patiently waiting for this to solve rollup’s memory usage during sourcemap generation with vite
from my understanding we could save ~1.5gb of ram for our build
i should probably try building our project with rolldown-based vite
being able to `import { telemetry } from 'https://foo.bar/telemetry@1.2.3'` is pretty cool and surprisingly powerful
we'll probably use import maps in the future to make it even more seamless but its good enough for now
if we really want to share something common we externalize the package during build and replace with a native esm import from cdn. that way widgets can reuse stuff but still upgrade independently
the entire thing is hand-rolled and probably makes sense only inside the company
each widgets ships a small loader package. it accepts some reusable deps, configs and provides mount/unmount functions
it can also declare npm deps. if the consumer uses same version great, if not we get a duplicate and move on with our life
we wanted to prioritize independence over bundle size. each "widget" is used in multiple apps and some apps have a lot of widgets. some apps are 10+ year b2b slop using react 16, some are 19 already, some use angular.
basically having a single dependency version for us is close to impossible
we isolate so each fronted has its own copy of react/vue/angular. we deduplicate loading it from cdn. middle ground between excessive duplication and ease of upgrading
you don’t if you don’t want to suffer upgrading
renders should only happen in components that use the data
as long as the react team keeps ignoring state management story in react and keeps adding more memoisation performance will stay subpar
not using automatic formatting tools
they would still have to process a bunch of unnecessary stuff
no. it will not be okay because bundles will still have to process everything, especially in dev mode
if you put a barrel file with hundreds of imports in your library im uninstalling it
couldn’t migrate so now i have biome.jsonc
i have like 3 alternative providers and a dozen of protocols to click every time something breaks
well makes sense since core bluesky experience is mostly read-only and not interaction heavy
update performance is still important though, b2b style apps suffer heavily from react state management and one of the best things we do is migrating all the state outside of react
this survey being 50% ai and zero percent firefox tells me everything about current state of mozilla
nah, you can have fun with sveltekit though
