feat: rpc2: cbor codec

[lucix-aws]

CBOR codec implementation.

[isaiahvita]

nit: what is meant by specification? does this package implement a subset of the smithy rpcv2-cbor protocol or a subset of CBOR as described in RFC 8949?

[kstich]

Agreed, I think being clear in this preamble documentation on what is/isn't in line with the CBOR spec is important (half-precision, etc.)

[kstich]

Agreed, I think being clear in this preamble documentation on what is/isn't in line with the CBOR spec is important (half-precision, etc.)

[kstich]

How are customers expected to use this value in a negative context?

[lucix-aws]

They don't ever see it - it's just an intermediate value for encode/decode (like all of these Value implementations are). Behind the scenes we figure out whether this value fits into the type of the modeled field when we deserialize.

[kstich]

What about if this were to end in 0xff?

[lucix-aws]

So essentially I chose to build in the -1 bias from type 1 into the codec layer. Since the range of type1 is [-2^64,-1], the wrap to zero that comes from a type1 of 64 1s would basically be that implicit 65-bit left bound.

In practice remember the caller doesn't ever see it, it's something the generated deserializer would reject since -2nd ^64 doesn't fit in long.

We could drop the implicit bias here and instead have the generated deserializers reintroduce it but I think we'd be trading up in terms of code size (it's a special case on NegInt(0) vs a bunch of +1s).

[lucix-aws]

Maybe to clarify, this code exists in the in-progress deserializer, this is basically what happens when we're deserializing to something that's modeled as a long.

// AsInt64 coerces a Value to its int64 representation if possible.
func AsInt64(v Value) (int64, error) {
    const max64 = 0x7fffffff_ffffffff

    switch vv := v.(type) {
    case Uint:
        if vv > max64 {
            return 0, fmt.Errorf("cbor uint %d exceeds max int64 value", vv) 
        }   
        return int64(vv), nil 
    case NegInt:
        if vv > max64+1 || vv == 0 { // -1 bias means NegInt(0) is actually -2^64
            return 0, fmt.Errorf("cbor negint %d exceeds min int64 value", vv) 
        }   
        return -int64(vv), nil 
    }   
    return 0, fmt.Errorf("unexpected value type %T", v)
}

[kstich]

Max value of a uint64 would be 2^64 - 1, but this would add that one.

[lucix-aws]

See above