I am, for performance reason, porting a C# library to C++. During normal operation, this library needs, amongst other things, to parse about 150’000 math expressions (think excel formulas) with an average length of less than 150 characters.
In the C# version, I used GOLD parser to generate parsing code. It can parse all 150’000 expressions in under one second.
Because we were thinking about extending our language, I figured the move to C++ might be a good chance to change to ANTLR. I have ported the (simple) grammar to ANTLR and generated C code out of it. Parsing the 150’000 expressions takes over 12 seconds, because for each expression, I need to create a new ANTL3_INPUT_STREAM, token stream, lexer and parser – there is, at least in version 3.4, no way to reuse them.
I’d be grateful is someone could give me a recommendation what to use instead – GOLD is of course an option though generating C++ or C code seems a lot more complicated than the C# variety. My grammar is LALR and LL(1) compatible. Paramount concern is parsing performance on small inputs.
I would try boost::spirit. It is often extreamly fast (even for parsing simple things like an integer it can be faster than the C function atoi http://alexott.blogspot.com/2010/01/boostspirit2-vs-atoi.html)
http://boost-spirit.com/home/
It has nice things : header only, so dependency hell, liberal licence.
However be warned that the learning curve is difficult. It’s modern C++ (no pointer, but a lot of template and very frustrating compiling errors), so coming from C or C#, you might not be very comfortable.