Robert Hoffmann

@clairevo specs, docs, tests, and now accessibility (think playwright agents) are all becoming increasingly important !! x.com/i/status/20311…

@svpino @zquestz That attribution stuff is months old as ppl responsible for triaging pull requests have been banning AI committed stuff Also, everything should be a choice And yeah, even codex has it : commit_attribution developers.openai.com/codex/config-r…

@svpino So you don't believe an AI should be credited for their work? What about in an age when AGI exists?

It's been about a year, so I think it'd be fun to look back and see which of these complaints has been solved. 1. IDL infrastructure is even more of a mess now. There's now two sources of truth, the original PDA and now a metadata program. Unfortunately, the metadata program was built without first class support for IDL versioning. So, next time IDLs change major versions, expect more pain. 2. Parsing historical programs given the lack of IDL versioning/history is still a problem now, and with the added metadata program it's now slightly worse. 3. The read layer has improved substantially. Most rpc providers heavily index gPA, so it's not as bad (still bad). Indexers have gotten unbelievably reliable now, such that I trust running an entire UI off of indexed data. It's heavier, you have devops costs, but at least it works and you can customize to your heart's content. 4. Running an API powered by indexers is still somewhat pricey 5. Transaction sending and landing is more or less solved. I haven't had to think about this in months. Smooth as butter. 6. Anchor has gotten faster and they've fixed a lot of the issues that blow out memory. There's still a lot of improvements to be made here, and new frameworks/anchor v2 are coming. 7. CPI depth limit is set to be increased. So this will not be an issue for much longer. 8. TX size limits are set to be increased. So this will not be an issue for much longer. 9. Rent costs have gone down because SOL price shit the bed. Yay? They are also going to actually cut rent costs soon™ 10. All of the cNFT issues described still exist. These have mostly been solved by just not using cNFTs. Use core or token metadata nfts instead. 11. Errors have, for the most part, gotten better. There are still quite a few useless errors, though. 12. All of the top wallets are still closed source. No progress made there, and seems like no one cares. 13. Rust anchor client still sucks. 14. Explorers still have issues, but thanks to claude I built my own that solves most of the issues (explorer.chewing.glass) 15. The base RPC spec still sucks, but if you're okay with vendor lock-in, Helius has a ton of useful APIs. Transaction history. Ability to only fetch accounts that have changed since a given slot. Etc. 16. I actually haven't had a dependency hell issue in a few months, I also haven't upgraded anything. Maybe it's getting better? 17. I haven't had issues with preflight disabled transactions in a while. Not sure if they fixed these issues, or I just started simulating more often. 18. Running programs on a cron is largely fixed. I wrote tuktuk for this, and it's been running reliably for almost a year. Largely things seem to be getting better. I've noticed that a lot of the pain goes away if you shove all of your solana-related stuff into an API, and use that API from your various clients web2 style. You can leverage indexers for complex queries. You can change your program and IDLs without breaking all clients. You can record your own transaction history. Etc. I always had a dream that blockchain dev was a way to avoid devops costs and burden by running everything on decentralized infrastructure. We seem to have moved far from this ideal, web3 is just web2 with another kind of database. Doesn't look great for protocols outliving their parent companies, but as we've seen that almost never happens anyway.