John Cowan wrote: > If anyone actually exploits this R6RS innovation, I still hope to hear > of it.
The innovation allows Scheme compilers to use eta conversion for all procedures, without having to distinguish between procedures that were introduced by the compiler or are the value of compiler-generated temporary variables from procedures that are the values of variables that appear in the original code. With the R5RS semantics, sophisticated compilers had to include that extra complexity. Simpler compilers either avoided eta conversion, limited their optimizations to avoid using eta conversion in problematic situations, or ignored that detail of the R5RS semantics. In Larceny, for example, the compiler's lambda lifting and closure conversion optimizations apply only to procedures that are either anonymous or used in first order fashion. If you name a procedure and pass that procedure as an argument to another procedure, that procedure will not be optimized by Larceny. With eta conversion allowed by the R6RS semantics, Larceny's optimizations would not have to be restricted to first order procedures. I don't think Larceny takes advantage yet, but Larceny is an R5RS/R6RS implementation. If the R7RS were to allow eta conversion, so Larceny would no longer have to support R5RS semantics, then Larceny would do a better job of optimizing higher order procedures. PC Scheme definitely used eta conversion, to solve exactly the same problems that limit Larceny's optimization to first order programs. The PC Scheme compiler may not have been strictly legal (it was pre-R5RS, and I'm not going to bother to look up the relevant standards of its time), but there are some programmers who value performance more than Spanky-level conformance to some committee's arbitrary decisions. I don't know whether any of the other R6RS implementors have yet taken advantage of the eta rule. Chez Scheme and Racket are the most likely implementations to have done so, but you'd have to ask Kent Dybvig and Matthew Flatt. (I suspect that typed Racket and contracts rely on eta conversion, but those languages are not limited by R5RS or R6RS semantics.) Will _______________________________________________ r6rs-discuss mailing list r6rs-discuss@lists.r6rs.org http://lists.r6rs.org/cgi-bin/mailman/listinfo/r6rs-discuss
