1

i have a channel handler that contains code like the following:

/// Create a (generic) encoding view
var output:BSON.Output<ByteBufferView> = .init(
    preallocated: .init(context.channel.allocator.buffer(
        capacity: .init(message.header.size))))

/// Encode a message
output += message

/// Write the encoded ByteBuffer to the channel pipeline
context.writeAndFlush(self.wrapOutboundOut(ByteBuffer.init(output.destination)),
    promise: promise)

where

@frozen public
struct Output<Destination>:Sendable
    where   Destination:RangeReplaceableCollection<UInt8>,
            Destination.Index == Int,
            Destination:Sendable
{
    public
    var destination:Destination
}

i do not want BSON.Output to remain generic, this has caused a lot of (in hindsight, easily forseen) friction with the optimizer. because the BSON library has no relationship to SwiftNIO, i wanted to standardize it on ArraySlice<UInt8>.

for the channel handler, this means ByteBuffer.init(output.destination) is no longer an efficient way to write data to the channel pipeline, because it will now always allocate a new buffer and copy all the memory. i find this hard to justify, because the data already exists, contiguously, in the original ArraySlice<UInt8> destination buffer.

how do i send a normal ArraySlice<UInt8> (or even an unsafe pointer representation of it) over the network without allocating a second buffer for every message?

2

Fundamentally, you cannot. NIO expects ByteBuffer as the mechanism of transport to the network, and does not have any capacity to send or receive an ArraySlice<UInt8>.

What you can do is reuse the buffer you've allocated. This should remove the extra allocations, but will retain the memory copies.