return true;
1014     }
1015     return false;
1016   }
1017 
1018   // Get the handle of a header file entry.
1019   // Return HeaderHandleInvalid if not found.
1020   HeaderHandle findHeaderHandle(llvm::StringRef HeaderPath) const {
1021     std::string CanonicalPath = getCanonicalPath(HeaderPath);
1022     HeaderHandle H = 0;
1023     for (std::vector<StringHandle>::const_iterator I = HeaderPaths.begin(),
1024                                                    E = HeaderPaths.end();
1025          I != E; ++I, ++H) {
1026       if (**I == CanonicalPath)
1027         return H;
1028     }
1029     return HeaderHandleInvalid;
1030   }
1031 
1032   // Add a new header file entry, or return existing handle.
1033   // Return the header handle.
1034   HeaderHandle addHeader(llvm::StringRef HeaderPath) {
1035     std::string CanonicalPath = getCanonicalPath(HeaderPath);
1036     HeaderHandle H = findHeaderHandle(CanonicalPath);
1037     if (H == HeaderHandleInvalid) {
1038       H = HeaderPaths.size();
1039       HeaderPaths.push_back(addString(CanonicalPath));
1040     }
1041     return H;
1042   }
1043 
1044   // Return a header file path string given its handle.
1045   StringHandle getHeaderFilePath(HeaderHandle H) const {
1046     if ((H >= 0) && (H < (HeaderHandle)HeaderPaths.size()))
1047       return HeaderPaths[H];
1048     return StringHandle();
1049   }
1050 
1051   // Returns a handle to the inclusion path.
1052   InclusionPathHandle pushHeaderHandle(HeaderHandle H) {
1053     HeaderStack.push_back(H);
1054     return CurrentInclusionPathHandle = addInclusionPathHandle(HeaderStack);
1055   }
1056   // Pops the last header handle from the stack;
1057   void popHeaderHandle() {
1058     // assert((HeaderStack.size() != 0) && "Header stack already empty.");
1059     if (HeaderStack.size() != 0) {
1060       HeaderStack.pop_back();
1061       CurrentInclusionPathHandle = addInclusionPathHandle(HeaderStack);
1062     }
1063   }
1064   // Get the top handle on the header stack.
1065   HeaderHandle getCurrentHeaderHandle() const {
1066     if (HeaderStack.size() != 0)
1067       return HeaderStack.back();
1068     return HeaderHandleInvalid;
1069   }
1070 
1071   // Check for presence of header handle in the header stack.
1072   bool isHeaderHandleInStack(HeaderHandle H) const {
1073     for (std::vector<HeaderHandle>::const_iterator I = HeaderStack.begin(),
1074                                                    E = HeaderStack.end();
1075          I != E; ++I) {
1076       if (*I == H)
1077         return true;
1078     }
1079     return false;
1080   }
1081 
1082   // Get the handle of a header inclusion path entry.
1083   // Return InclusionPathHandleInvalid if not found.
1084   InclusionPathHandle
1085   findInclusionPathHandle(const std::vector<HeaderHandle> &Path) const {
1086     InclusionPathHandle H = 0;
1087     for (std::vector<HeaderInclusionPath>::const_iterator
1088              I = InclusionPaths.begin(),
1089              E = InclusionPaths.end();
1090          I != E; ++I, ++H) {
1091       if (I->Path == Path)
1092         return H;
1093     }
1094     return HeaderHandleInvalid;
1095   }
1096   // Add a new header inclusion path entry, or return existing handle.
1097   // Return the header inclusion path entry handle.
1098   InclusionPathHandle
1099   addInclusionPathHandle(const std::vector<HeaderHandle> &Path) {
1100     InclusionPathHandle H = findInclusionPathHandle(Path);
1101     if (H == HeaderHandleInvalid) {
1102       H = InclusionPaths.size();
1103       InclusionPaths.push_back(HeaderInclusionPath(Path));
1104     }
1105     return H;
1106   }
1107   // Return the current inclusion path handle.
1108   InclusionPathHandle getCurrentInclusionPathHandle() const {
1109     return CurrentInclusionPathHandle;
1110   }
1111 
1112   // Return an inclusion path given its handle.
1113   const std::vector<HeaderHandle> &
1114   getInclusionPath(InclusionPathHandle H) const {
1115     if ((H >= 0) && (H <= (InclusionPathHandle)InclusionPaths.size()))
1116       return InclusionPaths[H].Path;
1117     static std::vector<HeaderHandle> Empty;
1118     return Empty;
1119   }
1120 
1121   // Add a macro expansion instance.
