1//===- JSONCompilationDatabase.cpp ----------------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9//  This file contains the implementation of the JSONCompilationDatabase.
10//
11//===----------------------------------------------------------------------===//
12
13#include "clang/Tooling/JSONCompilationDatabase.h"
14#include "clang/Basic/LLVM.h"
15#include "clang/Tooling/CompilationDatabase.h"
16#include "clang/Tooling/CompilationDatabasePluginRegistry.h"
17#include "clang/Tooling/Tooling.h"
18#include "llvm/ADT/Optional.h"
19#include "llvm/ADT/STLExtras.h"
20#include "llvm/ADT/SmallString.h"
21#include "llvm/ADT/SmallVector.h"
22#include "llvm/ADT/StringRef.h"
23#include "llvm/ADT/Triple.h"
24#include "llvm/Support/Allocator.h"
25#include "llvm/Support/Casting.h"
26#include "llvm/Support/CommandLine.h"
27#include "llvm/Support/ErrorOr.h"
28#include "llvm/Support/Host.h"
29#include "llvm/Support/MemoryBuffer.h"
30#include "llvm/Support/Path.h"
31#include "llvm/Support/StringSaver.h"
32#include "llvm/Support/VirtualFileSystem.h"
33#include "llvm/Support/YAMLParser.h"
34#include "llvm/Support/raw_ostream.h"
35#include <cassert>
36#include <memory>
37#include <string>
38#include <system_error>
39#include <tuple>
40#include <utility>
41#include <vector>
42
43using namespace clang;
44using namespace tooling;
45
46namespace {
47
48/// A parser for escaped strings of command line arguments.
49///
50/// Assumes \-escaping for quoted arguments (see the documentation of
51/// unescapeCommandLine(...)).
52class CommandLineArgumentParser {
53 public:
54  CommandLineArgumentParser(StringRef CommandLine)
55      : Input(CommandLine), Position(Input.begin()-1) {}
56
57  std::vector<std::string> parse() {
58    bool HasMoreInput = true;
59    while (HasMoreInput && nextNonWhitespace()) {
60      std::string Argument;
61      HasMoreInput = parseStringInto(Argument);
62      CommandLine.push_back(Argument);
63    }
64    return CommandLine;
65  }
66
67 private:
68  // All private methods return true if there is more input available.
69
70  bool parseStringInto(std::string &String) {
71    do {
72      if (*Position == '"') {
73        if (!parseDoubleQuotedStringInto(String)) return false;
74      } else if (*Position == '\'') {
75        if (!parseSingleQuotedStringInto(String)) return false;
76      } else {
77        if (!parseFreeStringInto(String)) return false;
78      }
79    } while (*Position != ' ');
80    return true;
81  }
82
83  bool parseDoubleQuotedStringInto(std::string &String) {
84    if (!next()) return false;
85    while (*Position != '"') {
86      if (!skipEscapeCharacter()) return false;
87      String.push_back(*Position);
88      if (!next()) return false;
89    }
90    return next();
91  }
92
93  bool parseSingleQuotedStringInto(std::string &String) {
94    if (!next()) return false;
95    while (*Position != '\'') {
96      String.push_back(*Position);
97      if (!next()) return false;
98    }
99    return next();
100  }
101
102  bool parseFreeStringInto(std::string &String) {
103    do {
104      if (!skipEscapeCharacter()) return false;
105      String.push_back(*Position);
106      if (!next()) return false;
107    } while (*Position != ' ' && *Position != '"' && *Position != '\'');
108    return true;
109  }
110
111  bool skipEscapeCharacter() {
112    if (*Position == '\\') {
113      return next();
114    }
115    return true;
116  }
117
118  bool nextNonWhitespace() {
119    do {
120      if (!next()) return false;
121    } while (*Position == ' ');
122    return true;
123  }
124
125  bool next() {
126    ++Position;
127    return Position != Input.end();
128  }
129
130  const StringRef Input;
131  StringRef::iterator Position;
132  std::vector<std::string> CommandLine;
133};
134
135std::vector<std::string> unescapeCommandLine(JSONCommandLineSyntax Syntax,
136                                             StringRef EscapedCommandLine) {
137  if (Syntax == JSONCommandLineSyntax::AutoDetect) {
138    Syntax = JSONCommandLineSyntax::Gnu;
139    llvm::Triple Triple(llvm::sys::getProcessTriple());
140    if (Triple.getOS() == llvm::Triple::OSType::Win32) {
141      // Assume Windows command line parsing on Win32 unless the triple
142      // explicitly tells us otherwise.
