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/STLExtras.h"
19#include "llvm/ADT/SmallString.h"
20#include "llvm/ADT/SmallVector.h"
21#include "llvm/ADT/StringRef.h"
22#include "llvm/ADT/Triple.h"
23#include "llvm/Support/Allocator.h"
24#include "llvm/Support/Casting.h"
25#include "llvm/Support/CommandLine.h"
26#include "llvm/Support/ErrorOr.h"
27#include "llvm/Support/Host.h"
28#include "llvm/Support/MemoryBuffer.h"
29#include "llvm/Support/Path.h"
30#include "llvm/Support/StringSaver.h"
31#include "llvm/Support/VirtualFileSystem.h"
32#include "llvm/Support/YAMLParser.h"
33#include "llvm/Support/raw_ostream.h"
34#include <cassert>
35#include <memory>
36#include <optional>
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#ifdef _WIN32
139    // Assume Windows command line parsing on Win32
140    Syntax = JSONCommandLineSyntax::Windows;
141#else
142    Syntax = JSONCommandLineSyntax::Gnu;
143#endif
144  }
145
146  if (Syntax == JSONCommandLineSyntax::Windows) {
147    llvm::BumpPtrAllocator Alloc;
148    llvm::StringSaver Saver(Alloc);
149    llvm::SmallVector<const char *, 64> T;
150    llvm::cl::TokenizeWindowsCommandLine(EscapedCommandLine, Saver, T);
151    std::vector<std::string> Result(T.begin(), T.end());
152    return Result;
153  }
154  assert(Syntax == JSONCommandLineSyntax::Gnu);
155  CommandLineArgumentParser parser(EscapedCommandLine);
156  return parser.parse();
157}
158
159// This plugin locates a nearby compile_command.json file, and also infers
160// compile commands for files not present in the database.
161class JSONCompilationDatabasePlugin : public CompilationDatabasePlugin {
162  std::unique_ptr<CompilationDatabase>
163  loadFromDirectory(StringRef Directory, std::string &ErrorMessage) override {
164    SmallString<1024> JSONDatabasePath(Directory);
165    llvm::sys::path::append(JSONDatabasePath, "compile_commands.json");
166    auto Base = JSONCompilationDatabase::loadFromFile(
167        JSONDatabasePath, ErrorMessage, JSONCommandLineSyntax::AutoDetect);
168    return Base ? inferTargetAndDriverMode(
169                      inferMissingCompileCommands(expandResponseFiles(
170                          std::move(Base), llvm::vfs::getRealFileSystem())))
171                : nullptr;
172  }
173};
174
175} // namespace
176
177// Register the JSONCompilationDatabasePlugin with the
178// CompilationDatabasePluginRegistry using this statically initialized variable.
179static CompilationDatabasePluginRegistry::Add<JSONCompilationDatabasePlugin>
180X("json-compilation-database", "Reads JSON formatted compilation databases");
181
182namespace clang {
183namespace tooling {
184
185// This anchor is used to force the linker to link in the generated object file
186// and thus register the JSONCompilationDatabasePlugin.
187volatile int JSONAnchorSource = 0;
188
189} // namespace tooling
190} // namespace clang
191
192std::unique_ptr<JSONCompilationDatabase>
193JSONCompilationDatabase::loadFromFile(StringRef FilePath,
194                                      std::string &ErrorMessage,
195                                      JSONCommandLineSyntax Syntax) {
196  // Don't mmap: if we're a long-lived process, the build system may overwrite.
