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,
108                        serialization::ModuleKind Kind) override {
109     DepCollector.maybeAddDependency(Filename, /*FromModule*/true,
110                                    /*IsSystem*/false, /*IsModuleFile*/true,
111                                    /*IsMissing*/false);
112   }
113   bool visitInputFile(StringRef Filename, bool IsSystem,
114                       bool IsOverridden, bool IsExplicitModule) override {
115     if (IsOverridden || IsExplicitModule)
116       return true;
117 
118     DepCollector.maybeAddDependency(Filename, /*FromModule*/true, IsSystem,
119                                    /*IsModuleFile*/false, /*IsMissing*/false);
120     return true;
121   }
122 };
123 } // end anonymous namespace
124 
125 void DependencyCollector::maybeAddDependency(StringRef Filename, bool FromModule,
126                                             bool IsSystem, bool IsModuleFile,
127                                             bool IsMissing) {
128   if (Seen.insert(Filename).second &&
129       sawDependency(Filename, FromModule, IsSystem, IsModuleFile, IsMissing))
130     Dependencies.push_back(Filename);
131 }
132 
133 static bool isSpecialFilename(StringRef Filename) {
134   return llvm::StringSwitch<bool>(Filename)
135       .Case("<built-in>", true)
136       .Case("<stdin>", true)
137       .Default(false);
138 }
139 
140 bool DependencyCollector::sawDependency(StringRef Filename, bool FromModule,
141                                        bool IsSystem, bool IsModuleFile,
142                                        bool IsMissing) {
143   return !isSpecialFilename(Filename) &&
144          (needSystemDependencies() || !IsSystem);
145 }
146 
147 DependencyCollector::~DependencyCollector() { }
148 void DependencyCollector::attachToPreprocessor(Preprocessor &PP) {
149   PP.addPPCallbacks(
150       llvm::make_unique<DepCollectorPPCallbacks>(*this, PP.getSourceManager()));
151   PP.getHeaderSearchInfo().getModuleMap().addModuleMapCallbacks(
152       llvm::make_unique<DepCollectorMMCallbacks>(*this));
153 }
154 void DependencyCollector::attachToASTReader(ASTReader &R) {
155   R.addListener(llvm::make_unique<DepCollectorASTListener>(*this));
156 }
157 
158 namespace {
159 /// Private implementation for DependencyFileGenerator
160 class DFGImpl : public PPCallbacks {
161   std::vector<std::string> Files;
162   llvm::StringSet<> FilesSet;
163   const Preprocessor *PP;
164   std::string OutputFile;
165   std::vector<std::string> Targets;
166   bool IncludeSystemHeaders;
167   bool PhonyTarget;
168   bool AddMissingHeaderDeps;
169   bool SeenMissingHeader;
170   bool IncludeModuleFiles;
171   DependencyOutputFormat OutputFormat;
172 
173 private:
174   bool FileMatchesDepCriteria(const char *Filename,
175                               SrcMgr::CharacteristicKind FileType);
176   void OutputDependencyFile();
177 
178 public:
179   DFGImpl(const Preprocessor *_PP, const DependencyOutputOptions &Opts)
180     : PP(_PP), OutputFile(Opts.OutputFile), Targets(Opts.Targets),
181       IncludeSystemHeaders(Opts.IncludeSystemHeaders),
182       PhonyTarget(Opts.UsePhonyTargets),
183       AddMissingHeaderDeps(Opts.AddMissingHeaderDeps),
184       SeenMissingHeader(false),
185       IncludeModuleFiles(Opts.IncludeModuleFiles),
186       OutputFormat(Opts.OutputFormat) {
187     for (auto ExtraDep : Opts.ExtraDeps) {
188       AddFilename(ExtraDep);
189     }
190   }
191 
192   void FileChanged(SourceLocation Loc, FileChangeReason Reason,
193                    SrcMgr::CharacteristicKind FileType,
194                    FileID PrevFID) override;
