1 //===--- Tooling.cpp - Running clang standalone tools ---------------------===//
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 file implements functions to run clang tools standalone instead
11 //  of running them as a plugin.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "clang/Tooling/Tooling.h"
16 #include "clang/AST/ASTConsumer.h"
17 #include "clang/Driver/Compilation.h"
18 #include "clang/Driver/Driver.h"
19 #include "clang/Driver/Tool.h"
20 #include "clang/Frontend/ASTUnit.h"
21 #include "clang/Frontend/CompilerInstance.h"
22 #include "clang/Frontend/FrontendDiagnostic.h"
23 #include "clang/Frontend/TextDiagnosticPrinter.h"
24 #include "clang/Tooling/ArgumentsAdjusters.h"
25 #include "clang/Tooling/CompilationDatabase.h"
26 #include "llvm/ADT/STLExtras.h"
27 #include "llvm/Config/llvm-config.h"
28 #include "llvm/Option/Option.h"
29 #include "llvm/Support/Debug.h"
30 #include "llvm/Support/FileSystem.h"
31 #include "llvm/Support/Host.h"
32 #include "llvm/Support/raw_ostream.h"
33 
34 // For chdir, see the comment in ClangTool::run for more information.
35 #ifdef LLVM_ON_WIN32
36 #  include <direct.h>
37 #else
38 #  include <unistd.h>
39 #endif
40 
41 #define DEBUG_TYPE "clang-tooling"
42 
43 namespace clang {
44 namespace tooling {
45 
~ToolAction()46 ToolAction::~ToolAction() {}
47 
~FrontendActionFactory()48 FrontendActionFactory::~FrontendActionFactory() {}
49 
50 // FIXME: This file contains structural duplication with other parts of the
51 // code that sets up a compiler to run tools on it, and we should refactor
52 // it to be based on the same framework.
53 
54 /// \brief Builds a clang driver initialized for running clang tools.
newDriver(clang::DiagnosticsEngine * Diagnostics,const char * BinaryName)55 static clang::driver::Driver *newDriver(clang::DiagnosticsEngine *Diagnostics,
56                                         const char *BinaryName) {
57   clang::driver::Driver *CompilerDriver = new clang::driver::Driver(
58       BinaryName, llvm::sys::getDefaultTargetTriple(), *Diagnostics);
59   CompilerDriver->setTitle("clang_based_tool");
60   return CompilerDriver;
61 }
62 
63 /// \brief Retrieves the clang CC1 specific flags out of the compilation's jobs.
64 ///
65 /// Returns NULL on error.
getCC1Arguments(clang::DiagnosticsEngine * Diagnostics,clang::driver::Compilation * Compilation)66 static const llvm::opt::ArgStringList *getCC1Arguments(
67     clang::DiagnosticsEngine *Diagnostics,
68     clang::driver::Compilation *Compilation) {
69   // We expect to get back exactly one Command job, if we didn't something
70   // failed. Extract that job from the Compilation.
71   const clang::driver::JobList &Jobs = Compilation->getJobs();
72   if (Jobs.size() != 1 || !isa<clang::driver::Command>(*Jobs.begin())) {
73     SmallString<256> error_msg;
74     llvm::raw_svector_ostream error_stream(error_msg);
75     Jobs.Print(error_stream, "; ", true);
76     Diagnostics->Report(clang::diag::err_fe_expected_compiler_job)
77         << error_stream.str();
78     return nullptr;
79   }
80 
81   // The one job we find should be to invoke clang again.
82   const clang::driver::Command &Cmd =
83       cast<clang::driver::Command>(*Jobs.begin());
84   if (StringRef(Cmd.getCreator().getName()) != "clang") {
85     Diagnostics->Report(clang::diag::err_fe_expected_clang_command);
86     return nullptr;
87   }
88 
89   return &Cmd.getArguments();
90 }
91 
92 /// \brief Returns a clang build invocation initialized from the CC1 flags.
