TWIL May 1st 2026

• You cannot convert a Webpack project to Vite in ~2 hours by making a plan with Copilot and letting it rip
  • Tomorrow we'll see if it made any meaningful progress and I just need to clean up, or it was a complete waste of time
  • Next day managed to clean up some errors in an hour, now stuck on not copying the dict files for kuromoji
• Don't always take requirements from a single engineer on a team at face value
  • Or you'll end up with two major refactors of what should have been a simple PR, the final one of which doesn't even need your brilliant idea
• Revisited the cache clearing issue from 10s to 100ms today
  • Turns out clearing the cache key for every client is still a little slow (~2s at worst)
  • So I narrowed it to only clients with whitelist rules, which doesn't even need a join
  • Since we only need the client ID, I can just pluck distinct client IDs from the whitelist table and clear the cache for each of those
• It's best practice to dynamically fetch SAML metadata from the IdP's /metadata endpoint, not persist it in your DB/code
  • Makes me wonder why we don't do that; it seems easier
  • Though I guess to avoid making a bunch of network requests you wanna cache it anyway?
  • Maybe also for the mappings
• Don't just assume something being used like a foreign key guarantees it isn't null

Links

• The End of Responsive Images - Mat Marquis

I admit to feeling frustrated I don't have more control over what srcset and sizes actually result in on an img element from time to time; this is a very good case for why I shouldn't from the person who got that W3C standard ratified. Also contains a helpful little feature I hadn't hear about; sizes: auto, which lets the browser decide which source to load from srcset for you (as long as the image is lazy loaded).