Skip to content

keiichiw/constexpr-8cc

Repository files navigation

constexpr-8cc: Compile-time C Compiler Build Status

constexpr-8cc is a compile-time C compiler implemented as C++14 constant expressions. This enables you to compile while you compile! This project is a port of 8cc built on ELVM Infrastructure.

Constant expressions in C++ are expressions that can be evaluated at compile-time. In C++14, by relaxing constrains, constant expressions became so powerful that a C compiler can be implemented in!

In constexpr-8cc, the main routine for compilations of C programs is implemented in a C++14 constexpr function. Therefore, if you compile 8cc.cpp to a binary file by g++, compilation of a C program will be performed as a compile-time computation and the result of this C compilation will be embedded into the generated binary. In this sense, constexpr-8cc is a compile-time C compiler.

The following is the main function in 8cc.cpp.

int main() {
  // Compile-time
  constexpr buffer buf = eight_cc(); // Compile C code into ELVM IR
  constexpr unsigned int output_size = buf.size;

  static_assert(0 <= output_size && output_size < EIGHT_CC_OUTPUT_LIMIT, "8cc: Error");

  // Run-time
  for(int i = 0; i < output_size; ++i) {
    putchar(buf.b[i]);
  }
}

In this program, the return value of eight_cc is stored into the variable buf with a constexpr specifier. Thus, you will find that the compilation of a C program is done in compile-time.

Requirements

constexpr-8cc requires Linux with >g++-6.2. I confirmed ./test/hello.c can be compiled with g++-6.2, g++-8.3 and g++-9.3 at least.

  • g++-6.2 worked without any extra flag related to constexpr as there was no limitation of constant's loop counts at this version.
  • With g++-8.3, I needed to enlarge constexpr's loop count with -fconstexpr-loop-limit. The maximum number we can specify is 2**31 - 1.
  • With g++-9.3, in addition to -fconstexpr-loop-limit, enlarging -fconstexpr-ops-limit was needed.
  • There is no guarantee that it works with other versions of g++.
  • I couldn't make it work with clang++ as clang++ has more strict limitation of constexpr loop counts.

How to run

Compilation by run_8cc.py

In order to try constexpr-8cc easily, use run_8cc.py.

$ ./run_8cc.py x86 ./test/hello.c -o ./hello.exe # It takes about 3 minutes on my laptop
$ chmod +x ./hello.exe                           # 'hello.exe' is i386-linux binary
$ ./hello.exe
Hello, world!

You can change the target language of compilations like the following:

$ ./run_8cc.py py ./test/hello.c -o ./hello.py # target language is Python
$ python ./hello.py
Hello, world!

For more information about this script, type $ ./run_8cc.py -h.

Compilation by hand

If you want to compile 8cc.cpp manually, please look at config.hpp. In this file, the variable EIGHT_CC_INPUT_FILE is defined. EIGHT_CC_INPUT_FILE should be a name of a file that contains a source C program as a C++ string literal. This string will be embedded in 8cc.cpp at pre-processing-time and used as an input of the compile-time computation.

So, before compiling 8cc.cpp manually, you have to convert a raw program to a string literal like the following:

$ sed '1s/^/R"(/' ./test/hello.c | sed '$s/$/\n)"/' > ./test/hello.c.txt # Convert C to string literal
$ g++-6 ./8cc.cpp -o eir_gen.out
$ ./eir_gen.out > ./test/hello.eir       # eir_gen.out outputs ELVM IR
$ sed -i '1s/^/R"(x86/' ./test/hello.eir # Convert IR to string literal
$ sed -i '$s/$/\n)"/' ./test/hello.eir
$ g++-6 ./elc.cpp -o exe_gen.out
$ ./exe_gen.out > ./hello.exe            # exe_gen.out outputs i386-linux binary
$ chmod +x ./hello.exe
$ ./hello.exe
Hello, world!

How was constexpr-8cc generated?

When you see 8cc.hpp, you will know this program was not written by hand. Actually, I used ELVM Compiler Infrastructure to generate it. I just implemented a translator from ELVM IR to C++14 constexpr here.

Author

Keiichi Watanabe (udon.watanabe [at] gmail.com)

Links

About

Compile-time C Compiler implemented as C++14 constant expressions

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages