• Ben Gordon's avatar
    Requesting extension id 1072 · f5362e11
    Ben Gordon authored
    This library supports an idiomatic proto3 protobuf generator for kotlin.
    The library will be open sourced by Toast Inc under the Apache2 license, and is currently used in production at Toast.
    The following is the readme.md that will be released with the code by the end of Q4 2019.
    
    Supports only the Protocol Buffers language version 3.
    
    #### Features
    - Clean data class generation
    - Oneof types handled as sealed classes
    - JavaDoc comments on generated code
    - Deprecation option pass-through to Kotlin's `@Deprecated` annotation
    - Protokt-specific options: non-null types, wrapper types, interface implementation,
    and more
    - Tight integration with Protobuf's Java library: compatibility with its well-known
    types and usage of CodedInputStream and CodedOutputStream for best performance
    
    #### Not yet implemented
    - Kotlin native support
    - Kotlin JS support
    - Support for gRPC service generation
    - Protobuf JSON support
    
    See examples in [protokt-testing](https://github.com/toasttab/protokt/tree/master/protokt-testing).
    
    ### Generated Code
    Generated code is placed in `<buildDir>/generated-sources/main/protokt`.
    
    A simple example:
    ```proto
    syntax = "proto3";
    
    package com.protokt.sample;
    
    message Sample {
      string sample_field = 1;
    }
    ```
    
    will produce:
    ```kotlin
    /*
     * Generated by protokt. Do not modify.
     */
    package com.protokt.sample
    
    import com.toasttab.protokt.rt.*
    
    data class Sample(
        val sampleField: String,
        val unknown: Map<Int, Unknown> = emptyMap()
    ) : KtMessage {
        @Suppress("UNUSED")
        constructor(
            sampleField: String = ""
        ) : this(
            sampleField,
            emptyMap()
        )
    
        override val messageSize by lazy { sizeof() }
    
        override fun serialize(serializer: KtMessageSerializer) {
            if (sampleField.isNotEmpty()) {
                serializer.write(Tag(10)).write(sampleField)
            }
            if (unknown.isNotEmpty()) {
                serializer.writeUnknown(unknown)
            }
        }
    
        private fun sizeof(): Int {
            var res = 0
            if (sampleField.isNotEmpty()) {
                res += sizeof(Tag(1)) + sizeof(sampleField)
            }
            res += unknown.entries.sumBy { it.value.sizeof() }
            return res
        }
    
        companion object Deserializer : KtDeserializer<Sample> {
            override fun deserialize(deserializer: KtMessageDeserializer): Sample {
                var sampleField = ""
                val unknown = mutableMapOf<Int, Unknown>()
                while (true) {
                    when (deserializer.readTag()) {
                        0 ->
                            return Sample(
                                sampleField,
                                unknown
                            )
                        10 -> sampleField = deserializer.readString()
                        else -> {
                            val unk = deserializer.readUnknown()
                            unknown[unk.fieldNum] = unknown[unk.fieldNum].let {
                                when (it) {
                                    null -> unk
                                    else ->
                                        when (val v = it.value) {
                                            is ListVal ->
                                                Unknown(unk.fieldNum, ListVal(v.value + unk.value))
                                            else ->
                                                Unknown(unk.fieldNum, ListVal(listOf(v, unk.value)))
                                        }
                                }
                            }
                        }
                    }
                }
            }
        }
    }
    ```
    
    #### Runtime Notes
    ##### Package
    The Kotlin package of a generated file can be overridden from protobuf package with the `(protokt).package` option:
    ```proto
    syntax = "proto3";
    
    import "protokt.proto";
    
    package com.example;
    
    option (protokt).package = "com.package";
    ```
    
    ##### Message
    Each protokt message implements the `KtMessage` interface. `KtMessage` defines the `serialize()`
    method and its overloads which can serialize to a byte array, a `KtMessageSerializer`, or on the JVM,
    an `OutputStream`.
    
    Each protokt message has a companion object `Deserializer` that implements the `KtDeserializer`
    interface, which provides the `deserialize()` method and its overloads to construct an
    instance of the message from a byte array, a Java InputStream, or others.
    
    In order to enjoy the full benefits of Kotlin data classes, byte arrays are wrapped in the
    protokt `Bytes` class, which provides appropriate `equals()` and `hashCode()` implementations.
    
