EOF + Verkle
Context & goals
- EOF considered for Prague
- Verkle considered for Osaka
- How can each EIP affect each other?
- Impact for:
- Ethereum users?
- VKT spec and implementation?
- Risks for EOF adoption?
TL;DR verkle trees
- Motivation: viable stateless clients
- Introduce execution witness, which has:
- Account data
- SC code
SC code is now in the state tree
TL;DR verkle trees
- Deploying bigger contracts: more gas (than today)
- More bytecodes required during exec: more gas (new gas cost dimension)
TL;DR EOF
- More structure in SC code (Header/Type sections)
- Execution efficiency gains
- Avoid JUMPDEST analysis
- Easier to add/remove features/instructions.
- Potential effects:
- Bigger code?
- Read more bytecodes during execution?
EOF + Verkle (gas risks)
- EOF might increase code size
- EOF could read more bytecodes in contract execution
- VKT makes those actions add more gas costs
- Caveat!:
- EOF bring new efficiencies that might offset new costs.
- Compilers can optimize code layout to account for the new code-chunk gas cost dimension and EOF sections.
- By how much? ->
EOF + Verkle (gas risks - idea)
- Requires a client with both EOF+VKT implemented
- Testnet with VKT + EOF
- Deploy contracts both compiled with legacy and EOF
- Tx executions to compare both
- With EOF-compilers accounting for both EOF and VKT optimizations? (ideally)
- Feedback for both EOF and Verkle specs to adjust!
EOF + Verkle (EOF adoption risk)
- [Depends if EOF vs legacy has a big gas difference!]
- Users might deploy EOF vs. legacy depending on which has a better gas cost for them.
- Hard to stop accepting new legacy SCs?
- Accepting two contract execution modes forever?
EOF + Verkle (VKT complexity risk)
- How does code-chunking for EOF work?
- Eventual double-spec for legacy vs EOF executions?
- EOF introduces ~19 new instructions: new witness gas extra cost for those? (e.g: BALANCE, BLOCKHASH)
- EOF could delay Osaka: state growth make MPT->VKT transition longer?
Final words
- EOF and VKT implementations looks like non-overlapping, but not their effects
- EOF original assumptions didn't expect code to be in the tree
- VKT needs that for stateless clients
- Described (potential) complications emerge!
- Can deploying a minimal-EOF version help?
- Not claiming there's a big problem. Claiming we don't know and need to find out!
EOF + Verkle
{"title":"EOF + Verkle","description":"EOF Verkle slides","slideOptions":"{\"theme\":\"moon\"}","contributors":"[{\"id\":\"46f4f11e-c5fd-4f0b-9417-1132b8336f80\",\"add\":8728,\"del\":6179},{\"id\":\"94149ceb-593b-4989-b5d1-0e0af0737310\",\"add\":7,\"del\":7}]"}