Alex RussellCome for the frank discussion of the limiting factors to the web's longevity, stay for the incontrovertible evidence that Next.js suuuuuuuuuuuuuuuucks:
https://infrequently.org/2024/10/platforms-are-competitions/
Jon@slightlyoff its weird seeing washingtonpost.com as all green in your footnotes. When I manually lighthouse them, their "field" crux is good, but the actual report is pretty bad.
@jon_nunan Interesting!
@jon_nunan Looking at this a little more closely, it's at least plausible that this effect is explained by audience selection. CrUX reports that half the site's traffic is from mobile[1] (likely larger, because iOS blindness in that data), but the sorts of folks who visit WaPo are likely wealthier, with faster phones and on faster networks. At the extreme end, a wealth bias effect would explain a divergence even this big.
@slightlyoff yeah, survivor bias if you will; but am not convinced its just that. And to be clear I wasn't accusing them of a VW dieselgate type situation either. Just that I've seen some performance tools get messed up by really bad sites and report incorrectly.
@slightlyoff like I have to look after a Next.js site, and we were using catchpoint to at least firefight performance regressions. We did a lot of work to get a fix out, just to find our LCP score jumped in the wrong direction. Turns out the site was so bad, it hadn't sent a network request in 2 seconds as it ploughed through the mass of JS we were sending, and the tool was taking that as the site had finished loading, so took the LCP from before that finished.
@slightlyoff once we got that work to less than 2s, the tool was able to start correctly identifying the LCP. So we were getting a more correct measurement for LCP, but for people only looking at dashboards it made our work look pointless.
@jon_nunan I recently saw a similar effect in California's (React-based) public benefits SPA; took a while to diagnose that the parade of loading screens was confusing the LCP algorithm:
https://infrequently.org/2024/08/object-lesson/#:~:text=On%20low%2Dend,the%20complete%20page.
Reckoning: Part 2 — Object Lesson
SNAP benefits sites for more than 20% of Americans are unusably slow. All of them would be significantly faster if states abandoned client-side-rendering, and along with it, the legacy JavaScript frameworks (React, Angular, etc.) built to enable the SPA model.