1 //===--- DependencyFile.cpp - Generate dependency file --------------------===//
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 // This code generates dependency files.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/Frontend/Utils.h"
15 #include "clang/Basic/FileManager.h"
16 #include "clang/Basic/SourceManager.h"
17 #include "clang/Frontend/DependencyOutputOptions.h"
18 #include "clang/Frontend/FrontendDiagnostic.h"
19 #include "clang/Lex/DirectoryLookup.h"
20 #include "clang/Lex/LexDiagnostic.h"
21 #include "clang/Lex/ModuleMap.h"
22 #include "clang/Lex/PPCallbacks.h"
23 #include "clang/Lex/Preprocessor.h"
24 #include "clang/Serialization/ASTReader.h"
25 #include "llvm/ADT/StringSet.h"
26 #include "llvm/ADT/StringSwitch.h"
27 #include "llvm/Support/FileSystem.h"
28 #include "llvm/Support/Path.h"
29 #include "llvm/Support/raw_ostream.h"
30 
31 using namespace clang;
32 
33 namespace {
34 struct DepCollectorPPCallbacks : public PPCallbacks {
35   DependencyCollector &DepCollector;
36   SourceManager &SM;
37   DepCollectorPPCallbacks(DependencyCollector &L, SourceManager &SM)
38       : DepCollector(L), SM(SM) { }
39 
40   void FileChanged(SourceLocation Loc, FileChangeReason Reason,
41                    SrcMgr::CharacteristicKind FileType,
42                    FileID PrevFID) override {
43     if (Reason != PPCallbacks::EnterFile)
44       return;
45 
46     // Dependency generation really does want to go all the way to the
47     // file entry for a source location to find out what is depended on.
48     // We do not want #line markers to affect dependency generation!
49     const FileEntry *FE =
50         SM.getFileEntryForID(SM.getFileID(SM.getExpansionLoc(Loc)));
51     if (!FE)
52       return;
53 
54     StringRef Filename = FE->getName();
55 
56     // Remove leading "./" (or ".//" or "././" etc.)
57     while (Filename.size() > 2 && Filename[0] == '.' &&
58            llvm::sys::path::is_separator(Filename[1])) {
59       Filename = Filename.substr(1);
60       while (llvm::sys::path::is_separator(Filename[0]))
61         Filename = Filename.substr(1);
62     }
63 
64     DepCollector.maybeAddDependency(Filename, /*FromModule*/false,
65                                    FileType != SrcMgr::C_User,
66                                    /*IsModuleFile*/false, /*IsMissing*/false);
67   }
68 
69   void InclusionDirective(SourceLocation HashLoc, const Token &IncludeTok,
70                           StringRef FileName, bool IsAngled,
71                           CharSourceRange FilenameRange, const FileEntry *File,
72                           StringRef SearchPath, StringRef RelativePath,
73                           const Module *Imported) override {
74     if (!File)
75       DepCollector.maybeAddDependency(FileName, /*FromModule*/false,
76                                      /*IsSystem*/false, /*IsModuleFile*/false,
77                                      /*IsMissing*/true);
78     // Files that actually exist are handled by FileChanged.
