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/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 namespace clang {
42 namespace tooling {
43 
44 ToolAction::~ToolAction() {}
45 
46 FrontendActionFactory::~FrontendActionFactory() {}
47 
48 // FIXME: This file contains structural duplication with other parts of the
49 // code that sets up a compiler to run tools on it, and we should refactor
50 // it to be based on the same framework.
51 
52 /// \brief Builds a clang driver initialized for running clang tools.
53 static clang::driver::Driver *newDriver(clang::DiagnosticsEngine *Diagnostics,
54                                         const char *BinaryName) {
55   const char *DefaultOutputName = "a.out";
56   clang::driver::Driver *CompilerDriver = new clang::driver::Driver(
57     BinaryName, llvm::sys::getDefaultTargetTriple(),
58     DefaultOutputName, *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.
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 NULL;
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 NULL;
87   }
88 
89   return &Cmd->getArguments();
90 }
91 
92 /// \brief Returns a clang build invocation initialized from the CC1 flags.
93 static 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 
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>
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 
124 bool runToolOnCodeWithArgs(clang::FrontendAction *ToolAction, const Twine &Code,
125                            const std::vector<std::string> &Args,
126                            const Twine &FileName) {
127   SmallString<16> FileNameStorage;
128   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
129   llvm::IntrusiveRefCntPtr<FileManager> Files(
130       new FileManager(FileSystemOptions()));
131   ToolInvocation Invocation(getSyntaxOnlyToolArgs(Args, FileNameRef), ToolAction,
132                             Files.getPtr());
133 
134   SmallString<1024> CodeStorage;
135   Invocation.mapVirtualFile(FileNameRef,
136                             Code.toNullTerminatedStringRef(CodeStorage));
137   return Invocation.run();
138 }
139 
140 std::string getAbsolutePath(StringRef File) {
141   StringRef RelativePath(File);
142   // FIXME: Should '.\\' be accepted on Win32?
143   if (RelativePath.startswith("./")) {
144     RelativePath = RelativePath.substr(strlen("./"));
145   }
146 
147   SmallString<1024> AbsolutePath = RelativePath;
148   llvm::error_code EC = llvm::sys::fs::make_absolute(AbsolutePath);
149   assert(!EC);
150   (void)EC;
151   llvm::sys::path::native(AbsolutePath);
152   return AbsolutePath.str();
153 }
154 
155 namespace {
156 
157 class SingleFrontendActionFactory : public FrontendActionFactory {
158   FrontendAction *Action;
159 
160 public:
161   SingleFrontendActionFactory(FrontendAction *Action) : Action(Action) {}
162 
163   FrontendAction *create() override { return Action; }
164 };
165 
166 }
167 
168 ToolInvocation::ToolInvocation(std::vector<std::string> CommandLine,
169                                ToolAction *Action, FileManager *Files)
170     : CommandLine(std::move(CommandLine)),
171       Action(Action),
172       OwnsAction(false),
173       Files(Files),
174       DiagConsumer(NULL) {}
175 
176 ToolInvocation::ToolInvocation(std::vector<std::string> CommandLine,
177                                FrontendAction *FAction, FileManager *Files)
178     : CommandLine(std::move(CommandLine)),
179       Action(new SingleFrontendActionFactory(FAction)),
180       OwnsAction(true),
181       Files(Files),
182       DiagConsumer(NULL) {}
183 
184 ToolInvocation::~ToolInvocation() {
185   if (OwnsAction)
186     delete Action;
187 }
188 
189 void ToolInvocation::setDiagnosticConsumer(DiagnosticConsumer *D) {
190   DiagConsumer = D;
191 }
192 
193 void ToolInvocation::mapVirtualFile(StringRef FilePath, StringRef Content) {
194   SmallString<1024> PathStorage;
195   llvm::sys::path::native(FilePath, PathStorage);
196   MappedFileContents[PathStorage] = Content;
197 }
198 
199 bool ToolInvocation::run() {
200   std::vector<const char*> Argv;
201   for (const std::string &Str : CommandLine)
202     Argv.push_back(Str.c_str());
203   const char *const BinaryName = Argv[0];
204   IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
205   TextDiagnosticPrinter DiagnosticPrinter(
206       llvm::errs(), &*DiagOpts);
207   DiagnosticsEngine Diagnostics(
208       IntrusiveRefCntPtr<clang::DiagnosticIDs>(new DiagnosticIDs()), &*DiagOpts,
209       DiagConsumer ? DiagConsumer : &DiagnosticPrinter, false);
210 
211   const std::unique_ptr<clang::driver::Driver> Driver(
212       newDriver(&Diagnostics, BinaryName));
213   // Since the input might only be virtual, don't check whether it exists.
