* refactor: switch `json-stable-stringify` to `fast-json-stable-stringify`
* chore: drop `@types/json-stable-stringify`
* Update CHANGELOG.md for #2065.
This also re-enables the Expiration tests for this library, though I think I
will continue to leave the testsuite decomposed into more granular methods
for readability and future growth.
Note that this doesn't implement the same default `300` second TTL as the
other Memcache and Redis implementations currently do (a very debatable
default we should re-consider in the future).
While it is certainly our current implementation, I'm not sure that the true
spirit of an in-memory key-value store is to only map strings to other strings.
While enforcing `<string, string>` might be necessary for some distributed
cache stores, it seems we shouldn't enforce it for `InMemoryLRUCache`.
By changing the `KeyValueCache` to default to `string` but allow other
options, we can allow the user to decide exactly how the RHS of this store
should be typed.
This does require a bit more flexible implementation of the `length`
calculator which is used for cache ejection, but that implementation will
become immediately useful when we start storing parsed ASTs in this KV store.
Rather than crutching on the `return`, this logic reads more concisely with
a full `if` / `else` pattern.
I say this because two people looked at it and weren't immediately aware,
which really shouldn't be the case from such a brief piece of code.
This test harness is intended to be relatively temporary, but it seems
fruitful to start picking from where these types are already defined rather
than re-implementing them in multiple places.
This change came up in a larger feature implementation and should otherwise be
a no-op but was worth surfacing as its own commit in case that feature never
lands.
This fixes a TypeScript `TypeError` which was encountered after the
`@types/ws` typings were updated[[1]] to more correctly identify valid URLs.
The error exhibited on CI[[2] was:
TypeError [ERR_INVALID_URL]: Invalid URL: ws://localhost:6666undefined
This had been previously reverted[[3]] but was auto-merged again for a
reason I don't quite have time to investigate right now, though I suspect
it's something I'll be able to prevent in the future with a bit more care.
[1]: https://github.com/apollographql/apollo-server/pull/1992
[2]: https://circleci.com/gh/apollographql/apollo-server/22416
[3]: https://github.com/apollographql/apollo-server/pull/2006
cc @martijnwalraven
* AER: Remove encodedTraces to prevent duplicates
When there are multiple instances of apollo-engine-reporting, the
`Trace.encode` method gets wrapped each time to add the
`encodedTraces`. In order to prevent them from being added to the
protobuf multiple times, we remove the encodedTraces after adding them
once
* Add changelog
* Move incremental Trace encoding to a-e-r-protobuf
To enable incrmental compilation of traces, we add a patch to the
Trace.encode method generated by protobujs to accept and store encoded
traces. Occassionally with multiple instances of apollo-engine-reporting
that share the same version of the protobuf, the wrapper method gets
applied more than once. In order to ensure that the wrapper only gets
applied once, we patch the Trace.encode method inside of
apollo-engine-protobuf.
tsc hangs on the processing the generated protobuf.js files, so the
tsconfig.json ignores the generated protobuf file. In order for the
typescript index.ts file to compile the generated protobuf.js file is
output to the src folder. To ensure the protobuf files are available to
the production build, `npm run compile` copies the protobuf file
manually from src to dist.
* Reexport protobuf import after modification
`export * from './protobuf'` exports the unmodified reference
* Update comment on Trace.encode to point at a-e-r-p
The override now occurs inside of apollo-engine-reporting-protobuf due
to the case of having multiple reporting challenges, so we need to
update the comments to help indicate that
* Remove typescript build step
In order to simplify the generation of this library, we move the change
the index.ts file into index.js and remove the typescript build step.
Since the type safety is created by the protobufjs generation, this
seems acceptable. In general this portion of the code should remain
relatively stable, so generating and copying the code with `prepare`
remains reasonable.
This commit again looks quite complicated, but's merely an over-complication
inflicted by Prettification. Disabling whitespace differences when viewing
this commit, the functional change here is that we no longer skip many file
upload tests when the (semver) major segment of `process.version` is `10`.
I couldn't be happier to get rid of this exception for file upload tests on
Node.js 10, which was an unfortunate reality of the non-updated
`graphql-upload` module world we previously lived in.
Thanks, @jaydenseric, for the newfound (to us!) Node.js upload support!
This commit looks way more complicated than it really is thanks to Prettier
trying to be helpful.
All I've done is add `NODE_MAJOR_VERSION === 6` as a version NOT to test
uploads for, in an effort to fix the failing tests (failing appropriately
since Node.js 6 with file uploads throws an error right now and cannot run
uploads anymore.).
I hope to actually remove this limitation, but I'm going to iterate on it
first. This also just switches to a `NODE_MAJOR_VERSION` constant from the
`apollo-server-integration-testsuite` rather than using the GTE function on
the same input since...math.
