Add fixtures and migrate to serde_with

[aumetra]

No description provided.

[aumetra]

@tesaguri There are two failures right now and I'm not quite sure why.
Since you wrote the code, I thought you could maybe look over it. Because a lot of the migration from raw serde types to the serde_with traits worked flawless.

There is just some weird bug with FirstOk and with the Id deserializer..

[aumetra]

Never mind, fixed one issue, and replaced the other with a different one :^)
Really fun!

[aumetra]

@tesaguri just one little question, I can't quite wrap my head around that yet, why do we need this SeqAccess indirection here?
I get that it's required for it to work (I tried other things), but not quite sure why. Why do we need this indirection instead of doing something with seq.next_element() and some mapping operations?

kitsune/crates/kitsune-type/src/jsonld/serde/id.rs

Lines 87 to 115 in 7dea8fd

	struct SeqAccess<A>(A);
	impl<'de, A> de::SeqAccess<'de> for SeqAccess<A>
	where
	A: de::SeqAccess<'de>,
	{
	type Error = A::Error;
	fn next_element<T>(&mut self) -> Result<Option<T>, Self::Error>
	where
	T: Deserialize<'de>,
	{
	let value = self.0.next_element::<DeserializeAsWrap<_, Id>>()?;
	Ok(value.map(DeserializeAsWrap::into_inner))
	}
	fn next_element_seed<T>(&mut self, _seed: T) -> Result<Option<T::Value>, Self::Error>
	where
	T: DeserializeSeed<'de>,
	{
	unreachable!();
	}
	fn size_hint(&self) -> Option<usize> {
	self.0.size_hint()
	}
	}
	T::deserialize(SeqAccessDeserializer::new(SeqAccess(seq)))

[tesaguri]

@tesaguri just one little question, I can't quite wrap my head around that yet, why do we need this SeqAccess indirection here?
I get that it's required for it to work (I tried other things), but not quite sure why. Why do we need this indirection instead of doing something with seq.next_element() and some mapping operations?

I don't remember my thought at that time exactly so it's going to be a wild guess, but with seq.next_element::<T>(), I think <T as Deserialize>::deserialize would see map values like {"id": "https://example.com/", …} as-is. On the other hand, our goal here is to show the deserialize method a string value like "https://example.com/".

In addition to the map type, we also need to be able to deserialise verbatim string inputs like "https://example.com/". I think (unverified) the only way of handling both the types via seq.next_element() without the SeqAccess indirection would be to deserialise the next element with seq.next_element::<Value>() and matching on Value::String and Value::Object, which wouldn't be very efficient (if we don't need efficiency, we would be deserialising the entire input JSON into a DOM instead of bothering with the whole jsonld::serde megalith in the first place).

Also, Id::deserialize_as(SeqAccessDeserializer::new(seq)) would run into infinite recursion since the SeqAccessDeserializer::deserialize_any method would call Visitor::visit_seq.