1122   void addMacroExpansionInstance(clang::Preprocessor &PP, HeaderHandle H,
1123                                  clang::SourceLocation InstanceLoc,
1124                                  clang::SourceLocation DefinitionLoc,
1125                                  clang::IdentifierInfo *II,
1126                                  llvm::StringRef MacroUnexpanded,
1127                                  llvm::StringRef MacroExpanded,
1128                                  InclusionPathHandle InclusionPathHandle) {
1129     if (InNestedHeader)
1130       return;
1131     StringHandle MacroName = addString(II->getName());
1132     PPItemKey InstanceKey(PP, MacroName, H, InstanceLoc);
1133     PPItemKey DefinitionKey(PP, MacroName, H, DefinitionLoc);
1134     MacroExpansionMapIter I = MacroExpansions.find(InstanceKey);
1135     // If existing instance of expansion not found, add one.
1136     if (I == MacroExpansions.end()) {
1137       std::string InstanceSourceLine =
1138           getSourceLocationString(PP, InstanceLoc) + ":\n" +
1139           getSourceLine(PP, InstanceLoc) + "\n";
1140       std::string DefinitionSourceLine =
1141           getSourceLocationString(PP, DefinitionLoc) + ":\n" +
1142           getSourceLine(PP, DefinitionLoc) + "\n";
1143       MacroExpansions[InstanceKey] = MacroExpansionTracker(
1144           addString(MacroUnexpanded), addString(MacroExpanded),
1145           addString(InstanceSourceLine), DefinitionKey,
1146           addString(DefinitionSourceLine), InclusionPathHandle);
1147     } else {
1148       // We've seen the macro before.  Get its tracker.
1149       MacroExpansionTracker &CondTracker = I->second;
1150       // Look up an existing instance value for the macro.
1151       MacroExpansionInstance *MacroInfo =
1152           CondTracker.findMacroExpansionInstance(addString(MacroExpanded),
1153                                                  DefinitionKey);
1154       // If found, just add the inclusion path to the instance.
1155       if (MacroInfo)
1156         MacroInfo->addInclusionPathHandle(InclusionPathHandle);
1157       else {
1158         // Otherwise add a new instance with the unique value.
1159         std::string DefinitionSourceLine =
1160             getSourceLocationString(PP, DefinitionLoc) + ":\n" +
1161             getSourceLine(PP, DefinitionLoc) + "\n";
1162         CondTracker.addMacroExpansionInstance(
1163             addString(MacroExpanded), DefinitionKey,
1164             addString(DefinitionSourceLine), InclusionPathHandle);
1165       }
1166     }
1167   }
1168 
1169   // Add a conditional expansion instance.
1170   void
1171   addConditionalExpansionInstance(clang::Preprocessor &PP, HeaderHandle H,
1172                                   clang::SourceLocation InstanceLoc,
1173                                   clang::tok::PPKeywordKind DirectiveKind,
1174                                   clang::PPCallbacks::ConditionValueKind ConditionValue,
1175                                   llvm::StringRef ConditionUnexpanded,
1176                                   InclusionPathHandle InclusionPathHandle) {
1177     // Ignore header guards, assuming the header guard is the only conditional.
1178     if (InNestedHeader)
1179       return;
1180     StringHandle ConditionUnexpandedHandle(addString(ConditionUnexpanded));
1181     PPItemKey InstanceKey(PP, ConditionUnexpandedHandle, H, InstanceLoc);
1182     ConditionalExpansionMapIter I = ConditionalExpansions.find(InstanceKey);
1183     // If existing instance of condition not found, add one.
1184     if (I == ConditionalExpansions.end()) {
1185       std::string InstanceSourceLine =
1186           getSourceLocationString(PP, InstanceLoc) + ":\n" +
1187           getSourceLine(PP, InstanceLoc) + "\n";
1188       ConditionalExpansions[InstanceKey] =
1189           ConditionalTracker(DirectiveKind, ConditionValue,
1190                              ConditionUnexpandedHandle, InclusionPathHandle);
1191     } else {
1192       // We've seen the conditional before.  Get its tracker.
1193       ConditionalTracker &CondTracker = I->second;
1194       // Look up an existing instance value for the condition.
1195       ConditionalExpansionInstance *MacroInfo =
1196           CondTracker.findConditionalExpansionInstance(ConditionValue);
1197       // If found, just add the inclusion path to the instance.
1198       if (MacroInfo)
1199         MacroInfo->addInclusionPathHandle(InclusionPathHandle);
1200       else {
1201         // Otherwise add a new instance with the unique value.
