From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-11-21 20:54:46
At 06:02 PM 11/21/2001, David Abrahams wrote:
>Mike,
>
>Your test was practically /designed/ to favor your approach.
>
>I agree that for some (perhaps even many) applications, your approach
might
>be most appropriate. There are, however, many applications where the user
>wants to use indirect references. How many times have you heard the
>question
>about whether you can put an auto_ptr in a container? For many
>applications,
>you already have a pointer or smart pointer to the object; in that case
>insertion is cheap. Another example arises when you need to store many
>element references. There are lots of trade-offs.
>
>If you want to prove that one approach is better than another in general,
>you have to try much a more comprehensive suite of tests.
True, but at least Corwin and Mike are testing efficiency, rather than just
making unsupported assertions.
In addition to several more usage scenarios being tested, it would be
interesting to see the tests run with at least one other compiler. If I
understand the test results so far, the ratio was 52/133. That is close
enough that a different compiler might eliminate some or most of the time
difference.
Also, please refresh my memory. Do the differences in approach use the
same interface, or different interfaces? If different, what are the key
usage differences?
Thanks,
--Beman