    ##### Enums
    Enum fields are generated as data classes with a single integer field. Kotlin enum classes are
    closed and cannot retain unknown values, and protobuf requires that unknown enum values are
    preserved for reserialization. This compromise exposes a constructor taking an integer, but the
    `from(value: Int)` on an enum's `Deserializer` should be preferred as it avoids instantiation
    when possible.
    
    
    Other notes:
    - `optimize_for` is ignored.
    - `repeated` fields are deserialized to Lists.
    - `map` fields are deserialized to Maps.
    - `oneof` fields are represented as data class subtypes of a sealed base class with a single property.
    
    ### Extensions
    See examples of each option in the [protokt-options](https://github.com/toasttab/protokt/tree/master/protokt-testing/protokt-options/src/main/proto)
    module. All protokt-specific options require importing `protokt.proto` in the protocol file.
    #### Wrapper Types
    Sometimes a field on a protobuf message corresponds to a concrete nonprimitive type. In
    standard protobuf the user would be responsible for this extra transformation, but the
    protokt wrapper type option allows specification of a converter that will automatically
    encode and decode custom types to protobuf primitives and well-known types. Some standard
    types are implemented in
    [protokt-extensions](https://github.com/toasttab/protokt/tree/master/protokt-extensions/src/main/kotlin/com/toasttab/protokt/ext).
    
    Wrap a field by invoking the `(protokt_property).wrap` option:
    ```proto
    message DateWrapperMessage {
      int64 date = 1 [
        (protokt_property).wrap = "java.util.Date"
      ];
    }
    ```
    
    Converters implement the `Converter` interface:
    ```kotlin
    interface Converter<S: Any, T: Any> {
        val wrapper: KClass<S>
    
        fun wrap(unwrapped: T): S
    
        fun unwrap(wrapped: S): T
    }
    ```
    
    and protokt will reference the converter's methods to wrap and unwrap from protobuf primitives:
    ```kotlin
    object DateConverter : Converter<Date, Long> {
        override val wrapper = Date::class
    
        override fun wrap(unwrapped: Long) =
            Date(unwrapped)
    
        override fun unwrap(wrapped: Date) =
            wrapped.time
    }
    ```
    
    ```kotlin
    data class WrapperModel(
        val date: java.util.Date,
        ...
    ) : KtMessage {
        ...
        override fun serialize(serializer: KtMessageSerializer) {
            serializer.write(Tag(10)).write(Int64(DateConverter.unwrap(date)))
            ...
        }
    
        override fun deserialize(deserializer: KtMessageDeserializer): WrapperModel {
            var date = 0L
    
            while (true) {
                when (deserializer.readTag()) {
                    0 ->
                        return WrapperModel(
                            DateConverter.wrap(date),
                            ...
                        )
                    ...
                }
            }
        }
    }
    ```
    
    Converters can also implement the `OptimizedSizeofConverter` interface adding `sizeof()`,
    which allows them to optimize the calculation of the wrapper's size rather than unwrap
    the object twice. For example, a UUID is always 16 bytes:
    
    ```kotlin
    object UuidConverter : OptimizedSizeofConverter<UUID, ByteArray> {
        override val wrapper = UUID::class
    
        private val sizeofProxy = ByteArray(16)
    
        override fun sizeof(wrapped: UUID) =
            sizeof(sizeofProxy)
    
        override fun wrap(unwrapped: ByteArray): UUID {
            require(unwrapped.size == 16) {
                "input must have size 16; had ${unwrapped.size}"
            }
    
            return ByteBuffer.wrap(unwrapped)
                .run { UUID(long, long) }
        }
    
        override fun unwrap(wrapped: UUID) =
            ByteBuffer.allocate(16)
                .putLong(wrapped.mostSignificantBits)
                .putLong(wrapped.leastSignificantBits)
                .array()
    }
    ```
    
    Rather than convert a UUID to a byte array both for size calculation and for serialization
    (which is what a naïve implementation would do), UuidConverter always returns the size of a
    constant 16-byte array.
    
    If the wrapper type is in the same package as the generated protobuf message, then it
    does not need a fully-qualified name. Custom wrapper type converters can be in the same module as
    protobuf types that reference them. In order to use any wrapper type defined in
    `protokt-extensions`, the module must be included as a dependency:
    
    ```groovy
    dependencies {
        implementation 'com.toasttab.protokt:protokt-extensions:0.0.3'
    }
    ```
    
    #### Interface implementation
    To avoid the need to create domain-specific objects from protobuf messages you can declare
    that a protobuf message implements a custom interface with properties and default methods.
    