195   void InclusionDirective(SourceLocation HashLoc, const Token &IncludeTok,
196                           StringRef FileName, bool IsAngled,
197                           CharSourceRange FilenameRange, const FileEntry *File,
198                           StringRef SearchPath, StringRef RelativePath,
199                           const Module *Imported) override;
200 
201   void EndOfMainFile() override {
202     OutputDependencyFile();
203   }
204 
205   void AddFilename(StringRef Filename);
206   bool includeSystemHeaders() const { return IncludeSystemHeaders; }
207   bool includeModuleFiles() const { return IncludeModuleFiles; }
208 };
209 
210 class DFGMMCallback : public ModuleMapCallbacks {
211   DFGImpl &Parent;
212 public:
213   DFGMMCallback(DFGImpl &Parent) : Parent(Parent) {}
214   void moduleMapFileRead(SourceLocation Loc, const FileEntry &Entry,
215                          bool IsSystem) override {
216     if (!IsSystem || Parent.includeSystemHeaders())
217       Parent.AddFilename(Entry.getName());
218   }
219 };
220 
221 class DFGASTReaderListener : public ASTReaderListener {
222   DFGImpl &Parent;
223 public:
224   DFGASTReaderListener(DFGImpl &Parent)
225   : Parent(Parent) { }
226   bool needsInputFileVisitation() override { return true; }
227   bool needsSystemInputFileVisitation() override {
228     return Parent.includeSystemHeaders();
229   }
230   void visitModuleFile(StringRef Filename,
231                        serialization::ModuleKind Kind) override;
232   bool visitInputFile(StringRef Filename, bool isSystem,
233                       bool isOverridden, bool isExplicitModule) override;
234 };
235 }
236 
237 DependencyFileGenerator::DependencyFileGenerator(void *Impl)
238 : Impl(Impl) { }
239 
240 DependencyFileGenerator *DependencyFileGenerator::CreateAndAttachToPreprocessor(
241     clang::Preprocessor &PP, const clang::DependencyOutputOptions &Opts) {
242 
243   if (Opts.Targets.empty()) {
244     PP.getDiagnostics().Report(diag::err_fe_dependency_file_requires_MT);
245     return nullptr;
246   }
247 
248   // Disable the "file not found" diagnostic if the -MG option was given.
249   if (Opts.AddMissingHeaderDeps)
250     PP.SetSuppressIncludeNotFoundError(true);
251 
252   DFGImpl *Callback = new DFGImpl(&PP, Opts);
253   PP.addPPCallbacks(std::unique_ptr<PPCallbacks>(Callback));
254   PP.getHeaderSearchInfo().getModuleMap().addModuleMapCallbacks(
255       llvm::make_unique<DFGMMCallback>(*Callback));
256   return new DependencyFileGenerator(Callback);
257 }
258 
259 void DependencyFileGenerator::AttachToASTReader(ASTReader &R) {
260   DFGImpl *I = reinterpret_cast<DFGImpl *>(Impl);
261   assert(I && "missing implementation");
262   R.addListener(llvm::make_unique<DFGASTReaderListener>(*I));
263 }
264 
265 /// FileMatchesDepCriteria - Determine whether the given Filename should be
266 /// considered as a dependency.
267 bool DFGImpl::FileMatchesDepCriteria(const char *Filename,
268                                      SrcMgr::CharacteristicKind FileType) {
269   if (isSpecialFilename(Filename))
270     return false;
271 
272   if (IncludeSystemHeaders)
273     return true;
274 
275   return FileType == SrcMgr::C_User;
276 }
277 
278 void DFGImpl::FileChanged(SourceLocation Loc,
279                           FileChangeReason Reason,
280                           SrcMgr::CharacteristicKind FileType,
281                           FileID PrevFID) {
282   if (Reason != PPCallbacks::EnterFile)
283     return;
284 
285   // Dependency generation really does want to go all the way to the
286   // file entry for a source location to find out what is depended on.
287   // We do not want #line markers to affect dependency generation!