1202         CondTracker.addConditionalExpansionInstance(ConditionValue,
1203                                                     InclusionPathHandle);
1204       }
1205     }
1206   }
1207 
1208   // Report on inconsistent macro instances.
1209   // Returns true if any mismatches.
1210   bool reportInconsistentMacros(llvm::raw_ostream &OS) {
1211     bool ReturnValue = false;
1212     // Walk all the macro expansion trackers in the map.
1213     for (MacroExpansionMapIter I = MacroExpansions.begin(),
1214                                E = MacroExpansions.end();
1215          I != E; ++I) {
1216       const PPItemKey &ItemKey = I->first;
1217       MacroExpansionTracker &MacroExpTracker = I->second;
1218       // If no mismatch (only one instance value) continue.
1219       if (!MacroExpTracker.hasMismatch())
1220         continue;
1221       // Tell caller we found one or more errors.
1222       ReturnValue = true;
1223       // Start the error message.
1224       OS << *MacroExpTracker.InstanceSourceLine;
1225       if (ItemKey.Column > 0)
1226         OS << std::string(ItemKey.Column - 1, ' ') << "^\n";
1227       OS << "error: Macro instance '" << *MacroExpTracker.MacroUnexpanded
1228          << "' has different values in this header, depending on how it was "
1229             "included.\n";
1230       // Walk all the instances.
1231       for (std::vector<MacroExpansionInstance>::iterator
1232                IMT = MacroExpTracker.MacroExpansionInstances.begin(),
1233                EMT = MacroExpTracker.MacroExpansionInstances.end();
1234            IMT != EMT; ++IMT) {
1235         MacroExpansionInstance &MacroInfo = *IMT;
1236         OS << "  '" << *MacroExpTracker.MacroUnexpanded << "' expanded to: '"
1237            << *MacroInfo.MacroExpanded
1238            << "' with respect to these inclusion paths:\n";
1239         // Walk all the inclusion path hierarchies.
1240         for (std::vector<InclusionPathHandle>::iterator
1241                  IIP = MacroInfo.InclusionPathHandles.begin(),
1242                  EIP = MacroInfo.InclusionPathHandles.end();
1243              IIP != EIP; ++IIP) {
1244           const std::vector<HeaderHandle> &ip = getInclusionPath(*IIP);
1245           int Count = (int)ip.size();
1246           for (int Index = 0; Index < Count; ++Index) {
1247             HeaderHandle H = ip[Index];
1248             OS << std::string((Index * 2) + 4, ' ') << *getHeaderFilePath(H)
1249                << "\n";
1250           }
1251         }
1252         // For a macro that wasn't defined, we flag it by using the
1253         // instance location.
1254         // If there is a definition...
1255         if (MacroInfo.DefinitionLocation.Line != ItemKey.Line) {
1256           OS << *MacroInfo.DefinitionSourceLine;
1257           if (MacroInfo.DefinitionLocation.Column > 0)
1258             OS << std::string(MacroInfo.DefinitionLocation.Column - 1, ' ')
1259                << "^\n";
1260           OS << "Macro defined here.\n";
1261         } else
1262           OS << "(no macro definition)"
1263              << "\n";
1264       }
1265     }
1266     return ReturnValue;
1267   }
1268 
1269   // Report on inconsistent conditional instances.
1270   // Returns true if any mismatches.
1271   bool reportInconsistentConditionals(llvm::raw_ostream &OS) {
1272     bool ReturnValue = false;
1273     // Walk all the conditional trackers in the map.
1274     for (ConditionalExpansionMapIter I = ConditionalExpansions.begin(),
1275                                      E = ConditionalExpansions.end();
1276          I != E; ++I) {
1277       const PPItemKey &ItemKey = I->first;
1278       ConditionalTracker &CondTracker = I->second;
1279       if (!CondTracker.hasMismatch())
1280         continue;
1281       // Tell caller we found one or more errors.
1282       ReturnValue = true;
1283       // Start the error message.
1284       OS << *HeaderPaths[ItemKey.File] << ":" << ItemKey.Line << ":"
1285          << ItemKey.Column << "\n";
1286       OS << "#" << getDirectiveSpelling(CondTracker.DirectiveKind) << " "
1287          << *CondTracker.ConditionUnexpanded << "\n";
1288       OS << "^\n";
1289       OS << "error: Conditional expression instance '"
1290          << *CondTracker.ConditionUnexpanded
1291          << "' has different values in this header, depending on how it was "
1292             "included.\n";
1293       // Walk all the instances.
