Choosing the Right Place for your Check¶

If you have an idea of a check, you should decide whether it should be implemented as a:

• Clang diagnostic: if the check is generic enough, targets code patterns that most probably are bugs (rather than style or readability issues), can be implemented effectively and with extremely low false positive rate, it may make a good Clang diagnostic.
• Clang static analyzer check: if the check requires some sort of control flow analysis, it should probably be implemented as a static analyzer check.
• clang-tidy check is a good choice for linter-style checks, checks that are related to a certain coding style, checks that address code readability, etc.

Preparing your Workspace¶

If you are new to LLVM development, you should read the Getting Started with the LLVM System, Using Clang Tools and How To Setup Clang Tooling For LLVM documents to check out and build LLVM, Clang and Clang Extra Tools with CMake.

Once you are done, change to the llvm/tools/clang/tools/extra directory, and let’s start!

The Directory Structure¶

clang-tidy source code resides in the llvm/tools/clang/tools/extra directory and is structured as follows:

clang-tidy/                       # Clang-tidy core.
|-- ClangTidy.h                   # Interfaces for users.
|-- ClangTidyCheck.h              # Interfaces for checks.
|-- ClangTidyModule.h             # Interface for clang-tidy modules.
|-- ClangTidyModuleRegistry.h     # Interface for registering of modules.
   ...
|-- google/                       # Google clang-tidy module.
|-+
  |-- GoogleTidyModule.cpp
  |-- GoogleTidyModule.h
        ...
|-- llvm/                         # LLVM clang-tidy module.
|-+
  |-- LLVMTidyModule.cpp
  |-- LLVMTidyModule.h
        ...
|-- objc/                         # Objective-C clang-tidy module.
|-+
  |-- ObjCTidyModule.cpp
  |-- ObjCTidyModule.h
        ...
|-- tool/                         # Sources of the clang-tidy binary.
        ...
test/clang-tidy/                  # Integration tests.
    ...
unittests/clang-tidy/             # Unit tests.
|-- ClangTidyTest.h
|-- GoogleModuleTest.cpp
|-- LLVMModuleTest.cpp
|-- ObjCModuleTest.cpp
    ...

Writing a clang-tidy Check¶

So you have an idea of a useful check for clang-tidy.

First, if you’re not familiar with LLVM development, read through the Getting Started with LLVM document for instructions on setting up your workflow and the LLVM Coding Standards document to familiarize yourself with the coding style used in the project. For code reviews we mostly use LLVM Phabricator.

Next, you need to decide which module the check belongs to. Modules are located in subdirectories of clang-tidy/ and contain checks targeting a certain aspect of code quality (performance, readability, etc.), certain coding style or standard (Google, LLVM, CERT, etc.) or a widely used API (e.g. MPI). Their names are same as user-facing check groups names described above.

After choosing the module and the name for the check, run the clang-tidy/add_new_check.py script to create the skeleton of the check and plug it to clang-tidy. It’s the recommended way of adding new checks.

If we want to create a readability-awesome-function-names, we would run:

$ clang-tidy/add_new_check.py readability awesome-function-names

The add_new_check.py script will:

• create the class for your check inside the specified module’s directory and register it in the module and in the build system;
• create a lit test file in the test/clang-tidy/ directory;
• create a documentation file and include it into the docs/clang-tidy/checks/list.rst.

Let’s see in more detail at the check class definition:

...

#include "../ClangTidyCheck.h"

namespace clang {
namespace tidy {
namespace readability {

...
class AwesomeFunctionNamesCheck : public ClangTidyCheck {
public:
  AwesomeFunctionNamesCheck(StringRef Name, ClangTidyContext *Context)
      : ClangTidyCheck(Name, Context) {}
  void registerMatchers(ast_matchers::MatchFinder *Finder) override;
  void check(const ast_matchers::MatchFinder::MatchResult &Result) override;
};

} // namespace readability
} // namespace tidy
} // namespace clang

...

Constructor of the check receives the Name and Context parameters, and must forward them to the ClangTidyCheck constructor.

