Re: [boost] Re: Review Request: cyclic_buffer

From: Howard Hinnant (hinnant_at_[hidden])
Date: 2003-06-10 10:52:59

On Tuesday, June 10, 2003, at 10:48  AM, Jan Gaspar wrote:

>> insert in general should have the basic exception safety guarantee.
>> But when inserting only one element at either end, the strong 
>> guarantee
>> is in place.
>
> I think the strong guarantee can reached only in case the 
> cyclic_buffer is not
> full (if there is no reallocation).

You can do it with push_back/front too regardless.  vector::push_back 
has the strong guarantee.  If you're using copy semantics (which I 
would expect instead of move semantics), you basically have to build a 
whole new vector in a temp (with the push_back), and then swap it in 
when everything is set (just describing the realloc branch here).

If you're looking ahead to use move semantics to perform this 
operation, it is a little more tricky, but not too much.  In fact it is 
easier with cyclic_buffer than it is with vector! :-)  You have to 
push_back the new element into the temporary first.  If it throws, the 
temporary cleans up the new buffer.  If it doesn't, *then* move the old 
data over (move must be nothrow).

-Howard