Serialize numeric types with smallest lossless format

[timothy-shields]

What do you think about serializing numeric types with the smallest lossless format? For example, if I have a double[] of length N containing 0's and 1's, its contents (after the header) would be stored using N bytes, because 0 and 1 fit losslessly into a positive fixnum. If I try to then deserialize to a double[], that would work.

msgpack-cli is already doing this for ints and uints internally.

The implementation of DoubleMsgPackValue.Pack would become something logically similar to this:

public override void Pack(Packer packer)
{
    if (this.Value == (int)this.Value)
    {
        packer.Pack(this.Value < 0 ? (int)this.Value : (uint)this.Value);
    }
    else if (this.Value == (float)this.Value)
    {
        packer.Pack((float)this.Value);
    }
    else
    {
        packer.Pack(this.Value);
    }
}

What are your thoughts? This could potentially be an optional configuration on the MessagePackWriter. I'd be willing to help out with this.

[timothy-shields]

I've actually tried this locally now, and it's working great. Deserialization appears to already handle this without any changes. That is, if you try to deserialize a uint fixnum containing the value 0 (hex 00) into a C# double that works correctly.

[darkl]

Hi,

The feature sounds nice,
You are welcome to propose a pull request with your changes. I think it is preferred this feature can be configurable.

I encourage you to propose further changes/improvements to this library. This project could use a little love.