newInvocation(clang::DiagnosticsEngine * Diagnostics,const llvm::opt::ArgStringList & CC1Args)93 clang::CompilerInvocation *newInvocation(
94     clang::DiagnosticsEngine *Diagnostics,
95     const llvm::opt::ArgStringList &CC1Args) {
96   assert(!CC1Args.empty() && "Must at least contain the program name!");
97   clang::CompilerInvocation *Invocation = new clang::CompilerInvocation;
98   clang::CompilerInvocation::CreateFromArgs(
99       *Invocation, CC1Args.data() + 1, CC1Args.data() + CC1Args.size(),
100       *Diagnostics);
101   Invocation->getFrontendOpts().DisableFree = false;
102   Invocation->getCodeGenOpts().DisableFree = false;
103   Invocation->getDependencyOutputOpts() = DependencyOutputOptions();
104   return Invocation;
105 }
106 
runToolOnCode(clang::FrontendAction * ToolAction,const Twine & Code,const Twine & FileName)107 bool runToolOnCode(clang::FrontendAction *ToolAction, const Twine &Code,
108                    const Twine &FileName) {
109   return runToolOnCodeWithArgs(
110       ToolAction, Code, std::vector<std::string>(), FileName);
111 }
112 
113 static std::vector<std::string>
getSyntaxOnlyToolArgs(const std::vector<std::string> & ExtraArgs,StringRef FileName)114 getSyntaxOnlyToolArgs(const std::vector<std::string> &ExtraArgs,
115                       StringRef FileName) {
116   std::vector<std::string> Args;
117   Args.push_back("clang-tool");
118   Args.push_back("-fsyntax-only");
119   Args.insert(Args.end(), ExtraArgs.begin(), ExtraArgs.end());
120   Args.push_back(FileName.str());
121   return Args;
122 }
123 
runToolOnCodeWithArgs(clang::FrontendAction * ToolAction,const Twine & Code,const std::vector<std::string> & Args,const Twine & FileName,const FileContentMappings & VirtualMappedFiles)124 bool runToolOnCodeWithArgs(clang::FrontendAction *ToolAction, const Twine &Code,
125                            const std::vector<std::string> &Args,
126                            const Twine &FileName,
127                            const FileContentMappings &VirtualMappedFiles) {
128 
129   SmallString<16> FileNameStorage;
130   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
131   llvm::IntrusiveRefCntPtr<FileManager> Files(
132       new FileManager(FileSystemOptions()));
133   ToolInvocation Invocation(getSyntaxOnlyToolArgs(Args, FileNameRef),
134                             ToolAction, Files.get());
135 
136   SmallString<1024> CodeStorage;
137   Invocation.mapVirtualFile(FileNameRef,
138                             Code.toNullTerminatedStringRef(CodeStorage));
139 
140   for (auto &FilenameWithContent : VirtualMappedFiles) {
141     Invocation.mapVirtualFile(FilenameWithContent.first,
142                               FilenameWithContent.second);
143   }
144 
145   return Invocation.run();
146 }
147 
getAbsolutePath(StringRef File)148 std::string getAbsolutePath(StringRef File) {
149   StringRef RelativePath(File);
150   // FIXME: Should '.\\' be accepted on Win32?
151   if (RelativePath.startswith("./")) {
152     RelativePath = RelativePath.substr(strlen("./"));
153   }
154 
155   SmallString<1024> AbsolutePath = RelativePath;
156   std::error_code EC = llvm::sys::fs::make_absolute(AbsolutePath);
157   assert(!EC);
158   (void)EC;
159   llvm::sys::path::native(AbsolutePath);
160   return AbsolutePath.str();
161 }
162 
163 namespace {
164 
165 class SingleFrontendActionFactory : public FrontendActionFactory {
166   FrontendAction *Action;
167 
168 public:
SingleFrontendActionFactory(FrontendAction * Action)169   SingleFrontendActionFactory(FrontendAction *Action) : Action(Action) {}
170 
create()171   FrontendAction *create() override { return Action; }
172 };
173 
174 }
175 
ToolInvocation(std::vector<std::string> CommandLine,ToolAction * Action,FileManager * Files)176 ToolInvocation::ToolInvocation(std::vector<std::string> CommandLine,
177                                ToolAction *Action, FileManager *Files)
178     : CommandLine(std::move(CommandLine)),
179       Action(Action),
180       OwnsAction(false),
181       Files(Files),
