In a message dated 5/7/00 3:54:06 PM Central Daylight Time,
d96xxxxxx@d.kth.se writes:

> But my opinion is that the SRFI should not mention ANY algorithm, it
>  should leave it up to implementors. A SRFI is a specification, not an
>  implementation, isn't it?

I think I disagree with this part. So far, a SRFI is a specification,
INCLUDING a _reference implementation_. While I agree that the specification,
per se, should not require a specific algorithm, the reference implementation
MUST--in my opinion. (My impression was that Olin agreed to remove the
reference to KMP in the body of the specification.) In particular, any SRFI
which claims to be implementable under a R5RS-compliant implementation should
be complete--and runnable under such an implementation. In the case of this
strings SRFI, that means it must include some algorithm like KMP.

With "library SRFIs" such as Olin's list and string SRFIs, I would be very
disappointed if the SRFI (at least the reference implementation) "left things
out" which would prevent me from using the SRFI as is in the two Schemes I
use.

Jim Bender