143      if (!Triple.hasEnvironment() ||
144          Triple.getEnvironment() == llvm::Triple::EnvironmentType::MSVC)
145        Syntax = JSONCommandLineSyntax::Windows;
146    }
147  }
148
149  if (Syntax == JSONCommandLineSyntax::Windows) {
150    llvm::BumpPtrAllocator Alloc;
151    llvm::StringSaver Saver(Alloc);
152    llvm::SmallVector<const char *, 64> T;
153    llvm::cl::TokenizeWindowsCommandLine(EscapedCommandLine, Saver, T);
154    std::vector<std::string> Result(T.begin(), T.end());
155    return Result;
156  }
157  assert(Syntax == JSONCommandLineSyntax::Gnu);
158  CommandLineArgumentParser parser(EscapedCommandLine);
159  return parser.parse();
160}
161
162// This plugin locates a nearby compile_command.json file, and also infers
163// compile commands for files not present in the database.
164class JSONCompilationDatabasePlugin : public CompilationDatabasePlugin {
165  std::unique_ptr<CompilationDatabase>
166  loadFromDirectory(StringRef Directory, std::string &ErrorMessage) override {
167    SmallString<1024> JSONDatabasePath(Directory);
168    llvm::sys::path::append(JSONDatabasePath, "compile_commands.json");
169    auto Base = JSONCompilationDatabase::loadFromFile(
170        JSONDatabasePath, ErrorMessage, JSONCommandLineSyntax::AutoDetect);
171    return Base ? inferTargetAndDriverMode(
172                      inferMissingCompileCommands(expandResponseFiles(
173                          std::move(Base), llvm::vfs::getRealFileSystem())))
174                : nullptr;
175  }
176};
177
178} // namespace
179
180// Register the JSONCompilationDatabasePlugin with the
181// CompilationDatabasePluginRegistry using this statically initialized variable.
182static CompilationDatabasePluginRegistry::Add<JSONCompilationDatabasePlugin>
183X("json-compilation-database", "Reads JSON formatted compilation databases");
184
185namespace clang {
186namespace tooling {
187
188// This anchor is used to force the linker to link in the generated object file
189// and thus register the JSONCompilationDatabasePlugin.
190volatile int JSONAnchorSource = 0;
191
192} // namespace tooling
193} // namespace clang
194
195std::unique_ptr<JSONCompilationDatabase>
196JSONCompilationDatabase::loadFromFile(StringRef FilePath,
197                                      std::string &ErrorMessage,
198                                      JSONCommandLineSyntax Syntax) {
199  // Don't mmap: if we're a long-lived process, the build system may overwrite.
200  llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> DatabaseBuffer =
201      llvm::MemoryBuffer::getFile(FilePath, /*IsText=*/false,
202                                  /*RequiresNullTerminator=*/true,
203                                  /*IsVolatile=*/true);
204  if (std::error_code Result = DatabaseBuffer.getError()) {
205    ErrorMessage = "Error while opening JSON database: " + Result.message();
206    return nullptr;
207  }
208  std::unique_ptr<JSONCompilationDatabase> Database(
209      new JSONCompilationDatabase(std::move(*DatabaseBuffer), Syntax));
210  if (!Database->parse(ErrorMessage))
211    return nullptr;
212  return Database;
213}
214
215std::unique_ptr<JSONCompilationDatabase>
216JSONCompilationDatabase::loadFromBuffer(StringRef DatabaseString,
217                                        std::string &ErrorMessage,
218                                        JSONCommandLineSyntax Syntax) {
219  std::unique_ptr<llvm::MemoryBuffer> DatabaseBuffer(
220      llvm::MemoryBuffer::getMemBufferCopy(DatabaseString));
221  std::unique_ptr<JSONCompilationDatabase> Database(
