Elliott Johnson
ell.iott.dev
Elliott Johnson
@ell.iott.dev
Svelte Core Team @ Vercel
Yeah, for this use case it doesn’t really matter
November 18, 2025 at 1:16 AM
Wow, this predates me -- I started web development (and my Svelte involvement) juuuust at the start of very early SvelteKit -- late 2021/early 2022.
November 17, 2025 at 11:00 PM
LOL this is amazing
November 17, 2025 at 10:41 PM
Ahh, rough. I’ve always stuck a spending limit of like $10 just to catch runaway silliness. Haven’t needed it yet but someday maybe I’ll be popular enough 😂
November 17, 2025 at 9:39 PM
That's a lot of function duration, especially considering fluid compute -- what do you have going on that's generating that much?
November 17, 2025 at 8:18 PM
ok fine I guess that means it's more like 3 or 4 years to go 😂
November 17, 2025 at 6:15 PM
Only *checks watch* 8 more years until I can say the same 😂
November 16, 2025 at 2:57 PM
I finally feel like I’ve made it in life
November 15, 2025 at 2:22 AM
...takes a single-digit number of milliseconds. Last time I checked it was around 2.
November 14, 2025 at 9:44 PM
But anyway, you can test this pretty easily. If you enable OpenTelemetry on SvelteKit (which itself adds some overhead, making things slower) and then go look at the spans, the entire server request (including all of the stuff SvelteKit does on top of just the server render)...
November 14, 2025 at 9:44 PM
you may be confusing caching with cold starts, which:
- at any appreciable scale are a very small percentage of requests
- do not affect some platforms at all, just depends on your architecture
November 14, 2025 at 9:42 PM
it is not cached
November 14, 2025 at 9:40 PM
I would call into question your timings here. On Vercel, a nontrivial SSRed SvelteKit app can serve a response in under 200ms, easily. This includes network traversal, proxy on both sides, and actually rendering the content on the server. Unless your SPA can serve content in -300-0ms...
November 14, 2025 at 5:07 PM
It's likely something to do with the experimental nature -- if you can narrow it down and file a bug (to kit or svelte, whichever it ends up being) we'd love it!
November 13, 2025 at 9:37 PM
Ouch, that's painful. I had some heartburn upgrading an app from Svelte 4 to Svelte 5 as well. Thankfully it was just like... half a day of painful boring work
November 13, 2025 at 9:36 PM
Yeah, it’s treeshaking. They should be able to remove this code, but they don’t.
November 12, 2025 at 5:07 PM
Yeah rollup has the best treeshaking we’re aware of and rolldown isn’t far behind.
November 12, 2025 at 2:01 PM
Don’t forget Apple Music / App Store!
November 8, 2025 at 10:02 PM
Just Say No to platforms with backslash path delimiters
November 5, 2025 at 5:57 PM
Even cooler, if you don't use the clsx syntax, you don't end up with clsx as a dependency. 100% value if you want it, 0% cost if you dont
November 3, 2025 at 7:27 PM
JSDoc has significant benefits (no build step being the primary one), but its unreadability and verbosity are strong marks against it. In any context where I already have to have a build step (eg working on webapp that needs to run through something like Vite/rollup), I'm always picking TypeScript.
October 27, 2025 at 3:32 PM
...to the process of software development. TypeScript's benefit has way, way outweighed its negatives.

As someone who spends all day, every day working on a JSDoc codebase, getting to write regular TS syntax is a breath of fresh air. The JSDoc version is just so, so ugly and annoying to write.
October 27, 2025 at 3:31 PM