288   SourceManager &SM = PP->getSourceManager();
289 
290   const FileEntry *FE =
291     SM.getFileEntryForID(SM.getFileID(SM.getExpansionLoc(Loc)));
292   if (!FE) return;
293 
294   StringRef Filename = FE->getName();
295   if (!FileMatchesDepCriteria(Filename.data(), FileType))
296     return;
297 
298   // Remove leading "./" (or ".//" or "././" etc.)
299   while (Filename.size() > 2 && Filename[0] == '.' &&
300          llvm::sys::path::is_separator(Filename[1])) {
301     Filename = Filename.substr(1);
302     while (llvm::sys::path::is_separator(Filename[0]))
303       Filename = Filename.substr(1);
304   }
305 
306   AddFilename(Filename);
307 }
308 
309 void DFGImpl::InclusionDirective(SourceLocation HashLoc,
310                                  const Token &IncludeTok,
311                                  StringRef FileName,
312                                  bool IsAngled,
313                                  CharSourceRange FilenameRange,
314                                  const FileEntry *File,
315                                  StringRef SearchPath,
316                                  StringRef RelativePath,
317                                  const Module *Imported) {
318   if (!File) {
319     if (AddMissingHeaderDeps)
320       AddFilename(FileName);
321     else
322       SeenMissingHeader = true;
323   }
324 }
325 
326 void DFGImpl::AddFilename(StringRef Filename) {
327   if (FilesSet.insert(Filename).second)
328     Files.push_back(Filename);
329 }
330 
331 /// Print the filename, with escaping or quoting that accommodates the three
332 /// most likely tools that use dependency files: GNU Make, BSD Make, and
333 /// NMake/Jom.
334 ///
335 /// BSD Make is the simplest case: It does no escaping at all.  This means
336 /// characters that are normally delimiters, i.e. space and # (the comment
337 /// character) simply aren't supported in filenames.
338 ///
339 /// GNU Make does allow space and # in filenames, but to avoid being treated
340 /// as a delimiter or comment, these must be escaped with a backslash. Because
341 /// backslash is itself the escape character, if a backslash appears in a
342 /// filename, it should be escaped as well.  (As a special case, $ is escaped
343 /// as $$, which is the normal Make way to handle the $ character.)
344 /// For compatibility with BSD Make and historical practice, if GNU Make
345 /// un-escapes characters in a filename but doesn't find a match, it will
346 /// retry with the unmodified original string.
347 ///
348 /// GCC tries to accommodate both Make formats by escaping any space or #
349 /// characters in the original filename, but not escaping backslashes.  The
350 /// apparent intent is so that filenames with backslashes will be handled
351 /// correctly by BSD Make, and by GNU Make in its fallback mode of using the
352 /// unmodified original string; filenames with # or space characters aren't
353 /// supported by BSD Make at all, but will be handled correctly by GNU Make
354 /// due to the escaping.
355 ///
356 /// A corner case that GCC gets only partly right is when the original filename
357 /// has a backslash immediately followed by space or #.  GNU Make would expect
358 /// this backslash to be escaped; however GCC escapes the original backslash
359 /// only when followed by space, not #.  It will therefore take a dependency
360 /// from a directive such as
361 ///     #include "a\ b\#c.h"
362 /// and emit it as
363 ///     a\\\ b\\#c.h
364 /// which GNU Make will interpret as
365 ///     a\ b\
366 /// followed by a comment. Failing to find this file, it will fall back to the
367 /// original string, which probably doesn't exist either; in any case it won't
368 /// find
369 ///     a\ b\#c.h
370 /// which is the actual filename specified by the include directive.
371 ///
372 /// Clang does what GCC does, rather than what GNU Make expects.
373 ///
374 /// NMake/Jom has a different set of scary characters, but wraps filespecs in
375 /// double-quotes to avoid misinterpreting them; see
376 /// https://msdn.microsoft.com/en-us/library/dd9y37ha.aspx for NMake info,
377 /// https://msdn.microsoft.com/en-us/library/windows/desktop/aa365247(v=vs.85).aspx
378 /// for Windows file-naming info.
