> In retrospect, perhaps we should have considered introducing "let"
> bindings automatically to preserve left-to-right semantics within the
> push-enter model (like Moscow ML does, I think), although this entails
> a performance hit for the bytecode interpreter.

Let me put in a vote for left-to-right ordering.  I've also
been bitten by the evaluation order (with almost exactly the
same kind of code -- reading a record's values from some file).
But more importantly, the primary reasons I chose SML over 
O'Caml for my class this semester is that the evaluation 
model of SML is more uniform.  Left-to-right in O'Caml might
well change my mind.  

On this note, I find the discussion of syntax very interesting.
Teaching either variant of ML to a group of students raised
on Visual Basic, Java, and Javascript is not easy, and today's
ML implementations are not very student-friendly when it comes
to either parse- or type-error messages...

-Greg