Page:AITR-474.djvu/11



The first part of this report is a slightly revised version of a dissertation submitted in May 1977. Where it was of historical interest to reflect changes in the SCHEME language which ocurredoccurred [sic] in the following year and the effect they had on RABBIT, the text was left intact, with notes added of the form, "Since the dissertation was written, thus-and-so occurred." The second part, the Appendix, was not part of the dissertation, and is a complete listing of the source code for RABBIT, with extensive commentary.

It is intended that the first part should be self-contained, and provide a qualitative overview of the compilation methods used in RABBIT. The second part is provided for those readers who would like to examine the precise mechanisms used to carry out the general methods.

Thus there are five levels of thoroughness at which the reader may consume this document:

(1) The reader who wishes only to skim is advised to read sections 1, 5, 6, possibly 7, 8A, 8B, 8C, 10, 11, and 12. This will give a basic overview, including the use of macros and the optimizing techniques. (2) The reader who also wants to know about the details of SCHEME, the run-time system, and a long example is advised to read the entire main text (about a third of the document). (3) The reader who wants to understand the low-level organization of the algorithms, and read about the more tricky special cases, should read the main text and then the commentary on the code. (4) The reader who additionally wants to understand the nit-picking details should read the code along with the commentary. (5) The reader who wants a real feel for the techniques involved should read the entire document, invent three new SCHEME constructs and write macros for them, and then reimplement the compiler for another run-time environment. (He ought please also to send a copy of any documents on such a project to this author, who would be very interested!)