Memory manager/class allocation question

[jmacc]

Hi,
I'm thinking about implementing a c++ memory manager that would be based on pooling, or bucketing, objects of the same size. Thus, when a new request came in for some size_t chunk of memory, I'd try to allocate a chunk, of the same size, from the appropriate bucket.

My main concern with this approach is re-using memory as the footprint for a class that was not the original class when the memory was first allocated. In other words, will I end up with a corrupted object?

Ideally, a user would:
(1) call new <class>, as normal
(2) and the constructors would do the rest (that is, re-initialize all members).

My question is, would the compiler do anything wacky like store metadata particular to the original class such that, when the memory is re-used for a different class, memory corruption occurs?
And, if no memory corruption occurs -is this behavior guaranteed on future compilers/platforms?

Any help would be appreciated.
Thanks.

[Danny]

You have to distinguish between raw memory for which C++ store no special metadata (i.e., RTTI, exception stuff etc.) and objects, which are pieces of raw memory that have been initialized by a class's ctor. Your aim is obviously to preallocate a pool of the former type, and then recycle that memory, not reconstruct an object over the memory of an existing object( doing so would cause many nasty surprises such as a destructor of the original object being called on the newly constructed object etc.).
What you need is some form of placement new. This operaor allows you to construct an object on a pre-allocated memory address, under the assumption that address contains properly aligned memory with the appropriate size. You can find more info about placement new here:
http://www.devx.com/DevX/LegacyLink/9485
Gee, this was my first 10 Minute Solution way back in 1999 and it's still useful today!

[jmacc]

Danny, thanks for the response. I'll give this a go around and see.