Ethereum Core Devs Meeting 51 Notes
Meeting Date/Time: Fri, December 7, 2018 14:00 UTC
Meeting Duration: 1.5 hours
GitHub Agenda Page
Audio/Video of the meeting
Agenda
- Testing
- Client Updates
- Research Updates
- Working Group Updates
- Constantinople HF
- Openness of meetings
- ProgPoW update
- Quick announcement re: coordinator roles
- Date for next meeting?
Testing
- Brian intro
- Recently hired to work full time on making testing better
- Not sure what that means yet - easy to generate and use tests
- I’ve talked to Trinity, have an idea what they want
- Please reach out and tell me what you’d like to see improved
- Dimitry
- Finishing EXTCODEHASH implementation
- Discovered that every EXTCODEHASH test can be used to test *SIZE, *COPY, and *CALL
- Not related to Const.
- Each could be enhanced like that
- Received a lot of help writing these tests from [Andrei?] and Hugo
- Finishing things still on list
- Still waiting for contribution on bitshift test review
- Waiting for hive results to see if there are any issues we need to fix, or new tests we need to write
- Martin said Hive has some maintenance
- CREATE2, SSTORE tests ready, bitshift ready but needs review, EXTCODEHASH mostly covered
- Martin
- We’re in process of rewriting Hive and making it more robust
- In the meantime it’s kind of flakey
- Will also see some additional tests, P2P stuff I mentioned previously
- Regarding tests
- Updated all tests for geth less than two weeks ago
- All tests pass
- Kind of expected client devs to import and run tests in native settings, despite having Hive
- But hoping to have Hive up and running again within next couple of days
- No new failures on fuzz testers running for at least last couple of weeks
- So I’m fairly confidence in code in geth and parity
Client updates
- Harmony/EthereumJ (Mikhail)
- No major updates
- Mostly working on Beacon chain impl., been major spec changes lately, still catching up over next couple of weeks
- Parity (Fred)
- Released 2.2.2 with new improvements, some exciting stuff
- Improves sync performance
- Pairing stability, fixed pairing issues where it now finds a lot of pairs but fluctuation in pair counts
- Block propagation fixed - waiting to see effect on network
- Other minor things
- Decided to move jsonrpc APIs not yet accepted in an EIP but implemented anyway (eth.person or web3 namespaces) - now disabled by default, enable with flag
-jsonrpc-experimental
- Geth (Peter)
- Database optimizations
- Just trying out ideas and looking at charts
- Turbogeth (Alexey)
- Fixed bugs found with Ropsten sync, syncs now
- Started to work on test suite
- Never run tests, have issues, fixing them
- In order to make it work, had to modify boltDB
- Now can run boltDB in in-memory mode, nothing on disk, wanted to try this for a long time
- Client still not ready for Const., CREATE2 revival still not addressed, will address this
- Found myself using turbogeth a lot recently as it’s useful for extracting data I used for state rent proposal
- 300gb database on external DB, made some copies of it and running data analysis
- Aleth (Pawel)
- Fixes and improvements to network code
- Probably today will have a new tool that will work as a bootnode, can be used similarly to geth bootnode, want to put somewhere on mainnet or testnet as alternate implementation
- Nimbus (Jacek)
- Been focusing on parts of VM related to DevEx, e.g. debug and tracing
- Let us know if you have any ideas, this is a good time to discuss
- First Eth 2.0 testing framework and PoC things are landing right now, discussion on gitter
- Good time to influence direction of Eth 2.0 testing
- https://gitter.im/eth2-0-tests/Lobby - Ethereum 2 testing working group looking for feedback on the first proof-of-concept test vectors - now is a good time to join :)
- Pantheon (Danno)
- Planning on minor release next week
- Includes contribs from external contributors
- Including Goerli support
- Looking into [missed this]
- Beacon chain impl.
