[boost] [shared] : container : any interest

Subject: [boost] [shared] : container : any interest
From: Ingo.Loehken_at_[hidden]
Date: 2009-03-05 08:42:27

Hi,

as a consequence of my daily work in COM design, it revealed, that most 
often there are
only views into existing containers required, so copies can be delayed, 
until container
invariants are viloated - elements are already shared, as they reference 
pointers with
embedded reference counts.

Thus, the design fulfills the following objectives :
+ Every existing container can be adopted to be a shareable container. 
Shareable
  means, concurrent read and write access to elements. Container invariant 
methods, such
  like insert for sorted associative containers, trigger a copy of the 
container.
+ Iterators are managed. This guarantees valid iterators, even if the 
underlying container
  is copied, because of a container invariant modifying operation.
+ The existing concepts for containers are mapped to support shared 
containers. This avoids
  any redundant code and provides maximum flexibility in combining 
concepts via base class
  chaining.
+ Every shared container behaves simply like its stl or boost counterpart 
(except unsorted,
  because of a current unsolved issue regarding the iterators in buckets, 
requiring a second
  layer of managed iterators - imho inefficient), so that usage is as 
simple as with original
  containers.

A brief example

using boost::shared;

set<int> x, y = x;    // y is attached to the same container as x, no copy
x.insert(3);          // y gets detached from x, because set invariant 
(sort order) is modified

list<int> x;
x.push_back(3);
list<int> y = x;
y.front() = 8;        // also modifies the value in x

list<int>::const_iterator cit(x.begin());   // attach iterator to 
container shared by x and y
y.pop_back();                               // detaches y from container 
shared with x (container invariant)
std::cout << *cit << std::endl;             // output : 8 (iterator still 
valid)

Is there any interest for boost, to support such a library or is the 
number of use
cases to small to introduce something like that ?

Thx, ILo.