Re: auto_vector - vector with auto_ptr semantics

From: mfdylan (dylan_at_[hidden])
Date: 2002-02-04 21:45:13

--- In boost_at_y..., "Rainer Deyke" <root_at_r...> wrote:
> 
> std::vector<int> g()
> {
>   std::vector<int> first, second;
>   // Do various operations on 'first' and 'second' here.
>   return (first.size() > second.size()) ? first : second;
> }
>
> std::vector<int> v(g());
> 
I would love to see a compiler that could compile that into efficient 
code.  In theory of course it can - the semantics of std::vector are 
part of the language, so it should essential be able to reduce the 
above to only 2 memory allocations and one free.
But if you wanted to be able to write your own classes that the 
compiler could generate similarly efficient code you essentially need 
a way of knowing in what is currently the "copy" constructor that the 
parameter being passed is going out of scope, so you can safely steal 
its resources.  I'm pretty sure this can only be done with language 
help.  One possibility I guess is an extra type-qualification, call 
it, um "expiring":

vector(expiring vector& other)
{
 swap(other);
}

But looking at the implementation, why not just have the compiler 
call swap directly?  That is, if your class implements a "swap" 
member, and you are constructing or assigning from an 'expiring' 
object, the swap() call is used instead of the copy 
constructor/assignment operator.  Giving a named member special 
status is perhaps a little anathema to the existing standard 
("operator=" almost fits this description), and I believe Java does 
this sort of thing for "to_string" etc.  I'm not sure if the same 
should be done for a "swap" global, especially if it isn't compilable 
for that type.

Anyway just a thought (shouldn't this discussion be in c.l.c++.m or 
even comp.std.c++?)

Dylan