- Mana (Andrew)
- Syncing Const. blocks on Ropsten, anticipating full sync shortly
- Adam (Swarm)
- Swarm is a part of geth, in same repo with own semantic versioning
- Last geth release contains swarm improvements such as access control on feeds
- Will be announced with a blog post coming with next major release of geth
- Last two weeks, introduced user perception tests [?] and benchmarks
- Ewasm (Lane)
- Working towards testnet spec v3
- Released #Eth1x proposal, working towards this
Research
- Danny update
- Revisions of phase 0 - Serenity beacon chain spec
- Readability, reorg., polishing data structures - major edits of spec
- Tough to keep up, homing in on first version of release candidate
- Phase 1 - sharded data chain algo. - data structures forming there
- General algorithms known
- Being specified
- State execution, account structures - lots of thought going into this
- Handful of proposals on ethresear.ch that V posted that we’re looking for feedback on
- Justin and VFD alliance: increasing number of blockchains interested in adding VFDs in various ways
- Begun to do Beacon chain implementation in py-evm
- Justin did talk on VDFs at Devcon, video online
- Other DevCon IV talks should be online today
Working group updates
- Whiteblocks (Zak)
- Putting together specs for testnet
- Working with 1x in simulation group
- Relevant for testers as well
- Will schedule separate call today that I’ll post in issue that I opened
- Please join and provide feedback, we can start getting specs for testnet and building it out
- Will be powered by Whiteblock framework
- Alexey
- I have a writeup on state rent - first version of proposal published, received a lot of feedback
- Most interesting one: V suggested using CREATE2 opcode and some other things to implement something I proposed to do with linear cross-contract storage
- I managed to implement ERC-20 contract based on these ideas
- It actually works, can mint and transfer tokens
- Interesting to get feel of how you’d use CREATE2
- Commented on issue in solidity, created thread on EthMagicians for ppl interested in creating new primitives to make it easier to work with CREATE2
- Right now you have to copy and paste bytecode into source code
- Linear cross-contract storage will most likely be dropped from next version of proposal
- Priority queue too
- But I’ll still look into linear storage elsewhere e.g. Ewasm integration
- I’ve asked PegaSys team to help with PoC, Adrian Sutton created first prototype, reviewing and hope to get my prototype running as well - so this is based on pantheon for now
- Done a lot of data analysis, not all included in proposal
- Trying to run heuristics to identify all ERC20 tokens in state
- Looking for successful token transfers
- Around 71k contracts
- Take around 53% of all contract storage
- Cryptokitties is also ERC-20 - has both this and 721
- Very important class of contract, that’s why I did sample implementation using V’s ideas
- Next step will be to look into on-chain order books
- Put some ideas about how to identify them
- Will do some analysis about token nexus contracts - tokens go in and out - likely will be on-chain order books
- Might be able to compress state using generations
- Looking into how we can have a very compact representation of current state
- E.g. all contracts that GasToken creates could be generated using minimal seed data
- So it would not take a lot of space in the snapshot
- Not sure how well suited other clients are for data analysis - but someone could look at Google Ethereum dataset - see if it’s possible to reproduce the data analysis I did, some in state rent proposal, plus this ERC20 analysis
- If someone gets the hang of this, might be a good way to split up this work, there’s so much to do in this workstream and I can’t do it all alone
- So trying to enlist people to do PoC, data analysis, maybe mods to Solidity and Vyper to make all this easier
- Ewasm
- See proposal and FEM thread
- Peter “working group” on state pruning
- FEM thread: https://ethereum-magicians.org/t/ethereum-chain-pruning-for-long-term-1-0-scalability-and-viability/2074
- Long Github gist about two weeks ago (proposal)
- In blog post I mentioned that I think there are two viable approaches we should experiment with
- IPFS and Bittorrent
- But both kinda of suck
- Want to experiment with Felix’s ENR work
- Tiny extension to discovery protocol, really elegant
- Nodes can advertise certain capabilities
- Most infra. in place already
- Want to play around with it where nodes advertise that they’re a light or full node with certain datasets available
- Trying to figure out whether we could do the whole historical state retrieval in protocol without needing to hack another decentralized protocol for it
- Not sure if it’s possible, but this is a third direction that may be valuable, need it anyway for light servers
- Q: is this discovery V5?
