Hector Santos wrote:
(i) Two different reasonable people could interpret the reply
differently.
More like 1 in a million. :-)
No; see John's example:
250-OK
550 Not OK.
(ii) It seems like a confusing hack.
eh, hence why we want to legitimate it.
We do NOT want to legitimize hacks. As I said, your use-case relies on
accidents of implementation.
(iii) The only use-case proposed so far relies on client implementation
details upon which the server SHOULD NOT rely.
Then why bother with with 45x and 55x in tying to control clients? In
reality, you can't rely on that either.
You can rely on the fact that an RFC-Compliant client will react as
expected to 4xx and 5xx codes. You can't predict at all what an
RFC-Compliant client will do with a multiline response with different
codes.
(Obviously, if the client is evil ratware, then you cannot predict how
it will react to 4xx and 5xx codes, but that's outside the scope of the RFC.)
The fact is, it hasn't shown to be a problem. Now is the time to make
it OFFICIAL. Who want to wait until 5-10 years?
I don't think evil hacks should be baked into RFCs. That's the process
that leads to abominations like the Win32 API.
Regards,
David.