This library provides facilities to derive JSON codec instances for Circe using Magnolia macros.
⚠️ Early development status warning
Although this project is extensively tested and seems to work fine, it's still at early development stages. It's not advised to use this in production without proper test coverage of related code.
There are still some things, that are different from circe-generic, including a critical issue with auto derivation.
To play around with circe-magnolia, add it to your build:
libraryDependencies += "io.circe" %%% "circe-magnolia-derivation" % "0.3.0"
After that, as in
circe-generic, you can use one of two derivation modes.
Note, that at the moment for both auto and semiauto modes you have to import encoder and decoder machinery separately (see examples below).
Works in the same way as
import io.circe.magnolia.derivation.decoder.auto._ import io.circe.magnolia.derivation.encoder.auto._ case class Foo(i: Int, s: String) implicitly[Encoder[Foo]] implicitly[Decoder[Foo]]
Works in the same way as
circe-generic, but the method names differ so that you can theoretically use both in the same scope:
import io.circe.magnolia.derivation.decoder.semiauto._ import io.circe.magnolia.derivation.encoder.semiauto._ case class Foo(i: Int, s: String) val encoder = deriveMagnoliaEncoder[Foo] val decoder = deriveMagnoliaDecoder[Foo]
circe-magnolia derivation and codecs work in the same way as in
circe-generic, several test suites from original circe repository were adapted and added to this project. These tests validate the derivation semantics and also the lawfulness of derived codecs (example).
There's another set of tests, that validate the equivalence of JSON and decoding logic, produced by
Test suite is currently green, but couple of cases are worked around or ignored, and waiting to be fixed. See the issue tracker for outstanding issues.
Overall, semiauto derivation works pretty well. All laws are satisfied and compatibility tests are passing.
There is a subtle difference from circe-generic semiauto in what can and what can not be derived. Circe-magnolia deriver has more relaxed requirements on what has to be in scope, so you might not even notice this. Basically, circe-magnolia doesn't require any pre-defined codecs for intermediate types.
In essense: current version of
circe-magnolia semiauto is as powerful as
circe-generic's auto under the hood. It just doesn't provide any top-level implicits out of the box - you have to call
deriveMagnolia[Encoder|Decoder] to start derivation process. See more here
Auto derivation also works, but there's a twist: at the moment, for default codecs (for example,
Encoder[List]) to be picked up, they have to be imported explicitly.
Otherwise, deriver uses Magnolia to derive codecs for any types which are either case classes or sealed traits. These derived codecs then override the default ones.
This is definitely going to be fixed in future, but if you want to switch from circe-generic's auto to circe-magnolia's auto today, you would have to add additional imports to every place where derivation takes place:
// import all default codecs, defined in circe import io.circe.Encoder._ import io.circe.Decoder._ // also import all codecs, defined in the companion objects of your own data definitions import Foo._, Bar._
Another outstanding issue with auto is that it doesn't pick up default instances for tagged types.
- Facilitate magnolia development to make auto derivation work the same way as in
- Add derivation of partial/patch codecs.
- Configurable derivation should be very simple to do with Magnolia. Potentially this can provide huge flexibility to the
Circe-magnolia is currently developed and maintained by Vladimir Pavkin.
I really welcome any kind of contributions, including test/bug reports and benchmarks.
I really appreciate all the people who contributed to the project:
I also want to say "Thank you!" to