Nice! If you want something really powerful I’d co...

# arrow

[pakoito]

Nice! If you want something really powerful I’d consider supporting unions/sealed classes

[sam]

Great idea. I'll add that for 1.1

[sam]

How do you see that working? Defining a field as the sealed class supertype, and then marshalling to one of the sealed instances? But which one to pick, look at the fields available?

[pakoito]

It’s implementation-dependent, let me show you JSON

[pakoito]

type my_sealed = Left of int | Right of string

type bla = { a: my_sealed }

"bla": {
  "a" : {
    "left" : 1
  }
}

"bla": {
  "a" : {
    "right" : "2"
  }
}

[sam]

so lookup the type based on the field name

[sam]

In avro4s what I do is check which fields are defined, and find the sealed type that matches it

[sam]

although that wouldn’t work if you have two types with the same field names

[pakoito]

yes, so the validation is just seeing that there’s only one field, and the field name is one of the union

[pakoito]

you shouldn’t

[pakoito]

in most languages it isn’t possible to have the same name for a variant

[sam]

I mean, if you did Left(a:String) and Right(a:String) then the presence of a field “a” isn’t enough to determine which to pick

[pakoito]

and the values of a variant aren’t types themselves

[pakoito]

yeah, that’s a frequent design mistake

[pakoito]

the “dependent type”

[pakoito]

happened in FlowJS a lot

[sam]

So I guess the only way to do it is to have a nested block that informs the type

[sam]

or in yaml, you could ahve another field like _type as a sibling

[pakoito]

that’s the tag, it makes validating harder AFAIK

[pakoito]

"bla": {
  "a" : {
     "type": "right",
    "content" : "2"
  }
}

"bla": {
  "a" : {
     "type": "left",
    "inside" : 2
  }
}

[pakoito]

problem here: you have to parse the value, see the tag, then reparse/revalidate the contents

[pakoito]

whereas with the other one you can directly access the deserializer based off the field

[pakoito]

because each field corresponds to a monomorphic deserializer

[pakoito]

same for serialization

[sam]

yes agreed, good