Re: [boost] [shared_ptr] dangerous implementation of moveconstructor

From: Peter Dimov (pdimov_at_[hidden])
Date: 2007-04-17 09:48:19

Greer, Joe wrote:

> [Joe] Tossing in my own $.02, this Joe tends to agree somewhat with
> the other Joe.  :)  If I wanted to swap p and q, I can do that today
> without the move.

Not in a generic way, though. Move is about performance. For some types, 
swap is slower than copy. There is also the use case of

    a = f();

where you automatically get the move assignment if f() returns an rvalue, 
and an ordinary copy otherwise.

> On the other hand, if I move a value from q to p, I expect
> p's contents to be gone, not to be transferred to q.

In 99.4% of the use cases for move, q is a temporary and you don't care what 
it holds, since it will be destroyed anyway at the end of the expression. 
Decreasing the performance for the majority to appease a minority which 
would like to move from lvalues and have them in a predictable state 
afterwards seems like a questionable tradeoff to me. Not necessarily wrong, 
just questionable. :-)