EOF Implementers Call #63

[poojaranjan]

Meeting Info

Dec 11, 2024 , 15:00 UTC

Duration: 60 minutes

Zoom: https://us02web.zoom.us/j/88940506383?pwd=aTdsbHVyMTNDSUFHYmhTWlI2ZEVldz09

📅 Subscribe to the Ethereum Protocol Call calendar for calendar invites

Resources

• Current specs
  • EIP-7692: EVM Object Format (EOFv1) Meta
  • EIPs Included
    
    EIP-3540: EOF - EVM Object Format v1
    EIP-3670: EOF - Code Validation
    EIP-4200: EOF - Static relative jumps
    EIP-4750: EOF - Functions
    EIP-5450: EOF - Stack Validation
    EIP-6206: EOF - JUMPF and non-returning functions
    EIP-7480: EOF - Data section access instructions
    EIP-663: SWAPN, DUPN and EXCHANGE instructions
    EIP-7069: Revamped CALL instructions
    EIP-7620: EOF Contract Creation
    EIP-7698: EOF - Creation transaction
• EOF Meeting Playlist
• EVM (EOF) Taxonomy
• EOF explainers on PEEPanEIP
  
  EVM Object Format (EOF) with Danno Ferrin
  EVM Object Format (EIP-3540 & EIP-3670) with Andrei Maiboroda, Axic & Pawel
• EOF Testing Coverage Tracker
  • Implementation Matrix
• Solidity + EOF POC summary
• Matrix on evmobjectformat.org
• Prev meeting notes

Agenda

• Testing updates
• Client updates
• Compiler updates
• Tracing:
  • Updated EIP-7756
    • PC=0 is always the first byte of the container
    • Number and HexNumber are interchangable
    • OpCode now defaults to hex
    • gas members now default to number
• Fuzzing:
  • branch with WIP fuzzing tool shemnon/eof-fuzz
    • Needs a patch to goevmlab to work - Non-osaka EOF changes holiman/goevmlab#179
  • Corpus is the single test set of reference tests
  • 4 mutators currently (more to come)
    • Add PUSH0/POP
    • Replace PUSH value with in-test address
    • Replace PUSH value with "magic" number (mostly 2^x and friends)
    • Replace PUSH with random bytes (biased to shorter strings)
  • Found 3 bugs so far
    • Besu 32/64 bug in EOFCREATE
    • Geth DATACOPY overflow
    • Nethermind EXT*CALL gas costing issues
  • plan is to write more fuzzers to bring it on par with FuzzyVM, but for EOF.
• Feedback and next steps for EIP-7834: Separate Metadata Section for EOF Add EIP: Separate Metadata Section for EOF EIPs#9100

Please add other agenda items or links to discuss.

Next call on Jan 08, 2025

[frangio]

Discussion topic postponed from last call:

• Generic factories and deterministic deployment (probably relevant to counterfactual deployments)

[shemnon]

I'd like to talk about tracing and fuzzing.

Tracing:

• Updated EIP-7756
  • PC=0 is always the first byte of the container
  • Number and HexNumber are interchangable
  • OpCode now defaults to hex
  • gas members now default to number

reth and evmone will need to adjust to PC=0, geth, besu, and nethermind have already adjusted

Fuzzing:

• branch with WIP fuzzing tool shemnon/eof-fuzz
  • Needs a patch to goevmlab to work - Non-osaka EOF changes holiman/goevmlab#179
• Corpus is the single test set of reference tests
• 4 mutators currently (more to come)
  • Add PUSH0/POP
  • Replace PUSH value with in-test address
  • Replace PUSH value with "magic" number (mostly 2^x and friends)
  • Replace PUSH with random bytes (biased to shorter strings)
• Found 3 bugs so far
  • Besu 32/64 bug in EOFCREATE
  • Geth DATACOPY overflow
  • Nethermind EXT*CALL gas costing issues
• plan is to write more fuzzers to bring it on par with FuzzyVM, but for EOF.

[kuzdogan]

Feedback and next steps for EIP-7834: Separate Metadata Section for EOF

ethereum/EIPs#9100

[charles-cooper]

feedback on ethereum/EIPs#9074 -- it should probably be accompanied by instruction + container changes

[poojaranjan]

Recording: https://youtu.be/2Z5YPfOnb74

[poojaranjan]

Closing in favor of #1217

[shemnon]

Agenda

• Clients and Compilers
• Spec Updates
  • EIP-7834 - EOF Metadata section
  • Generic factories and deterministic deployment (probably relevant to counterfactual deployments)
  • EOFCREATE Hashing
  • code size EIP
  • EIP-7765 tracing
• Testing
  • EOF Execution Fuzzing
• Other
  • Cancel 25 Dec meeting, next meeting 8 Jan

Meeting Notes

• Clients and Compilers

  • Solidity
    • some functionality merged,
    • updating solidity tests (proves correct compilation)
    • working on stability next
  • Frangio is working on a new EVM assembler for EOF and Legacy
    • Do any assemblers output EOF as well? Other than YUL?
    • EXCHANGE opcode textual representation, how? [n, m-n] or [n, m]?
    • (mixed prefs, closer to bytecode/Internal encoding vs user meaning)
    • proposal will be made for review in 2 weeks

• Spec Changes

  • Metadata
    • Possibly change data to 0xff (and reserve 0x04)
    • Not all containers use metadata in runtime containers
      • initcode only,
      • tx data only
      • Event during initcode
    • Keeping initcode separate from runtime code makes sourcify's job harder
      • metadata in tx only requires archive data of all deployment transactions
    • There is an appeal to unreadable code in runtime, but perhaps we need different conventions for compiler metadata
    • Mixing metadata with code produces runtime issues with indexes into the data and makes metadata discovery difficult.
  • Generic Contracts and Deterministic / Counterfactual Deployments
    • There is a desire for contracts at the same address on multiple chains
    • Ways it's done
      • Nick's method is used to create a single-use TX to make a fixed address (chain #0, high fees)
      • CREATE2 deployer deployed w/ Nicks Method (most used factory)
      • createx - https://github.com/pcaversaccio/createx
    • createx factory is not implementable in EOF right now.
      • Will TXCREATE fix this? (https://github.com/ipsilon/eof/blob/main/spec/eof_future_upgrades.md#txcreate-and-initcodetransaction)
    • App devs use both salted create2 and nick's method
      • Nick's method is hard on some other chains, no toehold.
    • Can only deploy code in containers, not arbitrary contracts
    • General question on how to get reliable addresses in Eth w/o "toehold process."
  • Unsalted deployed - nonce derived address
    • EOFCREATE and TXCREATE require a salt, no nonce.
    • Could the factory maintain this?
    • Possibly remove the "witness" from the address. (EOFCREATE: Don't hash the init-container ipsilon/eof#162)
      • Witnesses include: initcode, code address (sender or caller), initcode container index
      • Removing the witness makes counterfactual addresses more difficult.
  • Time boxed at 15 minutes

• Code Size EIP

  • Needs to be accompanied by EOF changes to support > 64KiB containers / code / etc.

• Testing

  • Devnet 5 is the focus, no EOF work due to priority
  • Fuzzing and Tracing overview