222      new JSONCompilationDatabase(std::move(DatabaseBuffer), Syntax));
223  if (!Database->parse(ErrorMessage))
224    return nullptr;
225  return Database;
226}
227
228std::vector<CompileCommand>
229JSONCompilationDatabase::getCompileCommands(StringRef FilePath) const {
230  SmallString<128> NativeFilePath;
231  llvm::sys::path::native(FilePath, NativeFilePath);
232
233  std::string Error;
234  llvm::raw_string_ostream ES(Error);
235  StringRef Match = MatchTrie.findEquivalent(NativeFilePath, ES);
236  if (Match.empty())
237    return {};
238  const auto CommandsRefI = IndexByFile.find(Match);
239  if (CommandsRefI == IndexByFile.end())
240    return {};
241  std::vector<CompileCommand> Commands;
242  getCommands(CommandsRefI->getValue(), Commands);
243  return Commands;
244}
245
246std::vector<std::string>
247JSONCompilationDatabase::getAllFiles() const {
248  std::vector<std::string> Result;
249  for (const auto &CommandRef : IndexByFile)
250    Result.push_back(CommandRef.first().str());
251  return Result;
252}
253
254std::vector<CompileCommand>
255JSONCompilationDatabase::getAllCompileCommands() const {
256  std::vector<CompileCommand> Commands;
257  getCommands(AllCommands, Commands);
258  return Commands;
259}
260
261static llvm::StringRef stripExecutableExtension(llvm::StringRef Name) {
262  Name.consume_back(".exe");
263  return Name;
264}
265
266// There are compiler-wrappers (ccache, distcc, gomacc) that take the "real"
267// compiler as an argument, e.g. distcc gcc -O3 foo.c.
268// These end up in compile_commands.json when people set CC="distcc gcc".
269// Clang's driver doesn't understand this, so we need to unwrap.
270static bool unwrapCommand(std::vector<std::string> &Args) {
271  if (Args.size() < 2)
272    return false;
273  StringRef Wrapper =
274      stripExecutableExtension(llvm::sys::path::filename(Args.front()));
275  if (Wrapper == "distcc" || Wrapper == "gomacc" || Wrapper == "ccache" ||
276      Wrapper == "sccache") {
277    // Most of these wrappers support being invoked 3 ways:
278    // `distcc g++ file.c` This is the mode we're trying to match.
279    //                     We need to drop `distcc`.
280    // `distcc file.c`     This acts like compiler is cc or similar.
281    //                     Clang's driver can handle this, no change needed.
282    // `g++ file.c`        g++ is a symlink to distcc.
283    //                     We don't even notice this case, and all is well.
284    //
285    // We need to distinguish between the first and second case.
286    // The wrappers themselves don't take flags, so Args[1] is a compiler flag,
287    // an input file, or a compiler. Inputs have extensions, compilers don't.
288    bool HasCompiler =
289        (Args[1][0] != '-') &&
290        !llvm::sys::path::has_extension(stripExecutableExtension(Args[1]));
291    if (HasCompiler) {
292      Args.erase(Args.begin());
293      return true;
294    }
295    // If !HasCompiler, wrappers act like GCC. Fine: so do we.
296  }
297  return false;
298}
299
300static std::vector<std::string>
301nodeToCommandLine(JSONCommandLineSyntax Syntax,
302                  const std::vector<llvm::yaml::ScalarNode *> &Nodes) {
303  SmallString<1024> Storage;
304  std::vector<std::string> Arguments;
305  if (Nodes.size() == 1)
306    Arguments = unescapeCommandLine(Syntax, Nodes[0]->getValue(Storage));
307  else
308    for (const auto *Node : Nodes)
309      Arguments.push_back(std::string(Node->getValue(Storage)));
310  // There may be multiple wrappers: using distcc and ccache together is common.
