"Arkadiy Vertleyb" <vertleyb@hotmail.com> writes:
"David Abrahams" <dave@boost-consulting.com> wrote
I believe I have substantially simplified the implementation of typeof and fixed the vc8 compatibility problems. See the enclosed typeof_impl.hpp
Right. That's exactly the simplification I meant in my previous post. Unfortunately it means at() is instantiated BOOST_TYPEOF_LIMIT_SIZE times, whereas with more complicated expression it is instantiated only n times, where n is the actual size of the encoded vector. I introduced this complexity about half a year ago, and it cut the time of my tests more than in half...
Okay, how about the one enclosed here, then?
I am a little concerned about the implementation. When I saw the complexity of what the BOOST_TYPEOF macro expanded into, I was a bit shocked. Aside from what I addressed, the fact that the expression gets repeated 100 times was really surprising to me -- it seems to me that asking the compiler to evaluate the type of the expression over and over could get to be quite expensive.
But this is on average seems to be compensated by less times the template is instantiated.
Okay.
The simplification I've made results in the expression being repeated "only" 50 times. Is it possible to do better?
I don't believe it's possible with conventional metaprogramming. The problem is that, in order to calculate the size of the vector at compile time, we have to pass the expression inside a metafunction, which doesn't seem possible. Therefore we have to do it at preprocessing time, which means BOOST_TYPEOF_LIMIT_SIZE times. This is the price of crossing boundaries.
Also, on a stylistic note, the paucity of whitespace in
namespace boost{namespace type_of{
seems a bit draconian (yes, I'm exercising my vocabulary today ;->).
Well, there are two whitespaces :-)
Why? Does it make it unreadable?
Just less readable than it might be. -- Dave Abrahams Boost Consulting www.boost-consulting.com