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/Driver/ToolChain.h"
21 #include "clang/Frontend/ASTUnit.h"
22 #include "clang/Frontend/CompilerInstance.h"
23 #include "clang/Frontend/FrontendDiagnostic.h"
24 #include "clang/Frontend/TextDiagnosticPrinter.h"
25 #include "clang/Tooling/ArgumentsAdjusters.h"
26 #include "clang/Tooling/CompilationDatabase.h"
27 #include "llvm/ADT/STLExtras.h"
28 #include "llvm/Config/llvm-config.h"
29 #include "llvm/Option/Option.h"
30 #include "llvm/Support/Debug.h"
31 #include "llvm/Support/FileSystem.h"
32 #include "llvm/Support/Host.h"
33 #include "llvm/Support/raw_ostream.h"
34 
35 #define DEBUG_TYPE "clang-tooling"
36 
37 namespace clang {
38 namespace tooling {
39 
40 ToolAction::~ToolAction() {}
41 
42 FrontendActionFactory::~FrontendActionFactory() {}
43 
44 // FIXME: This file contains structural duplication with other parts of the
45 // code that sets up a compiler to run tools on it, and we should refactor
46 // it to be based on the same framework.
47 
48 /// \brief Builds a clang driver initialized for running clang tools.
49 static clang::driver::Driver *newDriver(
50     clang::DiagnosticsEngine *Diagnostics, const char *BinaryName,
51     IntrusiveRefCntPtr<vfs::FileSystem> VFS) {
52   clang::driver::Driver *CompilerDriver = new clang::driver::Driver(
53       BinaryName, llvm::sys::getDefaultTargetTriple(), *Diagnostics, VFS);
54   CompilerDriver->setTitle("clang_based_tool");
55   return CompilerDriver;
56 }
57 
58 /// \brief Retrieves the clang CC1 specific flags out of the compilation's jobs.
59 ///
60 /// Returns NULL on error.
61 static const llvm::opt::ArgStringList *getCC1Arguments(
62     clang::DiagnosticsEngine *Diagnostics,
63     clang::driver::Compilation *Compilation) {
64   // We expect to get back exactly one Command job, if we didn't something
65   // failed. Extract that job from the Compilation.
66   const clang::driver::JobList &Jobs = Compilation->getJobs();
67   if (Jobs.size() != 1 || !isa<clang::driver::Command>(*Jobs.begin())) {
68     SmallString<256> error_msg;
69     llvm::raw_svector_ostream error_stream(error_msg);
70     Jobs.Print(error_stream, "; ", true);
71     Diagnostics->Report(clang::diag::err_fe_expected_compiler_job)
72         << error_stream.str();
73     return nullptr;
74   }
75 
76   // The one job we find should be to invoke clang again.
77   const clang::driver::Command &Cmd =
78       cast<clang::driver::Command>(*Jobs.begin());
79   if (StringRef(Cmd.getCreator().getName()) != "clang") {
80     Diagnostics->Report(clang::diag::err_fe_expected_clang_command);
81     return nullptr;
82   }
83 
84   return &Cmd.getArguments();
85 }
86 
87 /// \brief Returns a clang build invocation initialized from the CC1 flags.
