Re: [boost] string_algo Formal Review

From: Pavol Droba (droba_at_[hidden])
Date: 2003-10-22 09:36:46

On Wed, Oct 22, 2003 at 08:57:42AM -0400, Beman Dawes wrote:
> At 02:23 AM 10/22/2003, Pavol Droba wrote:

[snip]
> >...
> >I have used C++ syntax, but I haven't mind it as a piece of code. Try to
> >look at it as table of required feature.
> 
> That's the problem I'm having with the library's specification. Far too 
> much has to be imagined. Hopefully the final docs will contain a real 
> requirements table.

There are already some concept specifications in docs. Is that format alright 
for you?
 
> >...
> >Well, the Sequence is sufficient model for all algorithms in string_algo
> >library, where mutable operations are need.
> >Currently I don't see a major benefit in a refinement of this concept.
> 
> Short term that's probably OK. Long term users may want to use certain 
> algorithms with containers which don't full meet the Sequence requirement. 
> Why shouldn't some of the algorithms work on an Associative container for 
> example? Or a home-made container which supports insert but not erase or 
> visa versa.

Sure, that is a reason to lower the requirements where possible. There is an
idea floating around to add more power for sequence_traits. With this at hand,
it would be possible to lower requirements for some containers.
However, in the current state, without additional information from the traits,
lowering the requirements would imply some degradation in performance.
Therefore I would suggest to postpone this issue until the other, more important
(like docs) will be solved.

> >...
> >So let me sumarize the requirements for the container arguments:
> >
> >1. for non-result type container arguments (see my previous posts)
> >InputContainerConcept is required.
> >2. For all others SequenceConcept is required.
> >
> >There are some exceptions to point 2, but given the fact, that the
> >requirements are lower, it is safe
> >to assume the SequenceConcept as requirement.
> 
> Minor point: since "SequenceConcept" refers to what the Standard Library 
> calls a "Sequence", maybe it would be better not to tack the word "Concept" 
> on the end of these requirement names.
> 
Point noted. I will use wording "Sequence concept" and "Container concept". Is it ok?

Regards,

Pavol