Re: [Boost-users] Proposal mpl::switch_case

Subject: Re: [Boost-users] Proposal mpl::switch_case
From: TONGARI (tongari95_at_[hidden])
Date: 2011-10-19 14:47:24

2011/10/20 Hite, Christopher <Christopher.Hite_at_[hidden]>

>  My use case is a fast protocol decoder.  I've described the protocol in
> mpl and generate the instructions I need to decode/encode/store packets.
> During decoding I have to be able to take a run time id and map it to the
> correct compile time generated code.  Traditionally such code was written as
> a switch/case.
>
> The obvious solution is to use metaprogramatic recursion, for example
> mpl::for_each to check the id against each value.  You would hope the
> compiler would inline all the recursion, detect the ifs on an id, and do
> general switch/case optimization.  Unfortunately the code generated by gcc
> and it's not as good as I hoped.  They actually generate a function for
> every ~6 levels of recursion and just do a bunch of if/equal/else.
>
> boost::variant has this same problem.  They use BOOST_PP_REPEAT to generate
> an actual switch/case statement(visitation_impl.hpp).  The compiler does a
> good job with the swich/case.  It removes redundant cases and then optimizes
> using array lookups and binary searches depending on size.  Good job!
>
> 1) Could we factor this logic out into a utility like mpl::switch_case?
>
> 2) What would the best interface look like?
>
> 3) What other bits of code could benefit from it?
> phoenix::switch_?
> Meta State Machine?
>

Is this what you want?
http://dancinghacker.com/switch/