json: Support representing enum variants as tags

[flying-sheep]

Comment moved over from facet-rs/facet#326:

My use case that the serde ecosystem is struggling with is this:

#[derive(serde_derive::Serialize, serde_derive::Deserialize)]
#[serde(tag = "type")]
enum Outer {
  OVar1,
  #[serde(untagged)]
  Inner(Inner),
}

#[derive(serde_derive::Serialize, serde_derive::Deserialize)]
#[serde(tag = "type")]
enum Inner {
  IVar1,
  IVar2,
}

Which should just work like

• Outer::OVar1 ⭤ {"type":"OVar1"}
• Outer::Inner(Inner::Ivar1) ⭤ {"type":"Ivar1"} ⭤ Inner::Ivar1
• Outer::Inner(Inner::Ivar2) ⭤ {"type":"Ivar2"} ⭤ Inner::Ivar2

But also strange things happen like [0] deserializing to Outer::OVar1.

I feel like we can do better.

[lmmx]

This issue was opened a week before the following issue/PR:

• facet-core: define a dedicated tag attribute facet#634
• Add support for Shape::type_tag facet#688

That is to say, this would appear to have some of the groundwork in place

Note that it was introduced as #[facet(type_tag = "...")] not tag = "..."

[lmmx]

Having looked through the code my understanding is that the type_tag on an enum Shape specifies only that the enum uses external tagging but doesn't specify which field name contains the tag.

ASN.1

Note

ASN.1 embeds this in the protocol:

To ensure that encodings are not ambiguous, every ASN.1 type is associated with a tag.

...the newly added tag appears as the value's encoded identifier. This identifier is followed by octets encoding the value's length and contents. The contents octets of tagged types completely encode the value of the type being tagged

— source

They're known as the "TLV" bytes (source), so from the looks of it they are doing both type annotation and discrimination.

Since JSON does not have such a protocol then facet-json would need to store this mapping.

serde

The corresponding docs for serde are here where they refer to it as "internally tagged"

#[serde(tag = "type")]

• Update: draft begun in Tagged enum #38