    ```kotlin
    package com.protokt.sample
    
    interface Model {
        val id: String
    }
    ```
    
    ```proto
    package com.protokt.sample;
    
    message ImplementsSampleMessage {
      option (protokt_class).implements = "Model";
    
      string id = 1;
    }
    ```
    
    If the wrapper interface is in the same package as the generated protobuf message, then it
    does not need a fully-qualified name. Wrapper interfaces cannot be used by protobuf messages
    in the same module that defines them; the dependency must be declared with`protoktExtensions`
    in `build.gradle`:
    
    ```groovy
    dependencies {
        protoktExtensions project(':api-module')
    }
    ```
    
    #### Nonnull fields
    If there is a message that has no meaning whatsoever when a particular field is missing, you
    can emulate proto2's `required` key word by using the `(protokt_oneof).non_null` option:
    
    ```proto
    message Sample {
    }
    
    message NonNullSampleMessage {
      Sample non_null_sample = 1 [
        (protokt_property).non_null = true
      ];
    }
    ```
    
    Generated code will not have a nullable type so the field can be referenced without using
    Kotlin's `!!`.
    
    Oneof fields can also be declared non-null:
    
    ```proto
    message NonNullSampleMessage {
      oneof non_null_oneof {
        option (protokt_oneof).non_null = true;
    
        string message = 2;
      }
    }
    ```
    
    Note that deserialization of a message with a non-nullable field will fail if the
    message being decoded does not contain an instance of the required field.
    
    #### BytesSlice
    When reading messages that contain other serialized messages as `bytes` fields, protokt can
    keep a reference to the originating byte array to prevent a large copy operation on
    deserialization. This can be desirable when the wrapping message is a thin metadata shim and
    doesn't include much memory overhead:
    
    ```proto
    message SliceModel {
      int64 version = 1;
    
      bytes encoded_message = 2 [
        (protokt_property).bytes_slice = true
      ];
    }
    ```
    
    ### Usage
    
    #### Gradle
    
    ```groovy
    buildscript {
        dependencies {
            classpath "com.toasttab.protokt:protokt-gradle-plugin:0.0.3"
        }
    }
    
    apply plugin: 'com.toasttab.protokt'
    ```
    
    This will automatically download and install protokt, apply the Google protobuf plugin,
    and configure all the necessary boilerplate. By default it will also add `protokt-runtime`
    to the api scope of the project, and `protobuf-java` to the implementation scope.
    
    If your project is pure Kotlin you may run into the following error:
    
    ```
    Execution failed for task ':compileJava'.
    > error: no source files
    ```
    
    To work around it, disable all `JavaCompile` tasks in the project:
    
    ```groovy
    tasks.withType(JavaCompile) {
        enabled = false
    }
    ```
    
    or:
    ```groovy
    compileJava.enabled = false
    ```
    
    #### Command line code generation
    
    ```bash
    protokt-codegen$ ./gradlew assemble [OR ./gradlew installDist]
    
    protokt-codegen$ ./run-protokt.sh -h
    
    protokt-codegen$ ./run-protokt.sh \
      -out=../kotlin \
      -pkg=com.toasttab.protokt.conformance \
      -file=conformance.proto \
      -cp=../build/libs/protokt-codegen-0.0.3-SNAPSHOT-all.jar \
      -plugin=../bin/protokt.sh
    ```
    
    ### Contribution
    To enable rapid development of the code generator, the protobuf conformance tests have been 
    compiled and included in the protokt-testing project. They run on Mac OS 10.14+ and Ubuntu
    16.04 x86-64.
    
    Publish the plugin to the integration repository:
    ```bash
    protokt$ ./gradlew publishToIntegrationRepository
    ```
    
    Then run the tests from `protokt-testing`:
    ```bash
    protokt-testing$ ./gradlew protokt-conformance-tests:test
    ```
    
    All integration tests can be run with:
    ```
    protokt-testing$ ./gradlew test
    ```
    f5362e11
Name
Last commit
Last update
..
options.md Loading commit data...
performance.md Loading commit data...
third_party.md Loading commit data...