The rise in big data analytics has impelled the popularity of sparse tensors in High-Performance Computing (HPC). Meanwhile, I/O limitations in HPC have motivated efforts towards in situ and in transit software composition, particularly for visualization, but also for sparse tensors. The wide variety of sparse tensor memory formats and their associated high-performance sparse tensor algorithms lend a basis for considering software composition systems. This work surveys the diverse approaches to HPC software composition with respect to a novel set of four categories, or realms: workflow managers, middleware, discrete processing services, and distributed data services. Each realm is motivated and represented by several software composition systems. Each system is compared with its rivals regarding fitness for a particular purpose. Finally, this paper highlights opportunities for future work, based on the unique challenges and the traditional challenges of software composition for sparse tensor decomposition in HPC.