Releases: Kotlin/kotlinx.serialization
1.4.0
1.4.0-RC
This is a candidate for the next big release with many new exciting features to try.
It uses Kotlin 1.7.10 by default.
Integration with Okio's BufferedSource and BufferedSink
Okio library by Square is a popular solution for fast and efficient IO operations on JVM, K/N and K/JS.
In this version, we have added functions that parse/write JSON directly to Okio's input/output classes, saving you the overhead of copying data to String
beforehand.
These functions are called Json.decodeFromBufferedSource
and Json.encodeToBufferedSink
, respectively.
There's also decodeBufferedSourceToSequence
that behaves similarly to decodeToSequence
from Java streams integration, so you can lazily decode multiple objects the same way as before.
Note that these functions are located in a separate new artifact, so users who don't need them wouldn't find themselves dependent on Okio.
To include this artifact in your project, use the same group id org.jetbrains.kotlinx
and artifact id kotlinx-serialization-json-okio
.
To find out more about this integration, check new functions' documentation and corresponding pull requests:
#1901 and #1982.
Inline classes and unsigned numbers do not require experimental annotations anymore
Inline classes and unsigned number types have been promoted to a Stable feature in Kotlin 1.5,
and now we are promoting support for them in kotlinx.serialization to Stable status, too.
To be precise, we've removed all @ExperimentalSerializationApi
annotations from functions related to inline classes encoding and decoding,
namely SerialDescriptor.isInline
, Encoder.encodeInline
, and some others. We've also updated related documentation article.
Additionally, all @ExperimentalUnsignedTypes
annotations were removed completely,
so you can freely use types such as UInt
and their respective serializers as a stable feature
without opt-in requirement.
Part of SerializationException's hierarchy is public now
When kotlinx.serialization 1.0 was released, all subclasses of SerializationException
were made internal,
since they didn't provide helpful information besides the standard message.
Since then, we've received a lot of feature requests with compelling use-cases for exposing some of these internal types to the public.
In this release, we are starting to fulfilling these requests by making MissingFieldException
public.
One can use it in the catch
clause to better understand the reasons of failure — for example, to return 400 instead of 500 from an HTTP server — and then use its fields
property to communicate the message better.
See the details in the corresponding PR.
In future releases, we'll continue work in this direction, and we aim to provide more useful public exception types & properties.
In the meantime, we've revamped KDoc for some methods regarding the exceptions — all of them now properly declare which exception types are allowed to be thrown.
For example, KSerializer.deserialize
is documented to throw IllegalStateException
to indicate problems unrelated to serialization, such as data validation in classes' constructors.
@MetaSerializable annotation
This release introduces a new @MetaSerializable
annotation that adds @Serializable
behavior to user-defined annotations — i.e., those annotations would also instruct the compiler plugin to generate a serializer for class. In addition, all annotations marked with @MetaSerializable
are saved in the generated @SerialDescriptor
as if they are annotated with @SerialInfo
.
This annotation will be particularly useful for various format authors who require adding some metadata to the serializable class — this can now be done using a single annotation instead of two, and without the risk of forgetting @Serializable
. Check out details & examples in the KDoc and corresponding PR.
Note: Kotlin 1.7.0 or higher is required for this feature to work.
Moving documentation from GitHub pages to kotlinlang.org
As a part of a coordinated effort to unify kotlinx libraries users' experience, Dokka-generated documentation pages (KDoc) were moved from https://kotlin.github.io/kotlinx.serialization/ to https://kotlinlang.org/api/kotlinx.serialization/. No action from you is required — there are proper redirects at the former address, so there is no need to worry about links in your blogpost getting obsolete or broken.
Note that this move does not affect guides written in Markdown in the docs
folder. We aim to move them later, enriching text with runnable examples as in the Kotlin language guides.
Other improvements
- Allow Kotlin's null literal in JSON DSL (#1907) (thanks to Lukellmann)
- Stabilize EmptySerializersModule (#1921)
- Boost performance of polymorphic deserialization in optimistic scenario (#1919)
- Added serializer for the
kotlin.time.Duration
class (plugin support comes in Kotlin 1.7.20) (#1960) - Support tagged not null marks in TaggedEncoder/Decoder (#1954) (thanks to EdwarDDay)
Bugfixes
- Support quoting unsigned integers when used as map keys (#1969)
- Fix protocol buffer enum schema generation (#1967) (thanks to mogud)
- Support diamond inheritance of sealed interfaces in SealedClassSerializer (#1958)
- Support retrieving serializer for sealed interface (#1968)
- Fix misleading token description in JSON errors (#1941) (thanks to TheMrMilchmann)
1.3.3
This release contains support for Protocol Buffers packed fields, as well as several bugfixes.
It uses Kotlin 1.6.21 by default.
Protobuf packed fields
It is now possible to encode and decode Kotlin classes to/from Protobuf messages with packed repeated fields.
To mark the field as packed, use @ProtoPacked
annotation on it.
Note it affects only List
and primitive collection such as IntArray
types.
