dan

восемь суток на тракторе по снежной степи

🌉 bridged from 🦋 danabra.mov, follow @bsky.brid.gy to interact

Public Key

npub1l26jhuzntdzct6p9vprrs33j70ds5d9utk05n28sx78790sg2s5sm07c36

NIP-05 Address

danabra.mov@bsky-brid-gy.mostr.pub

Profile Code

nprofile1qqs04dft7pf4k3v9aqjkq33cgce08kc2xj79m86f4rcr0rlzhcy9g2gpz4mhxue69uhhyetvv9ujuerfw36x7tnsw43qayalvq

Publishing to

relay.ditto.pub

Author Public Key

npub1l26jhuzntdzct6p9vprrs33j70ds5d9utk05n28sx78790sg2s5sm07c36

Show more details

Last Notes

2026-03-31 13:11:03 UTC

i still think posting this was a huge fucking L. i struggle to explain why exactly but the current backlash feels earned because of this posturing
https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:fpruhuo22xkm5o7ttr2ktxdo&cid=bafkreifnw666e2aubxvickrlh2dgujmkefessu2sdyouyqrxm2dq54zxge

2025-10-28 14:09:42 UTC

could just be a forked codepath for reduced motion or something

2025-10-28 12:38:46 UTC

got @slices.network OAuth session working with pg! for some reason didn't work initially but this fixed the crash for me tangled.org/@slices.netw...
Fix crashes in Postgres OAuth ...

2025-10-28 12:06:31 UTC

figured it out. the fix:
this ... was not obvious. would be nice if there a way to "rebuild without cache" or some env variable to turn it off
https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:fpruhuo22xkm5o7ttr2ktxdo&cid=bafkreibxj5tlrahadq2mk6xzit3hfvkmrmd7zlelsqh3iilmyygboufgfe

2025-10-28 07:35:07 UTC

if a feed operator served you a list of items, and then you're requesting the next 20 items, i think it's safe to assume you've somehow scrolled past them. so i don't know what kind of private information is expected there. whether you scrolled slightly slower or slightly faster?

2025-10-28 06:36:20 UTC

is patch-package supposed to work well with railway? getting bizarre issues where it seems to be running unpatched code despite patch log showing up before next bundling. tried turning off build caching but not sure it worked

2025-10-28 03:20:23 UTC

i could probably work around it by creating the database before running Next build but i’m not sure what i’ll do when i want >1 instance doing writes

2025-10-28 02:38:59 UTC

i wouldn’t *normally* have more than one process (yet) so maybe i could fix the special case… but what if i do want multiple processes or instances using same db? i’m not sure what i’m supposed to do to “properly” fix other than avoiding sqlite

2025-10-28 02:31:03 UTC

i have a Node.js app deployed to Railway. it uses a sqlite database. initially the database is empty
i'm getting "database is locked" on deploy because my deploy script ends up trying to create the database tables concurrently from two processes, and sqlite doesn't like that
what's the proper fix?

2025-10-27 18:33:45 UTC

you *do* have to ask for it to root cause and deflake though, otherwise it'll keep inserting random hacks or "document" existing issues

2025-10-27 18:32:20 UTC

pretty cool that you can get claude to deflake tests if you specifically ask for that. it can run small scripts to do stuff like stress-test 100 times in parallel, then work to narrow down the cause of flakiness

2025-10-27 05:03:14 UTC

i’m not working on AT myself so i hope fixing issues isn’t up to me personally. for feedback on oauth, @matthieu.bsky.team is probably the right person.

2025-10-26 08:57:17 UTC

some things i want to work on
- more atproto writing
- a few atproto apps
- fixing react docs to have examples for newer features
- writing new reference for lean for normal people
- fixing papercuts in lean ide tooling
- prototyping a ui framework in lean
- RSC course
- beyond vibecoding course

2025-10-26 08:51:27 UTC

i feel like if i directed my effort towards the exact projects that need it most in the capacity and amount that would be most productive, it would be impossible to get paid for it. that’s not exactly news but it’s still strange

2025-10-26 08:47:47 UTC

right now i’m still on sabbatical and i really like that i could spend a few months just doing math, and now a few weeks just doing atproto / app stuff, with periods of active writing in between. these complete switches feel very natural but i can’t do the same with “having a job” and it sucks

