Code Loading & Linking

Code linking is a process of mapping identifiers used in the code to an actual implementations for execution. Different kinds and types of linking feature speed of execution, variability trade-offs and greatly influence possibilities of code reloading in the platform. Code loading is the process of referencing, locating and making the virtual machine level code accessible to the caller which is already loaded into the same virtual machine. Code loading mechanisms heavily depend on the representation of units of compilation that machine is using. Linking and loading techniques define code reloading capabilities and influence the existence of an industrial code reloading solutions for the given virtual machine. Why the problem is important Code binding, loading and linking are essential concepts for building multi-module applications, libraries and in fact any non-trivial program calls external libraries. Even if the call is to functions defined at the core of the language like to Java Class Library, which forms an essential part of the language, it should be considered an external call and require a compiler or a runtime to make a mapping between an identifier and an exact set of instructions that are be referenced by that identifier. We consider every identifier to be global, so any unit of compilation is able to call another unit of compilation by its identifier. The exact mechanism of distinguishing unit’s identifier from all others like: local variables, local function names, etc is not important and is omitted from the discussion. Once a virtual machine, VM, locates the exact instructions that represent an unit of compilation to load it begins the loading process. The usual activities done during that phase is: verification of the code, verification of the types and making sure the code is safe (in the VM sense), loading the code into the memory and making that unit of compilation available to already existing code. After that the code must be initialized and made ready to use by the caller. What is specific to this paper and what are expected results In this survey paper we describe the principles and implementations of several real-life choices about code binding, loading and reloading in several virtual machine platforms. When the solutions are dissected and investigated together, different trade-offs become clear and distinguishing features of every approach are seen clearer. Theoretical introduction, terms and principles, trade-offs For the better understanding of the code loading process we need to get back to the roots of the modern programming languages and describe how a program executes in a physical machine. A computer processor has a set of instructions that are implemented in hardware and can be executed sequentially by this hardware. So essentially any program is a specification which instruction should be executed and in which order. A program specified as a sequence of such instructions is referred as a machine code. As the hardware is the only physical entity involved into the computation, it is the only entity which is capable of executing a program. In spite of this any program, written in any representation must be converted to a machine code prior the execution. It is perfectly valid to create and write programs in the machine code directly, however this approach tends to be very cumbersome, as modern programs consist of thousands of machine code instructions. To solve this problem a compiler is used. A compiler is a tool for translation process from a higher, more abstract programming language into a lower level set of instructions, which is called compilation.