There were many missing releases and bullet points as well as inaccurate
ones. For example, it was impossible to figure out that users of Hapi 16 who
want to use Apollo Cache Control needed to use precisely version 1.2.0 of
apollo-server-hapi (which wasn't even mentioned).
Link to CHANGELOG from all READMEs. Add READMEs for the graphql-server-*
packages. Add an explanation of our versioning system to the top of
CHANGELOG.md.
For reasons unbeknownst to me, the changes below were only being complained
about in the CI environment but _not_ when I ran `npm run lint` locally.
It seems partially related to the OS its ran on, so I imagine there might be
some other sub-dependency at play here. In an effort to fix this, I just
spawned a Ubuntu docker image, checked out this repository and ran the same
`npm run lint`. This produced identical results to Travis, so I ran
`npm run lint-fix`, then saved the `git diff` results and have applied this
locally.
This should allow us to re-enable `prettier` in CI, though I have plans to
separate that from the actual `npm test` runs. This should result in a better
workflow for managing PRs.
As reported in #313 and #384, an error thrown that isn't a
`HttpQueryError` will result in an "unhandled promise rejection",
leading to infinitely hanging requests.
While exploring ways to reproduce #313, I thought that validation errors
coming from GraphQL's internal type checking would reproduce the issue.
It didn't. But instead of throwing out the test I wrote, I thought it
might be good to keep it around to prevent future possible regressions.
I'll keep exploring ways to reproduce the issue I'm experiencing in a
subsequent commit(s).
