Very interesting.
I may be wrong, and haven’t gone over the RFC, but practically speaking I think the value prop UUIDv7 is uniqueness with “best effort” monotony guarantee.
Relying on time sync and avoiding drift is probably not advisable and unreliable, otherwise it would be impractical in terms of perf to generate lots of these in a hot loop etc.
Bit of a mess of an attempted rationale here, hope it makes sense
@curtissummers@ohmypy Because not all systems are distributed and the monotony guarantee is crucial if you’re using UUIDv7. But agree with you generally, IMO it should be on the DB level anyway
@realTomBenton@Microsoft@surface Not sure about the buy part, but corp IT would absolutely love for people to reduce attack surface and DLP risk, not to mention running internal compliant AI infra, if the maintenance overhead is reasonable.
Yep totally makes sense and sounds good friend, feel free to do that. I got inspired to do Protobuf after seeing @jonhoo’s stream where he generated Avro IDL, so had similar goals.
One of the problems I tried to solve other than the “one source of truth” thing, is the verbosity and no sane defaults for many of those IDLs, so tried to make it simple by default and extensible for power users.
People say with AI it doesn’t matter anymore, but obviously disagree with that premise 😅
Wrote about it a bit here if you have patience for my rambles:
https://t.co/dCodJ0GKL9
@frogtoss And to clarify, I have no interest in competing and Glue is totally open source. Just think it has value and didn’t invest any time into marketing it, so happy for new adopters. It’s not aimed at game engines at all, so the problem space may be different
@frogtoss Hi! This is a really cool project, and if you’re interested I developed this a few months ago:
https://t.co/PXpViqms4x
Would be happy to support new targets for Glue if you’re up to it
@rfleury 100% agree, DRY can be extremely overdone. LLMs love doing this especially…
There is also a huge price for cognitive complexity, for anyone, human or otherwise, that works on a codebase.
Something that is really fun sometimes as a training exercise is to take a piece of complex tech and break it down by building it from scratch.
Had some time off last week, so I wrote the first post in a series I will call "from scratch", where each time we will focus on a particular tool/tech and build it from the ground up (within a reasonable layer of abstraction).
Starting with - containers! Let's build "Docker" from scratch.
Next up - AI agents/harnesses (a session I presented internally at Palo a few months ago) and then probably TLS.
https://t.co/539PUG1m7C
Agree with @igalklebanov and not from a place of AI hate, most maintainers don’t want to be hit by 100% AI generated PRs with no human behind them that understands the project well enough.
Some notable projects have banned people that did this or adjusted their contribution policies.
The goal is noble, but I would suggest to use LLMs to accelerate yourself or research if you wish, but this just seems wrong to me personally.
If you want to contribute tokens for the greater good, then maybe ask in some dev forum if any maintainers have issues they want AI to deal with.