Skip to content

Latest commit

 

History

History
233 lines (197 loc) · 8.48 KB

README.md

File metadata and controls

233 lines (197 loc) · 8.48 KB
Raw

nacro

Build Status

A better macro extension for C/C++. Implemented in Clang plugins. Inspired by Rust's macro system.

#pragma nacro rule myPrint
(val:$expr) -> $stmt {
    printf("%d\n", val * 2)
}

#pragma nacro rule genColors
(cases:$expr*) -> {
    $loop(c in cases) {
        case c:
        printf("the color is %s\n", $str(c));
        break;
    }
}

enum Color { RED, BLUE, YELLOW };
void printColor(enum Color color) {
    switch(color) genColors(RED, BLUE, YELLOW)
}

int main() {
    myPrint(1 + 3)   // print out '8'
    printColor(RED); // print 'the color is RED'
    return 0;
}

Note: This project is still under heavy development. Please checkout this wiki page for work-in-progress items.

Table of Contents

How to Build and Use

Supported Platforms

  • Linux x86_64
  • Mac OSX (partial support [1])

Prepare

Here are the tool and version requirements:

Tool Name Version Requirements Notes
CMake >=3.4.4
GCC >=5.1.0 Used to compile the plugin
LLVM + Clang 10.0.0 Used to build and run the plugin. Need to be exactly this version
Python >=3.0 Only required if end-to-end tests are enabled
Google Test [2] Any Only required if unit tests are enabled
Ninja Any Optional. Feel free to use any build tool you prefer

Since LLVM project neither has a stable Clang plugin interface nor has ABI compatibilities (even among minor version releases!). The LLVM + Clang bundle need to be exactly the version as stated above. We recommend to download the prebuilt one from offical website.

If you would like to enable end-to-end tests, several python packages are also required:

pip3 install lit filecheck

In addition to requirements above, you might also need to install libtinfo on Linux. It can be installed by libtinfo-dev package in Debian-family distros.

Build

First, generate build folder using CMake:

mkdir .build
cd .build
cmake -G Ninja \
      -DLLVM_DIR=${LLVM_INSTALL_PATH}/lib/cmake/llvm \
      -DClang_DIR=${LLVM_INSTALL_PATH}/lib/cmake/clang \
      ../

The ${LLVM_INSTALL_PATH} shell variable points to the pre-built bundle folder you download or your home-built LLVM build dir.

If you want to enable unit tests and/or end-to-end tests, use the following configurations:

mkdir .build
cd .build
cmake -G Ninja \
      -DLLVM_DIR=${LLVM_INSTALL_PATH}/lib/cmake/llvm \
      -DClang_DIR=${LLVM_INSTALL_PATH}/lib/cmake/clang \
      -DNACRO_ENABLE_UNITTESTS=ON \
      -DNACRO_ENABLE_TESTS=ON \
      ../

To build it:

ninja all

To run unit tests and end-to-end tests, respectively.

ninja check-units
ninja check

Getting Started

As shown in the snippet at the top of this page, nacro allows you to embed a small DSL that acts like normal C/C++ function macros but with safer and more powerful features.

The DSL always starts with a pragma directive:

#pragma nacro rule foo

The line above will create a nacro rule, which is the only supported nacro kinds, with the name foo.

Then we add the body for foo:

#pragma nacro rule foo
(a:$expr) -> $expr {
    1 + a
}

The syntax here should be pretty straight forward: a is the nacro argument, and the $expr directive right after it shows that this argument is an expression. Outside the argument list, another $expr following the arrow tell us that eventually the nacro body (i.e. 1 + a) will be an expression after it is expanded.

#pragma nacro rule foo
(a:$expr) -> $expr {
    1 + a
}

int caller(int x) {
    return foo(x << 2) * 3;
}

Using foo is no different than using a normal C/C++ macro, as shown above. It will also got expanded during pre-processing time just like the old-fashion one. The only difference here is that for caller, instead of being expanded to this:

int caller(int x) {
    return 1 + x << 2 * 3;
}

The nacro one will be expanded to this:

int caller(int x) {
    return (1 + (x << 2)) * 3;
}

This is one of the protections that we will mentioned in the following chapters.

Another cool feature is loop:

#pragma nacro rule bar
(items:$expr*) -> {
    $loop(i in items) {
        printf("hello %s\n", $str(i));
    }
}

void the_caller() {
    bar(some, random, stuff)
}

(Here we omit the "generated type" after the arrow, this will implicitly use $block as the generated type)

As you can guess, the $str directive will perform the stringify (just like the '#') and the $loop directive here will decompose items, which is a list of expression annotated by the '*' suffix on its type, and generate one copy of loop body for each of the element. Just like loop unrolling.

// After bar got expanded...
void the_caller() {
    printf("hello %s\n", "some");
    printf("hello %s\n", "random");
    printf("hello %s\n", "stuff");
}

Finally, we're gonna compile the code. To use this plugin in an easier way without adding bunch of -Xclang options to clang, we've created a simple wrapper script, clang-nacro, to do all the heavy liftings for you. Just use it like a normal clang:

$ cat input.c
#include <stdio.h>

#pragma nacro rule bar
(items:$expr*) -> {
    $loop(i in items) {
        printf("hello %s\n", $str(i));
    }
}

int main() {
    bar(some, random, stuff)
    return 0;
}
$ /your/build/dir/clang-nacro -o out_exe input.c
$ ./out_exe
hello some
hello random
hello stuff
$

Of course, this is not the full story. Other features like Invalid Capture Detection are waiting for you to explore in the wiki!

Motivations

The C/C++ macro system adopts a copy-and-paste style text replacement that leads to many problems. For example, values might unintentionally be mixed up with adjacent operands in an expression.

#define doubleUp(x) x * 2

// ERROR: foo(2) will print out 8 rather than 10
void foo(int a) {
    printf("%d\n", doubleUp(a + 3));
}

Of course this can be solved by adding parentheses around every possible expressions but it will make code less readable and it's still pretty easy for programmers to make mistakes.

Another common problem is 'variable capture':

#define foo(arg) {\
    int x = 0;  \
    return arg + x; \
}

// ERROR: caller will always return 0 
// regardless the argument value
int caller(int x) {
    foo(x)
}

The root cause is that substitution of arg, which is happens to be x, will be bind to the declaration within the macro (i.e. int x = 0;) rather than the one in caller's argument list.

Last but not the least, C/C++ macro system lacks a clean way to express repeatance text generation or substitution (i.e. loops). Making it less powerful than its counterpart in other languages (e.g. Rust).

Syntax and Features

For detail syntax and list of features, please checkout the wiki pages.

FAQ

Q: What does the name 'Nacro' come from?

A: It's a combination of "New Macro" or "Novel Macro" (whatever). And it was intended to make a pun on 'Necromancer'. Since in LLVM project there has been DWARF (the debug format) and ORC (On-Request-Compilation JIT), why not adding another D&D reference?

Notes

[1]: Actually most parts of the plugin can work without any problem in Mac OSX. Utilities and testing infrastructures are the parts that hasn't supported yet.

[2]: Unfortunately pre-built google test is not widely available in mainstream Linux distributions. If that's the case, please follow the instructions from here. And make sure it installed in CMake's default search path. That is, /usr/lib or /usr/local/lib in most cases.