Awesome

fccf: Fast C/C++ Code Finder

fccf is a command-line tool that quickly searches through C/C++ source code in a directory based on a search string and prints relevant code snippets that match the query.

Highlights

• Quickly identifies source files that contain a search string.
• For each candidate source file, builds an abstract syntax tree (AST).
• Visits the nodes in the AST, looking for function declarations, classes, enums, variables etc., that match the user's request.
• Pretty-prints the identified snippet of source code to the terminal.
• MIT License

Searching the Linux kernel source tree

The following video shows fccf searching and finding snippets of code in torvalds/linux.

https://user-images.githubusercontent.com/8450091/165400381-9ba49a62-97fb-4f4a-890a-0dc9b20dfe75.mp4

Searching the fccf source tree (Modern C++)

The following video shows fccf searching the fccf C++ source code.

Note that search results here include:

1. Class declaration
2. Functions and function templates
3. Variable declarations, including lambda functions

https://user-images.githubusercontent.com/8450091/165402206-65d9ed43-b9dd-4528-92bd-0b4ce76b6468.mp4

Search for any of --flag that matches

Provide an empty query to match any --flag, e.g., any enum declaration.

... or any class constructor

Searching a for statement

Use --for-statement to search for statements. fccf will try to find for statements (including C++ ranged for) that contain the query string.

Searching expressions

Use the --include-expressions option to find expressions that match the query.

The following example shows fccf find calls to isdigit().

The following example shows fccf finding references to the clang_options variable.

Searching for using declarations

Use the --using-declaration option to find using declarations, using directives, and type alias declarations.

Searching for namespace aliases

Use the --namespace-alias option to find namespace aliases.

Searching for throw expressions

Use --throw-expression with a query to search for specific throw expressions that contain the query string.

As presented earlier, an empty query here will attempt to match any throw expression in the code base:

Build Instructions

Build fccf using CMake. For more details, see BUILDING.md.

NOTE: fccf requires libclang and LLVM installed.

# Install libclang and LLVM
# sudo apt install libclang-dev llvm

git clone https://github.com/p-ranav/fccf
cd fccf

# Build
cmake -S . -B build -D CMAKE_BUILD_TYPE=Release
cmake --build build

# Install
sudo cmake --install build

fccf Usage

foo@bar:~$ fccf --help
Usage: fccf [--help] [--version] [--help] [--exact-match] [--json] [--filter VAR] [-j VAR] [--enum] [--struct] [--union] [--member-function] [--function] [--function-template] [-F] [--class] [--class-template] [--class-constructor] [--class-destructor] [-C] [--for-statement] [--namespace-alias] [--parameter-declaration] [--typedef] [--using-declaration] [--variable-declaration] [--verbose] [--include-expressions] [--static-cast] [--dynamic-cast] [--reinterpret-cast] [--const-cast] [-c] [--throw-expression] [--ignore-single-line-results] [--include-dir VAR]... [--language VAR] [--std VAR] [--no-color] query [path]...

Positional arguments:
  query                                
  path                                 [nargs: 0 or more] 