182       DiagConsumer(nullptr) {}
183 
ToolInvocation(std::vector<std::string> CommandLine,FrontendAction * FAction,FileManager * Files)184 ToolInvocation::ToolInvocation(std::vector<std::string> CommandLine,
185                                FrontendAction *FAction, FileManager *Files)
186     : CommandLine(std::move(CommandLine)),
187       Action(new SingleFrontendActionFactory(FAction)),
188       OwnsAction(true),
189       Files(Files),
190       DiagConsumer(nullptr) {}
191 
~ToolInvocation()192 ToolInvocation::~ToolInvocation() {
193   if (OwnsAction)
194     delete Action;
195 }
196 
mapVirtualFile(StringRef FilePath,StringRef Content)197 void ToolInvocation::mapVirtualFile(StringRef FilePath, StringRef Content) {
198   SmallString<1024> PathStorage;
199   llvm::sys::path::native(FilePath, PathStorage);
200   MappedFileContents[PathStorage] = Content;
201 }
202 
run()203 bool ToolInvocation::run() {
204   std::vector<const char*> Argv;
205   for (const std::string &Str : CommandLine)
206     Argv.push_back(Str.c_str());
207   const char *const BinaryName = Argv[0];
208   IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
209   TextDiagnosticPrinter DiagnosticPrinter(
210       llvm::errs(), &*DiagOpts);
211   DiagnosticsEngine Diagnostics(
212       IntrusiveRefCntPtr<clang::DiagnosticIDs>(new DiagnosticIDs()), &*DiagOpts,
213       DiagConsumer ? DiagConsumer : &DiagnosticPrinter, false);
214 
215   const std::unique_ptr<clang::driver::Driver> Driver(
216       newDriver(&Diagnostics, BinaryName));
217   // Since the input might only be virtual, don't check whether it exists.
218   Driver->setCheckInputsExist(false);
219   const std::unique_ptr<clang::driver::Compilation> Compilation(
220       Driver->BuildCompilation(llvm::makeArrayRef(Argv)));
221   const llvm::opt::ArgStringList *const CC1Args = getCC1Arguments(
222       &Diagnostics, Compilation.get());
223   if (!CC1Args) {
224     return false;
225   }
226   std::unique_ptr<clang::CompilerInvocation> Invocation(
227       newInvocation(&Diagnostics, *CC1Args));
228   for (const auto &It : MappedFileContents) {
229     // Inject the code as the given file name into the preprocessor options.
230     std::unique_ptr<llvm::MemoryBuffer> Input =
231         llvm::MemoryBuffer::getMemBuffer(It.getValue());
232     Invocation->getPreprocessorOpts().addRemappedFile(It.getKey(),
233                                                       Input.release());
234   }
235   return runInvocation(BinaryName, Compilation.get(), Invocation.release());
236 }
237 
runInvocation(const char * BinaryName,clang::driver::Compilation * Compilation,clang::CompilerInvocation * Invocation)238 bool ToolInvocation::runInvocation(
239     const char *BinaryName,
240     clang::driver::Compilation *Compilation,
241     clang::CompilerInvocation *Invocation) {
242   // Show the invocation, with -v.
243   if (Invocation->getHeaderSearchOpts().Verbose) {
244     llvm::errs() << "clang Invocation:\n";
245     Compilation->getJobs().Print(llvm::errs(), "\n", true);
246     llvm::errs() << "\n";
247   }
248 
249   return Action->runInvocation(Invocation, Files, DiagConsumer);
250 }
251 
runInvocation(CompilerInvocation * Invocation,FileManager * Files,DiagnosticConsumer * DiagConsumer)252 bool FrontendActionFactory::runInvocation(CompilerInvocation *Invocation,
253                                           FileManager *Files,
254                                           DiagnosticConsumer *DiagConsumer) {
255   // Create a compiler instance to handle the actual work.
256   clang::CompilerInstance Compiler;
257   Compiler.setInvocation(Invocation);
258   Compiler.setFileManager(Files);
259 
260   // The FrontendAction can have lifetime requirements for Compiler or its
261   // members, and we need to ensure it's deleted earlier than Compiler. So we
262   // pass it to an std::unique_ptr declared after the Compiler variable.
263   std::unique_ptr<FrontendAction> ScopedToolAction(create());
264 
265   // Create the compiler's actual diagnostics engine.