In our case the check needs to operate on the AST level and it overrides the registerMatchers and check methods. If we wanted to analyze code on the preprocessor level, we’d need instead to override the registerPPCallbacks method.

In the registerMatchers method we create an AST Matcher (see AST Matchers for more information) that will find the pattern in the AST that we want to inspect. The results of the matching are passed to the check method, which can further inspect them and report diagnostics.

using namespace ast_matchers;

void AwesomeFunctionNamesCheck::registerMatchers(MatchFinder *Finder) {
  Finder->addMatcher(functionDecl().bind("x"), this);
}

void AwesomeFunctionNamesCheck::check(const MatchFinder::MatchResult &Result) {
  const auto *MatchedDecl = Result.Nodes.getNodeAs<FunctionDecl>("x");
  if (MatchedDecl->getName().startswith("awesome_"))
    return;
  diag(MatchedDecl->getLocation(), "function %0 is insufficiently awesome")
      << MatchedDecl
      << FixItHint::CreateInsertion(MatchedDecl->getLocation(), "awesome_");
}

(If you want to see an example of a useful check, look at clang-tidy/google/ExplicitConstructorCheck.h and clang-tidy/google/ExplicitConstructorCheck.cpp).

Registering your Check¶

(The add_new_check.py takes care of registering the check in an existing module. If you want to create a new module or know the details, read on.)

The check should be registered in the corresponding module with a distinct name:

class MyModule : public ClangTidyModule {
 public:
  void addCheckFactories(ClangTidyCheckFactories &CheckFactories) override {
    CheckFactories.registerCheck<ExplicitConstructorCheck>(
        "my-explicit-constructor");
  }
};

Now we need to register the module in the ClangTidyModuleRegistry using a statically initialized variable:

static ClangTidyModuleRegistry::Add<MyModule> X("my-module",
                                                "Adds my lint checks.");

When using LLVM build system, we need to use the following hack to ensure the module is linked into the clang-tidy binary:

Add this near the ClangTidyModuleRegistry::Add<MyModule> variable:

// This anchor is used to force the linker to link in the generated object file
// and thus register the MyModule.
volatile int MyModuleAnchorSource = 0;

And this to the main translation unit of the clang-tidy binary (or the binary you link the clang-tidy library in) clang-tidy/tool/ClangTidyMain.cpp:

// This anchor is used to force the linker to link the MyModule.
extern volatile int MyModuleAnchorSource;
static int MyModuleAnchorDestination = MyModuleAnchorSource;

Configuring Checks¶

If a check needs configuration options, it can access check-specific options using the Options.get<Type>("SomeOption", DefaultValue) call in the check constructor. In this case the check should also override the ClangTidyCheck::storeOptions method to make the options provided by the check discoverable. This method lets clang-tidy know which options the check implements and what the current values are (e.g. for the -dump-config command line option).

class MyCheck : public ClangTidyCheck {
  const unsigned SomeOption1;
  const std::string SomeOption2;

public:
  MyCheck(StringRef Name, ClangTidyContext *Context)
    : ClangTidyCheck(Name, Context),
      SomeOption(Options.get("SomeOption1", -1U)),
      SomeOption(Options.get("SomeOption2", "some default")) {}

  void storeOptions(ClangTidyOptions::OptionMap &Opts) override {
    Options.store(Opts, "SomeOption1", SomeOption1);
    Options.store(Opts, "SomeOption2", SomeOption2);
  }
  ...

Assuming the check is registered with the name “my-check”, the option can then be set in a .clang-tidy file in the following way:

CheckOptions:
  - key: my-check.SomeOption1
    value: 123
  - key: my-check.SomeOption2
    value: 'some other value'

If you need to specify check options on a command line, you can use the inline YAML format:

$ clang-tidy -config="{CheckOptions: [{key: a, value: b}, {key: x, value: y}]}" ...

