Modelling a JVM for polymorphic bytecode.

Abstract

In standard compilation of Java-like languages, the bytecode generated for a given source depends on both the source itself and the compilation environment. This latter dependency poses some unnecessary restrictions on which execution environments can be used to run the code. When using polymorphic bytecode, a binary depends only on its source and can be dynamically adapted to run on diverse environments. Dynamic linking is particularly suited to polymorphic bytecode, because it can be adapted to an execution environment as late as possible, maximizing the flexibility of the approach. We analyze how polymorphic bytecode can be dynamically linked presenting a deterministic model of a Java Virtual Machine which interleaves loading and linking steps with execution. In our model, loading and execution phases are basically standard, whereas verification handles also type constraints, which are part of polymorphic bytecode, and resolution blends in verification.

Flexible Type-Safe Linking of Components for Java-Like Languages

“…First, our binary components are linkable units, but not loadable units, that is, they cannot be replaced or serviced after application execution has started. Hence, we plan to study the possibility of considering a different semantics for the composition operators based on dynamic rather static linking, following the approach taken in [11,16] where models for virtual machines able to execute polymorphic bytecode have been defined. Another limitation of the approach is that mutual consistency of components only means that type correctness is guaranteed, but of course does not imply that components satisfy some expected behaviour.…”