Op 20-9-2011 0:53, Barry Leiba schreef:
I personally prefer the simplicity of doing the action and test together. That
said, I'm happy to specify that the test not take the action, if that's the WG
consensus.
I see a problem with that approach: the test can determine whether the
conversion is supported without performing it. But it can't determine whether
it will succeed until it performs it. If the input is corrupted or
mislabelled, how do we have a condition on that? Would it always have to make
the script fail with a runtime error?
So I'm looking for others to weigh in. Do you prefer the action to be taken
along with the test? Or do you prefer having to repeat the text, once for the
test and again to do the action?
I'd like the test to have the action side-effect. That is indeed also
what I understood from your initial query. I agree that first testing
and then performing the action is much of a hassle, both for the
implementer of the Sieve interpreter and the Sieve script author.
From what I see in RFC5228 section 14 and in the old RFC3028, tests
having side-effects were originally forbidden explicitly, but are
allowed now as per RFC5228. I guess there is a reason for this. However,
to my knowledge, this is the first time this line is truly crossed, so I
understand Alexey's reservations. Anyway, I do think the advantages
outweigh the concerns in this case.
Regards,
Stephan
_______________________________________________
sieve mailing list
sieve(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/sieve