79   }
80 
81   void EndOfMainFile() override {
82     DepCollector.finishedMainFile();
83   }
84 };
85 
86 struct DepCollectorMMCallbacks : public ModuleMapCallbacks {
87   DependencyCollector &DepCollector;
88   DepCollectorMMCallbacks(DependencyCollector &DC) : DepCollector(DC) {}
89 
90   void moduleMapFileRead(SourceLocation Loc, const FileEntry &Entry,
91                          bool IsSystem) override {
92     StringRef Filename = Entry.getName();
93     DepCollector.maybeAddDependency(Filename, /*FromModule*/false,
94                                     /*IsSystem*/IsSystem,
95                                     /*IsModuleFile*/false,
96                                     /*IsMissing*/false);
97   }
98 };
99 
100 struct DepCollectorASTListener : public ASTReaderListener {
101   DependencyCollector &DepCollector;
102   DepCollectorASTListener(DependencyCollector &L) : DepCollector(L) { }
103   bool needsInputFileVisitation() override { return true; }
104   bool needsSystemInputFileVisitation() override {
105     return DepCollector.needSystemDependencies();
106   }
107   void visitModuleFile(StringRef Filename) override {
108     DepCollector.maybeAddDependency(Filename, /*FromModule*/true,
109                                    /*IsSystem*/false, /*IsModuleFile*/true,
110                                    /*IsMissing*/false);
111   }
112   bool visitInputFile(StringRef Filename, bool IsSystem,
113                       bool IsOverridden) override {
114     if (IsOverridden)
115       return true;
116 
117     DepCollector.maybeAddDependency(Filename, /*FromModule*/true, IsSystem,
118                                    /*IsModuleFile*/false, /*IsMissing*/false);
119     return true;
120   }
121 };
122 } // end anonymous namespace
123 
124 void DependencyCollector::maybeAddDependency(StringRef Filename, bool FromModule,
125                                             bool IsSystem, bool IsModuleFile,
126                                             bool IsMissing) {
127   if (Seen.insert(Filename).second &&
128       sawDependency(Filename, FromModule, IsSystem, IsModuleFile, IsMissing))
129     Dependencies.push_back(Filename);
130 }
131 
132 static bool isSpecialFilename(StringRef Filename) {
133   return llvm::StringSwitch<bool>(Filename)
134       .Case("<built-in>", true)
135       .Case("<stdin>", true)
136       .Default(false);
137 }
138 
139 bool DependencyCollector::sawDependency(StringRef Filename, bool FromModule,
140                                        bool IsSystem, bool IsModuleFile,
141                                        bool IsMissing) {
142   return !isSpecialFilename(Filename) &&
143          (needSystemDependencies() || !IsSystem);
144 }
145 
146 DependencyCollector::~DependencyCollector() { }
147 void DependencyCollector::attachToPreprocessor(Preprocessor &PP) {
148   PP.addPPCallbacks(
149       llvm::make_unique<DepCollectorPPCallbacks>(*this, PP.getSourceManager()));
150   PP.getHeaderSearchInfo().getModuleMap().addModuleMapCallbacks(
151       llvm::make_unique<DepCollectorMMCallbacks>(*this));
152 }
153 void DependencyCollector::attachToASTReader(ASTReader &R) {
154   R.addListener(llvm::make_unique<DepCollectorASTListener>(*this));
155 }
156 
157 namespace {
158 /// Private implementation for DependencyFileGenerator
159 class DFGImpl : public PPCallbacks {
160   std::vector<std::string> Files;
161   llvm::StringSet<> FilesSet;
162   const Preprocessor *PP;
163   std::string OutputFile;
164   std::vector<std::string> Targets;
165   bool IncludeSystemHeaders;
166   bool PhonyTarget;
167   bool AddMissingHeaderDeps;
168   bool SeenMissingHeader;
169   bool IncludeModuleFiles;
170   DependencyOutputFormat OutputFormat;
171 
172 private:
173   bool FileMatchesDepCriteria(const char *Filename,
174                               SrcMgr::CharacteristicKind FileType);
175   void OutputDependencyFile();
176 
177 public:
178   DFGImpl(const Preprocessor *_PP, const DependencyOutputOptions &Opts)
179     : PP(_PP), OutputFile(Opts.OutputFile), Targets(Opts.Targets),
180       IncludeSystemHeaders(Opts.IncludeSystemHeaders),
