Kotlin

Edit on GitHub

Kotlin API migration guide for 1.0.0

The API has been extensively modified with the following goals in mind:

  • Enhancing the API to be more idiomatic to Kotlin.
  • Match the API rework done through the Rust Zenoh API.
  • Abstracting users from the underlying native mechanisms

Default arguments

Throughout the 0.11.0 API we were exposing builders, however we decided to replace all of them with the more Kotlin-idiomatic way of the default arguments.

For instance in 0.11.0:

session.put(keyExpr, value)
  .congestionControl(CongestionControl.BLOCK)
  .priority(Priority.REALTIME)
  .res()

Becomes the following in 1.0.0:

session.put(keyExpr, 
  value, 
  congestionControl = CongestionControl.BLOCK, 
  priority = Priority.REALTIME,
)

Notice that the .res() function has been removed. Now functions exposed by the API execute imediately, without the need of .res(). i.e. When doing session.put(...) the put is run immediately.

This applies to all the builders being present on 0.11.0. Generally 1 all the parameters present on each builder have now their corresponding argument available in the functions.

Session opening

Session.open(config: Config) has now been replaced with Zenoh.open(config: Config).

Encoding rework

The Encoding class has been modified. In 0.11.0. it had the signature

class Encoding(val knownEncoding: KnownEncoding, val suffix: String = "")

where KnownEncoding was an enum.

In 0.11.0. an encoding instance would be created as follows:

val encoding = Encoding(knownEncoding = KnownEncoding.TEXT_JSON)

In 1.0.0. we have implemented the following changes:

  • KnownEncoding enum is removed, instead we provide static Encoding instances containing an ID and a description.
  • Custom encodings can be created
  • The list of pre-defined encodings has been expanded.

In 1.0.0. the previous example would instead now become:

val encoding = Encoding.TEXT_JSON

Session-managed declarations

Up until 0.11.0, it was up to the user to keep track of their variable declarations to keep them alive, because once the variable declarations were garbage collected, the declarations were closed. This was because each Kotlin variable declaration is associated with a native Rust instance, and in order to avoid leaking the memory of that Rust instance, it was necessary to free it upon dropping the declaration instance. However, this behavior could be counterintuitive, as users were expecting the declaration to keep running despite losing track of the reference to it.

In this release we introduce a change in which any session declaration is internally associated to the session from which it was declared. Users may want to keep a reference to the declaration in case they want to undeclare it earlier before the session is closed, otherwise, the declaration is kept alive.

For instance:

val keyExprA = "A/B/C".intoKeyExpr().getOrThrow()
val subscriber = session.declareSubscriber(keyExprA, 
  callback = { sample -> println("Receiving sample on 'A/B/C': ${sample.payload}") }).getOrThrow()

val keyExprB = "A/C/D".intoKeyExpr().getOrThrow()
session.declareSubscriber(keyExprB, 
  callback = { sample -> println("Receiving sample on 'A/C/D': ${sample.payload}") }) // No variable is associated to the declared session, on 0.11.0 it would have been instantly dropped

Therefore, when receiving a ‘hello’ message on A/** we would still see:

>> Receiving sample on 'A/B/C': hello
>> Receiving sample on 'A/C/D': hello

since both declarations are still alive.

Now the question is, for how long? What happens first, either when:

  • you call undeclare() or close() to the declaration
  • the session is closed, then all the associated declarations are automatically undeclared.

Key Expression rework

KeyExpr instances are not bound to a native key expression anymore, unless they are declared from a session. It is safe to drop the reference to the key expression instance, but the memory management associated to a key expression will differ:

  • If the KeyExpr was not declared from a session, then the garbage collector simply claims back the memory.
  • If it was declared from a session, then the session keeps track of it and frees the native memory upon closing the session.

Declaring a KeyExpr on a session results in better performance, since the session is informed that we intend to use a key expression repeatedly. We also associate a native key expression to a Kotlin key expression instance, avoiding copies.

Config loading

When opening a session, it’s now mandatory to provide a configuration to the session, even for a default config:

val config = Config.default()
Zenoh.open(config).onSuccess {
  // ...

}

The Config class has been modified

  • Config.default(): returns the default config
  • Config.fromFile(file: File): Result<Config>: allows to load a config file.
  • Config.fromPath(path: Path): Result<Config>: allows to load a config file from a path.
  • Config.fromJson(json: String): Result<Config>: loads the config from a string literal with json format
  • Config.fromJson5(json5: String): Result<Config>: loads the config from a string literal with json5 format
  • Config.fromYaml(yaml: String): Result<Config>: loads the config from a string literal with yaml format
  • Config.fromJsonElement(jsonElement: JsonElement): Result<Config>: loads the config from a kotlin JsonElement.

In case of failure loading the config, a Result.Error is returned.

Packages rework

The package structure of the API is now aligned with Zenoh Rust package structure.

Changes:

  • Removing the “prelude” package
  • QoS package now contains:
    • CongestionCOntrol
    • Priority
    • Reliability
    • QoS
  • Bytes package is created with:
    • ZBytes, IntoZBytes, Encoding
  • Config package:
    • Config, ZenohId
  • Session package:
    • SessionInfo
  • Query package:
    • contains Query and Queryable
    • removing queryable package

Reliability

The Reliability config parameter used on when declaring a subscriber, has been moved. It now must be specified when declaring a Publisher or when performing a Put or a Delete operation.