1294       for (std::vector<ConditionalExpansionInstance>::iterator
1295                IMT = CondTracker.ConditionalExpansionInstances.begin(),
1296                EMT = CondTracker.ConditionalExpansionInstances.end();
1297            IMT != EMT; ++IMT) {
1298         ConditionalExpansionInstance &MacroInfo = *IMT;
1299         OS << "  '" << *CondTracker.ConditionUnexpanded << "' expanded to: '"
1300            << ConditionValueKindStrings[MacroInfo.ConditionValue]
1301            << "' with respect to these inclusion paths:\n";
1302         // Walk all the inclusion path hierarchies.
1303         for (std::vector<InclusionPathHandle>::iterator
1304                  IIP = MacroInfo.InclusionPathHandles.begin(),
1305                  EIP = MacroInfo.InclusionPathHandles.end();
1306              IIP != EIP; ++IIP) {
1307           const std::vector<HeaderHandle> &ip = getInclusionPath(*IIP);
1308           int Count = (int)ip.size();
1309           for (int Index = 0; Index < Count; ++Index) {
1310             HeaderHandle H = ip[Index];
1311             OS << std::string((Index * 2) + 4, ' ') << *getHeaderFilePath(H)
1312                << "\n";
1313           }
1314         }
1315       }
1316     }
1317     return ReturnValue;
1318   }
1319 
1320   // Get directive spelling.
1321   static const char *getDirectiveSpelling(clang::tok::PPKeywordKind kind) {
1322     switch (kind) {
1323     case clang::tok::pp_if:
1324       return "if";
1325     case clang::tok::pp_elif:
1326       return "elif";
1327     case clang::tok::pp_ifdef:
1328       return "ifdef";
1329     case clang::tok::pp_ifndef:
1330       return "ifndef";
1331     default:
1332       return "(unknown)";
1333     }
1334   }
1335 
1336 private:
1337   llvm::SmallVector<std::string, 32> HeaderList;
1338   // Only do extern, namespace check for headers in HeaderList.
1339   bool BlockCheckHeaderListOnly;
1340   llvm::StringPool Strings;
1341   std::vector<StringHandle> HeaderPaths;
1342   std::vector<HeaderHandle> HeaderStack;
1343   std::vector<HeaderInclusionPath> InclusionPaths;
1344   InclusionPathHandle CurrentInclusionPathHandle;
1345   llvm::SmallSet<HeaderHandle, 128> HeadersInThisCompile;
1346   std::vector<PPItemKey> IncludeDirectives;
1347   MacroExpansionMap MacroExpansions;
1348   ConditionalExpansionMap ConditionalExpansions;
1349   bool InNestedHeader;
1350 };
1351 
1352 // PreprocessorTracker functions.
1353 
1354 // PreprocessorTracker desctructor.
1355 PreprocessorTracker::~PreprocessorTracker() {}
1356 
1357 // Create instance of PreprocessorTracker.
1358 PreprocessorTracker *PreprocessorTracker::create(
1359     llvm::SmallVector<std::string, 32> &Headers,
1360     bool DoBlockCheckHeaderListOnly) {
1361   return new PreprocessorTrackerImpl(Headers, DoBlockCheckHeaderListOnly);
1362 }
1363 
1364 // Preprocessor callbacks for modularize.
1365 
1366 // Handle include directive.
1367 void PreprocessorCallbacks::InclusionDirective(
1368     clang::SourceLocation HashLoc, const clang::Token &IncludeTok,
1369     llvm::StringRef FileName, bool IsAngled,
1370     clang::CharSourceRange FilenameRange, const clang::FileEntry *File,
1371     llvm::StringRef SearchPath, llvm::StringRef RelativePath,
1372     const clang::Module *Imported) {
1373   int DirectiveLine, DirectiveColumn;
1374   std::string HeaderPath = getSourceLocationFile(PP, HashLoc);
1375   getSourceLocationLineAndColumn(PP, HashLoc, DirectiveLine, DirectiveColumn);
1376   PPTracker.handleIncludeDirective(HeaderPath, DirectiveLine, DirectiveColumn,
1377                                    FileName);
1378 }
1379 
1380 // Handle file entry/exit.
1381 void PreprocessorCallbacks::FileChanged(
1382     clang::SourceLocation Loc, clang::PPCallbacks::FileChangeReason Reason,
1383     clang::SrcMgr::CharacteristicKind FileType, clang::FileID PrevFID) {
1384   switch (Reason) {
1385   case EnterFile:
1386     PPTracker.handleHeaderEntry(PP, getSourceLocationFile(PP, Loc));
1387     break;
1388   case ExitFile: {
1389     const clang::FileEntry *F =
1390         PP.getSourceManager().getFileEntryForID(PrevFID);
1391     if (F)
1392       PPTracker.handleHeaderExit(F->getName());
1393   } break;
1394   case SystemHeaderPragma:
1395   case RenameFile:
1396     break;
1397   }
1398 }
1399 
1400 // Handle macro expansion.