2025-10-26 08:46:17 UTC

i guess that’s some forms of consulting but consulting is usually for saving from a mess or for non-critical stuff. always feels second-class. whereas i mean more like doing some good work here, doing some good work there. ehh idk what i mean

2025-10-26 08:43:59 UTC

there should be a way to work for 20 companies at once and rotate them like once a month so you don’t get bored

2025-10-26 05:54:41 UTC

also nitpicking but localStorage is just a web browser API, its shape isn’t javascript’s fault as some in the thread imply. the language’s design has no relation to it

2025-10-26 05:25:58 UTC

is there a tldr somewhere on how to make mutations reflect latest state in Next.js
should i
- revalidatePage/revalidateTag in every server action
- or should i router.refresh next to their calls
- both?
i haven't tried cacheComponents yet so curious how to work with the existing model

2025-10-25 16:10:25 UTC

the advantage of having it be local is iteration speed. can write tests against them, can quickly change schema without invalidating old records, can have claude generate sample records, don’t need to actually have many handles to work on features with aggregation, trivial to reset state to mocks

2025-10-25 15:11:20 UTC

@tangled.org is sending PRs broken? this is a very confusing and unhelpful error. "failed" why?
https://morel.us-east.host.bsky.network/xrpc/com.atproto.sync.getBlob?did=did:plc:fpruhuo22xkm5o7ttr2ktxdo&cid=bafkreidp7eu7lwsnzhs5krmbtjbp4qoyg2sgrkaq2vtcspgqyw7jdoc76a

2025-10-25 14:35:20 UTC

is 3m3zm7eurxk26 a valid tid? i got it from TID.nextStr(); but @slices.network validation says it "does not match format 'Tid'"

2025-10-25 13:43:11 UTC

i'm on 16 but with cacheComponents off so gonna bump into some walls first and then see if cacheComponents is more intuitive

2025-10-25 05:59:16 UTC

there isn’t much in the atproto department yet because i only made a mock backend so far. but i have the lexicons drafted up and in theory hooking up to @slices.network should work. i’ll try to actually get it running in the next week or two

2025-10-24 04:48:04 UTC

claude ios has an annoying bug where you have to send every message twice because it fails the first time. anyone else seeing this? been broken for weeks

2025-10-24 00:53:50 UTC

you can then stack multiple of those and have a many-to-many

2025-10-23 16:07:25 UTC

i recall "instant transitions" as an idea from very early conversations (predating even first RSC demo, not to speak of App Router) so the desire definitely isn't new but this specific set of constraints seems like it may square prefetching with worries about CPU usage. prefetching cached is freeish

2025-10-23 15:54:49 UTC

very good talk by @samselikoff.com about App Router instant navigations and "use cache", clarified my mental model a lot
Next.js Conf 2025

2025-10-23 01:52:54 UTC

people saying ai is a fad don't know what they're talking about. i'm definitely not giving up my claude now; the ability to do structural refactors "change this thing to be layered like this thing in this other project" is fantastic. i *do* want the model to be local and hope we'll get there in time

2025-10-23 00:39:30 UTC

sigh i guess we wait another year
#nevent1q…lj3e

2025-10-22 19:03:42 UTC

fun idea: a bluesky custom feed with posts from contributors of the @tangled.org repos you’ve starred

2025-10-22 18:55:46 UTC

web (1 + i)

2025-10-22 16:11:30 UTC

there’s a bunch of good stuff in there. “hosted web, posted web and signed web” is another banger.
#nevent1q…h7a2

2025-10-22 16:06:38 UTC

i mean… jay literally tried to redefine it to mean something broader. still one of my favorite articles even though it didn’t catch on
Web3 is Self-Certifying

2025-10-22 13:11:39 UTC

i know RSC-in-browser doesn’t have great perf compared to real RSC but i think it’s compelling to be able to write a pure SPA using the same model
#nevent1q…amwv

2025-10-22 13:09:32 UTC

mostly just code organization! i vastly prefer the RSC model to “loaders” etc. and i do want the option of later moving all my “loader” execution to the server. so there’s a natural upgrade path — start with RSC SPA that enforces server/client split, later move the server part to the actual server

2025-10-21 01:44:59 UTC

funnily enough avoiding scroll restoration on some changes was my first contribution to react-router, back in uhhh 2014! github.com/remix-run/re...
[changed] Remove preserveScrol...

