Alessandro Decina

[Alessandro Decina @alessandrod]

✨ Announcing Aya, a new eBPF library for Rust ✨

https://t.co/NL4ZW3G8aJ

#rustlang #ebpf

[Alessandro Decina @alessandrod]

this would fix the EU

[Aberaeron 🇪🇺🇺🇦🇨🇦🇩🇰🇫🇷 @ItsFullOfFellas]

Winner!

[Alessandro Decina @alessandrod]

@Dola_porr it's not really available, even if you reconstruct the DAG of the block you don't know when a transaction was ingested, so you don't know if it waited or if it arrived after its predecessor

[Alessandro Decina @alessandrod]

what 20ms saved on full blocks looks like

[Alessandro Decina @alessandrod]

@Dola_porr example

[Alessandro Decina @alessandrod]

@Dola_porr I need more context I'm not sure what you're trying to do

In general (logical) account locks are the biggest source of landing latency (you have to wait your turn)

[Alessandro Decina @alessandrod]

@Dola_porr

[Alessandro Decina @alessandrod]

@cavemanloverboy many such cases

[Alessandro Decina @alessandrod]

@_jhunsaker @monad someone's been busy

[Alessandro Decina @alessandrod]

@diman_io Yeah that’s totally our fault and we need to do a lot better

[Alessandro Decina @alessandrod]

This can be avoided by communicating better on both sides, but there’s one important thing that many people often don’t consider: what you said about our impl being more canonical.

Most code that floats around and isn’t merged is like 80% complete. But we can’t merge 80% complete. And often the original author isn’t interested in finishing the final 20% because it doesn’t apply to their use case or whatever, and we aren’t incentivized to take their code and do the final 20% because picking up someone else’s incomplete work is a chore and often harder than doing the whole work to begin with

[Alessandro Decina @alessandrod]

@5prairie I believe scientists call this the having that dawg in you syndrome

[Alessandro Decina @alessandrod]

@cavemanloverboy @trentdotsol (I actually agree on the things that do make sense, and don’t understand why they’re not prioritized)

[Alessandro Decina @alessandrod]

@cavemanloverboy @trentdotsol Yes we should just have coded rev strat by default the cluster knows best

[Alessandro Decina @alessandrod]

@qkniep > hack the computer

< I hacked the computer

> oh my god 😲

desperate to get fable-d, sad to see. Luckily commie AI is gonna save us

[Alessandro Decina @alessandrod]

> anza has had plenty of time and opportunity to upstream many things that operators use (gre+multicast,

brother, 1 week before 4.0 becomes mainnet candidate helius is doing the final XDP tests, they rebase jito, pickup a stray DZ patch and suddenly broadcast regresses double digit milliseconds - that was multicast. Rush to fix.

In january I spent 1 month reviewing the initial implementation of GRE, it finally lands, a week later they change how GRE is configured and I have to rage rewrite it all in a week or we miss the 4.0 cut.

We play the "anza mean muh" part but the truth is that a lot of the code is half assed and is a huge waste of our time.

[Alessandro Decina @alessandrod]

@cavemanloverboy @trentdotsol wild thing to say considering that the whole reason bindings exists is to mitigate mod risks

> nobody runs your client as is brother

is this a good thing or a bad thing? should we try to have everyone use it? or you think we don't know how?

[Alessandro Decina @alessandrod]

@toly @lvboudre The GPL is infringed all the time by commercial entities and nobody gets sued. Still infringed, still ethically wrong. Exactly as wrong as kimi distilling.

Say no to US propaganda

[Alessandro Decina @alessandrod]

@toly @lvboudre it does? if a license is infringed in the forest when nobody is reading the code, has it still been infringed?

[Alessandro Decina @alessandrod]

@toly @lvboudre > so it’s on them to check if it’s infringing

lmao how much of percolator have you actually read?

[Alessandro Decina @alessandrod]

lol no way that ingesting whole GPL codebases to train a commercial product is fair use, and you could easily make the argument that if an LLM spits code that is similar to the training data then it's derivative work

LLMs are great but we don't have to pretend that scraping the whole internet is not problematic

[Alessandro Decina @alessandrod]

@toly @lvboudre people who published their code under the GPL would like to have a word