1401 void PreprocessorCallbacks::MacroExpands(const clang::Token &MacroNameTok,
1402                                          const clang::MacroDirective *MD,
1403                                          clang::SourceRange Range,
1404                                          const clang::MacroArgs *Args) {
1405   clang::SourceLocation Loc = Range.getBegin();
1406   // Ignore macro argument expansions.
1407   if (!Loc.isFileID())
1408     return;
1409   clang::IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
1410   const clang::MacroInfo *MI = PP.getMacroInfo(II);
1411   std::string MacroName = II->getName().str();
1412   std::string Unexpanded(getMacroUnexpandedString(Range, PP, MacroName, MI));
1413   std::string Expanded(getMacroExpandedString(PP, MacroName, MI, Args));
1414   PPTracker.addMacroExpansionInstance(
1415       PP, PPTracker.getCurrentHeaderHandle(), Loc, MI->getDefinitionLoc(), II,
1416       Unexpanded, Expanded, PPTracker.getCurrentInclusionPathHandle());
1417 }
1418 
1419 void PreprocessorCallbacks::Defined(const clang::Token &MacroNameTok,
1420                                     const clang::MacroDirective *MD,
1421                                     clang::SourceRange Range) {
1422   clang::SourceLocation Loc(Range.getBegin());
1423   clang::IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
1424   const clang::MacroInfo *MI = PP.getMacroInfo(II);
1425   std::string MacroName = II->getName().str();
1426   std::string Unexpanded(getSourceString(PP, Range));
1427   PPTracker.addMacroExpansionInstance(
1428       PP, PPTracker.getCurrentHeaderHandle(), Loc,
1429       (MI ? MI->getDefinitionLoc() : Loc), II, Unexpanded,
1430       (MI ? "true" : "false"), PPTracker.getCurrentInclusionPathHandle());
1431 }
1432 
1433 void PreprocessorCallbacks::If(clang::SourceLocation Loc,
1434                                clang::SourceRange ConditionRange,
1435                                clang::PPCallbacks::ConditionValueKind ConditionResult) {
1436   std::string Unexpanded(getSourceString(PP, ConditionRange));
1437   PPTracker.addConditionalExpansionInstance(
1438       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_if,
1439       ConditionResult, Unexpanded, PPTracker.getCurrentInclusionPathHandle());
1440 }
1441 
1442 void PreprocessorCallbacks::Elif(clang::SourceLocation Loc,
1443                                  clang::SourceRange ConditionRange,
1444                                  clang::PPCallbacks::ConditionValueKind ConditionResult,
1445                                  clang::SourceLocation IfLoc) {
1446   std::string Unexpanded(getSourceString(PP, ConditionRange));
1447   PPTracker.addConditionalExpansionInstance(
1448       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_elif,
1449       ConditionResult, Unexpanded, PPTracker.getCurrentInclusionPathHandle());
1450 }
1451 
1452 void PreprocessorCallbacks::Ifdef(clang::SourceLocation Loc,
1453                                   const clang::Token &MacroNameTok,
1454                                   const clang::MacroDirective *MD) {
1455   clang::PPCallbacks::ConditionValueKind IsDefined =
1456     (MD ? clang::PPCallbacks::CVK_True : clang::PPCallbacks::CVK_False );
1457   PPTracker.addConditionalExpansionInstance(
1458       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_ifdef,
1459       IsDefined, PP.getSpelling(MacroNameTok),
1460       PPTracker.getCurrentInclusionPathHandle());
1461 }
1462 
1463 void PreprocessorCallbacks::Ifndef(clang::SourceLocation Loc,
1464                                    const clang::Token &MacroNameTok,
1465                                    const clang::MacroDirective *MD) {
1466   clang::PPCallbacks::ConditionValueKind IsNotDefined =
1467     (!MD ? clang::PPCallbacks::CVK_True : clang::PPCallbacks::CVK_False );
1468   PPTracker.addConditionalExpansionInstance(
1469       PP, PPTracker.getCurrentHeaderHandle(), Loc, clang::tok::pp_ifndef,
1470       IsNotDefined, PP.getSpelling(MacroNameTok),
1471       PPTracker.getCurrentInclusionPathHandle());
1472 }
1473 } // end namespace Modularize
1474