Jaz

i think they need to rewrite rust-analyzer in golang so it is actually fast

claude can one shot atproto oauth when it uses your own atproto oauth library that you manually wrote

they must've watched the climate town video...

TikTok alternative Skylight soars to 380K+ users after TikTok U.S. deal finalized | TechCrunch Skylight, a TikTok alternative built on open protocols, has had a busy weekend.

I know social graphs almost always pull people back to the largest mass, but if you make shortform content, please consider trying Skylight as a TikTok alternative right now techcrunch.com/2026/01/26/t...

there were so many blog posts

iiuc lots of new uploads are stuck in review forever right now and also the storm is causing some kinda incident. combination of fucked ops handover and weather is my guess.

true, i don't understand why japanese napkins are so awful. why would you make oleophobic and hydrophobic napkins???

i try to

some user named patriotmocha in my replies: Please let me know if there is a Signal group I can join, since the others were deleted. Thank you so much.

some user named patriotmocha in someone else’s replies: What is your best recommendation to find people who's in charged?

How do you do, fellow neighbors?

let's gooooooooooooooooo

let’s fucking goooooooooo

i ended up enjoying the series as a whole, brando sando does a good job closing it out but the middle 4-5 books drag on a lot ngl

rip

no

its the video metadata service, not the transcode workers. just stores and serves video and transcode job metadata to the workers and CDN edges

same

will put out more details/timeline soon, but we are finally preparing to push through the operational changes to switch over bsky.network to the "new" relay.

have already started switching over some of our internal systems

wow incredible

forking entire applications with copy on write point in time snapshots of the block device, and even copy on write snapshots of RAM like Codesandbox to allow for <1 second forking

additionally, imagine containers built on EroFS that share a common base layer and that common base layer can only be stored once and then gets forked every time that you deploy something on top of it, and then forking process takes less than a second because all you're doing is adding metadata

and the forking of the device happens instantly and you're able to spin up an entire copy of your database in seconds, no matter how large it is

imagine you have a production database and you want to test a data migration. so you just fork the entire fucking block device and run your migration in a development environment with your real production data. you can run migrations and deploy new versions of your software with 100% confidence

I really just want to build EBS but without nitro cards and without specialized hardware, and I really think it is capable to do with first class performance through a lot of modern innovations we've seen in the past 5 years. anything is possible

there's also the added bonus that this looks a lot closer to things like Ceph RADOS from an operational standpoint and allows you to reuse a lot of that institutional knowledge

though I do admit that I've been back and forth on whether we should be doing this at the fs level or at the block device level, I think at the end of the day if you want to support more complex virtualization runtimes like microvms, and support encryption you probably want to virtualize the device

something like what I've been designing supports trivial point in time snapshotting and data forking of entire block devices, which I think is an incredibly powerful primitive

I do think trying to do it at the file system level has some advantages but it also definitely locks down the possible use cases. especially considering situations where we want to support people operating databases, I have no idea what file system is going to be the right fit for the database

I'm trying to build the future on the hardware of today (not yesterday) which is dramatically more achievable now with access to machines that can do a lot of the implementation heavy lifting with the right guidance and oversight.

again this is running on hardware we own and operate on OSs we install, we control the whole stack here

I'm not trying to build "cloud native" anything on someone else's cloud. I'm architecting for my own cloud running on my own metal where i own every piece of the stack (except the hypervisor for now).