181       PhonyTarget(Opts.UsePhonyTargets),
182       AddMissingHeaderDeps(Opts.AddMissingHeaderDeps),
183       SeenMissingHeader(false),
184       IncludeModuleFiles(Opts.IncludeModuleFiles),
185       OutputFormat(Opts.OutputFormat) {
186     for (auto ExtraDep : Opts.ExtraDeps) {
187       AddFilename(ExtraDep);
188     }
189   }
190 
191   void FileChanged(SourceLocation Loc, FileChangeReason Reason,
192                    SrcMgr::CharacteristicKind FileType,
193                    FileID PrevFID) override;
194   void InclusionDirective(SourceLocation HashLoc, const Token &IncludeTok,
195                           StringRef FileName, bool IsAngled,
196                           CharSourceRange FilenameRange, const FileEntry *File,
197                           StringRef SearchPath, StringRef RelativePath,
198                           const Module *Imported) override;
199 
200   void EndOfMainFile() override {
201     OutputDependencyFile();
202   }
203 
204   void AddFilename(StringRef Filename);
205   bool includeSystemHeaders() const { return IncludeSystemHeaders; }
206   bool includeModuleFiles() const { return IncludeModuleFiles; }
207 };
208 
209 class DFGMMCallback : public ModuleMapCallbacks {
210   DFGImpl &Parent;
211 public:
212   DFGMMCallback(DFGImpl &Parent) : Parent(Parent) {}
213   void moduleMapFileRead(SourceLocation Loc, const FileEntry &Entry,
214                          bool IsSystem) override {
215     if (!IsSystem || Parent.includeSystemHeaders())
216       Parent.AddFilename(Entry.getName());
217   }
218 };
219 
220 class DFGASTReaderListener : public ASTReaderListener {
221   DFGImpl &Parent;
222 public:
223   DFGASTReaderListener(DFGImpl &Parent)
224   : Parent(Parent) { }
225   bool needsInputFileVisitation() override { return true; }
226   bool needsSystemInputFileVisitation() override {
227     return Parent.includeSystemHeaders();
228   }
229   void visitModuleFile(StringRef Filename) override;
230   bool visitInputFile(StringRef Filename, bool isSystem,
231                       bool isOverridden) override;
232 };
233 }
234 
235 DependencyFileGenerator::DependencyFileGenerator(void *Impl)
236 : Impl(Impl) { }
237 
238 DependencyFileGenerator *DependencyFileGenerator::CreateAndAttachToPreprocessor(
239     clang::Preprocessor &PP, const clang::DependencyOutputOptions &Opts) {
240 
241   if (Opts.Targets.empty()) {
242     PP.getDiagnostics().Report(diag::err_fe_dependency_file_requires_MT);
243     return nullptr;
244   }
245 
246   // Disable the "file not found" diagnostic if the -MG option was given.
247   if (Opts.AddMissingHeaderDeps)
248     PP.SetSuppressIncludeNotFoundError(true);
249 
250   DFGImpl *Callback = new DFGImpl(&PP, Opts);
251   PP.addPPCallbacks(std::unique_ptr<PPCallbacks>(Callback));
252   PP.getHeaderSearchInfo().getModuleMap().addModuleMapCallbacks(
253       llvm::make_unique<DFGMMCallback>(*Callback));
254   return new DependencyFileGenerator(Callback);
255 }
256 
257 void DependencyFileGenerator::AttachToASTReader(ASTReader &R) {
258   DFGImpl *I = reinterpret_cast<DFGImpl *>(Impl);
259   assert(I && "missing implementation");
260   R.addListener(llvm::make_unique<DFGASTReaderListener>(*I));
261 }
262 
263 /// FileMatchesDepCriteria - Determine whether the given Filename should be
264 /// considered as a dependency.
265 bool DFGImpl::FileMatchesDepCriteria(const char *Filename,
266                                      SrcMgr::CharacteristicKind FileType) {
267   if (isSpecialFilename(Filename))
268     return false;
269 
270   if (IncludeSystemHeaders)
271     return true;
272 
273   return FileType == SrcMgr::C_User;
274 }
275 
276 void DFGImpl::FileChanged(SourceLocation Loc,
277                           FileChangeReason Reason,
278                           SrcMgr::CharacteristicKind FileType,
279                           FileID PrevFID) {
280   if (Reason != PPCallbacks::EnterFile)
281     return;
282 
283   // Dependency generation really does want to go all the way to the
284   // file entry for a source location to find out what is depended on.
285   // We do not want #line markers to affect dependency generation!
