Most of the errors can be fixed by retrying (e.g Minix 3 or Erlang)

"That works, to a degree, for random faults that are typical for hardware. Such retry approaches are fundamentally based on three assumptions:

Faults are rare
Faults are stochastic (truly random)
Multiple faults are uncorrelated.

For software, and especially attacks against security of software systems, *all* three are wrong:
faults are common – 1–3/kSLOC, see above
faults are completely deterministic, once you know how to trigger a bug, you can always trigger it again
once you found a way in by triggering a bug, you can look for the next one to get further, and you can chain them up deterministically at any time.

In short, retrying does nothing against security exploits." -- Gernot

"Retry is great, separate state is great, Erlang and Joe Armstrong are great, we steal a lot of the ideas including each state is sort of functional" -- Phil