Nuxt is pretty nice to use in most ways, and the migration was largely
straightforward. Its convention-heavy approach is nice and fairly
intuitive, and it makes data-fetching a breeze with useFetch. It
does have some drawbacks, as well, namely:
The unit tests for forms aren't as comprehensive anymore, because setErrors
doesn't work properly in the Nuxt tests like it did with vanilla
Vue. This isn't horrible, because I do have e2e coverage over these
features, but having faster, cheaper unit tests would be nice when
I'm working on the components, because I would get feedback in a
fraction of the time.
The e2e tests now need a delay. This breaks my heart to have to do, because it means we are often not running the tests as quickly as we can (the slowdown is currently only a few seconds off from where it was before, but will increase as we have more tests), but it's necessary for now because on page load Nuxt swaps out the server-rendered DOM nodes with new ones on the client. This breaks Playwright's auto-waiting for inputs, because what tends to happen is:
There may be some way to detect once this swap has happened, and it happens too fast to bother a real, human user, but so far I've decided to just settle with the easy way out.
I no longer have OpenTelemtry traces working in the frontend. The old instrumentation code doesn't work anymore, and I couldn't figure out how to get it working after an hour or two of trying. I found an open PR about adding support, but it seems like a neglected issue. It looks like others may have gotten it working, so there is some hope, but I'll return to it later.
I wish Nuxt weren't so all-or-nothing--it's a heavy framework and can be kind of a black box at times--but overall I think it's useful. Until Vue's from-scratch SSR support gets smoother (which I don't believe is a priority for them, because things like Nuxt exist), I think I will probably start new Vue projects using Nuxt.