286   SourceManager &SM = PP->getSourceManager();
287 
288   const FileEntry *FE =
289     SM.getFileEntryForID(SM.getFileID(SM.getExpansionLoc(Loc)));
290   if (!FE) return;
291 
292   StringRef Filename = FE->getName();
293   if (!FileMatchesDepCriteria(Filename.data(), FileType))
294     return;
295 
296   // Remove leading "./" (or ".//" or "././" etc.)
297   while (Filename.size() > 2 && Filename[0] == '.' &&
298          llvm::sys::path::is_separator(Filename[1])) {
299     Filename = Filename.substr(1);
300     while (llvm::sys::path::is_separator(Filename[0]))
301       Filename = Filename.substr(1);
302   }
303 
304   AddFilename(Filename);
305 }
306 
307 void DFGImpl::InclusionDirective(SourceLocation HashLoc,
308                                  const Token &IncludeTok,
309                                  StringRef FileName,
310                                  bool IsAngled,
311                                  CharSourceRange FilenameRange,
312                                  const FileEntry *File,
313                                  StringRef SearchPath,
314                                  StringRef RelativePath,
315                                  const Module *Imported) {
316   if (!File) {
317     if (AddMissingHeaderDeps)
318       AddFilename(FileName);
319     else
320       SeenMissingHeader = true;
321   }
322 }
323 
324 void DFGImpl::AddFilename(StringRef Filename) {
325   if (FilesSet.insert(Filename).second)
326     Files.push_back(Filename);
327 }
328 
329 /// Print the filename, with escaping or quoting that accommodates the three
330 /// most likely tools that use dependency files: GNU Make, BSD Make, and
331 /// NMake/Jom.
332 ///
333 /// BSD Make is the simplest case: It does no escaping at all.  This means
334 /// characters that are normally delimiters, i.e. space and # (the comment
335 /// character) simply aren't supported in filenames.
336 ///
337 /// GNU Make does allow space and # in filenames, but to avoid being treated
338 /// as a delimiter or comment, these must be escaped with a backslash. Because
339 /// backslash is itself the escape character, if a backslash appears in a
340 /// filename, it should be escaped as well.  (As a special case, $ is escaped
341 /// as $$, which is the normal Make way to handle the $ character.)
342 /// For compatibility with BSD Make and historical practice, if GNU Make
343 /// un-escapes characters in a filename but doesn't find a match, it will
344 /// retry with the unmodified original string.
345 ///
346 /// GCC tries to accommodate both Make formats by escaping any space or #
347 /// characters in the original filename, but not escaping backslashes.  The
348 /// apparent intent is so that filenames with backslashes will be handled
349 /// correctly by BSD Make, and by GNU Make in its fallback mode of using the
350 /// unmodified original string; filenames with # or space characters aren't
351 /// supported by BSD Make at all, but will be handled correctly by GNU Make
352 /// due to the escaping.
353 ///
354 /// A corner case that GCC gets only partly right is when the original filename
355 /// has a backslash immediately followed by space or #.  GNU Make would expect
356 /// this backslash to be escaped; however GCC escapes the original backslash
357 /// only when followed by space, not #.  It will therefore take a dependency
358 /// from a directive such as
359 ///     #include "a\ b\#c.h"
360 /// and emit it as
361 ///     a\\\ b\\#c.h
362 /// which GNU Make will interpret as
363 ///     a\ b\
364 /// followed by a comment. Failing to find this file, it will fall back to the
365 /// original string, which probably doesn't exist either; in any case it won't
366 /// find
367 ///     a\ b\#c.h
368 /// which is the actual filename specified by the include directive.
369 ///
370 /// Clang does what GCC does, rather than what GNU Make expects.
371 ///
372 /// NMake/Jom has a different set of scary characters, but wraps filespecs in
373 /// double-quotes to avoid misinterpreting them; see
374 /// https://msdn.microsoft.com/en-us/library/dd9y37ha.aspx for NMake info,
375 /// https://msdn.microsoft.com/en-us/library/windows/desktop/aa365247(v=vs.85).aspx
376 /// for Windows file-naming info.
