While FFPF is language neutral, some languages are better suited to exploit the strengths of FFPF than others. BPF, for instance, can not by itself take advantage of FFPF's persistent state, extensibility, etc. For this reason, we developed two new languages for filter expressions, known as FPL-1 and FPL-2 (FFPF packet languages 1 and 2). They were designed to exploit all of FFPF's features. The main distinctions are that FPL-1 is a fairly slow interpreted stack language, while FPL-2 is fully optimised native code (based on registers), and that FPL-1 code can be self-modifying, while FPL-2 code is fixed. Also, the syntax in FPL-2 is much improved.
Given that FPL-1 has been around for a year now, why did we develop FPL-2? The reason is that although FPL-1 bytecode is fairly efficient, running it in an interpreter hurts performance. Moreover, as observed by McCanne and Van Jacobson: for modern processor architectures, stack-based languages are less efficient than register-based approaches [25]. For this reason, we developed a language that (1) compiles to fully optimised object code, and (2) is based on registers and memory, and (3) has a more readable syntax.
Apart from self-modification, there is little functional difference between FPL-1 and FPL-2. For this reason, we discuss `FPL' as a general concept, using FPL-2 language constructs for illustration. A detailed explanation of FPL-1 and FPL-2 can be found in [8] and [12].