[I was serious when I thanked you and not joking.

  > At least you didn't bring address-of calculations from C.

  I know I'm not the Scheme insider you are, but was that really
  called for?

Thanks for the compliment but I am not a co-author of R5RS and indeed, I am
not even referenced in the report. The honor goes to others. -- I am truly
concerned about adding features to Scheme that violate basic safety
concerns. Your proposal could easily slide down this slippery slope.]

;; ---

R5RS isn't a perfect specification but it nails down what kind of syntactic
things can appear where, which behavior is specified, and which one isn't,
when errors must be signaled, when they can be signaled, and so on.

SRFIs (and not just yours) should respect this language, and I'd like to
encourage the editors to assume some editorial powers in this direction. It
will help convince system authors to add these features.

;; ---

In this specific case, I think we would like to see something like

    "
     (set! exp exp)			  expression

     The lhs is evaluated with a left-hand side
     evaluator. If the evaluation does not yield a
     reference, an error is signaled. If it does,
     the reference is modified.
    "

if this is what you intend. As of now, I am sorry, but I simply don't
understand the proposal.

-- Matthias

P.S. define-syntax lets you use the patterns you want and you can shadow
keywords. So there should be no problem using it for your specs. (But I
must admit, I haven't implemented a define-syntax system.)