Context:

I'm trying to make a cached pool the gives out std::shared_ptr<T>. Even if I cache the memory for T, I still hit memory allocations for the std::shared_ptr control block and that slows things down.

Right now to get around this, I use std::allocate_shared with a custom allocator that caches the allocation and checks to make sure that the same size/alignment is always asked for. This requires two allocations: 1) for the memory used for allocator_shared and 2) representing each pool entry with the pointer to 1) and other bookkeeping data.

If I could statically know the layout of 1), I could make it part of the allocation for 2).

This would make the standard require that std::allocate_shared only uses one allocation (it current recommends it but doesn't force it).