214   Driver->setCheckInputsExist(false);
215   const std::unique_ptr<clang::driver::Compilation> Compilation(
216       Driver->BuildCompilation(llvm::makeArrayRef(Argv)));
217   const llvm::opt::ArgStringList *const CC1Args = getCC1Arguments(
218       &Diagnostics, Compilation.get());
219   if (CC1Args == NULL) {
220     return false;
221   }
222   std::unique_ptr<clang::CompilerInvocation> Invocation(
223       newInvocation(&Diagnostics, *CC1Args));
224   for (const auto &It : MappedFileContents) {
225     // Inject the code as the given file name into the preprocessor options.
226     auto *Input = llvm::MemoryBuffer::getMemBuffer(It.getValue());
227     Invocation->getPreprocessorOpts().addRemappedFile(It.getKey(), Input);
228   }
229   return runInvocation(BinaryName, Compilation.get(), Invocation.release());
230 }
231 
232 bool ToolInvocation::runInvocation(
233     const char *BinaryName,
234     clang::driver::Compilation *Compilation,
235     clang::CompilerInvocation *Invocation) {
236   // Show the invocation, with -v.
237   if (Invocation->getHeaderSearchOpts().Verbose) {
238     llvm::errs() << "clang Invocation:\n";
239     Compilation->getJobs().Print(llvm::errs(), "\n", true);
240     llvm::errs() << "\n";
241   }
242 
243   return Action->runInvocation(Invocation, Files, DiagConsumer);
244 }
245 
246 bool FrontendActionFactory::runInvocation(CompilerInvocation *Invocation,
247                                           FileManager *Files,
248                                           DiagnosticConsumer *DiagConsumer) {
249   // Create a compiler instance to handle the actual work.
250   clang::CompilerInstance Compiler;
251   Compiler.setInvocation(Invocation);
252   Compiler.setFileManager(Files);
253 
254   // The FrontendAction can have lifetime requirements for Compiler or its
255   // members, and we need to ensure it's deleted earlier than Compiler. So we
256   // pass it to an std::unique_ptr declared after the Compiler variable.
257   std::unique_ptr<FrontendAction> ScopedToolAction(create());
258 
259   // Create the compilers actual diagnostics engine.
260   Compiler.createDiagnostics(DiagConsumer, /*ShouldOwnClient=*/false);
261   if (!Compiler.hasDiagnostics())
262     return false;
263 
264   Compiler.createSourceManager(*Files);
265 
266   const bool Success = Compiler.ExecuteAction(*ScopedToolAction);
267 
268   Files->clearStatCaches();
269   return Success;
270 }
271 
272 ClangTool::ClangTool(const CompilationDatabase &Compilations,
273                      ArrayRef<std::string> SourcePaths)
274     : Files(new FileManager(FileSystemOptions())), DiagConsumer(NULL) {
275   ArgsAdjusters.push_back(new ClangStripOutputAdjuster());
276   ArgsAdjusters.push_back(new ClangSyntaxOnlyAdjuster());
277   for (const auto &SourcePath : SourcePaths) {
278     std::string File(getAbsolutePath(SourcePath));
279 
280     std::vector<CompileCommand> CompileCommandsForFile =
281       Compilations.getCompileCommands(File);
282     if (!CompileCommandsForFile.empty()) {
283       for (CompileCommand &CompileCommand : CompileCommandsForFile) {
284         CompileCommands.push_back(
285             std::make_pair(File, std::move(CompileCommand)));
286       }
287     } else {
288       // FIXME: There are two use cases here: doing a fuzzy
289       // "find . -name '*.cc' |xargs tool" match, where as a user I don't care
290       // about the .cc files that were not found, and the use case where I
291       // specify all files I want to run over explicitly, where this should
292       // be an error. We'll want to add an option for this.
293       llvm::outs() << "Skipping " << File << ". Command line not found.\n";
294     }
295   }
296 }
297 
298 void ClangTool::setDiagnosticConsumer(DiagnosticConsumer *D) {
299   DiagConsumer = D;
300 }
301 
302 void ClangTool::mapVirtualFile(StringRef FilePath, StringRef Content) {
303   MappedFileContents.push_back(std::make_pair(FilePath, Content));
304 }
305 
306 void ClangTool::setArgumentsAdjuster(ArgumentsAdjuster *Adjuster) {
307   clearArgumentsAdjusters();
308   appendArgumentsAdjuster(Adjuster);
309 }
310 
311 void ClangTool::appendArgumentsAdjuster(ArgumentsAdjuster *Adjuster) {
312   ArgsAdjusters.push_back(Adjuster);
313 }
314 
315 void ClangTool::clearArgumentsAdjusters() {
316   for (unsigned I = 0, E = ArgsAdjusters.size(); I != E; ++I)
317     delete ArgsAdjusters[I];
318   ArgsAdjusters.clear();
319 }
320 
321 int ClangTool::run(ToolAction *Action) {
322   // Exists solely for the purpose of lookup of the resource path.