379 static void PrintFilename(raw_ostream &OS, StringRef Filename,
380                           DependencyOutputFormat OutputFormat) {
381   if (OutputFormat == DependencyOutputFormat::NMake) {
382     // Add quotes if needed. These are the characters listed as "special" to
383     // NMake, that are legal in a Windows filespec, and that could cause
384     // misinterpretation of the dependency string.
385     if (Filename.find_first_of(" #${}^!") != StringRef::npos)
386       OS << '\"' << Filename << '\"';
387     else
388       OS << Filename;
389     return;
390   }
391   assert(OutputFormat == DependencyOutputFormat::Make);
392   for (unsigned i = 0, e = Filename.size(); i != e; ++i) {
393     if (Filename[i] == '#') // Handle '#' the broken gcc way.
394       OS << '\\';
395     else if (Filename[i] == ' ') { // Handle space correctly.
396       OS << '\\';
397       unsigned j = i;
398       while (j > 0 && Filename[--j] == '\\')
399         OS << '\\';
400     } else if (Filename[i] == '$') // $ is escaped by $$.
401       OS << '$';
402     OS << Filename[i];
403   }
404 }
405 
406 void DFGImpl::OutputDependencyFile() {
407   if (SeenMissingHeader) {
408     llvm::sys::fs::remove(OutputFile);
409     return;
410   }
411 
412   std::error_code EC;
413   llvm::raw_fd_ostream OS(OutputFile, EC, llvm::sys::fs::F_Text);
414   if (EC) {
415     PP->getDiagnostics().Report(diag::err_fe_error_opening) << OutputFile
416                                                             << EC.message();
417     return;
418   }
419 
420   // Write out the dependency targets, trying to avoid overly long
421   // lines when possible. We try our best to emit exactly the same
422   // dependency file as GCC (4.2), assuming the included files are the
423   // same.
424   const unsigned MaxColumns = 75;
425   unsigned Columns = 0;
426 
427   for (std::vector<std::string>::iterator
428          I = Targets.begin(), E = Targets.end(); I != E; ++I) {
429     unsigned N = I->length();
430     if (Columns == 0) {
431       Columns += N;
432     } else if (Columns + N + 2 > MaxColumns) {
433       Columns = N + 2;
434       OS << " \\\n  ";
435     } else {
436       Columns += N + 1;
437       OS << ' ';
438     }
439     // Targets already quoted as needed.
440     OS << *I;
441   }
442 
443   OS << ':';
444   Columns += 1;
445 
446   // Now add each dependency in the order it was seen, but avoiding
447   // duplicates.
448   for (std::vector<std::string>::iterator I = Files.begin(),
449          E = Files.end(); I != E; ++I) {
450     // Start a new line if this would exceed the column limit. Make
451     // sure to leave space for a trailing " \" in case we need to
452     // break the line on the next iteration.
453     unsigned N = I->length();
454     if (Columns + (N + 1) + 2 > MaxColumns) {
455       OS << " \\\n ";
456       Columns = 2;
457     }
458     OS << ' ';
459     PrintFilename(OS, *I, OutputFormat);
460     Columns += N + 1;
461   }
462   OS << '\n';
463 
464   // Create phony targets if requested.
465   if (PhonyTarget && !Files.empty()) {
466     // Skip the first entry, this is always the input file itself.
467     for (std::vector<std::string>::iterator I = Files.begin() + 1,
468            E = Files.end(); I != E; ++I) {
469       OS << '\n';
470       PrintFilename(OS, *I, OutputFormat);
471       OS << ":\n";
472     }
473   }
474 }
475 
476 bool DFGASTReaderListener::visitInputFile(llvm::StringRef Filename,
477                                           bool IsSystem, bool IsOverridden,
478                                           bool IsExplicitModule) {
479   assert(!IsSystem || needsSystemInputFileVisitation());
480   if (IsOverridden || IsExplicitModule)
481     return true;
482 
483   Parent.AddFilename(Filename);
484   return true;
485 }
486 
487 void DFGASTReaderListener::visitModuleFile(llvm::StringRef Filename,
488                                            serialization::ModuleKind Kind) {
489   if (Parent.includeModuleFiles())
490     Parent.AddFilename(Filename);
491 }
492