Reconsidering the STL Allocator Model

Presented at C++ Connections in Las Vegas in November, 2005

Memory allocators wee originally added to the Standard Template Library (STL) for the purpose of supporting Intel x86 mixed-model (near and far) pointers and other non-flat memory models. Most people, however, consider allocators to be a way to achieve type-specific memory allocation for generic containers, generally for improved efficiency. This session will show that allocators, as defined in the STL, do not serve either purpose well. This session will suggest specific refinements that require significant changes in the way allocators work but yield tremendous payoffs in the power and usability of allocators. You’ll learn how these changes were merged with the original STL model and produced a working, effective memory allocation system that retains compatibility with the 1998 standard. These experiences are the basis for a proposed enhancement to the C++ Library for the next revision of the standard.