Skip to content

21.0

Compare
Choose a tag to compare
@bbakerman bbakerman released this 11 Jul 00:54
· 697 commits to master since this release
c1395c8

We are pleased to announce the release of graphql-java v21.0. This is a breaking change release.

Thanks to everyone in the community who contributed to the release, whether that was code, helping to report issues, or participating in discussions.

And Happy 8th Birthday to graphql-java, who celebrated their birthday last week!

Breaking Changes

Upgraded to Java 11

graphql-java now requires Java 11 as a minimum version. See the blog announcing the change.

For those who need time to upgrade to Java 11, keep in mind we will support graphql-java 20.x (with Java 8) for a short period as per our release policy.

If you are wondering why we are not on a later version, graphql-java has always been conservative on its base JVM version to allow the widest possible set of consumers.

Reverted stricter scalar parseValue coercion, added monitoring and interceptor callback

v20.0 introduced a stricter set of scalar parseValue coercions - for example previously an Integer would accept a string if it parsed into a number but that was removed and a more strict system was put in place.

While technically more correct, and consistent with the graphql-js reference implementation, in practice this proved problematic for some consumers. So this more stricter parseValue coercion was reverted in v20.3.

We would like to re-introduce this more strict scalar parseValue conversion in the future and to that end we have introduced a graphql.execution.values.InputInterceptor callback that allows you to observe what values you are receiving and potentially do special tweaking of those values.

A graphql.execution.values.legacycoercing.LegacyCoercingInputInterceptor implementation will convert old less strict values into then more strict values for example.

If you had problems with scalar values we urge you to use the new InputInterceptor to learn what less strict values are coming into your systems and fix them up. That way, when a future version re-introduces the more strict (and more correct) coercion then you will be prepared.

Static recordLike() methods no longer supported

In v20, the PropertyDataFetcher would read property values from recordLike() methods on objects even if they were static methods. This caused problems for some users and after considering how to fix it and talking to some our major consumers like the Spring team, we decided to remove this behavior. On balance we think this will lead to a better outcome over the long term.

This is a breaking change for those who might have relied on a static recordLike() method being called for a property.

Removal of old deprecated methods and classes

The following PRs removed old deprecated methods and class. The changes are breaking ones but these have been deprecated for a long time.

Other small breaking changes

A very minor breaking change is that graphql.execution.ExecutionStrategy had a protected method protected Iterable<Object> toIterable(Object result) which really is a utility method and not designed for overriding. graphql.util.FpKit#toIterable is the preferred replacement.

What's new in v21

ExecutableNormalisedXXX is now public API

The graphql.normalized.ExecutableNormalizedOperation and graphql.normalized.ExecutableNormalizedField code is now public API.

This API allows you to represent what MAY be executed given a schema and a valid GraphQL query.

This code is not intended for general consumption but perhaps you are writing a framework based on graphql-java and need to have a powerful representation of what would be executed, then these classes are for you.

This allows you to write specialized code (such as a new execution engine or perhaps a federated GraphQL engine like say Nadel) based on these tree like representations of a normalized and executable query.

Building extensions in data fetchers

There is a new graphql.extensions.ExtensionsBuilder that allows data fetcher callbacks to add extension values into the final result. Since extensions are a map and there could be merge conflicts on values, a graphql.extensions.ExtensionsMerger interface is provided to handle these conflicts and a default graphql.extensions.DefaultExtensionsMerger is provided.

This is available via the graphql.GraphQLContext and is put in there by default so data fetchers can rely on it being present. At the end of the request the ExtensionsBuilder is called to build out a final map of extensions which is placed in the graphql.ExecutionResult.

A smarter schema visitor API

A new graphql.schema.visitor.GraphQLSchemaVisitor has been created that is more domain specific around visiting GraphQL schemas. The old graphql.schema.GraphQLTypeVisitor worked however it is very generic in nature and is not domain specific to schemas.

The new API improves how you can visit schemas and the callbacks have better schema domain information provided on them. Also the graphql.schema.visitor.GraphQLSchemaVisitorEnvironment is better than older alternative with clearer return methods like changeNode() or deleteNode() and so on for controlling how the visitor works.

This is an adaptor to GraphQLTypeVisitor and hence can be used by the existing graphql.schema.SchemaTraverser and graphql.schema.SchemaTransformer classes (which expect a GraphQLTypeVisitor) via a small call to graphql.schema.visitor.GraphQLSchemaVisitor#toTypeVisitor.

Performance improvements

As always, we have tried to include some performance improvements in the release. One area of note is avoiding unnecessary CompletableFuture allocations when they are not needed.

Other things

The QueryComplexity calculator has been broken out into its own class and can be used outside the original graphql.analysis.MaxQueryComplexityInstrumentation context.

The graphql.execution.DataFetcherResult#map method was added to allow better functional mapping of results.

All Changes

New Contributors

Full Changelog: v20.2...v21.0