197  llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> DatabaseBuffer =
198      llvm::MemoryBuffer::getFile(FilePath, /*IsText=*/false,
199                                  /*RequiresNullTerminator=*/true,
200                                  /*IsVolatile=*/true);
201  if (std::error_code Result = DatabaseBuffer.getError()) {
202    ErrorMessage = "Error while opening JSON database: " + Result.message();
203    return nullptr;
204  }
205  std::unique_ptr<JSONCompilationDatabase> Database(
206      new JSONCompilationDatabase(std::move(*DatabaseBuffer), Syntax));
207  if (!Database->parse(ErrorMessage))
208    return nullptr;
209  return Database;
210}
211
212std::unique_ptr<JSONCompilationDatabase>
213JSONCompilationDatabase::loadFromBuffer(StringRef DatabaseString,
214                                        std::string &ErrorMessage,
215                                        JSONCommandLineSyntax Syntax) {
216  std::unique_ptr<llvm::MemoryBuffer> DatabaseBuffer(
217      llvm::MemoryBuffer::getMemBufferCopy(DatabaseString));
218  std::unique_ptr<JSONCompilationDatabase> Database(
219      new JSONCompilationDatabase(std::move(DatabaseBuffer), Syntax));
220  if (!Database->parse(ErrorMessage))
221    return nullptr;
222  return Database;
223}
224
225std::vector<CompileCommand>
226JSONCompilationDatabase::getCompileCommands(StringRef FilePath) const {
227  SmallString<128> NativeFilePath;
228  llvm::sys::path::native(FilePath, NativeFilePath);
229
230  std::string Error;
231  llvm::raw_string_ostream ES(Error);
232  StringRef Match = MatchTrie.findEquivalent(NativeFilePath, ES);
233  if (Match.empty())
234    return {};
235  const auto CommandsRefI = IndexByFile.find(Match);
236  if (CommandsRefI == IndexByFile.end())
237    return {};
238  std::vector<CompileCommand> Commands;
239  getCommands(CommandsRefI->getValue(), Commands);
240  return Commands;
241}
242
243std::vector<std::string>
244JSONCompilationDatabase::getAllFiles() const {
245  std::vector<std::string> Result;
246  for (const auto &CommandRef : IndexByFile)
247    Result.push_back(CommandRef.first().str());
248  return Result;
249}
250
251std::vector<CompileCommand>
252JSONCompilationDatabase::getAllCompileCommands() const {
253  std::vector<CompileCommand> Commands;
254  getCommands(AllCommands, Commands);
255  return Commands;
256}
257
258static llvm::StringRef stripExecutableExtension(llvm::StringRef Name) {
259  Name.consume_back(".exe");
260  return Name;
261}
262
263// There are compiler-wrappers (ccache, distcc, gomacc) that take the "real"
264// compiler as an argument, e.g. distcc gcc -O3 foo.c.
265// These end up in compile_commands.json when people set CC="distcc gcc".
266// Clang's driver doesn't understand this, so we need to unwrap.
267static bool unwrapCommand(std::vector<std::string> &Args) {
268  if (Args.size() < 2)
269    return false;
270  StringRef Wrapper =
271      stripExecutableExtension(llvm::sys::path::filename(Args.front()));
272  if (Wrapper == "distcc" || Wrapper == "gomacc" || Wrapper == "ccache" ||
273      Wrapper == "sccache") {
274    // Most of these wrappers support being invoked 3 ways:
275    // `distcc g++ file.c` This is the mode we're trying to match.
276    //                     We need to drop `distcc`.
277    // `distcc file.c`     This acts like compiler is cc or similar.
278    //                     Clang's driver can handle this, no change needed.
279    // `g++ file.c`        g++ is a symlink to distcc.
280    //                     We don't even notice this case, and all is well.
281    //
282    // We need to distinguish between the first and second case.
283    // The wrappers themselves don't take flags, so Args[1] is a compiler flag,
284    // an input file, or a compiler. Inputs have extensions, compilers don't.
285    bool HasCompiler =
286        (Args[1][0] != '-') &&
287        !llvm::sys::path::has_extension(stripExecutableExtension(Args[1]));
288    if (HasCompiler) {
289      Args.erase(Args.begin());
290      return true;
291    }
292    // If !HasCompiler, wrappers act like GCC. Fine: so do we.
293  }
294  return false;
295}
296
297static std::vector<std::string>
298nodeToCommandLine(JSONCommandLineSyntax Syntax,
299                  const std::vector<llvm::yaml::ScalarNode *> &Nodes) {
300  SmallString<1024> Storage;
301  std::vector<std::string> Arguments;
302  if (Nodes.size() == 1)
303    Arguments = unescapeCommandLine(Syntax, Nodes[0]->getValue(Storage));
304  else
305    for (const auto *Node : Nodes)
306      Arguments.push_back(std::string(Node->getValue(Storage)));
307  // There may be multiple wrappers: using distcc and ccache together is common.
