machine-code [Brown, 1977]. Once this has been done the intention is that the quality of the code so produced can be improved at leisure. While this approach has been very successful and relatively error-free, it has been the experience of several implementors that it is difficult to adapt the scheme to produce highly optimised code [Russell, 1974]; apparently considerations of portability and machine-independence have caused the problems of optimisation to be overlooked. The aspect of intermediate-code design which has received most debate concerns the level of the code: low-level with a fairly simple code-generator, or high-level with a more complex code-generator [Brown, 1972],
This thesis attempts to put machine-independence and optimisation on an equal footing, and describes the use of an intermediate code which takes a novel view of the process. Instead of the intermediate code describing the computation to be performed, it describes the operation of a code-generator which will produce a program to perform the required computation. This effectively adds an extra level of indirection into the compilation, weakening any linkage between the form of the intermediate code and the object code required for a particular implementation. In essence I-code attempts to describe the results required in a way which does not constrain the method of achieving those results.