alain@sophia.cnrs.fr (Alain Miniussi) wrote:
> I am not sure that the purpose of a standard is to decide the best
> implementation for a general purpose template.

Well, in my view the STL is generally over-specified. For example, there
are a great many ways to implement a map. Even if we rule out hash-tables,
many ways remain - eg a vector with linear search, or a sorted vector with
binary search, or a binary tree, or a multi-way tree. Most of the time I
don't care which is used; I just want a map. The STL's complexity
requirements more or less mandate a red-black tree, which is a highly
specific implementation, not suited to such a general purpose name as
map<>. (If it were called red_black_tree_map<> I wouldn't mind so much.)

Similarly the STL list<> requires that size() be O(1) which means the
length must be stored rather than calculated as needed. This doubles the
memory overhead for a zero-length list and means splice cannot be O(1).
Again these are highly specific, and controversial, choices.

So the STL containers are already highly constrained and not very
"encapsulated". I don't think requiring vector to use contiguous storage
makes them worse. In fact, I find it such a pragmatically useful property
that I'll be glad to have the guarantee.

  Dave Harris, Nottingham, UK | "Weave a circle round him thrice,
      brangdon@cix.co.uk      |   And close your eyes with holy dread,
                              |  For he on honey dew hath fed
 http://www.bhresearch.co.uk/ |   And drunk the milk of Paradise."
---
[ comp.std.c++ is moderated.  To submit articles, try just posting with ]
[ your news-reader.  If that fails, use mailto:std-c++@ncar.ucar.edu    ]
[              --- Please see the FAQ before posting. ---               ]
[ FAQ: http://reality.sgi.com/austern_mti/std-c++/faq.html              ]

In article <memo.19990727033220.38663C@btinternet.com> ,
scorp@btinternet.com (Dave Harris) wrote:

> Similarly the STL list<> requires that size() be O(1) which means the
> length must be stored rather than calculated as needed. This doubles the
> memory overhead for a zero-length list and means splice cannot be O(1).
> Again these are highly specific, and controversial, choices.

Actually, the exact opposite is the case.

-Dave
---
[ comp.std.c++ is moderated.  To submit articles, try just posting with ]
[ your news-reader.  If that fails, use mailto:std-c++@ncar.ucar.edu    ]
[              --- Please see the FAQ before posting. ---               ]
[ FAQ: http://reality.sgi.com/austern_mti/std-c++/faq.html              ]

Dave Harris wrote:
>
> alain@sophia.cnrs.fr (Alain Miniussi) wrote:
> > I am not sure that the purpose of a standard is to decide the best
> > implementation for a general purpose template.
[...]

> So the STL containers are already highly constrained and not very
> "encapsulated". I don't think requiring vector to use contiguous storage
> makes them worse.

Thanks, these are the best arguments I have read so far !
But why this wasn't frankly admited in the standard, instead
of the pseudo highh level stuff used in the vector introduction ?
and why vector does not provide a real accessor for the buffer ?

Alain
---
[ comp.std.c++ is moderated.  To submit articles, try just posting with ]
[ your news-reader.  If that fails, use mailto:std-c++@ncar.ucar.edu    ]
[              --- Please see the FAQ before posting. ---               ]
[ FAQ: http://reality.sgi.com/austern_mti/std-c++/faq.html              ]