llvm-clang: incremental or online parser?

Is there anyway to use the llvm-clang parser in an incremental/online manner?

Say I'm writing an editor and I want to be able to parse the C++ code I have in front of me.

I don't want to write my own hacked up parser.

I'd like to use something full featured, like llvm-clang.

Is there an easy way to hij开发者_如何学运维ack the llvm-clang parser? (And is it fast enough to run it continuously in the background)?

Thanks!

I don't think clang can incrementally parse C++ files, but it's one of this project goals: http://clang.llvm.org/features.html

I've written something similar for my final year project. It wasn't C++ editor, but a Visual Studio plugin, which main task was improving C++ intellisense (like Visual Assist X).

When I was writing this project I've been also thinking about C++ incremental parser, but I haven't found any suitable solution. To solve the C++ intellisense problem I used normal C++ parser from GCC. However it was to slow, to parse file after each code completion request (ctrl+space), just try including boost::spirit. To make this project work properly I parsed files in the background and after each code completion request I compared current file with it's previous version (via diff) to detect changes made from last parsing. Having those changes I updated syntax tree, mostly by adding or removing variables.

Except incremental parsing, there is also another problem with projects like this. Mostly you'll be parsing C++ code which is being edited so it's invalid code. Given the complex C++ grammar, sometimes parser won't be able to recover from syntax errors, so it won't detect correctly some symbols in code.

Another issue are C++ parsers / compilers differences. Let's say I'm using working in Visual Studio and I have used some VC++ compiler specific contruction in my code. Clang parser won't be able to parse it correctly.

For writing something similair to IntelliSense, I would advise you to write your own parser using the LALR parsing algorithm. Since you can save its state in each line so you don't have to reparse the whole file when a file has been editted, which is very fast!
Note that C++ can't be fully expressed in BNF, but I think you could get pretty far with some adjustments. It's ofcourse a lot more work than using Clang's frontend, but you could still use Clang for analysing header files in coöperation with you own written parser.