88 clang::CompilerInvocation *newInvocation(
89     clang::DiagnosticsEngine *Diagnostics,
90     const llvm::opt::ArgStringList &CC1Args) {
91   assert(!CC1Args.empty() && "Must at least contain the program name!");
92   clang::CompilerInvocation *Invocation = new clang::CompilerInvocation;
93   clang::CompilerInvocation::CreateFromArgs(
94       *Invocation, CC1Args.data() + 1, CC1Args.data() + CC1Args.size(),
95       *Diagnostics);
96   Invocation->getFrontendOpts().DisableFree = false;
97   Invocation->getCodeGenOpts().DisableFree = false;
98   Invocation->getDependencyOutputOpts() = DependencyOutputOptions();
99   return Invocation;
100 }
101 
102 bool runToolOnCode(clang::FrontendAction *ToolAction, const Twine &Code,
103                    const Twine &FileName,
104                    std::shared_ptr<PCHContainerOperations> PCHContainerOps) {
105   return runToolOnCodeWithArgs(ToolAction, Code, std::vector<std::string>(),
106                                FileName, "clang-tool", PCHContainerOps);
107 }
108 
109 static std::vector<std::string>
110 getSyntaxOnlyToolArgs(const Twine &ToolName,
111                       const std::vector<std::string> &ExtraArgs,
112                       StringRef FileName) {
113   std::vector<std::string> Args;
114   Args.push_back(ToolName.str());
115   Args.push_back("-fsyntax-only");
116   Args.insert(Args.end(), ExtraArgs.begin(), ExtraArgs.end());
117   Args.push_back(FileName.str());
118   return Args;
119 }
120 
121 bool runToolOnCodeWithArgs(
122     clang::FrontendAction *ToolAction, const Twine &Code,
123     const std::vector<std::string> &Args, const Twine &FileName,
124     const Twine &ToolName,
125     std::shared_ptr<PCHContainerOperations> PCHContainerOps,
126     const FileContentMappings &VirtualMappedFiles) {
127 
128   SmallString<16> FileNameStorage;
129   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
130   llvm::IntrusiveRefCntPtr<vfs::OverlayFileSystem> OverlayFileSystem(
131       new vfs::OverlayFileSystem(vfs::getRealFileSystem()));
132   llvm::IntrusiveRefCntPtr<vfs::InMemoryFileSystem> InMemoryFileSystem(
133       new vfs::InMemoryFileSystem);
134   OverlayFileSystem->pushOverlay(InMemoryFileSystem);
135   llvm::IntrusiveRefCntPtr<FileManager> Files(
136       new FileManager(FileSystemOptions(), OverlayFileSystem));
137   ToolInvocation Invocation(getSyntaxOnlyToolArgs(ToolName, Args, FileNameRef),
138                             ToolAction, Files.get(), PCHContainerOps);
139 
140   SmallString<1024> CodeStorage;
141   InMemoryFileSystem->addFile(FileNameRef, 0,
142                               llvm::MemoryBuffer::getMemBuffer(
143                                   Code.toNullTerminatedStringRef(CodeStorage)));
144 
145   for (auto &FilenameWithContent : VirtualMappedFiles) {
146     InMemoryFileSystem->addFile(
147         FilenameWithContent.first, 0,
148         llvm::MemoryBuffer::getMemBuffer(FilenameWithContent.second));
149   }
150 
151   return Invocation.run();
152 }
153 
154 std::string getAbsolutePath(StringRef File) {
155   StringRef RelativePath(File);
156   // FIXME: Should '.\\' be accepted on Win32?
157   if (RelativePath.startswith("./")) {
158     RelativePath = RelativePath.substr(strlen("./"));
159   }
160 
161   SmallString<1024> AbsolutePath = RelativePath;
162   std::error_code EC = llvm::sys::fs::make_absolute(AbsolutePath);
163   assert(!EC);
164   (void)EC;
165   llvm::sys::path::native(AbsolutePath);
166   return AbsolutePath.str();
167 }
168 
169 void addTargetAndModeForProgramName(std::vector<std::string> &CommandLine,
170                                     StringRef InvokedAs) {
171   if (!CommandLine.empty() && !InvokedAs.empty()) {
172     bool AlreadyHasTarget = false;
173     bool AlreadyHasMode = false;
174     // Skip CommandLine[0].
