Arkadiy Vertleyb wrote:
I believe (and please correct me if I am wrong) that what you actually mean is the call to foo need to be defined as many times as actually needed as opposed to always BOOST_MPL_LIMIT_VECTOR_SIZE times in my implementation. I don't see how you can get away with defining call to foo just "once" 'cause you still need to do it for every "pos".
When I said defined, I was talking about the code generated by the macro. I think you mean the number of times it is instantiated, so you're right.
This introduces the problem of naming such classes. You employ __LINE__ to solve it, but I don't believe it's enough, since problems can easily pop up when envoking typeof from different files.
I don't just use __LINE__, I also use the name of the typedef. I think this should be unique within the current namespace. The line number probably isn't needed at all. But, if this isn't acceptable then it could define a named struct, with the type as a member, i.e. the user would name the struct, and then get the type using 'name::type'.
So I still better like the approach where TYPEOF macro expands to a single call to a template meta-function, with no need to automatically create intermediate classes. This way it can be used in any context, and we don't introduce additional issues related to automatic class generation at the call site.
I'm actually inclined to agree here. I came up with the idea a little while ago when I first tried your typeof. I didn't think it was that useful so I left it on the shelf. But when I saw that Peder was doing, I saw a use for it. Daniel