Testing Checks¶

To run tests for clang-tidy use the command:

$ ninja check-clang-tools

clang-tidy checks can be tested using either unit tests or lit tests. Unit tests may be more convenient to test complex replacements with strict checks. Lit tests allow using partial text matching and regular expressions which makes them more suitable for writing compact tests for diagnostic messages.

The check_clang_tidy.py script provides an easy way to test both diagnostic messages and fix-its. It filters out CHECK lines from the test file, runs clang-tidy and verifies messages and fixes with two separate FileCheck invocations: once with FileCheck’s directive prefix set to CHECK-MESSAGES, validating the diagnostic messages, and once with the directive prefix set to CHECK-FIXES, running against the fixed code (i.e., the code after generated fix-its are applied). In particular, CHECK-FIXES: can be used to check that code was not modified by fix-its, by checking that it is present unchanged in the fixed code. The full set of FileCheck directives is available (e.g., CHECK-MESSAGES-SAME:, CHECK-MESSAGES-NOT:), though typically the basic CHECK forms (CHECK-MESSAGES and CHECK-FIXES) are sufficient for clang-tidy tests. Note that the FileCheck documentation mostly assumes the default prefix (CHECK), and hence describes the directive as CHECK:, CHECK-SAME:, CHECK-NOT:, etc. Replace CHECK by either CHECK-FIXES or CHECK-MESSAGES for clang-tidy tests.

An additional check enabled by check_clang_tidy.py ensures that if CHECK-MESSAGES: is used in a file then every warning or error must have an associated CHECK in that file. Or, you can use CHECK-NOTES: instead, if you want to also ensure that all the notes are checked.

To use the check_clang_tidy.py script, put a .cpp file with the appropriate RUN line in the test/clang-tidy directory. Use CHECK-MESSAGES: and CHECK-FIXES: lines to write checks against diagnostic messages and fixed code.

It’s advised to make the checks as specific as possible to avoid checks matching to incorrect parts of the input. Use [[@LINE+X]]/[[@LINE-X]] substitutions and distinct function and variable names in the test code.

Here’s an example of a test using the check_clang_tidy.py script (the full source code is at test/clang-tidy/google-readability-casting.cpp):

// RUN: %check_clang_tidy %s google-readability-casting %t

void f(int a) {
  int b = (int)a;
  // CHECK-MESSAGES: :[[@LINE-1]]:11: warning: redundant cast to the same type [google-readability-casting]
  // CHECK-FIXES: int b = a;
}

To check more than one scenario in the same test file use -check-suffix=SUFFIX-NAME on check_clang_tidy.py command line or -check-suffixes=SUFFIX-NAME-1,SUFFIX-NAME-2,.... With -check-suffix[es]=SUFFIX-NAME you need to replace your CHECK-* directives with CHECK-MESSAGES-SUFFIX-NAME and CHECK-FIXES-SUFFIX-NAME.

Here’s an example:

// RUN: %check_clang_tidy -check-suffix=USING-A %s misc-unused-using-decls %t -- -- -DUSING_A
// RUN: %check_clang_tidy -check-suffix=USING-B %s misc-unused-using-decls %t -- -- -DUSING_B
// RUN: %check_clang_tidy %s misc-unused-using-decls %t
...
// CHECK-MESSAGES-USING-A: :[[@LINE-8]]:10: warning: using decl 'A' {{.*}}
// CHECK-MESSAGES-USING-B: :[[@LINE-7]]:10: warning: using decl 'B' {{.*}}
// CHECK-MESSAGES: :[[@LINE-6]]:10: warning: using decl 'C' {{.*}}
// CHECK-FIXES-USING-A-NOT: using a::A;$
// CHECK-FIXES-USING-B-NOT: using a::B;$
// CHECK-FIXES-NOT: using a::C;$

There are many dark corners in the C++ language, and it may be difficult to make your check work perfectly in all cases, especially if it issues fix-it hints. The most frequent pitfalls are macros and templates:

1. code written in a macro body/template definition may have a different meaning depending on the macro expansion/template instantiation;
2. multiple macro expansions/template instantiations may result in the same code being inspected by the check multiple times (possibly, with different meanings, see 1), and the same warning (or a slightly different one) may be issued by the check multiple times; clang-tidy will deduplicate _identical_ warnings, but if the warnings are slightly different, all of them will be shown to the user (and used for applying fixes, if any);
3. making replacements to a macro body/template definition may be fine for some macro expansions/template instantiations, but easily break some other expansions/instantiations.

