While Next.js can handle static sites, we've had a few issues with it, notably a performance hit at runtime and some bugginess around routing and SSR. Gatsby being fully intended for high-performance static sites makes it a great technical fit for the Shields frontend. The `createPages()` API should be a really nice way to add a page for each service family, for example.
This migrates the frontend from Next.js to Gatsby. Gatsby is a powerful tool, which has a bit of downside as there's a lot to dig through. Overall I found configuration easier than Next.js. There are a lot of plugins and for the most part they worked out of the box. The documentation is good.
Links are cleaner now: there is no #. This will break old links though perhaps we could add some redirection to help with that. The only one I’m really concerned about `/#/endpoint`. I’m not sure if folks are deep-linking to the category pages.
There are a lot of enhancements we could add, in order to speed up the site even more. In particular we could think about inlining the SVGs rather than making separate requests for each one.
While Gatsby recommends GraphQL, it's not required. To keep things simple and reduce the learning curve, I did not use it here.
Close#1943Fix#2837Fix#2616
The suggest code was an exception to our usual organization pattern. There was a service test, but it's not a service. The code would sometimes regress because it wasn't being tested all the time.
This makes them no longer run as service tests, which is good because they run as part of every build. Some of them are smaller-bracket tests which is good too, because it will make them easier to test, especially as this code grows.
I'd have liked to keep using frisby for the ones that make requests to the server, though I ran into some issues with sequencing of setup that I think will require upstream changes.
In #2698 we decided to put legacy helper functions in `core/legacy`. I think that’s a fine idea, though if we’re going to have a bunch of badge helper functions in there, it seems like it is probably better to keep these two important but esoteric helper functions with the core code to which they are most coupled. So I added `legacy-` to the name, and put them in `core/base-service`.
This test is being weirdly flaky in #2809. The problem seems to be in the test helper code, so I rewrote this using Joi.
I imagine the change has to do with a change to the test ordering. It's a bit puzzling.
However, the new test seems fine (and the endpoint is rarely used; not critical to begin with).
