No Thread Safety

A memory context is not thread-safe. Use each context from one thread at a time, for example one context per thread. Locking support is not provided and is unlikely to be implemented, since this project is intended for use within a thread. Consider nedmalloc instead if you need a thread-safe allocator.

Alignment

Every pointer returned by allocate () of a memory context is aligned to cookmem::ALLOCATION_ALIGNMENT, which is twice the size of std::size_t: 16 bytes on 64-bit platforms and 8 bytes on 32-bit platforms.

For a larger alignment, use allocateAligned (size, alignment). The alignment must be a power of 2. The memory is freed with deallocate (). reallocate () keeps the memory in place when it shrinks, but if it has to move the memory, the new memory has only the normal alignment.

To find an aligned address, the pool takes a chunk that is larger than the request by the alignment, then returns the unused memory before and after the result to the pool. A large alignment therefore needs that much free space, and may fail under a footprint limit that a normal allocation of the same size would fit in.

cookmem::Allocator uses allocateAligned () automatically for types that require a larger alignment, such as types declared with alignas(64).

Best-Effort Error Detection

The checks for memory errors are designed to be cheap, so they cannot catch everything.

Double free detection is best effort. If the memory has already been reused by another allocation, the second free cannot be detected.
Padding bytes only detect writes just past the end of an allocation, and only when the allocation is deallocated explicitly.
Passing a pointer that was not allocated by the memory context to deallocate is not detected and is undefined behavior.