Logging

There are two main changes regarding logging, the interface and the mechanism.

Lets look at the following example, where we want to run the ZPub example with debug logging. On 1.0.0 we’ll do:

RUST_LOG=debug gradle ZPub

If we wanted to enable debug logging and tracing for some specific package, for instance zenoh::net::routing, we’d do:

RUST_LOG=debug,zenoh::net::routing=trace gradle ZPub

However, this is not enabled by default. In order to enable logging, one of these newly provided functions must be called:

Zenoh.tryInitLogFromEnv()

and

Zenoh.initLogFromEnvOr(fallbackFilter: String): Result<Unit>

This last function allows to programmatically specify the logs configuration if not provided as an environment variable.

ZBytes serialization / deserialization & replacement of Value

We have created a new abstraction with the name of ZBytes. This class represents the bytes received through the Zenoh network. This new approach has the following implications:

  • Attachment class is replaced by ZBytes.
  • Value is replaced by the combination of ZBytes and Encoding.
  • Replacing ByteArray to represent payloads

With ZBytes we have also introduced a Serialization and Deserialization for convenient conversion between ZBytes and Kotlin types.

Serialization & Deserialization

We can serialize primitive types into a ZBytes instance, that is, converting the data into bytes processed by the zenoh network:

Primitive types

(De)Serialization is supported by the following primitive types:

  • Numeric: Byte, Short, Int, Long, Float, Double, UByte, UShort, UInt and ULong.
  • String
  • ByteArray

For instance:

val exampleInt: Int = 256
val zbytes: ZBytes = zSerialize<Int>(exampleInt).getOrThrow()
val deserialization = zDeserialize<Int>(zbytes).getOrThrow()
check(exampleInt == deserialization)

This approach works as well for the other aforementioned types.

For serialization, String and ByteArray the functions ZBytes::from(string: String) and ZBytes::from(bytes: ByteArray) can be used respectively. Analogously, deserialization, ZBytes::toString() and ZBytes::toByteArray() can be used.

Lists

Lists are supported, but they must be either:

  • List of numeric types : (Byte, Short, Int, Long, Float, Double, UByte, UShort, UInt and ULong)
  • List of String
  • List of ByteArray
  • List of another supported type

The serialize syntax must be used:

val myList = listOf(1, 2, 5, 8, 13, 21)
val zbytes = zSerialize<List<Int>>(myList).getOrThrow()
val outputList = zDeserialize<List<Int>>(zbytes).getOrThrow()
check(myList == outputList)

Maps

Maps are supported as well, with the restriction that their inner types must supported primitives:

  • Numeric types
  • String
  • ByteArray
  • Map of another supported types
val myMap: Map<String, Int> = mapOf("foo" to 1, "bar" to 2)
val zbytes = zSerialize<Map<String, Int>>(myMap).getOrThrow()
val outputMap = zDeserialize<Map<String, Int>>(zbytes).getOrThrow()
output(myMap == outputMap)

Pair & Triple

Similarly,

  • Pair:
val input: Pair<String, Int> = Pair("one", 1)
val zbytes = zSerialize(input).getOrThrow()
val output: Pair<String, Int> = zDeserialize(zbytes).getOrThrow()
check(input == output)
  • Triple:
val input: Triple<String, Int, Boolean> = Triple("one", 1, true)
val zbytes = zSerialize(input).getOrThrow()
val output: Triple<String, Int, Boolean> = zDeserialize(zbytes).getOrThrow()
check(input == output)

Parameterized types combinations

Combinations of all the above types is supported. For instance:

  • List of lists
val input: List<List<Int>> = listOf(listOf(1, 2, 3))
val zbytes = zSerialize(input).getOrThrow()
val output = zDeserialize<List<List<Int>>>(zbytes).getOrThrow()
check(input == output)
  • List of maps
val input: List<Map<String, Int>> = listOf(mapOf("a" to 1, "b" to 2))
val zbytes = zSerialize(input).getOrThrow()
val output = zDeserialize<List<Map<String, Int>>>(zbytes).getOrThrow()
check(input == output)

Reply handling

Previously on 0.11.0 we were exposing the following classes:

  • Reply.Success
  • Reply.Error

In 0.11.0 executing a get on a session was done in the following way.

session.get(selector).res()
  .onSuccess { receiver ->
      runBlocking {
          for (reply in receiver!!) {
              when (reply) {
                  is Reply.Success -> {println("Received ('${reply.sample.keyExpr}': '${reply.sample.value}')")}
                  is Reply.Error -> println("Received (ERROR: '${reply.error}')")
              }
          }
      }
  }

Now, in 1.0.0, the Reply instances contain a result attribute, which is of type Result<Sample>.

session.get(selector, channel = Channel())
  .onSuccess { channelReceiver ->
      runBlocking {
          for (reply in channelReceiver) {
              reply.result.onSuccess { sample ->
                  when (sample.kind) {
                      SampleKind.PUT -> println("Received ('${sample.keyExpr}': '${sample.payload}')")
                      SampleKind.DELETE -> println("Received (DELETE '${sample.keyExpr}')")
                  }
              }.onFailure { error ->
                  println("Received (ERROR: '${error.message}')")
              }
          }
      }
  }

  1. Some parameters have been either moved (e.g. Reliability) or removed (e.g. SampleKind). ↩︎