Pattern matching languages are programming languages. If we were looking to speed up long-running conventional programs an obvious approach would be to use a virtual machine with a JIT to end up with efficient native code at the cost of slower startup.

I’ve been experimenting with this and the LLVM compiler infrastructure project. LLVM is a set of libraries and UNIX utilities that provide an assembler, optimizer, interpreter, and native compiler for a high level SSA intermediate form. So far I’ve implemented a simplified subset of the POSIX fnmatch function’s functionality as a proof of concept. It’s pretty hacky, but it’s up on github.

The performance isn’t great, it’s 20% slower than GLIBC in my trivial testcase, but so far all I’ve got is a really naiive implementation. I’m going to implement an NFA / DFA based algorithm which should be more efficient, and easier extend to full regular expressions. Hopefully it’ll look more useful then.