How to make a clean clang front-end?

Question

I'm working on a C++ source analyzer project and it seems that clang is nice candidate for the parsing work. The problem is that clang heavily depends on the infrastructure "llvm" project, How do I configure it to get a clean front-end without any concrete machine oriented backend? Just like LCC does, they provide a "null" backend for people who focus on parser parts. Any suggestion is appreciated.

Answer 1

I recently did this on Windows.

Download the clang and llvm source from here.

Install cmake and Python (contrary to the docs, you do need Python just to build clang; at least, cmake gives up if it can't find a Python runtime).

You also need VS2008 or VS2010.

One thing that's not entirely obvious is the required directory structure:

projectRoot     build  <- intermediate build files and DLLs, etc. will go here     llvm  <- contents of llvm-3.0.src from llvm-3.0.tar go here         tools             clang  <- contents of clang-3.0.src from clang-3.0.tar go here 

And follow the windows build instructions from step 4 onwards. Don't attempt to use the cmake GUI, it's a horror; just use the commands given in the build instructions.

Once the build is complete (which takes a while) you'll have:

projectRoot     build         bin             Release  <- libclang.dll will be here         lib             Release  <- libclang.lib will be here     llvm         tools             clang                 include                     clang-c  <- Index.h is here 

Index.h defines the API to access information about your source code; it contains quite a bit of documentation about the APIs.

To get started using clang you need something like:

CXIndex index = clang_createIndex(1, 1);  // Support Microsoft extensions char *args[] = {"-fms-extensions"};  CXTranslationUnit tu = clang_parseTranslationUnit(index, "mySource.c", args, ARRAY_SIZE(args), 0, 0, 0);  if (tu) {     CXCursor cursor = clang_getTranslationUnitCursor(tu);      // Use the cursor functions to navigate through the AST } 

Answer 2

Unfortunately, you cannot get "pure" front-end without machine-specific details. C/C++ are inherently machine-tied languages. Think about preprocessor and built-in defines, the sizes of the builtin types, etc. Some of these can be abstracted out, but not e.g. preprocessor.