| 摘要 |
In a program data stack in a computer system, every stack is implemented as two substacks, one to contain references and one to contain primitive data. In this manner, whether a piece of information on a stack is a reference or a primitive value is easily determined according to which substack it resides in. Each substack is itself a full-fledged stack data structure with a stack base, a stack pointer, and a stack limit. In the preferred embodiment, there is also a frame pointer for each substack. The normal instruction set is modified so that instructions that operate with references use the reference stack whereas instructions that operate with data use the primitive data stack. Specifically, during compilation, or loading, all instructions which can operate indiscriminately with either references or data are replaced by equivalent, but reference-specific or data-specific instructions. With this arrangement, garbage collection algorithms can locate all references stored within a stack, since the references are all in a more or less contiguous region of memory without interspersed primitive data.
|