175     for (auto Token = ++CommandLine.begin(); Token != CommandLine.end();
176          ++Token) {
177       StringRef TokenRef(*Token);
178       AlreadyHasTarget |=
179           (TokenRef == "-target" || TokenRef.startswith("-target="));
180       AlreadyHasMode |= (TokenRef == "--driver-mode" ||
181                          TokenRef.startswith("--driver-mode="));
182     }
183     auto TargetMode =
184         clang::driver::ToolChain::getTargetAndModeFromProgramName(InvokedAs);
185     if (!AlreadyHasMode && !TargetMode.second.empty()) {
186       CommandLine.insert(++CommandLine.begin(), TargetMode.second);
187     }
188     if (!AlreadyHasTarget && !TargetMode.first.empty()) {
189       CommandLine.insert(++CommandLine.begin(), {"-target", TargetMode.first});
190     }
191   }
192 }
193 
194 namespace {
195 
196 class SingleFrontendActionFactory : public FrontendActionFactory {
197   FrontendAction *Action;
198 
199 public:
200   SingleFrontendActionFactory(FrontendAction *Action) : Action(Action) {}
201 
202   FrontendAction *create() override { return Action; }
203 };
204 
205 }
206 
207 ToolInvocation::ToolInvocation(
208     std::vector<std::string> CommandLine, ToolAction *Action,
209     FileManager *Files, std::shared_ptr<PCHContainerOperations> PCHContainerOps)
210     : CommandLine(std::move(CommandLine)), Action(Action), OwnsAction(false),
211       Files(Files), PCHContainerOps(PCHContainerOps), DiagConsumer(nullptr) {}
212 
213 ToolInvocation::ToolInvocation(
214     std::vector<std::string> CommandLine, FrontendAction *FAction,
215     FileManager *Files, std::shared_ptr<PCHContainerOperations> PCHContainerOps)
216     : CommandLine(std::move(CommandLine)),
217       Action(new SingleFrontendActionFactory(FAction)), OwnsAction(true),
218       Files(Files), PCHContainerOps(PCHContainerOps), DiagConsumer(nullptr) {}
219 
220 ToolInvocation::~ToolInvocation() {
221   if (OwnsAction)
222     delete Action;
223 }
224 
225 void ToolInvocation::mapVirtualFile(StringRef FilePath, StringRef Content) {
226   SmallString<1024> PathStorage;
227   llvm::sys::path::native(FilePath, PathStorage);
228   MappedFileContents[PathStorage] = Content;
229 }
230 
231 bool ToolInvocation::run() {
232   std::vector<const char*> Argv;
233   for (const std::string &Str : CommandLine)
234     Argv.push_back(Str.c_str());
235   const char *const BinaryName = Argv[0];
236   IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
237   TextDiagnosticPrinter DiagnosticPrinter(
238       llvm::errs(), &*DiagOpts);
239   DiagnosticsEngine Diagnostics(
240       IntrusiveRefCntPtr<clang::DiagnosticIDs>(new DiagnosticIDs()), &*DiagOpts,
241       DiagConsumer ? DiagConsumer : &DiagnosticPrinter, false);
242 
243   const std::unique_ptr<clang::driver::Driver> Driver(
244       newDriver(&Diagnostics, BinaryName, Files->getVirtualFileSystem()));
245   // Since the input might only be virtual, don't check whether it exists.
246   Driver->setCheckInputsExist(false);
247   const std::unique_ptr<clang::driver::Compilation> Compilation(
248       Driver->BuildCompilation(llvm::makeArrayRef(Argv)));
249   const llvm::opt::ArgStringList *const CC1Args = getCC1Arguments(
250       &Diagnostics, Compilation.get());
251   if (!CC1Args) {
252     return false;
253   }
254   std::unique_ptr<clang::CompilerInvocation> Invocation(
255       newInvocation(&Diagnostics, *CC1Args));
256   // FIXME: remove this when all users have migrated!
257   for (const auto &It : MappedFileContents) {
258     // Inject the code as the given file name into the preprocessor options.
259     std::unique_ptr<llvm::MemoryBuffer> Input =
260         llvm::MemoryBuffer::getMemBuffer(It.getValue());
261     Invocation->getPreprocessorOpts().addRemappedFile(It.getKey(),
262                                                       Input.release());
263   }
264   return runInvocation(BinaryName, Compilation.get(), Invocation.release(),
265                        PCHContainerOps);
266 }
267 
268 bool ToolInvocation::runInvocation(
269     const char *BinaryName, clang::driver::Compilation *Compilation,
270     clang::CompilerInvocation *Invocation,
271     std::shared_ptr<PCHContainerOperations> PCHContainerOps) {
272   // Show the invocation, with -v.