311  while (unwrapCommand(Arguments))
312    ;
313  return Arguments;
314}
315
316void JSONCompilationDatabase::getCommands(
317    ArrayRef<CompileCommandRef> CommandsRef,
318    std::vector<CompileCommand> &Commands) const {
319  for (const auto &CommandRef : CommandsRef) {
320    SmallString<8> DirectoryStorage;
321    SmallString<32> FilenameStorage;
322    SmallString<32> OutputStorage;
323    auto Output = std::get<3>(CommandRef);
324    Commands.emplace_back(
325        std::get<0>(CommandRef)->getValue(DirectoryStorage),
326        std::get<1>(CommandRef)->getValue(FilenameStorage),
327        nodeToCommandLine(Syntax, std::get<2>(CommandRef)),
328        Output ? Output->getValue(OutputStorage) : "");
329  }
330}
331
332bool JSONCompilationDatabase::parse(std::string &ErrorMessage) {
333  llvm::yaml::document_iterator I = YAMLStream.begin();
334  if (I == YAMLStream.end()) {
335    ErrorMessage = "Error while parsing YAML.";
336    return false;
337  }
338  llvm::yaml::Node *Root = I->getRoot();
339  if (!Root) {
340    ErrorMessage = "Error while parsing YAML.";
341    return false;
342  }
343  auto *Array = dyn_cast<llvm::yaml::SequenceNode>(Root);
344  if (!Array) {
345    ErrorMessage = "Expected array.";
346    return false;
347  }
348  for (auto &NextObject : *Array) {
349    auto *Object = dyn_cast<llvm::yaml::MappingNode>(&NextObject);
350    if (!Object) {
351      ErrorMessage = "Expected object.";
352      return false;
353    }
354    llvm::yaml::ScalarNode *Directory = nullptr;
355    llvm::Optional<std::vector<llvm::yaml::ScalarNode *>> Command;
356    llvm::yaml::ScalarNode *File = nullptr;
357    llvm::yaml::ScalarNode *Output = nullptr;
358    for (auto& NextKeyValue : *Object) {
359      auto *KeyString = dyn_cast<llvm::yaml::ScalarNode>(NextKeyValue.getKey());
360      if (!KeyString) {
361        ErrorMessage = "Expected strings as key.";
362        return false;
363      }
364      SmallString<10> KeyStorage;
365      StringRef KeyValue = KeyString->getValue(KeyStorage);
366      llvm::yaml::Node *Value = NextKeyValue.getValue();
367      if (!Value) {
368        ErrorMessage = "Expected value.";
369        return false;
370      }
371      auto *ValueString = dyn_cast<llvm::yaml::ScalarNode>(Value);
372      auto *SequenceString = dyn_cast<llvm::yaml::SequenceNode>(Value);
373      if (KeyValue == "arguments") {
374        if (!SequenceString) {
375          ErrorMessage = "Expected sequence as value.";
376          return false;
377        }
378        Command = std::vector<llvm::yaml::ScalarNode *>();
379        for (auto &Argument : *SequenceString) {
380          auto *Scalar = dyn_cast<llvm::yaml::ScalarNode>(&Argument);
381          if (!Scalar) {
382            ErrorMessage = "Only strings are allowed in 'arguments'.";
383            return false;
384          }
385          Command->push_back(Scalar);
386        }
387      } else {
388        if (!ValueString) {
389          ErrorMessage = "Expected string as value.";
390          return false;
391        }
392        if (KeyValue == "directory") {
393          Directory = ValueString;
394        } else if (KeyValue == "command") {
395          if (!Command)
396            Command = std::vector<llvm::yaml::ScalarNode *>(1, ValueString);
397        } else if (KeyValue == "file") {
398          File = ValueString;
399        } else if (KeyValue == "output") {
400          Output = ValueString;
401        } else {
402          ErrorMessage =
403              ("Unknown key: \"" + KeyString->getRawValue() + "\"").str();
404          return false;
405        }
406      }
407    }
408    if (!File) {
409      ErrorMessage = "Missing key: \"file\".";
410      return false;
411    }
412    if (!Command) {
413      ErrorMessage = "Missing key: \"command\" or \"arguments\".";
414      return false;
415    }
416    if (!Directory) {
417      ErrorMessage = "Missing key: \"directory\".";
418      return false;
419    }
420    SmallString<8> FileStorage;
421    StringRef FileName = File->getValue(FileStorage);
422    SmallString<128> NativeFilePath;
423    if (llvm::sys::path::is_relative(FileName)) {
424      SmallString<8> DirectoryStorage;
425      SmallString<128> AbsolutePath(
426          Directory->getValue(DirectoryStorage));
427      llvm::sys::path::append(AbsolutePath, FileName);
428      llvm::sys::path::remove_dots(AbsolutePath, /*remove_dot_dot=*/ true);
429      llvm::sys::path::native(AbsolutePath, NativeFilePath);
430    } else {
431      llvm::sys::path::native(FileName, NativeFilePath);
432    }
433    auto Cmd = CompileCommandRef(Directory, File, *Command, Output);
434    IndexByFile[NativeFilePath].push_back(Cmd);
435    AllCommands.push_back(Cmd);
436    MatchTrie.insert(NativeFilePath);
437  }
438  return true;
439}
440