[boost] Re: Review Request: cyclic_buffer

From: Nigel Stewart (nigels_at_[hidden])
Date: 2003-06-06 20:43:51

Hi Jan,

>>1.	How about push_front()?  (in std::list style)
>>	And pop_front()?
> 
> Imagine you have a cyclic buffer which is full. When you add (push_back or
> insert) a new item into this buffer the first (oldest) item will be removed.

        Ahh, I see that your conceptualisation is based on
        age, or order of insertion.  Mine was more circular-
        list oriented:  Adding to the front would overwite the
        backmost item if the buffer is already full. Being a
        circle, the buffer slides either left or right, as
        needed.

        In the case that the buffer is full and items are
        alternatively added to the front and back, nothing
        much happens.  Admittedly, a special case.

        Perhaps our concepts are incompatible.  Looking at
        Knuth's coverage of Circular Lists, insertion
        is allowed at both ends.  Knuth also mentions
        FIFO (first-in first-out) lists as being
        sometimes called "circular stores".  There
        seems to be no particular implication of circular
        meaning one way or the other, based on Knuth.

        So the issue here seems to be whether a cyclic_buffer
        should be circular-list-like or FIFO-like.  Given
        that there is already a queue container adapter,
        perhaps it's worth listing some applications
        of cyclic_buffer that depend on either
        circular-list-like or FIFO-like semantics.

        Could a queue adaptor using a cyclic_buffer instead
        of a deque provide the explicitly FIFO-like mode of
        operation?  (While leaving cyclic_buffer to be
        circular-list-like?)

>>	Semantics of resize should probably follow vector:
>>	1. The capacity is implicitly adjusted to fit the
>>	   requested size.  

        http://www.sgi.com/tech/stl/Vector.html

        vector::resize

        Inserts or erases elements at the end such that
        the size becomes n.

        (I think a developer may find it surprising for a
         cyclic_buffer::resize to be clamped to the capacity.)

>>	2. Items are lost from the end if the container is
>>            to be shrunk.
> 
> Are you sure you want to lost the items at the end ? Not at the beginning (the
        
        In the FIFO-view, the leftmost/oldest should be removed, agreed.
        In the circular-list view, it is not so obvious - so follow
        vector and erase the rightmost... :-)

        Have a nice weekend, I'm off to the country side...

        Nigel