With this feature, it is now possible to decode Proto3 messages, where all repeated fields are packed by default.
Protobuf schema generator also supports new @ProtoPacked
annotation.
Many thanks to Paul de Vrieze for his valuable contribution!
Other improvements & small features
- Incorporate JsonPath into exception messages (#1841)
- Mark block in corresponding encodeStructure/decodeStructure extensions as crossinline to reduce amount of bytecode (#1917)
- Support serialization of compile-time
Collection<E>
properties that are not lists at the runtime (#1821) - Best-effort kotlin reflect avoidance in serializer(Type) (#1819)
Bugfixes
1.3.2
This release contains several features and bugfixes for core API as well as for HOCON format.
It uses Kotlin 1.6.10 by default.
Serializing objects to HOCON
It's now possible to encode Kotlin objects to Config
values with new Hocon.encodeToConfig
function.
This feature may help edit existing configs inside Kotlin program or generate new ones.
Big thanks to Osip Fatkullin for implementing this.
Polymorphic default serializers
As of now, polymorphicDefault
clause inside SerializersModule { }
builder specifies a
fallback serializer to be used only during deserialization process. A new function has been introduced to allow setting
fallback serializer for serialization: polymorphicDefaultSerializer
.
This function should ease serializing vast hierarchies of third-party or Java classes.
Note that there are two new experimental functions, polymorphicDefaultSerializer
and polymorphicDefaultDeserializer
.
To avoid naming confusion, we are going to deprecate polymorphicDefault
in favor of polymorphicDefaultDeserializer
in the next minor release (1.4.0).
Credit for the PR goes to our contributor Joseph Burton.
Other improvements
- HOCON: parse strings into integers and booleans if possible (#1795) (thanks to tobiaslieber)
- Add an encodeCollection extensions (#1749) (thanks to Nicklas Ansman Giertz)
Bugfixes
1.3.1
This release mainly contains bugfixes for 1.3.0 and provides new experimental Json.decodeToSequence
function.
Improvements
- Provide decodeToSequence to read multiple objects from stream lazily (#1691)
Bugfixes
- Correctly handle buffer boundaries while decoding escape sequences from json stream (#1706)
- Properly skip unknown keys for objects and structures with zero properties (#1720)
- Fix merging for maplikeSerializer when the map is not empty (by using the actual size * 2). (#1712) (thanks to pdvrieze)
- Fix lookup of primitive array serializers by Java type token (#1708)
1.3.0
This release contains all of the cool new features from 1.3.0-RC as well as minor improvements.
It uses Kotlin 1.5.31 by default.
Bugfixes and improvements
- Promote JsonConfiguration and its usages to stable (#1690)
- Remove opt-in annotations from SerialFormat, StringFormat, BinaryFormat (#1688)
- Correctly throw SerializationException instead of IOOBE for some cases with EOF in streams (#1677)
- CBOR: ignore tags when reading (#1614) (thanks to David Robertson)
1.3.0-RC
This is a release candidate for the next version. It contains a lot of interesting features and improvements,
so we ask you to evaluate it and share your feedback.
Kotlin 1.5.30 is used by default.
Java IO stream-based JSON serialization
Finally, in kotlinx.serialization
1.3.0 we’re presenting the first experimental version of the serialization API for IO streams:
Json.encodeToStream
and Json.decodeFromStream
extension functions.
With this API, you can decode objects directly from files, network connections, and other data sources without reading the data to strings beforehand.
The opposite operation is also available: you can send encoded objects directly to files and other streams in a single API call.
IO stream serialization is available only on the JVM platform and for the JSON format for now.
Check out more in the PR.
Property-level control over defaults values encoding
Previous versions of the library allowed to specify whether to encode or drop default properties values with
format configuration flags such as Json { encodeDefaults = false }
.
In 1.3.0 we’re extending this feature by adding a new way to fine-tune the serialization of default values:
you can now control it on the property level using the new @EncodeDefault
annotation.
@EncodeDefault
annotation has a higher priority over the encodeDefaults
property and takes one of two possible values:
ALWAYS
(default value) encodes a property value even if it equals to default.NEVER
doesn’t encode the default value regardless of the format configuration.
Encoding of the annotated properties is not affected by encodeDefaults
format flag
and works as described for all serialization formats, not only JSON.
To learn more, check corresponding PR.
Excluding null values from JSON serialization
In 1.3.0, we’re introducing one more way to reduce the size of the generated JSON strings: omitting null values.
A new JSON configuration property explicitNulls
defines whether null
property values should be included in the serialized JSON string.
The difference from encodeDefaults
is that explicitNulls = false
flag drops null values even if the property does not have a default value.
Upon deserializing such a missing property, a null
or default value (if it exists) will be used.
To maintain backwards compatibility, this flag is set to true
by default.
You can learn more in the documentation or the PR.
Per-hierarchy polymorphic class discriminators
In previous versions, you could change the discriminator name using the
classDiscriminator property of the Json
instance.
In 1.3.0, we’re adding a way to set a custom discriminator name for each class hierarchy to enable more flexible serialization.