266   Compiler.createDiagnostics(DiagConsumer, /*ShouldOwnClient=*/false);
267   if (!Compiler.hasDiagnostics())
268     return false;
269 
270   Compiler.createSourceManager(*Files);
271 
272   const bool Success = Compiler.ExecuteAction(*ScopedToolAction);
273 
274   Files->clearStatCaches();
275   return Success;
276 }
277 
ClangTool(const CompilationDatabase & Compilations,ArrayRef<std::string> SourcePaths)278 ClangTool::ClangTool(const CompilationDatabase &Compilations,
279                      ArrayRef<std::string> SourcePaths)
280     : Compilations(Compilations), SourcePaths(SourcePaths),
281       Files(new FileManager(FileSystemOptions())), DiagConsumer(nullptr) {
282   appendArgumentsAdjuster(getClangStripOutputAdjuster());
283   appendArgumentsAdjuster(getClangSyntaxOnlyAdjuster());
284 }
285 
~ClangTool()286 ClangTool::~ClangTool() {}
287 
mapVirtualFile(StringRef FilePath,StringRef Content)288 void ClangTool::mapVirtualFile(StringRef FilePath, StringRef Content) {
289   MappedFileContents.push_back(std::make_pair(FilePath, Content));
290 }
291 
appendArgumentsAdjuster(ArgumentsAdjuster Adjuster)292 void ClangTool::appendArgumentsAdjuster(ArgumentsAdjuster Adjuster) {
293   if (ArgsAdjuster)
294     ArgsAdjuster = combineAdjusters(ArgsAdjuster, Adjuster);
295   else
296     ArgsAdjuster = Adjuster;
297 }
298 
clearArgumentsAdjusters()299 void ClangTool::clearArgumentsAdjusters() {
300   ArgsAdjuster = nullptr;
301 }
302 
run(ToolAction * Action)303 int ClangTool::run(ToolAction *Action) {
304   // Exists solely for the purpose of lookup of the resource path.
305   // This just needs to be some symbol in the binary.
306   static int StaticSymbol;
307   // The driver detects the builtin header path based on the path of the
308   // executable.
309   // FIXME: On linux, GetMainExecutable is independent of the value of the
310   // first argument, thus allowing ClangTool and runToolOnCode to just
311   // pass in made-up names here. Make sure this works on other platforms.
312   std::string MainExecutable =
313       llvm::sys::fs::getMainExecutable("clang_tool", &StaticSymbol);
314 
315   llvm::SmallString<128> InitialDirectory;
316   if (std::error_code EC = llvm::sys::fs::current_path(InitialDirectory))
317     llvm::report_fatal_error("Cannot detect current path: " +
318                              Twine(EC.message()));
319   bool ProcessingFailed = false;
320   for (const auto &SourcePath : SourcePaths) {
321     std::string File(getAbsolutePath(SourcePath));
322 
323     // Currently implementations of CompilationDatabase::getCompileCommands can
324     // change the state of the file system (e.g.  prepare generated headers), so
325     // this method needs to run right before we invoke the tool, as the next
326     // file may require a different (incompatible) state of the file system.
327     //
328     // FIXME: Make the compilation database interface more explicit about the
329     // requirements to the order of invocation of its members.
330     std::vector<CompileCommand> CompileCommandsForFile =
331         Compilations.getCompileCommands(File);
332     if (CompileCommandsForFile.empty()) {
333       // FIXME: There are two use cases here: doing a fuzzy
334       // "find . -name '*.cc' |xargs tool" match, where as a user I don't care
335       // about the .cc files that were not found, and the use case where I
336       // specify all files I want to run over explicitly, where this should
337       // be an error. We'll want to add an option for this.
338       llvm::errs() << "Skipping " << File << ". Compile command not found.\n";
339       continue;
340     }
341     for (CompileCommand &CompileCommand : CompileCommandsForFile) {
342       // FIXME: chdir is thread hostile; on the other hand, creating the same
343       // behavior as chdir is complex: chdir resolves the path once, thus
344       // guaranteeing that all subsequent relative path operations work
345       // on the same path the original chdir resulted in. This makes a
346       // difference for example on network filesystems, where symlinks might be
347       // switched during runtime of the tool. Fixing this depends on having a
348       // file system abstraction that allows openat() style interactions.
