Opt-in hard-fork without alternate transaction history?
IETF’s RFC 7282 is an eloquent document which describes important aspects on consensus, and worthwhile if you want a more nuanced interpretation than “widespread agreement and disagreements addressed (even if not acommodated)”.
Measuring Consensus
Once we have a concrete technical proposal, and it seems to have some traction, we need to figure out if we really have consensus before it gets deployed.
Moving from RFC 7282 style technical rough consensus to economical and political (rough) consensus is quite problematic. If you want to stay in the spirit of RFC 7282 then you should only use polls to see if there is any opposition. You then need to actively go out and figure out what people’s concerns are and make sure those are reasonably addressed. You have to go through all that before you accept anything below 100% support.
This seems impossible in many cases, as non-technical objections can go all over the place; you may end up having to refute all of Nitsche to adequately address some convoluted philosophical objection and finally reach rough consensus between all users. It gets even worse if you need to consider potential future users.
The most pragmatic way out of this problem seems to be to make changes opt-in, hence a preference for soft-forks (though not all kinds).
This doesn’t work when it comes to hard forks; you can’t guarantee they won’t be controversial. Once a hard fork is controversial, exchanges start trading it and users will get confused. Replay protection doesn’t solve this problem, because users still need to choose which chain they believe in which is an enormous burden. They might not have agreed to the code changes had they known this outcome.
You can certainly hold off on any hard fork while it’s controversial, but you can’t predict if it suddenly becomes controversial after the point of no return.
The easiest solution would be to never risk a hard fork. One problem with that solution is that you can’t stop others from doing a hard fork and persuading a large economic and hash power majority to join. There’s always someone willing to take more risk. When the scope of this fork is far outside technical rough consensus, perhaps ignoring it and informing users about the risks is the best approach. However when it’s close to rough consensus, pre-empting may be better than ignoring. Thus it may be prudent to have one or more well tested hard-fork candidates ready to go at any moment, even if the preference is to never deploy them.
A second solution, something I think is worth (re)considering, is to kill off the original chain during a hard fork. Perhaps through some sort of merged mining, where the old chain only gets empty blocks or through a soft fork which makes the entire UTXO set unspendable on the original chain. This requires being even more certain about non-technical consensus, which I’ve argued above is near impossible.
We may need to look for a third solution. Something that is opt-in but doesn’t create two alternate transaction histories.