On Wed, Jun 17, 2009 at 12:46 AM, Frank Mori Hess <frank.hess@nist.gov>wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Tuesday 16 June 2009, Christian Schladetsch wrote:
In truth, this is a non-issue. The code you presented broke the standard; monotonic allocator didn't. Allocators are indeed allowed to have non-static members.
I hate to contradict your prejudices, but std c++03 containers are not required by the standard to support allocators which contain state, as has been noted several times already in this thread.
I didn't claim that they did. Ross did: "Luckily it seems c++0x will eliminate that assumption about equal allocators". In truth, the standard states that STL implementations are free to treat allocators of the same type isomorphically. It does not state that allocators may not have non-static data. That is all that I have claimed from the start. What this means practically is that yes, you can write code that breaks one some platforms (but not all) if you use and attempt to intermix elements from containers that use allocators that are based on different storage. In reality, this is rarely a problem. In any case, the monotonic library as it stands can be and is being safely used with just one storage and many allocators. If necessary, this can be enforced by using a static global. Thanks, Christian.