You can do it by annotating a class with @JsonClassDiscriminator
with the discriminator name as its argument.
A custom discriminator is applied to the annotated class and its subclasses.
Only one custom discriminator can be used in each class hierarchy, thanks to the new @InheritableSerialInfo
annotation.
Check out corresponding PR for details.
Support for Java module system
Now all kotlinx.serialization runtime libraries are shipped as a multi-release JAR with module-info.class
file for Java versions 9 and higher.
This enables possibilities to use kotlinx.serialization with modern tools such as jlink
and various technologies such as TorandoFX
.
Many thanks to our contributor Gerard de Leeuw and his PR for making this possible.
Native targets for Apple Silicon
This release includes klibs for new targets, introduced in Kotlin/Native 1.5.30 —
macosArm64
, iosSimulatorArm64
, watchosSimulatorArm64
, and tvosSimulatorArm64
.
Bugfixes and improvements
- Properly handle quoted 'null' literals in lenient mode (#1637)
- Switch on deep recursive function when nested level of JSON is too deep (#1596)
- Support for local serializable classes in IR compiler
- Support default values for
@SerialInfo
annotations in IR compiler - Improve error message for JsonTreeReader (#1597)
- Add guide for delegating serializers and wrapping serial descriptor (#1591)
- Set target JVM version to 8 for Hocon module in Gradle metadata (#1661)
1.2.2
This release contains various bugfixes, some useful features and important performance improvements.
It also uses Kotlin 1.5.20 as default.
Features
- Support for
@JsonNames
andcoerceInputValues
inJson.decodeFromDynamic
(#1479) - Add factory function to wrap a serial descriptor with a custom name for custom delegating serializers (#1547) (thanks to Fadenfire)
- Allow contextually serialized types to be used as map keys in Json (#1552) (thanks to pdvrieze)
Bugfixes and performance improvements
- Update size in
JsonStringBuilder
slow-path to avoid excessive array-copies for large strings with escape symbols (#1491) - Optimize integer encoding length in CBOR (#1570) (thanks to davertay)
- Throw
JsonDecodingException
instead ofClassCastException
during unexpected null inTreeJsonDecoder
(#1550) - Prohibit 'null' strings in lenient mode in order to get rid of 'null' and "null" ambiguity (#1549)
- Avoid usage of reflective-like
serialDescriptor<KType>
in production sources (#1540) - Added correct error message when deserializing missing enum member for Properties format (#1539)
- Make
DescriptorSchemaCache
in Json thread-local on Native (#1484)
1.2.1
This release mainly contains bugfixes for various issues, including important broken thread-safety and improper encoding.
Features
- Added support for nullable values, nested and empty collections in protobuf (#1430)
Bugfixes
- Support @JsonNames for enum values (#1473)
- Handle EOF in skipElement correctly (#1475)
- Allow using value classes with primitive carriers as map keys (#1470)
- Read JsonNull only for non-string literals in JsonTreeReader (#1466)
- Properly reuse JsonStringBuilders in CharArrayPool (#1455)
- Properly ensure capacity of the string builder on the append slow-path (#1441)
1.2.0
This release has some known critical bugs, so we advise to use 1.2.1 instead.
This release contains a lot of new features and important improvements listed below;
Kotlin 1.5.0 is used as a default compiler and language version.
JSON performance improvements
JSON encoder and decoder were revisited and significantly rewritten,
which lead us to up to 2-3x times speedup in certain cases.
Additional details can be found in the corresponding pull requests: [1], [2].
Ability to specify alternative names during JSON decoding
The one of the most voted issues is fixed now — it is possible to specify multiple names for one property
using new @JsonNames
annotation.
Unlike @SerialName
, it only affects JSON decoding, so it is useful when dealing with different versions of the API.
We've prepared a documentation for you about it.
JsonConfiguration in public API
JsonConfiguration
is exposed as a property of Json
instance. You can use it to adjust behavior in
your custom serializers.
Check out more in the corresponding issue and the PR.
Generator for .proto files based on serializable Kotlin classes
Our implementation of Protocol Buffers format uses @Serializable
Kotlin classes as a source of schema.
This is very convenient for Kotlin-to-Kotlin communication, but makes interoperability between languages complicated.
To resolve this issue, we now have a
schema generator that can produce .proto files out of Kotlin classes. Using it, you can keep Kotlin
classes as a source of truth and use traditional protoc compilers for other languages at the same time.
To learn more, check out the documentation for the new ProtoBufSchemaGenerator
class or
visit the corresponding PR.
Note: this generator is on its experimental stage and any feedback is very welcomed.
Contextual serialization of generic classes
Before 1.2.0, it was impossible to register context serializer for generic class,
because contextual
function accepted a single serializer.
Now it is possible to register a provider — lambda that allows to construct a serializer for generic class
out of its type arguments serializers. See the details in the documentation.
Other features
- Support for watchosX64 target (#1366).
- Introduce kotlinx-serialization-bom (#1356).
- Support serializer on JS IR when T is an interface (#1431).