Running clang-tidy on LLVM¶

To test a check it’s best to try it out on a larger code base. LLVM and Clang are the natural targets as you already have the source code around. The most convenient way to run clang-tidy is with a compile command database; CMake can automatically generate one, for a description of how to enable it see How To Setup Clang Tooling For LLVM. Once compile_commands.json is in place and a working version of clang-tidy is in PATH the entire code base can be analyzed with clang-tidy/tool/run-clang-tidy.py. The script executes clang-tidy with the default set of checks on every translation unit in the compile command database and displays the resulting warnings and errors. The script provides multiple configuration flags.

• The default set of checks can be overridden using the -checks argument, taking the identical format as clang-tidy does. For example -checks=-*,modernize-use-override will run the modernize-use-override check only.
• To restrict the files examined you can provide one or more regex arguments that the file names are matched against. run-clang-tidy.py clang-tidy/.*Check\.cpp will only analyze clang-tidy checks. It may also be necessary to restrict the header files warnings are displayed from using the -header-filter flag. It has the same behavior as the corresponding clang-tidy flag.
• To apply suggested fixes -fix can be passed as an argument. This gathers all changes in a temporary directory and applies them. Passing -format will run clang-format over changed lines.

On checks profiling¶

clang-tidy can collect per-check profiling info, and output it for each processed source file (translation unit).

To enable profiling info collection, use the -enable-check-profile argument. The timings will be output to stderr as a table. Example output:

$ clang-tidy -enable-check-profile -checks=-*,readability-function-size source.cpp
===-------------------------------------------------------------------------===
                          clang-tidy checks profiling
===-------------------------------------------------------------------------===
  Total Execution Time: 1.0282 seconds (1.0258 wall clock)

   ---User Time---   --System Time--   --User+System--   ---Wall Time---  --- Name ---
   0.9136 (100.0%)   0.1146 (100.0%)   1.0282 (100.0%)   1.0258 (100.0%)  readability-function-size
   0.9136 (100.0%)   0.1146 (100.0%)   1.0282 (100.0%)   1.0258 (100.0%)  Total

It can also store that data as JSON files for further processing. Example output:

$ clang-tidy -enable-check-profile -store-check-profile=.  -checks=-*,readability-function-size source.cpp
$ # Note that there won't be timings table printed to the console.
$ ls /tmp/out/
20180516161318717446360-source.cpp.json
$ cat 20180516161318717446360-source.cpp.json
{
"file": "/path/to/source.cpp",
"timestamp": "2018-05-16 16:13:18.717446360",
"profile": {
  "time.clang-tidy.readability-function-size.wall": 1.0421266555786133e+00,
  "time.clang-tidy.readability-function-size.user": 9.2088400000005421e-01,
  "time.clang-tidy.readability-function-size.sys": 1.2418899999999974e-01
}
}

There is only one argument that controls profile storage:

• -store-check-profile=<prefix>

  By default reports are printed in tabulated format to stderr. When this option is passed, these per-TU profiles are instead stored as JSON. If the prefix is not an absolute path, it is considered to be relative to the directory from where you have run clang-tidy. All . and .. patterns in the path are collapsed, and symlinks are resolved.

  Example: Let’s suppose you have a source file named example.cpp, located in the /source directory. Only the input filename is used, not the full path to the source file. Additionally, it is prefixed with the current timestamp.

  • If you specify -store-check-profile=/tmp, then the profile will be saved to /tmp/<ISO8601-like timestamp>-example.cpp.json
  • If you run clang-tidy from within /foo directory, and specify -store-check-profile=., then the profile will still be saved to /foo/<ISO8601-like timestamp>-example.cpp.json