FWIW: My sieve interpreter does exactly what we're (not) allowing:
AFAIK nobodhy is talking about not allowing this. The question is whether or
not we should specify all of the nuances involved in supporting this mode of
operation, and if we're going to whether or not this is the right place to do
it. I remain to be convinced of the need to specify this and I'm opposed to
doing it in this document.
I
fetch the sieve script as part of verifying that the rcpt to address
exists, and evaluate until some test fails for lack of data. I think
the draft as it stands is fine.
Our implementation doesn't work this way but nevertheless allows some types of
Sieves to be evaluated at RCPT TO or even MAIL FROM time.
Ned