On Sat, Oct 15, 2005 at 06:07:43PM +0100, Alexey Melnikov wrote:
The case I had in mind is as follows: The notification URI is stored
elsewhere (as an LDAP attribute or an IMAP annotation) and Sieve scripts
uses Sieve variables extension to specify the notification method. Is it
desirable for an unrecognized notification method to cause runtime
error? I am open to suggestions.
People who use notifications usually consider some (or all) mails very
important. Silently ignoring errors, in particular when variables are
involved and the error only happens every now and then, will make them
think notifications were sent when they were not.
I agree that error notification is important and should be done when possible.
However, I also think it is important to keep some perspective here. Many
notification services are in fact NOTORIOUSLY unreliable. Even worse, error
reporting when failures occur is in many cases impossible as a result of how
notification services are set up or even as a result of the design of the
service.
SMS are not reliable
and they may complain about the service.
Right, but this cuts both ways. Errors should be reported when possible in
order to prevent blame from being laid at the wrong door - if it is SMS that
failed fine, but if there's a busted sieve script the user needs to know that
in order to distinguish it from an SMS failure. OTOH, people routinelly reach
bogus conclusions about where fault lies and no amount of persuasion will
convince them they're wrong. So, while requiring error reports may help in some
cases, in many others it will make no difference.
Ned