273   if (Invocation->getHeaderSearchOpts().Verbose) {
274     llvm::errs() << "clang Invocation:\n";
275     Compilation->getJobs().Print(llvm::errs(), "\n", true);
276     llvm::errs() << "\n";
277   }
278 
279   return Action->runInvocation(Invocation, Files, PCHContainerOps,
280                                DiagConsumer);
281 }
282 
283 bool FrontendActionFactory::runInvocation(
284     CompilerInvocation *Invocation, FileManager *Files,
285     std::shared_ptr<PCHContainerOperations> PCHContainerOps,
286     DiagnosticConsumer *DiagConsumer) {
287   // Create a compiler instance to handle the actual work.
288   clang::CompilerInstance Compiler(PCHContainerOps);
289   Compiler.setInvocation(Invocation);
290   Compiler.setFileManager(Files);
291 
292   // The FrontendAction can have lifetime requirements for Compiler or its
293   // members, and we need to ensure it's deleted earlier than Compiler. So we
294   // pass it to an std::unique_ptr declared after the Compiler variable.
295   std::unique_ptr<FrontendAction> ScopedToolAction(create());
296 
297   // Create the compiler's actual diagnostics engine.
298   Compiler.createDiagnostics(DiagConsumer, /*ShouldOwnClient=*/false);
299   if (!Compiler.hasDiagnostics())
300     return false;
301 
302   Compiler.createSourceManager(*Files);
303 
304   const bool Success = Compiler.ExecuteAction(*ScopedToolAction);
305 
306   Files->clearStatCaches();
307   return Success;
308 }
309 
310 ClangTool::ClangTool(const CompilationDatabase &Compilations,
311                      ArrayRef<std::string> SourcePaths,
312                      std::shared_ptr<PCHContainerOperations> PCHContainerOps)
313     : Compilations(Compilations), SourcePaths(SourcePaths),
314       PCHContainerOps(PCHContainerOps),
315       OverlayFileSystem(new vfs::OverlayFileSystem(vfs::getRealFileSystem())),
316       InMemoryFileSystem(new vfs::InMemoryFileSystem),
317       Files(new FileManager(FileSystemOptions(), OverlayFileSystem)),
318       DiagConsumer(nullptr) {
319   OverlayFileSystem->pushOverlay(InMemoryFileSystem);
320   appendArgumentsAdjuster(getClangStripOutputAdjuster());
321   appendArgumentsAdjuster(getClangSyntaxOnlyAdjuster());
322 }
323 
324 ClangTool::~ClangTool() {}
325 
326 void ClangTool::mapVirtualFile(StringRef FilePath, StringRef Content) {
327   MappedFileContents.push_back(std::make_pair(FilePath, Content));
328 }
329 
330 void ClangTool::appendArgumentsAdjuster(ArgumentsAdjuster Adjuster) {
331   if (ArgsAdjuster)
332     ArgsAdjuster = combineAdjusters(ArgsAdjuster, Adjuster);
333   else
334     ArgsAdjuster = Adjuster;
335 }
336 
337 void ClangTool::clearArgumentsAdjusters() {
338   ArgsAdjuster = nullptr;
339 }
340 
341 static void injectResourceDir(CommandLineArguments &Args, const char *Argv0,
342                               void *MainAddr) {
343   // Allow users to override the resource dir.
344   for (StringRef Arg : Args)
345     if (Arg.startswith("-resource-dir"))
346       return;
347 
348   // If there's no override in place add our resource dir.
349   Args.push_back("-resource-dir=" +
350                  CompilerInvocation::GetResourcesPath(Argv0, MainAddr));
351 }
352 
353 int ClangTool::run(ToolAction *Action) {
354   // Exists solely for the purpose of lookup of the resource path.
355   // This just needs to be some symbol in the binary.
356   static int StaticSymbol;
357 
358   llvm::SmallString<128> InitialDirectory;
359   if (std::error_code EC = llvm::sys::fs::current_path(InitialDirectory))
360     llvm::report_fatal_error("Cannot detect current path: " +
361                              Twine(EC.message()));
362 
363   // First insert all absolute paths into the in-memory VFS. These are global
364   // for all compile commands.
