Prototik / kotlinx-io

Kotlin multiplatform I/O library

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Official JetBrains project Download TeamCity Build GitHub License

kotlinx-io is a multiplatform library suitable for I/O primitives building and manipulations

Experimental

Please note that this library is experimental. Any API is a subject to change.

Best with kotlinx.coroutines-io

Setup

repositories {
    jcenter()
}

dependencies {
    compile "org.jetbrains.kotlinx:kotlinx-io-jvm:$kotlinx_io_version"
}

Use kotlinx-io-js with Kotlin JavaScript and kotlin-io for common code if you are doing multiplatform module.

Basic concepts

IoBuffer

A buffer view is a view to byte buffer (in JVM it could be direct ByteBuffer, on JS it could be ArrayBuffer). Comparing to java's NIO ByteBuffer, IoBuffer ...

  • should be released via release() invocation
  • could be copied via copy() that actually doesn't copy any bytes but makes a new view
    • copy should be released as well
  • could be used to read and write, no need to do flip
  • has a next property so it is suitable to make chains of buffer views with no need to allocate any lists or extra arrays
  • designed to work with buffer view pool

Note that IoBuffer is not concurrent safe however it is safe to do copy() and release() concurrently (when this makes sense).

ByteReadPacket

ByteReadPacket is a packet consist of a managed sequence of buffer views. So one can easily read from a packet and buffer view segments will be released on the way of reading. As far it contains buffer views, a ByteReadPacket instance should be released as well via release() invocation at the end however there is no need to do it (but it is allowed to do) if all the bytes were consumed and the packet is empty.

  • byte packet is read only
  • there is no way to reset/pushback already readen bytes
  • every buffer view will be released once it becomes empty
  • does support copy(), similar to IoBuffer.copy() it doesn't copy bytes
  • not reusable - once all bytes were read there is no way to reset it to read bytes again - make a copy instead
  • supports start gap hint (see byte packet builder)
  • provides java.io.Reader (reads characters as UTF-8) and java.io.InputStream compatibility
suspend fun parse(packet: ByteReadPacket, tee: SendChannel<ByteReadPacket>) {
    tee.send(packet.copy())

    while (!packet.isEmpty()) {
        val size = packet.readInt()
        for (i in 1 .. size) {
            println("number: ${packet.readInt()}")
        }
    }
}

BytePacketBuilder

A packet builder that consists of a sequence of buffer views. It borrows buffer views from a pool on demand and does nevery copy bytes on growth (as it does ByteArrayOutputStream).

  • write-only
  • has explicit release() function to discard all bytes
  • build() makes an instance of ByteReadPacket and resets builder's state to the initial one so builder becomes empty and ready to build another one packet
  • supports optimized write byte packet operation: could merge miltiple buffers into one if possible (only if bytes quantity is not too large), considers start gap hint as well
  • provides java.io.OutputStream and java.lang.Appendable (appends characters as UTF-8)
  • as was noted before it is reusable: another byte packet could be built once build() has been invoked to build a previous one or reset() to discard all previously written bytes
val packet = buildPacket {
    writeInt(size * 4)
    for (i in 1..size) {
        writeInt(i)
    }
}
private fun BytePacketBuilder.writeData() = TODO("write something")

suspend fun loop(destination: SendChannel<ByteReadPacket>) {
    val builder = BytePacketBuilder()

    try {
        while (true) {
            builder.writeData()
            destination.send(builder.build())
        }
    } finally {
        builder.release()
    }
}

ObjectPool

ObjectPool is a general purpose lock-free concurrent-safe object pool. It is leak-safe: all object that hasn't been recycled but collected by GC do not cause any issues with a pool but only allocation penalty. Note that it doens't mean that leaking object will not cause any issues at all as lost objects could hold some native or external resources. The only guarantee is that ObjectPool is not going to break if there are lost objects.

val ExampleIntArrayPool = object : DefaultPool<IntArray>(ARRAY_POOL_SIZE) {
    override fun produceInstance(): IntArray = IntArray(ARRAY_SIZE)
}
class ExampleDirectByteBufferPool(val bufferSize: Int, size: Int) : DefaultPool<ByteBuffer>(size) {
    override fun produceInstance(): ByteBuffer = java.nio.ByteBuffer.allocateDirect(bufferSize)

    override fun clearInstance(instance: ByteBuffer): ByteBuffer {
        instance.clear()
        return instance
    }

    override fun validateInstance(instance: ByteBuffer) {
        require(instance.isDirect)
        require(instance.capacity() == bufferSize)
    }
}

About

Kotlin multiplatform I/O library

License:Apache License 2.0


Languages

Language:Kotlin 100.0%