Optional arguments:
  -h, --help                           shows help message and exits 
  -v, --version                        prints version information and exits 
  -h, --help                           Shows help message and exits 
  -E, --exact-match                    Only consider exact matches 
  --json                               Print results in JSON format 
  -f, --filter                         Only evaluate files that match filter pattern [nargs=0..1] [default: "*.*"]
  -j                                   Number of threads [nargs=0..1] [default: 5]
  --enum                               Search for enum declaration 
  --struct                             Search for struct declaration 
  --union                              Search for union declaration 
  --member-function                    Search for class member function declaration 
  --function                           Search for function declaration 
  --function-template                  Search for function template declaration 
  -F                                   Search for any function or function template or class member function 
  --class                              Search for class declaration 
  --class-template                     Search for class template declaration 
  --class-constructor                  Search for class constructor declaration 
  --class-destructor                   Search for class destructor declaration 
  -C                                   Search for any class or class template or struct 
  --for-statement                      Search for `for` statement 
  --namespace-alias                    Search for namespace alias 
  --parameter-declaration              Search for function or method parameter 
  --typedef                            Search for typedef declaration 
  --using-declaration                  Search for using declarations, using directives, and type alias declarations 
  --variable-declaration               Search for variable declaration 
  --verbose                            Request verbose output 
  --ie, --include-expressions          Search for expressions that refer to some value or member, e.g., function, variable, or enumerator. 
  --static-cast                        Search for static_cast 
  --dynamic-cast                       Search for dynamic_cast 
  --reinterpret-cast                   Search for reinterpret_cast 
  --const-cast                         Search for const_cast 
  -c                                   Search for any static_cast, dynamic_cast, reinterpret_cast, orconst_cast expression 
  --throw-expression                   Search for throw expression 
  --isl, --ignore-single-line-results  Ignore forward declarations, member function declarations, etc. 
  -I, --include-dir                    Additional include directories [nargs=0..1] [default: {}] [may be repeated]
  -l, --language                       Language option used by clang [nargs=0..1] [default: "c++"]
  --std                                C++ standard to be used by clang [nargs=0..1] [default: "c++17"]
  --nc, --no-color                     Stops fccf from coloring the output 

How it works

1. fccf does a recursive directory search for a needle in a haystack - like grep or ripgrep - It uses an SSE2 strstr SIMD algorithm (modified Rabin-Karp SIMD search; see here) if possible to quickly find, in multiple threads, a subset of the source files in the directory that contain a needle.
2. For each candidate source file, it uses libclang to parse the translation unit (build an abstract syntax tree).
3. Then it visits each child node in the AST, looking for specific node types, e.g., CXCursor_FunctionDecl for function declarations.
4. Once the relevant nodes are identified, if the node's "spelling" (libclang name for the node) matches the search query, then the source range of the AST node is identified - source range is the start and end index of the snippet of code in the buffer
5. Then, it pretty-prints this snippet of code. I have a simple lexer that tokenizes this code and prints colored output.

Note on include_directories

For all this to work, fccf first identifies candidate directories that contain header files, e.g., paths that end with include/. It then adds these paths to the clang options (before parsing the translation unit) as -Ifoo -Ibar/baz etc. Additionally, for each translation unit, the parent and grandparent paths are also added to the include directories for that unit in order to increase the likelihood of successful parsing.

Additional include directories can also be provided to fccf using the -I or --include-dir option. Using verbose output (--verbose), errors in the libclang parsing can be identified and fixes can be attempted (e.g., adding the right include directories so that libclang is happy).

To run fccf on the fccf source code without any libclang errors, I had to explicitly provide the include path from LLVM-12 like so:

foo@bar:~$ fccf --verbose 'lexer' . --include-dir /usr/lib/llvm-12/include/
Checking ./source/lexer.cpp
Checking ./source/lexer.hpp
Checking ./source/searcher.cpp

// ./source/lexer.hpp (Line: 14 to 40)
class lexer
{
  std::string_view m_input;
  fmt::memory_buffer* m_out;
  std::size_t m_index {0};
  bool m_is_stdout {true};

  char previous() const;
  char current() const;
  char next() const;
  void move_forward(std::size_t n = 1);
  bool is_line_comment();
  bool is_block_comment();
  bool is_start_of_identifier();
  bool is_start_of_string();
  bool is_start_of_number();
  void process_line_comment();
  void process_block_comment();
  bool process_identifier(bool maybe_class_or_struct = false);
  void process_string();
  std::size_t get_number_of_characters(std::string_view str);

public:
  void tokenize_and_pretty_print(std::string_view source,
                                 fmt::memory_buffer* out,
                                 bool is_stdout = true);
}

References

1. SIMD-friendly algorithms for substring searching
2. libclang: C Interface to Clang