365   if (SeenWorkingDirectories.insert("/").second)
366     for (const auto &MappedFile : MappedFileContents)
367       if (llvm::sys::path::is_absolute(MappedFile.first))
368         InMemoryFileSystem->addFile(
369             MappedFile.first, 0,
370             llvm::MemoryBuffer::getMemBuffer(MappedFile.second));
371 
372   bool ProcessingFailed = false;
373   for (const auto &SourcePath : SourcePaths) {
374     std::string File(getAbsolutePath(SourcePath));
375 
376     // Currently implementations of CompilationDatabase::getCompileCommands can
377     // change the state of the file system (e.g.  prepare generated headers), so
378     // this method needs to run right before we invoke the tool, as the next
379     // file may require a different (incompatible) state of the file system.
380     //
381     // FIXME: Make the compilation database interface more explicit about the
382     // requirements to the order of invocation of its members.
383     std::vector<CompileCommand> CompileCommandsForFile =
384         Compilations.getCompileCommands(File);
385     if (CompileCommandsForFile.empty()) {
386       // FIXME: There are two use cases here: doing a fuzzy
387       // "find . -name '*.cc' |xargs tool" match, where as a user I don't care
388       // about the .cc files that were not found, and the use case where I
389       // specify all files I want to run over explicitly, where this should
390       // be an error. We'll want to add an option for this.
391       llvm::errs() << "Skipping " << File << ". Compile command not found.\n";
392       continue;
393     }
394     for (CompileCommand &CompileCommand : CompileCommandsForFile) {
395       // FIXME: chdir is thread hostile; on the other hand, creating the same
396       // behavior as chdir is complex: chdir resolves the path once, thus
397       // guaranteeing that all subsequent relative path operations work
398       // on the same path the original chdir resulted in. This makes a
399       // difference for example on network filesystems, where symlinks might be
400       // switched during runtime of the tool. Fixing this depends on having a
401       // file system abstraction that allows openat() style interactions.
402       if (OverlayFileSystem->setCurrentWorkingDirectory(
403               CompileCommand.Directory))
404         llvm::report_fatal_error("Cannot chdir into \"" +
405                                  Twine(CompileCommand.Directory) + "\n!");
406 
407       // Now fill the in-memory VFS with the relative file mappings so it will
408       // have the correct relative paths. We never remove mappings but that
409       // should be fine.
410       if (SeenWorkingDirectories.insert(CompileCommand.Directory).second)
411         for (const auto &MappedFile : MappedFileContents)
412           if (!llvm::sys::path::is_absolute(MappedFile.first))
413             InMemoryFileSystem->addFile(
414                 MappedFile.first, 0,
415                 llvm::MemoryBuffer::getMemBuffer(MappedFile.second));
416 
417       std::vector<std::string> CommandLine = CompileCommand.CommandLine;
418       if (ArgsAdjuster)
419         CommandLine = ArgsAdjuster(CommandLine, CompileCommand.Filename);
420       assert(!CommandLine.empty());
421 
422       // Add the resource dir based on the binary of this tool. argv[0] in the
423       // compilation database may refer to a different compiler and we want to
424       // pick up the very same standard library that compiler is using. The
425       // builtin headers in the resource dir need to match the exact clang
426       // version the tool is using.
427       // FIXME: On linux, GetMainExecutable is independent of the value of the
428       // first argument, thus allowing ClangTool and runToolOnCode to just
429       // pass in made-up names here. Make sure this works on other platforms.
430       injectResourceDir(CommandLine, "clang_tool", &StaticSymbol);
431 
432       // FIXME: We need a callback mechanism for the tool writer to output a
433       // customized message for each file.
434       DEBUG({ llvm::dbgs() << "Processing: " << File << ".\n"; });
435       ToolInvocation Invocation(std::move(CommandLine), Action, Files.get(),
436                                 PCHContainerOps);
437       Invocation.setDiagnosticConsumer(DiagConsumer);
438 
439       if (!Invocation.run()) {
440         // FIXME: Diagnostics should be used instead.
