From: Phil Nash (phil.nash.lists_at_[hidden])
Date: 2003-03-04 07:13:43
I'm interested in something like this, but I am not sure if your
implementation meets the needs I often seem to have.
1. I very often am given a raw buffer by pointer and wish to acquire it into
a more RAII-like container without copying.
2. Also I would like to provide a customer deleter.
Obviously these criteria in themselves are met well by any smart pointer
that has a customer deleter, but then you lose iterators, array syntax,
push/ pop etc. It would be nice to have the best of both worlds.
scoped_array and shared_array go some way towards this, but not all the way.
I don't know whether, again, this would all be better implemented as
policies of a generic smart pointer framework (or even managed_resource or
whatever the name might be).
----- Original Message -----
From: "Kevin Atkinson" <kevin_at_[hidden]>
Sent: Tuesday, March 04, 2003 11:05 AM
Subject: [boost] Any Interest In a Raw Memory Buffer
> Is there any interest in a vector like container which is designed to make
> working with raw memory easy. It is different from a vector<char> (with
> the additional assumption that the objects are stored in memory
> sequentially) is the following ways:
> The integrator is defined to be a pointer to make working with functions
> that use raw memory easy.
> It can use a specific range of user supplied memory for its storage.
> When using user memory it will never attempt to allocate more memory.
> capacity() == max_size().
> Any function which takes a range:
> (const_iterator start, const_iterator stop)
> can also take a void pointer and a size ie:
> (const void * data, size_type size)
> Any function which takes an iterator to represent a position within
> the exiting object can also take an offset. If two iterators are
> used to represent a range than an offset and a size can also be used.
> Like vector, push_back(), and pop_back() or constant time operations if
> storage is available. However, unlike vector, pop_front() is a
> constant time operation and push_front() can be made to be constant
> time operations.
> support for replacing one range with another while moving the minimal
> amount of memory possible:
> replace(iterator start, iterator_stop,
> const_iterator rstart, const_iterator rstop)
> A special member function
> template <typename U> U * datap(size_type pos = 0)
> which allows treating the memory as another type without an ugly
> reinterpret_cast. This statement is equivalent to
> reinterpret_cast<U *>(...) + pos
> My implementation is not a template. It uses new and delete for
> allocation and the data type is "unsigned char". Since the main goal for
> this class was working with raw memory and not an array of objects I
> decided not to make it a template.
> Some of the features of Buffer might be useful for a more general purpose
> vector, such as the replace method and the ability to use user memory.
> If people are really interested I will post my code.
> I wrote it to make working with Berkeley database and sending network
> messages as painless as possible.
> Let me know.
> Unsubscribe & other changes:
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk