Applications have to deal with more then just losing a
connection. They have to deal with the loss of state that occurs when
you lose a connection. In general you really don't know which
transactions finished and which ones didn't, so you have to re-sync
your state in some way.
I believe that no matter what we do to provide connection persistence,
connections will still break. Whether they break because of a total
loss of path for a long period of time, or more likely path is lost
for longer then the human timeout (i.e, the human starts typing ^C
[unix] or clicking CANCEL [gui] or whatever).
As long as connections break, applications will have to deal with
getting back to work when a connection can be made again. Given I
don't believe you will ever get "perfect" connection persistence (even
if you can do better then TCP does today) so therefore applications
will always have to have that re-syncing baggage.
-Jeff