- A: depends how you define it, it’s different, but Felix wants this ENR work to become new discovery v5
- Problem is that it was hacked into protocol but not scalable
- ENR approach is clear, there are a few EIPs open, quite a lot of published material
- Danny: We are discussing using mature version of Dv5 for advertising which shards you’re participating in
- Peter: Considering putting historical state on top of that mechanism
- Can’t yet see single proposal that will work
Constantinople hard fork
- Afri block time proposals
- Last time we decided to talk about a block number at this meeting
- I just put down some numbers
- Did calculations to see how much it could vary if we target a Weds - would most likely fall on a weekday
- Afri proposed block 7080042
- Lane: have we always used even numbers ending in -0000 in the past?
- Alexey: We don’t think someone will start super mining to speed up the fork do we?
- Danny: Would be very expensive on mainnet
- Afri: very unlikely because of high difficulty of mainnet. Worst case scenario, if price of Eth continues dropping fast and a lot of miners stop mining then it could slow us down
- Block times have been stable recently so this is not a concern
- Martin: We’ll put block number into next release with possible commandline option to delay it
- Only reason to change block number again would be if we find another consensus bug
- Hudson: All previous ones have been -000
- Hudson: Do people prefer the palindrome or sticking with -000?
- Peter: palindromes could be hard in the future
- Easy if block numbers still in the millions’
- Greg: four zeroes, keep it easy
- Palindromes for testnet, four zeroes for mainnet - let’s do this
- So let’s use 7080000
- Stireby update
- [missed a bit]
- Alexey
- Failure was that there was a GPU miner, we fell prey to this because spec was public
- Plan was to have a bit more mining, didn’t happen
- Not going to push to have another one
- I think this was worth doing
- Martin: Suggestion for new PoW testnet called Gangnam with more genesis alloc, can deprecate Ropsten
- For dapp testing
- Peter: Good time to bring up a few discussions
- We agreed we don’t want to continue Ropsten
- But do we want to relaunch a PoW or instead do a PoA network?
- Afri & co. have been working on Goerli testnet
- Would mean we don’t have a public PoW testnet but we could always do public PoW testnet forks just to verify PoW and difficulties when we need to do a hard fork
- Not too much point in running a PoW replacement for Ropsten
- Martin: We’d drop all Trinity, aleth, and those that haven’t yet implemented clique yet
- Afri: This can be done
- With all devs out there we can do this
- Agree with Peter that going fwd with PoW not feasible, this Stirby experience showed how unfeasible PoW testnets are
- Esp. having one for apps not for consensus needs
- Mikhail: Fresh test networks not good for testing difficulty bomb delays - block numbers too low, not yet activated
- Zak: join this testnet call since we can create testnets with relative ease that are provisioned and controlled
- Peter: We can create purpose-built testnets to test difficulty adjustment - for everything else PoA is a saner approach from dapp developer perspective
- Zak: Will probably want different consensus algos in different testnets for different reasons - purpose built
- Provisioning, ensuring adequate activity is the hard part
- Peter: if we want a purpose-built testnet to test difficult adjustment then we don’t need any activity
- For other stuff PoA works fine
- Mikhail: Could add configuration for difficult bomb and delays
- Peter: Don’t need to make a decision about this here
- If we want to replace Ropsten it might be a good idea to do a PoA testnet
- Everyone please think about it, if someone has a good reason why it’s a horrible idea we’ll consider that
- Zak: Activity, background traffic will affect consensus and various metrics within the network
- Peter: This is the point of having a public testnet, have traffic etc.
- Alexey: Let’s explore option of purpose-built PoW network with Zac and then come back and discuss more on next call, whether this is feasible, PoW for specific things and PoA for everything else
- Zak: These testnets can be ephemeral, not entirely public, will be permissioned, activity and behavior will be automated with our tools
- Hudson: This is a good topic for the next meeting
- Will Goerli be compatible with ETC? Is that a problem?
- Afri: No, it will not have ETC compatibility
- But we work on something similar for ETC
Openness for meetings
- Hudson: Enough people who said “over my dead body” re: any type of closed meetings
- So we will have open meetings for now
- Less inclination to do this for these calls
- Not sure about in person yet - technical feasibility is different
- Won’t have any more “Eth1x” / mainnet improvement calls as WG all created now
- So we’ll just keep doing what we’ve been doing
- If anyone has a differing opinion speak now or speak to me privately later
- Martin: There are WG, we will talk within them
- Is that frowned upon?
