The markup would be seperated by newlines, with the key-value pairs on the same line, always in quotes. I've solved it with regex and a crude parser before, but I've always heard of things like Bison and such but never found a 'good' way to impliment a real parser in a language like Python.
I would go with a simple recursive descent parser, unless you are interested in learning a DSL, the intricacies of the parsing algorithm your particular parser generator uses, and willing to debug one --- IMHO not worth it. Even "real", production-quality compilers like GCC, Clang, MSVC, and ICC use recursive-descent; and some, like GCC, switched away from parser generators.
Parser generators work really well for context-free grammars, but as soon as you start needing context-sensitive lexing (like C does) you enter World Of Pain time. And they avoid having to deal with lookahead, which is always messy in hand-written parsers.
But I think I'd agree that writing a recursive descent parser from scratch is probably the best way to learn about parsing; at least for a simple grammar.
BTW, fun fact: it turns out that BNF was invented by an Indian linguist called Pāṇini about 2500 years ago, in order to parse Sanskrit.