349       if (chdir(CompileCommand.Directory.c_str()))
350         llvm::report_fatal_error("Cannot chdir into \"" +
351                                  Twine(CompileCommand.Directory) + "\n!");
352       std::vector<std::string> CommandLine = CompileCommand.CommandLine;
353       if (ArgsAdjuster)
354         CommandLine = ArgsAdjuster(CommandLine);
355       assert(!CommandLine.empty());
356       CommandLine[0] = MainExecutable;
357       // FIXME: We need a callback mechanism for the tool writer to output a
358       // customized message for each file.
359       DEBUG({ llvm::dbgs() << "Processing: " << File << ".\n"; });
360       ToolInvocation Invocation(std::move(CommandLine), Action, Files.get());
361       Invocation.setDiagnosticConsumer(DiagConsumer);
362       for (const auto &MappedFile : MappedFileContents)
363         Invocation.mapVirtualFile(MappedFile.first, MappedFile.second);
364       if (!Invocation.run()) {
365         // FIXME: Diagnostics should be used instead.
366         llvm::errs() << "Error while processing " << File << ".\n";
367         ProcessingFailed = true;
368       }
369       // Return to the initial directory to correctly resolve next file by
370       // relative path.
371       if (chdir(InitialDirectory.c_str()))
372         llvm::report_fatal_error("Cannot chdir into \"" +
373                                  Twine(InitialDirectory) + "\n!");
374     }
375   }
376   return ProcessingFailed ? 1 : 0;
377 }
378 
379 namespace {
380 
381 class ASTBuilderAction : public ToolAction {
382   std::vector<std::unique_ptr<ASTUnit>> &ASTs;
383 
384 public:
ASTBuilderAction(std::vector<std::unique_ptr<ASTUnit>> & ASTs)385   ASTBuilderAction(std::vector<std::unique_ptr<ASTUnit>> &ASTs) : ASTs(ASTs) {}
386 
runInvocation(CompilerInvocation * Invocation,FileManager * Files,DiagnosticConsumer * DiagConsumer)387   bool runInvocation(CompilerInvocation *Invocation, FileManager *Files,
388                      DiagnosticConsumer *DiagConsumer) override {
389     // FIXME: This should use the provided FileManager.
390     std::unique_ptr<ASTUnit> AST = ASTUnit::LoadFromCompilerInvocation(
391         Invocation, CompilerInstance::createDiagnostics(
392                         &Invocation->getDiagnosticOpts(), DiagConsumer,
393                         /*ShouldOwnClient=*/false));
394     if (!AST)
395       return false;
396 
397     ASTs.push_back(std::move(AST));
398     return true;
399   }
400 };
401 
402 }
403 
buildASTs(std::vector<std::unique_ptr<ASTUnit>> & ASTs)404 int ClangTool::buildASTs(std::vector<std::unique_ptr<ASTUnit>> &ASTs) {
405   ASTBuilderAction Action(ASTs);
406   return run(&Action);
407 }
408 
buildASTFromCode(const Twine & Code,const Twine & FileName)409 std::unique_ptr<ASTUnit> buildASTFromCode(const Twine &Code,
410                                           const Twine &FileName) {
411   return buildASTFromCodeWithArgs(Code, std::vector<std::string>(), FileName);
412 }
413 
414 std::unique_ptr<ASTUnit>
buildASTFromCodeWithArgs(const Twine & Code,const std::vector<std::string> & Args,const Twine & FileName)415 buildASTFromCodeWithArgs(const Twine &Code,
416                          const std::vector<std::string> &Args,
417                          const Twine &FileName) {
418   SmallString<16> FileNameStorage;
419   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
420 
421   std::vector<std::unique_ptr<ASTUnit>> ASTs;
422   ASTBuilderAction Action(ASTs);
423   ToolInvocation Invocation(getSyntaxOnlyToolArgs(Args, FileNameRef), &Action,
424                             nullptr);
425 
426   SmallString<1024> CodeStorage;
427   Invocation.mapVirtualFile(FileNameRef,
428                             Code.toNullTerminatedStringRef(CodeStorage));
429   if (!Invocation.run())
430     return nullptr;
431 
432   assert(ASTs.size() == 1);
433   return std::move(ASTs[0]);
434 }
435 
436 } // end namespace tooling
437 } // end namespace clang
438