2025-10-21 01:07:13 UTC

it’s also not exactly how this works, your metaphor is a little flawed. there’s some kind of emergent computional gates embedded within it but they’re weird and unreliable. sure in a sense it’s “remixing” other theories but theories themselves are just remixes of axioms. good enough remixing can do

2025-10-21 00:35:04 UTC

i wrote about how to fix any bug
How to Fix Any Bug — overreact...

2025-10-20 21:15:55 UTC

yeah github.com/remix-run/re...
[Bug]: `revalidate` causes pag...

2025-10-20 21:15:00 UTC

ah shit it was <ScrollRestoration> from react-router-dom being overzealous about restoration. i have no idea why it's trying to restore scroll when i'm not doing any navigations!

2025-10-20 20:53:59 UTC

trying to teach claude to root cause bugs by removing things

2025-10-20 16:31:34 UTC

super excited for changes to prefetching, caching, and ofc "use cache" stuff. hoping this can be the "snow leopard" to the iOS 7-ness of App Router (pardon mixed metaphors).
fixing mistakes and gaps in the model with vast codebases to verify those changes on

2025-10-20 16:04:11 UTC

if i do port to Next i would need to choose hosting. i think either Cloudflare or Railway probably makes sense. one argument in favor of Railway is i can deploy just a normal Docker container and not worry about the quirks of serverless adapters

2025-10-20 15:36:31 UTC

i know she signs "safe up here" but my headcanon is "same again"

2025-10-20 15:12:50 UTC

my “For You” feed has gotten really good

2025-10-20 00:26:37 UTC

alas no the bug is still there. i also have no idea where it is coming from. really annoying
#nevent1q…h444

2025-10-19 13:05:45 UTC

the mind state after this work is something else — a bit of a headache, really. i haven't felt that way from coding

2025-10-19 13:04:50 UTC

in the end what helped the most was
- asking to *remove* all parts i obviously saw were buggy
- giving it very specific small refactoring tasks that were obvious to me ("this thing shouldn't be here", "this looks dodgy")
then after i made it untangle the knots, it could re-add those features

2025-10-19 13:02:16 UTC

man idk vibecoding is such a rollercoaster. yesterday:
- it recreated most of my project from scratch with a proper architecture, in a way that allowed proper testing and pretty much fixed all business logic bugs (yay)
- but then when i asked it to reskin it to old ui, it introduced so many bugs

2025-10-19 12:15:43 UTC

spent an hour hunting a bizarre css flicker bug, solved by restarting my computer. what

2025-10-18 20:02:58 UTC

i think specifically it’s a superpower to iterate on very simple ui mocks/wireframes/skeletons with a very lasagna style service layering.
you can always dress it up in state later but it can iterate so much better when there’s less client-side stuff

2025-10-18 19:25:53 UTC

when claude starts posting ✅ with exclamation marks you know it's time to kill the agent. confident and useless

2025-10-18 18:25:20 UTC

mostly i'm just glad i know a much more efficient way to iterate now

2025-10-18 17:31:16 UTC

realized i made a mistake with claude iterating on a realistic app ui. it is MUCH more effective to:
- start with data model
- generate mocks for it
- generate in-memory PDS mock
- generate routes and static wireframe uis
- generate loaders and mutations that ui needs
then iterate on everything

2025-10-18 16:26:16 UTC

really digging the "local PDS mock" approach @spacecowboy17.bsky.social used for prototyping. claude is pretty good at generating those, and it's way easier to iterate on the UI and the model without worrying about actual network

2025-10-18 15:29:36 UTC

conclusion
#nevent1q…xhrt

2025-10-18 14:07:33 UTC

in my case, the secondary record changes over time and is mutable so it's a bit odd that there can be multiple of those but only one of them is the designated source of truth.

2025-10-07 20:16:02 UTC

source tangled.org/@danabra.mov...
@danabra.mov/resolve-lexicon

2025-10-05 14:55:01 UTC

@tangled.org needs og:image yesterday. can yall stop whatever you doing, ship this three hour change, and resume as usual? thank u

2025-10-02 20:21:27 UTC

i wrote down the process of resolving an at:// URI step by step. turns out, it's a great way to learn how the AT protocol works!
Where It's at:// — overreacted