> > There is no way we can predict beforehand that a network
> > connection or a target machine will break on next use.
> > Whatever the quality of the input data....
> That information is external to the program, and enters it through
> function calls; thus, it is still input data.
Barry, I'm afraid something is not getting through. I am trying
to say that checking at time T that a connection is alive wont
tell us anything about its state at time T + 1 microsec, when we
perform a write.
Some faults just cannot be predicted. That's why the relevant field
is called fault tolerance and not fault avoidance.