308  while (unwrapCommand(Arguments))
309    ;
310  return Arguments;
311}
312
313void JSONCompilationDatabase::getCommands(
314    ArrayRef<CompileCommandRef> CommandsRef,
315    std::vector<CompileCommand> &Commands) const {
316  for (const auto &CommandRef : CommandsRef) {
317    SmallString<8> DirectoryStorage;
318    SmallString<32> FilenameStorage;
319    SmallString<32> OutputStorage;
320    auto Output = std::get<3>(CommandRef);
321    Commands.emplace_back(
322        std::get<0>(CommandRef)->getValue(DirectoryStorage),
323        std::get<1>(CommandRef)->getValue(FilenameStorage),
324        nodeToCommandLine(Syntax, std::get<2>(CommandRef)),
325        Output ? Output->getValue(OutputStorage) : "");
326  }
327}
328
329bool JSONCompilationDatabase::parse(std::string &ErrorMessage) {
330  llvm::yaml::document_iterator I = YAMLStream.begin();
331  if (I == YAMLStream.end()) {
332    ErrorMessage = "Error while parsing YAML.";
333    return false;
334  }
335  llvm::yaml::Node *Root = I->getRoot();
336  if (!Root) {
337    ErrorMessage = "Error while parsing YAML.";
338    return false;
339  }
340  auto *Array = dyn_cast<llvm::yaml::SequenceNode>(Root);
341  if (!Array) {
342    ErrorMessage = "Expected array.";
343    return false;
344  }
345  for (auto &NextObject : *Array) {
346    auto *Object = dyn_cast<llvm::yaml::MappingNode>(&NextObject);
347    if (!Object) {
348      ErrorMessage = "Expected object.";
349      return false;
350    }
351    llvm::yaml::ScalarNode *Directory = nullptr;
352    std::optional<std::vector<llvm::yaml::ScalarNode *>> Command;
353    llvm::yaml::ScalarNode *File = nullptr;
354    llvm::yaml::ScalarNode *Output = nullptr;
355    for (auto& NextKeyValue : *Object) {
356      auto *KeyString = dyn_cast<llvm::yaml::ScalarNode>(NextKeyValue.getKey());
357      if (!KeyString) {
358        ErrorMessage = "Expected strings as key.";
359        return false;
360      }
361      SmallString<10> KeyStorage;
362      StringRef KeyValue = KeyString->getValue(KeyStorage);
363      llvm::yaml::Node *Value = NextKeyValue.getValue();
364      if (!Value) {
365        ErrorMessage = "Expected value.";
366        return false;
367      }
368      auto *ValueString = dyn_cast<llvm::yaml::ScalarNode>(Value);
369      auto *SequenceString = dyn_cast<llvm::yaml::SequenceNode>(Value);
370      if (KeyValue == "arguments") {
371        if (!SequenceString) {
372          ErrorMessage = "Expected sequence as value.";
373          return false;
374        }
375        Command = std::vector<llvm::yaml::ScalarNode *>();
376        for (auto &Argument : *SequenceString) {
377          auto *Scalar = dyn_cast<llvm::yaml::ScalarNode>(&Argument);
378          if (!Scalar) {
379            ErrorMessage = "Only strings are allowed in 'arguments'.";
380            return false;
381          }
382          Command->push_back(Scalar);
383        }
384      } else {
385        if (!ValueString) {
386          ErrorMessage = "Expected string as value.";
387          return false;
388        }
389        if (KeyValue == "directory") {
390          Directory = ValueString;
391        } else if (KeyValue == "command") {
392          if (!Command)
393            Command = std::vector<llvm::yaml::ScalarNode *>(1, ValueString);
394        } else if (KeyValue == "file") {
395          File = ValueString;
396        } else if (KeyValue == "output") {
397          Output = ValueString;
398        } else {
399          ErrorMessage =
400              ("Unknown key: \"" + KeyString->getRawValue() + "\"").str();
401          return false;
402        }
403      }
404    }
405    if (!File) {
406      ErrorMessage = "Missing key: \"file\".";
407      return false;
408    }
409    if (!Command) {
410      ErrorMessage = "Missing key: \"command\" or \"arguments\".";
411      return false;
412    }
413    if (!Directory) {
414      ErrorMessage = "Missing key: \"directory\".";
415      return false;
416    }
417    SmallString<8> FileStorage;
418    StringRef FileName = File->getValue(FileStorage);
419    SmallString<128> NativeFilePath;
420    if (llvm::sys::path::is_relative(FileName)) {
421      SmallString<8> DirectoryStorage;
422      SmallString<128> AbsolutePath(Directory->getValue(DirectoryStorage));
423      llvm::sys::path::append(AbsolutePath, FileName);
424      llvm::sys::path::native(AbsolutePath, NativeFilePath);
425    } else {
426      llvm::sys::path::native(FileName, NativeFilePath);
427    }
428    llvm::sys::path::remove_dots(NativeFilePath, /*remove_dot_dot=*/true);
429    auto Cmd = CompileCommandRef(Directory, File, *Command, Output);
430    IndexByFile[NativeFilePath].push_back(Cmd);
431    AllCommands.push_back(Cmd);
432    MatchTrie.insert(NativeFilePath);
433  }
434  return true;
435}
436