- Hudson: That can be private or public, it’s not a core dev meeting
- There was some confusion around that
- Greg: There was useful discussion on where to make the call about what you open up and how much
- It’s a matter of judgment
- From chat: While we worry about not feeling free to speak our minds in public in the face of a clueless press, I used to help organize anti-war actions with FBI infiltrators at our open meetings, and a press that would accuse us of violence when the cops had finished beating us.
- When I did war demonstrations, we had FBI and CIA infiltrators and a press who would accuse us of violence after the police beat us up
- So I shrug and say, this is important but we don’t really have that much on the line, it just isn’t that hard
ProgPoW update
- Pawel: Last two weeks, there’s one reported issue left, reported at beginning of this week, need to resolve
- Solution is decided but requires some small changes to implementation to mitigate small flaw in algorithm
- Martin: Had implementations in C++, ethhash, verifiers for geth and parity, then spec changed, geth and cpp updated, parity hasn’t yet implemented latest changes
- Some more discussion about potentially changing spec
- Two clients in sync
- If parity implements most recent changes then we could launch a testnet
- Martin: We had discussion with various people in crypto communities
- One is David Warwick (?), creator of Zia
- Got his approval to quote him
- ProgPoW, if we were to replace hashimoto S/ethhash with this, it would probably keep ASICs away for 1-1.5 years
- If we increase the resistance towards ASICs then that could itself increase centralization because only a very advanced ASIC manufacturer could afford to R&D such an ASIC that would get efficiency gains
- So he believes it would buy us 1-2 yrs, but another approach is to use an algo which is extremely ASIC-friendly
- We could adopt ProgPoW now and if a year from now we still don’t have PoS, we might consider switching to something very ASIC friendly which even a small manufacturer can produce
- Regarding prevalence of ASICs on network today, it’s very hard to find numbers
- Have been reported from various sources but none can be taken as fact
- No records on units manufactured or sold so it’s hard to take that into account
- Some people think the fact that we’re implementing ProgPoW is already decided, this is not the case, we are open to other options
- Alexey: I started to think about ProgPoW more because of eth1x proposals, elephant in room is what are other things slated for eth1.0
- ProgPoW which will take some amount of work
- Second, Justin’s interview for Epicenter, he said the reason they want to develop a VDF ASIC so that no one else can get a major advantage by creating something much much better
- I’m glad Martin mentioned as well that there is a potential downside (re: centralization)
- I want people to explore both
- Hudson: Reminder that these eth1x roadmap working groups are still not finalized
- Lane: Would it make sense for the new app testnet to be a ProgPoW POW testnet?
- A: Not really
- Martin: Just envelope of blocks, doesn’t matter whether they are full or empty
- Pawel: We want a testnet with a switch from one algo to the other, so not pure ProgPoW, or switching back and forth, let’s keep them separate
- Martin: Try different configs with a series of testnets
Quick announcement re: coordinator roles
- Hudson, Danny, Lane, Afri, Piper, Casey, Jamie Pitts have been getting people together to discuss coordinator role
- Person/people in thids role could help out with hard forks with Hudson and Afri
- Also EIPs and core devs calls
- Will be announcing them over the next few months, starting in January
- You’ll start seeing more of that
- Main goal is to lighten load for Hudson, Lane, Afri and others who have been handling coordination for things like core devs calls
- Also to train people as future PMs for teams
- We’ll work hard to get them more integrated into these calls, do it in a way that supports the core devs
- Idea to make it as easy and stress-free as possible for core devs
Date for next meeting
- Anyone opposed to skipping next call for holidays?
- No opposition
- Next call will be Jan. 4
Attendees
- Hudson
- Lane
- Alexey
- Pawel
- Peter
- Fred
- Mikhail
- Zak Cole
- Fred Harryson
- Dimitry Khoklov
- Brian
- Jacek
- Danny
- Danno
- Andrew Gross
- Daniel Ellison
- Afri
- Martin Holst Swende
- Greg
- Adam Schmideg
- Shahan Khatchadourian
