Interoperation between Miniboxing and Other Generics Translations

Generics allow programmers to design algorithms and data structures that operate in the same way regardless of the data used by abstracting over data types. Generics are useful as they improve the programmer’s productivity by raising the level of abstraction, which in turn leads to reducing code duplication and uniform interfaces. However, as data on the low-level comes in different shapes and sizes, it is not a trivial job of compiler to bridge the gap between the uniform interface and the non-uniform low level implementation. Different approaches are used for generics translation and all of them can be categorized either into homogeneous or heterogeneous group. The characteristic of homogeneous translations is that all different data representations are transformed into an identical representation and use the same low-level code for this purpose. In the heterogeneous translations, code is duplicated and adapted for each incompatible data type. From a programmer’s point of view, there should be no difference between a generic method or class compiled using some homogeneous or heterogeneous translation. Therefore, the programmer can combine different types of translations together on different parts of the code and the program has to be correct. But, as different generics translations are implemented in different ways, interoperation between them introduces noticeable slowdowns as values need to be converted to the foreign object’s desired representation, incurring significant performance losses. In this thesis, it will be explored why slowdowns happen when different translations interact together and proposed the ways how they can interoperate more efficiently. Proposed approaches are implemented and their effectiveness is presented by benchmarking the implementation.