BUGFIX: the playground page is currently being served without a content type
<!--
Thanks for filing a pull request on GraphQL Server!
Please look at the following checklist to ensure that your PR
can be accepted quickly:
-->
TODO:
* [x] Update CHANGELOG.md with your change (include reference to issue & this PR)
* [ ] Make sure all of the significant new logic is covered by tests
* [x] Rebase your changes on master so that they can be merged easily
* [x] Make sure all tests and linter rules pass
Now that there are specific versions of Node.js which don't support file
uploads (namely, <= 8.5.0) we need to explicitly disable uploads on those
versions, similar to how those users must opt-in to that behavior by setting
`uploads: false` in their Apollo Server constructor options.
This effectively accomplishes exactly that, but only when necessary.
Due to changes in the third-party `graphql-upload` package which Apollo
Server utilizes to implement out-of-the-box file upload functionality, we
must drop support for file uploads in versions of the Node.js engine prior
to v8.5.0. Since file uploads are supported by default in Apollo Server 2.x,
and there is an explicit dependency on `graphql-upload`, we must
prevent users who are affected by this mid-major-release deprecation by
being surprised by the sudden lack of upload support.
By `throw`-ing an error at server startup for affected users, we certainly
are breaking a semantic versioning agreement for these users, however with a
relatively simple ergonomic (setting `uploads: false`) we allow those users
who are NOT utilizing file uploads (as we believe is the case with a
majority) to continue using their version of Node.js until it reaches the
end of its supported lifetime (as dictated by its Long Term Support
agreement with the Node.js Foundation). If we did not `throw` the error at
server start-up, those affected may not notice since they may update and start
their updated server without noticing the impending chance of failure when
someone tries updating!
Apollo Server 2.x has attempted to maintain full compatibility with versions
of Node.js which are still under Long Term Support agreements with the
Node.js Foundation. While this continues to mostly be true, file uploads
are an exception which we've now had to make.
Third-party open-source projects must absolutely do what's best for their
project. From an architecture standpoint, I suspect that we (the designers
behind Apollo Server) are mostly to blame for this. Namely, it's unfortunate
that we had made such an incredibly coupled integration with a third-party
package that we restricted our users from incrementally adopting the
changes (and new/improved functionality) of, in this particular case,
the `graphql-upload` package. I hope we can take better care with decisions
like this in the future!
Lastly, this commit also adds documentation to help those affected.
Follow-up on the update to `graphql-playground-html` in previous release by
also bumping the minor version of the `graphql-playground-react` dependency
to `1.7.10` — which is the version requested from the from the CDN bundle by
`graphql-playground-html`.
Ref: https://github.com/apollographql/apollo-server/pull/2037
With any luck, we will no longer necessitate our fork which removed the
`graphql-config` dependency thanks to the work done in:
https://github.com/prisma/graphql-playground/pull/874🎉
Most notably though, this fixes a documentation scrolling problem with
Safari.
This was something I just happened to spot, though further investigation is
likely necessary to determine if other similar cases can be discovered.
This fixes this test to `expect` the number of additional `expectations we
have expressed for it: 2.
Testing errors in `catch`es of `Promise` chains necessitates the use of
`expect.assertions(n)` where `n` is the number of assertions declared in
that particular test.
Without this `expect`ation clearly defined, a resolved `Promise` will not
result in the assertion being tested. For example, if the `query` passed to
this test was a query against a type in the schema (i.e. `query: 'foo'` was
replaced with `query: { query: '{ testString }' }`) the desired error would
not be produced, but the test would still pass. I'm not saying that's
likely, but there's certainly the possibility that (given fault code)
something else in the plumbing could cause that Promise to resolve.
Ref: https://jestjs.io/docs/en/tutorial-async#error-handling
This is mandated by `lru-cache` v5 and surfaced the CircleCI failures on #2004:
https://circleci.com/gh/apollographql/apollo-server/22537
Luckily, this is a private implementation detail of Apollo Server's
`InMemoryLRUCache` so no additional changes should be necessary and we should
be able to update to `lru-cache` 5.0.0 in a semver minor respectful way.
In an effort to maintain consistency and semantically correct meaning,
this changes (only in documentation and internal Apollo Server tests) the
name of the first argument in resolver signatures to use `parent` rather
than `root`.
While `root` certainly makes sense when the resolver is belonging to the
root `Query` or `Mutation` type, once nested field resolvers begin getting
called, the more semantically correct term would seem to be `parent`.
Since `parent` still makes sense at the root level, and since resolvers
frequently get copied and pasted into more deeply-nested positions, putting
this pattern in place for apps which are just beginning might just help
someone more clearly understand the relationship in the future — without
incorrectly thinking that a nested resolver is accessing the root of the
graph, rather than the parent.
Due to a number of other assignments within `createPlaygroundOptions`, the default is still maintained and the `playground` configuration parameter should be entirely optional.