441         llvm::errs() << "Error while processing " << File << ".\n";
442         ProcessingFailed = true;
443       }
444       // Return to the initial directory to correctly resolve next file by
445       // relative path.
446       if (OverlayFileSystem->setCurrentWorkingDirectory(InitialDirectory.c_str()))
447         llvm::report_fatal_error("Cannot chdir into \"" +
448                                  Twine(InitialDirectory) + "\n!");
449     }
450   }
451   return ProcessingFailed ? 1 : 0;
452 }
453 
454 namespace {
455 
456 class ASTBuilderAction : public ToolAction {
457   std::vector<std::unique_ptr<ASTUnit>> &ASTs;
458 
459 public:
460   ASTBuilderAction(std::vector<std::unique_ptr<ASTUnit>> &ASTs) : ASTs(ASTs) {}
461 
462   bool runInvocation(CompilerInvocation *Invocation, FileManager *Files,
463                      std::shared_ptr<PCHContainerOperations> PCHContainerOps,
464                      DiagnosticConsumer *DiagConsumer) override {
465     std::unique_ptr<ASTUnit> AST = ASTUnit::LoadFromCompilerInvocation(
466         Invocation, PCHContainerOps,
467         CompilerInstance::createDiagnostics(&Invocation->getDiagnosticOpts(),
468                                             DiagConsumer,
469                                             /*ShouldOwnClient=*/false),
470         Files);
471     if (!AST)
472       return false;
473 
474     ASTs.push_back(std::move(AST));
475     return true;
476   }
477 };
478 
479 }
480 
481 int ClangTool::buildASTs(std::vector<std::unique_ptr<ASTUnit>> &ASTs) {
482   ASTBuilderAction Action(ASTs);
483   return run(&Action);
484 }
485 
486 std::unique_ptr<ASTUnit>
487 buildASTFromCode(const Twine &Code, const Twine &FileName,
488                  std::shared_ptr<PCHContainerOperations> PCHContainerOps) {
489   return buildASTFromCodeWithArgs(Code, std::vector<std::string>(), FileName,
490                                   "clang-tool", PCHContainerOps);
491 }
492 
493 std::unique_ptr<ASTUnit> buildASTFromCodeWithArgs(
494     const Twine &Code, const std::vector<std::string> &Args,
495     const Twine &FileName, const Twine &ToolName,
496     std::shared_ptr<PCHContainerOperations> PCHContainerOps) {
497   SmallString<16> FileNameStorage;
498   StringRef FileNameRef = FileName.toNullTerminatedStringRef(FileNameStorage);
499 
500   std::vector<std::unique_ptr<ASTUnit>> ASTs;
501   ASTBuilderAction Action(ASTs);
502   llvm::IntrusiveRefCntPtr<vfs::OverlayFileSystem> OverlayFileSystem(
503       new vfs::OverlayFileSystem(vfs::getRealFileSystem()));
504   llvm::IntrusiveRefCntPtr<vfs::InMemoryFileSystem> InMemoryFileSystem(
505       new vfs::InMemoryFileSystem);
506   OverlayFileSystem->pushOverlay(InMemoryFileSystem);
507   llvm::IntrusiveRefCntPtr<FileManager> Files(
508       new FileManager(FileSystemOptions(), OverlayFileSystem));
509   ToolInvocation Invocation(getSyntaxOnlyToolArgs(ToolName, Args, FileNameRef),
510                             &Action, Files.get(), PCHContainerOps);
511 
512   SmallString<1024> CodeStorage;
513   InMemoryFileSystem->addFile(FileNameRef, 0,
514                               llvm::MemoryBuffer::getMemBuffer(
515                                   Code.toNullTerminatedStringRef(CodeStorage)));
516   if (!Invocation.run())
517     return nullptr;
518 
519   assert(ASTs.size() == 1);
520   return std::move(ASTs[0]);
521 }
522 
523 } // end namespace tooling
524 } // end namespace clang
525