Author: Potuz, Developer Source: X, @potuz_eth Translation: Shan Ouba, Jinse Finance
In the current climate, I am quite bearish on on-chain rollups, and I think L2 preconfirmations are nothing more than castles in the air (let’s just say that). But I think the vision of on-chain rollups is valid, and nothing more than a distant dream. This brief discussion below will explain why.
People complain about rollup UX and fragmented liquidity. They’re right. L2s are run by profit-seeking companies, and even if those companies are fully aligned with Ethereum and sincerely looking to scale L1, they’re incentivized to capture activity. They want Uniswap to run on their L2, and they want fast access to L1 state so that users have a better experience during the bridging process. They can even have different components that interoperate well with each other as long as you don’t leave L2. Think Arbitrum’s Orbit chain or Optimism’s Superchain.
But rollups shouldn’t need to drive activity, they should only drive execution. They shouldn’t be the place where people do Uniswap trades, Uniswap should send the intensive computation to rollups, but the actual activity (and its assets) should stay on L1.
In a sense, rollup should just be a coprocessor for L1 , which can move some computations that don't require atomicity and allow slightly weaker security to rollup (either trusting a zero-knowledge proof system or trusting the 1/N assumption of the validator). In this hypothetical world, users would send transactions on L1, their ETH in exchange for pixelated monkey JPEGs or something else, and any computation would be done on some L2, but both ETH and the monkey would remain on L1.
In this hypothetical world , L1 wants fast access to L2 state, but not the other way around. It wants callbacks from transactions on L2. It wants to have special opcodes to trigger execution on L2 (yes, this inevitably leads to a rigid rollup scheme). Flashbot’s SUAVE vision is not far off from this (but requires sidechains, some TEEs, and Intel’s trust assumptions). And, if anything, rollups based on on-chain data seem more compatible with this vision than general-purpose rollups that pursue activity.
If a rollup project openly built an L1 coprocessor based on on-chain data, I would be much more bullish on it, and I would even temporarily accept the lie about decentralization (all pre-confirmation based designs I have seen are worse than a centralized trusted sorter).
