I was under the impression that, moving forward, the process would be at least somewhat like the soft fork process.
There is no real safe way of doing it.
Quote
Sure, miners can't jam it through, but there has to be some sort of orderly rollout. If BC Core rolled in support for >1MB blocks and such blocks started appeared overnight, the results would be...ugly, to say the least.
With a hard fork, getting consensus first is important.
The problem is how to handle it if complete consensus isn't achievable. The 95% rule with soft forks recognises that getting everyone to agree is to high a standard, but it still requires near total agreement.
It is pretty clear that complete consensus isn't required, even for hard forks. If 95% of miners and 95% of merchants updated, then everyone else would have to update too.
Quote
Part of the problem is that there's no way to upgrade everybody, unlike the early days, when there were very few users and (IIRC) Satoshi induced some hard forks due to catastrophic bugs. The current soft fork standard seems fine to me since it's impossible to get all users on board; alerts can go out on the network, and somebody will always refuse to upgrade for any number of reasons.
Of course, I could be wrong about all this.
Also, as you pointed out, any BIP that doesn't require a fork can be implemented at any time. For example, BIP 32 is pretty widespread now, and that one didn't require a fork.
Of course, I could be wrong about all this.
Also, as you pointed out, any BIP that doesn't require a fork can be implemented at any time. For example, BIP 32 is pretty widespread now, and that one didn't require a fork.I think giving notice is the best bet. If 95% of the miners say that "starting with block 400,000 we will be using larger blocks", then that gives notice to everyone else to update.
Satoshi could have made the 1MB "anti-DOS" rule expire and I don't think anyone would have objected.
The suggestion on the mailing list is to look at the client names in the version message. If most nodes end up using the large-block version of the client, then that shows that there is wide support for larger blocks. It wouldn't be an automatic thing, but a way to show that there is consensus for larger blocks.
The large block debate could set the rules for how to handle hard forks that don't have complete agreement.
