something that works 90% of the time is unacceptably unreliable.
something that works 99.999% of the time for 90% of users might be
acceptable, provided that the 90% contains essentially everyone with a
reasonable expectation of having mail autoconfigured.
IP does not "work" when there is no data-link layer. ICMP does not
"work"
when firewalls filter it out.
those examples are irrelevant.
Moreover, SRV "works" in more situations than all of the others.
handwaving is no substitute for analysis.
I mean, DHCP is pretty cool but it doesn't "work" when I'm dialing up
from
an airplane and getting "local" PPP (that changes every 300 miles) and
no
localized proxy gateway to my home server.
DHCP doesn't work for lots of other cases too - it's fine for getting a
netmask and default route and a Bad Idea for nearly everything else
because most defaults should be independent of network location.
Unfortunately DHCP has been widely used because it works "in some
cases" or even "in 90% of cases" (where those cases are drawn from a
restricted set of cases that happens to look typical to the speaker).
So if you're trying to convince me that DHCP sucks, consider me
convinced. But that doesn't do a thing to convince me that SRV
configuration of MUAs doesn't suck.
SRV works in most situations. It's the lowest common denominator.
That's the same kind of argument made in favor of DHCP, and it's really
bogus. I'm perfectly willing to believe that a SRV-based solution
might be a good idea IF the argument is supported by extensive case
analysis rather than some unsupported blanket statements.
Keith