The EIP aims to strike a balance between on the one hand reversing the unexpected harmful consequences of gas refunds (namely, gas tokens and block size variance) and on the other hand maintaining the benefits of original motivation for refunds (good state hygiene).
Part of that balance is to maintain a strong incentive to restore storage slots to zero, if those same slots were zero at the beginning of the transaction. I would like to suggest modifying the EIP so that the analogous refund is available for creating and destroying the same contract in a single transaction.
Such “ephemeral” contracts are an elegant pattern for state channel contract architecture: a (mostly) private interaction between a small number of parties who unanimously agree should not need to leave any trace on the L1 chain. In such a pattern, parties pay assets to a predictable CREATE2 address, and after their off-chain interactions proceed to launch a transaction which i) deploys a contract at the CREATE2 address which then ii) verifies their signatures , iii) pays out the funds and iv) self destructs.
I’m a contributor to statechannels.org, where we are seriously considering the use of ephemeral contracts. I estimate that under this EIP gas costs for state channels using this pattern may increase by approximately 20%. It would also disincentivize our current efforts at good state hygiene and result in many contracts remaining in state unnecessarily. Furthermore, because (as is) the EIP is somewhat asymmetric with respect to modifying state-based refunds and contract based-refunds, the differential gas cost with respect to other state channel contract architectures (that we have considered) might lead to neglecting the ephemeral contract pattern entirely, which may lead to even more state bloat.
Under my proposed modification, the incentives and gas would remain mostly as they are now for the ephemeral contract pattern (since the NEW_MAX_REFUND_QUOTIENT allows for the 20% refund we currently enjoy). It does not re-allow gas tokens, since self-destructs would only be allowed for contracts created earlier in the same transaction. More gas will have been spent creating the contract than will be refunded.
I am unsure, however, if it is as straightforward for clients to avoid writing to storage at all in such transactions (the analogue of the “prudent optimisation” of the EIP). Perhaps someone with more knowledge can comment?