323   // This just needs to be some symbol in the binary.
324   static int StaticSymbol;
325   // The driver detects the builtin header path based on the path of the
326   // executable.
327   // FIXME: On linux, GetMainExecutable is independent of the value of the
328   // first argument, thus allowing ClangTool and runToolOnCode to just
329   // pass in made-up names here. Make sure this works on other platforms.
330   std::string MainExecutable =
331       llvm::sys::fs::getMainExecutable("clang_tool", &StaticSymbol);
332 
333   bool ProcessingFailed = false;
334   for (const auto &Command : CompileCommands) {
335     // FIXME: chdir is thread hostile; on the other hand, creating the same
336     // behavior as chdir is complex: chdir resolves the path once, thus
337     // guaranteeing that all subsequent relative path operations work
338     // on the same path the original chdir resulted in. This makes a difference
339     // for example on network filesystems, where symlinks might be switched
340     // during runtime of the tool. Fixing this depends on having a file system
341     // abstraction that allows openat() style interactions.
342     if (chdir(Command.second.Directory.c_str()))
343       llvm::report_fatal_error("Cannot chdir into \"" +
344                                Twine(Command.second.Directory) + "\n!");
345     std::vector<std::string> CommandLine = Command.second.CommandLine;
346     for (ArgumentsAdjuster *Adjuster : ArgsAdjusters)
347       CommandLine = Adjuster->Adjust(CommandLine);
348     assert(!CommandLine.empty());
349     CommandLine[0] = MainExecutable;
350     // FIXME: We need a callback mechanism for the tool writer to output a
351     // customized message for each file.
352     DEBUG({
353       llvm::dbgs() << "Processing: " << Command.first << ".\n";
354     });
355     ToolInvocation Invocation(std::move(CommandLine), Action, Files.getPtr());
356     Invocation.setDiagnosticConsumer(DiagConsumer);
357     for (const auto &MappedFile : MappedFileContents) {
358       Invocation.mapVirtualFile(MappedFile.first, MappedFile.second);
359     }
360     if (!Invocation.run()) {
361       // FIXME: Diagnostics should be used instead.
362       llvm::errs() << "Error while processing " << Command.first << ".\n";
363       ProcessingFailed = true;
364     }
365   }
366   return ProcessingFailed ? 1 : 0;
367 }
368 
369 namespace {
370 
371 class ASTBuilderAction : public ToolAction {
372   std::vector<ASTUnit *> &ASTs;
373 
374 public:
375   ASTBuilderAction(std::vector<ASTUnit *> &ASTs) : ASTs(ASTs) {}
376 
377   bool runInvocation(CompilerInvocation *Invocation, FileManager *Files,
378                      DiagnosticConsumer *DiagConsumer) override {
379     // FIXME: This should use the provided FileManager.
380     ASTUnit *AST = ASTUnit::LoadFromCompilerInvocation(
381         Invocation, CompilerInstance::createDiagnostics(
382                         &Invocation->getDiagnosticOpts(), DiagConsumer,
383                         /*ShouldOwnClient=*/false));
384     if (!AST)
385       return false;
386 
387     ASTs.push_back(AST);
388     return true;
389   }
390 };
391 
392 }
393 
394 int ClangTool::buildASTs(std::vector<ASTUnit *> &ASTs) {
395   ASTBuilderAction Action(ASTs);
396   return run(&Action);
397 }
398 
399 ASTUnit *buildASTFromCode(const Twine &Code, const Twine &FileName) {
400   return buildASTFromCodeWithArgs(Code, std::vector<std::string>(), FileName);
401 }
402 
403 ASTUnit *buildASTFromCodeWithArgs(const Twine &Code,
404                                   const std::vector<std::string> &Args,
405                                   const Twine &FileName) {
406   SmallString<16> FileNameStorage;
407   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
408 
409   std::vector<ASTUnit *> ASTs;
410   ASTBuilderAction Action(ASTs);
411   ToolInvocation Invocation(getSyntaxOnlyToolArgs(Args, FileNameRef), &Action, 0);
412 
413   SmallString<1024> CodeStorage;
414   Invocation.mapVirtualFile(FileNameRef,
415                             Code.toNullTerminatedStringRef(CodeStorage));
416   if (!Invocation.run())
417     return 0;
418 
419   assert(ASTs.size() == 1);
420   return ASTs[0];
421 }
422 
423 } // end namespace tooling
424 } // end namespace clang
425