To clarify, we've assembled quick rundown on how different forks work.
Before we get into the classifications, it's worth noting that bitcoin forks already occur quite regularly.
A byproduct of distributed consensus, forks happen anytime two miners find a block at nearly the same time. The ambiguity is resolved when subsequent blocks are added to one, making it the longest chain, while the other block gets "orphaned" (or abandoned) by the network.
But forks also can be willingly introduced to the network. This occurs when developers seek to change the rules the software uses to decide whether a transaction is valid or not.
When a block contains invalid transactions, that block is ignored by the network, and the miner who found that block loses out on a block reward. As such, miners generally want to mine only valid blocks and build on the longest chain.
Following are some of the more common forks and their traits.
What is it? A hard fork is a software upgrade that introduces a new rule to the network that isn't compatible with the older software. You can think of a hard fork as an expansion of the rules. (A new rule that allows block size to be 2MB instead of 1MB would require a hard fork).
What happens? Nodes that continue running the old version of the software will see the new transactions as invalid. So, to switch over to the new chain and to continue to mine valid blocks, all of the nodes in the network need to upgrade to the new rules.
What can go wrong? The problem comes when some sort of political impasse arises, and a portion of the community decides to stick by the old rules no matter what. The hash rate, or network computing power, behind the old chain is irrelevant. What matters is that its data (and ruleset) is still perceived to have value, meaning miners still want to mine a chain and developers still want to support it.
The ethereum DAO hard fork was a perfect case study of how a community can split over rules. Now, we have two blockchains using a variant of the software – ethereum and ethereum classic, both of which boast a different ethos and a different currency.
What is it? A soft fork, by contrast, is any change that's backward compatible. Say, instead of 1MB blocks, a new rule might only allow 500K blocks.
What happens? Non-upgraded nodes will still see the new transactions as valid (500k is less than 1MB in this example). However, if non-upgraded nodes continue to mine blocks, the blocks they mine will be rejected by the upgraded nodes. This is why soft forks need a majority of hash power in the network.
What can go wrong? When a soft fork is supported by only a minority of hash power in the network, it could become the shortest chain and get orphaned by the network. Or, it can act like a hard fork, and one chain can splinter off.
Soft forks have been the most commonly used option to upgrade the bitcoin blockchain so far because it's argued they present a lower risk of splitting the network. Past examples of successful soft forks include software upgrades like BIP 66 (which dealt with signature validation) and P2SH (which altered bitcoin's address formatting).
User-activated soft fork
What is it? A user-activated soft fork (UASF) is a controversial idea that explores how a blockchain might add an upgrade that is not directly supported by those who provide the network's hashing power.
The idea with UASF is that instead of waiting for a threshold of support from mining pools, the power to activate a soft fork goes to the exchanges, wallets and businesses who are running full nodes. (In bitcoin, a full node, even if it is not a mining node, is still responsible for validating blocks.)
What happens? The majority of major exchanges would need to publicly support the change before it could be written into a new version of code. After that, the new software (which has an activation point in the future) gets installed on nodes that want to participate in the soft fork.
What can go wrong? This method requires a much longer lead time to work than a hash-power-triggered soft fork. In fact, it's believed it may take as long as a year or more to write the code and get everyone ready.
Further, if the majority of miners end up not ‘falling in line’ and activating the new rules, they could use their overwhelming hash power to split the network.
Currently this idea is theoretical and has not been implemented.