pull down to refresh

Stopped working on code review early today and decided to find a solution to githubless git, as Goog setting an example with completely destroying AOSP (#1217284) makes me really worried.
Now testing git-bug to see if I can make it so that a .patch can get proposed as an issue and then have it picked up with some CI. The interface for server layer is... graphql and now I really feel for all poor souls that use that, while realizing that well designed REST APIs are the real luxury life, lol.
157 sats \ 3 replies \ @k00b 11 Sep
There's a steep benefit curve to graphql. It's awesome when your API is evolving rapidly or you are caching on the client a lot. For something like git I'm not sure you need either of those much.
I looked into git-bug when I was hacking on https://gitern.com. I probably would've forked or rewritten it from scratch but that's how I imagined implementing github-like issues while maintaining portability.
reply
I agree that it could be nice for rapid development in an ecosystem around a central service - iirc that's what FB developed it for - but it's indeed overkill for something simple that doesn't change much (and if it does, you need versioning) and you just want to install and run; especially if it's distributed by nature.
It's also funny that the caching is killing my experience with this particular web-ui because I need to literally hit refresh when navigating (using in-app navigation) to see new state on an issue - so in this instance it's also kind of awfully implemented.
I was thinking that git-bug could be nice for git-collab-over-decentralized-protocol (like the nostr thing) because it can ultimately reduce the code needed, and technically I shouldn't need to run it as a server other than for some management GUI. But to support github-like code collaboration, review and communication, issues need to evolve to pull requests, so I am looking to hack that in.
Also, it may be nice to have everything in a single repo when running transient containers for LLM coding workloads. I figured a while back that if we're going to let an LLM code like a pro, we should treat it like a pro: not some prompt and a log and that's it, but really integrate a fully tooled agile workflow into the process. No need a gui with an annoying chatbox then, and doing it via text files doesn't scale with all the bloat LLMs generate - something structured would be nice. Anyway I was thinking that if this works well, I could in theory automate a whole dev team fr.
reply
136 sats \ 1 reply \ @ek 12 Sep
I agree that it could be nice for rapid development in an ecosystem around a central service - iirc that's what FB developed it for
Yes, and I think they also developed it for mobile because GraphQL also improves bandwidth since you can select exactly what you need, not more, not less. I think I heard that in here:
It's also funny that the caching is killing my experience with this particular web-ui because I need to literally hit refresh when navigating (using in-app navigation) to see new state on an issue - so in this instance it's also kind of awfully implemented.
yeah caching is hard for real haha, we also had many issues with our caching, but maybe it's just a "skill issue"
reply
since you can select exactly what you need, not more, not less.
That's a benefit; there is something similar in bitcoind for getblockstat.
maybe it's just a "skill issue"
I don't think so. Cache expiry / cached state transition is hard and really easy to get wrong, I can't remember a single instance where it was easy to make it smooth - and then I'm talking about back-end caching, which is easier than on the front-end.
I think that caching is hardest to consistently deliver when you are still rapidly adding or refactoring features/API, but you could design to overcome that by abstracting the mechanism.
reply