377 static void PrintFilename(raw_ostream &OS, StringRef Filename,
378                           DependencyOutputFormat OutputFormat) {
379   if (OutputFormat == DependencyOutputFormat::NMake) {
380     // Add quotes if needed. These are the characters listed as "special" to
381     // NMake, that are legal in a Windows filespec, and that could cause
382     // misinterpretation of the dependency string.
383     if (Filename.find_first_of(" #${}^!") != StringRef::npos)
384       OS << '\"' << Filename << '\"';
385     else
386       OS << Filename;
387     return;
388   }
389   assert(OutputFormat == DependencyOutputFormat::Make);
390   for (unsigned i = 0, e = Filename.size(); i != e; ++i) {
391     if (Filename[i] == '#') // Handle '#' the broken gcc way.
392       OS << '\\';
393     else if (Filename[i] == ' ') { // Handle space correctly.
394       OS << '\\';
395       unsigned j = i;
396       while (j > 0 && Filename[--j] == '\\')
397         OS << '\\';
398     } else if (Filename[i] == '$') // $ is escaped by $$.
399       OS << '$';
400     OS << Filename[i];
401   }
402 }
403 
404 void DFGImpl::OutputDependencyFile() {
405   if (SeenMissingHeader) {
406     llvm::sys::fs::remove(OutputFile);
407     return;
408   }
409 
410   std::error_code EC;
411   llvm::raw_fd_ostream OS(OutputFile, EC, llvm::sys::fs::F_Text);
412   if (EC) {
413     PP->getDiagnostics().Report(diag::err_fe_error_opening) << OutputFile
414                                                             << EC.message();
415     return;
416   }
417 
418   // Write out the dependency targets, trying to avoid overly long
419   // lines when possible. We try our best to emit exactly the same
420   // dependency file as GCC (4.2), assuming the included files are the
421   // same.
422   const unsigned MaxColumns = 75;
423   unsigned Columns = 0;
424 
425   for (std::vector<std::string>::iterator
426          I = Targets.begin(), E = Targets.end(); I != E; ++I) {
427     unsigned N = I->length();
428     if (Columns == 0) {
429       Columns += N;
430     } else if (Columns + N + 2 > MaxColumns) {
431       Columns = N + 2;
432       OS << " \\\n  ";
433     } else {
434       Columns += N + 1;
435       OS << ' ';
436     }
437     // Targets already quoted as needed.
438     OS << *I;
439   }
440 
441   OS << ':';
442   Columns += 1;
443 
444   // Now add each dependency in the order it was seen, but avoiding
445   // duplicates.
446   for (std::vector<std::string>::iterator I = Files.begin(),
447          E = Files.end(); I != E; ++I) {
448     // Start a new line if this would exceed the column limit. Make
449     // sure to leave space for a trailing " \" in case we need to
450     // break the line on the next iteration.
451     unsigned N = I->length();
452     if (Columns + (N + 1) + 2 > MaxColumns) {
453       OS << " \\\n ";
454       Columns = 2;
455     }
456     OS << ' ';
457     PrintFilename(OS, *I, OutputFormat);
458     Columns += N + 1;
459   }
460   OS << '\n';
461 
462   // Create phony targets if requested.
463   if (PhonyTarget && !Files.empty()) {
464     // Skip the first entry, this is always the input file itself.
465     for (std::vector<std::string>::iterator I = Files.begin() + 1,
466            E = Files.end(); I != E; ++I) {
467       OS << '\n';
468       PrintFilename(OS, *I, OutputFormat);
469       OS << ":\n";
470     }
471   }
472 }
473 
474 bool DFGASTReaderListener::visitInputFile(llvm::StringRef Filename,
475                                           bool IsSystem, bool IsOverridden) {
476   assert(!IsSystem || needsSystemInputFileVisitation());
477   if (IsOverridden)
478     return true;
479 
480   Parent.AddFilename(Filename);
481   return true;
482 }
483 
484 void DFGASTReaderListener::visitModuleFile(llvm::StringRef Filename) {
485   if (Parent.includeModuleFiles())
486     Parent.AddFilename(Filename);
487 }
488