Suggestion(Seq): Map to Vector of structs about serde_with

[jonasbb]

From/Into is not the correct interface here. During serialization you only have references, so this might lead to cloning. The way you wrote the attributes you would get a TupSt<Same, Same>, which is not really a useful struct to have.

The SerializeAs/DeserializeAs traits do cover that use case. With them, this can work (It doesn't currently, because Seq is not flexible enough). It does make quite heavy use of the generics.

#[derive(serde::Serialize, serde::Deserialize)]
struct Custom<K, V> {
    custom_key: K,
    v: V,
}

impl<K, KAs, V, VAs> SerializeAs<(K, V)> for Custom<KAs, VAs>
where
    KAs: SerializeAs<K>,
    VAs: SerializeAs<V>,
{
    fn serialize_as<S>((k, v): &(K, V), serializer: S) -> Result<S::Ok, S::Error>
    where
        S: serde::Serializer,
    {
        (Custom {
            custom_key: SerializeAsWrap::<K, KAs>::new(k),
            v: SerializeAsWrap::<V, VAs>::new(v),
        })
        .serialize(serializer)
    }
}

impl<'de, K, KAs, V, VAs> DeserializeAs<'de, (K, V)> for Custom<KAs, VAs>
where
    KAs: DeserializeAs<'de, K>,
    VAs: DeserializeAs<'de, V>,
{
    fn deserialize_as<D>(deserializer: D) -> Result<(K, V), D::Error>
    where
        D: serde::Deserializer<'de>,
    {
        let c = <Custom<DeserializeAsWrap<K, KAs>, DeserializeAsWrap<V, VAs>>>::deserialize(
            deserializer,
        )?;
        Ok((c.custom_key.into_inner(), c.v.into_inner()))
    }
}

#[serde_as]
#[derive(Debug, Serialize, Deserialize, PartialEq)]
struct SM(#[serde_as(as = "Seq<Custom<_, _>>")] BTreeMap<u32, IpAddr>);

let map: BTreeMap<_, _> = vec![(1, ip), (10, ip), (200, ip2)].into_iter().collect();
is_equal(
    SM(map),
    expect![[r#"
        [
          {
            "custom_key": 1,
            "v": "1.2.3.4"
          },
          {
            "custom_key": 10,
            "v": "1.2.3.4"
          },
          {
            "custom_key": 200,
            "v": "255.255.255.255"
          }
        ]"#]],
);

Is that a good enough solution?

[ClSlaid]

Sorry this doesn't seem works, I got error:

error[E0277]: the trait bound `serde_with::Seq<KvProxy<Same, Same>>: SerializeAs<_>` is not satisfied
   --> src/model/manifest.rs:245:24
    |
244 | / #[serde_as]
245 | | #[derive(Debug, Clone, Serialize, Deserialize)]
    | |                        ^^^^^^^^-
    | |________________________|_______|
    |                          |       required by a bound introduced by this call
    |                          the trait `SerializeAs<_>` is not implemented for `serde_with::Seq<KvProxy<Same, Same>>`
    |
    = help: the following other types implement trait `SerializeAs<T>`:
              <serde_with::Seq<(KAs, VAs)> as SerializeAs<BTreeMap<K, V>>>
              <serde_with::Seq<(KAs, VAs)> as SerializeAs<HashMap<K, V>>>
    = note: required for `DefaultOnNull<serde_with::Seq<KvProxy<Same, Same>>>` to implement `SerializeAs<_>`
note: required by a bound in `serde_with::ser::<impl As<T>>::serialize`
   --> /home/cl/.cargo/registry/src/github.com-1ecc6299db9ec823/serde_with-2.2.0/src/ser/mod.rs:166:12
    |
166 |         T: SerializeAs<I>,
    |            ^^^^^^^^^^^^^^ required by this bound in `serde_with::ser::<impl As<T>>::serialize`
    = note: this error originates in the derive macro `Serialize` (in Nightly builds, run with -Z macro-backtrace for more info)

Looks like SerializeAs<BTreeMap<K,V>, Seq<KvProxy<KAs, VAs>>> cannot be deduced.

--- code ---

#[derive(Clone, Debug, Serialize, Deserialize)]
/// helper struct to deserialize a custom named struct into K-V pairs in Maps
struct KvProxy<KAs, VAs> {
    /// proxy typed key
    key: KAs,
    /// proxy typed value
    value: VAs,
}

impl<K, V, KAs, VAs> SerializeAs<(K, V)> for KvProxy<KAs, VAs>
where
    KAs: SerializeAs<K>,
    VAs: SerializeAs<V>,
{
    fn serialize_as<S>((k, v): &(K, V), serializer: S) -> Result<S::Ok, S::Error>
    where
        S: serde::Serializer,
    {
        let proxy = KvProxy {
            key: SerializeAsWrap::<K, KAs>::new(k),
            value: SerializeAsWrap::<V, VAs>::new(v),
        };
        proxy.serialize(serializer)
    }
}

impl<'de, K, V, KAs, VAs> DeserializeAs<'de, (K, V)> for KvProxy<KAs, VAs>
where
    KAs: DeserializeAs<'de, K>,
    VAs: DeserializeAs<'de, V>,
{
    fn deserialize_as<D>(deserializer: D) -> Result<(K, V), D::Error>
    where
        D: Deserializer<'de>,
    {
        let proxy = KvProxy::<DeserializeAsWrap<K, KAs>, DeserializeAsWrap<V, VAs>>::deserialize(
            deserializer,
        )?;
        Ok((proxy.key.into_inner(), proxy.value.into_inner()))
    }
}

//...

struct DataFile {
// other fields
    #[serde(default)]
    #[serde_as(as = "DefaultOnNull<Seq<KvProxy<_,_>>>")]
    column_sizes: BTreeMap<i32, i64>,
}

[jonasbb]

This still requires changes to Seq for it to work. I said as much in my last message

With them, this can work (It doesn't currently, because Seq is not flexible enough).

[ClSlaid]

This still requires changes to Seq for it to work. I said as much in my last message

With them, this can work (It doesn't currently, because Seq is not flexible enough).

Sorry, my mistake.

[jonasbb]

This doesn't actually work. You get this error message for this code snippet. Unfortunately, it wasn't caught with the tests.

#[serde_as(as = "Seq<(_, Vec<_>)>")]
BTreeMap<usize, Vec<usize>>,

error[E0277]: the trait bound `serde_with::Seq<(serde_with::Same, std::vec::Vec<serde_with::Same>)>: serde_with::SerializeAs<_>` is not satisfied

The problem is that during serialization you never get the key or value directly, you only get references to them. That is ok for the SerializeAs implementation of Seq where the two types are used separately. But you cannot make a SerializeAs which works on a (K, V). At best you can do a (&K, &V) but that doesn't really work well with SerializeAs. The SerializeAs trait unfortunately cannot handle different amounts of references. The type and the type pattern always must match. For the snippet above: After the PR #566 it tried to match the Vec of the type with a &Vec in the type pattern (the thing after "as = "). So there is a mismatch in the number of references and it no longer compiles.

---

Ultimately, this probably means it cannot be implemented properly. As mentioned before, From/Into does not seem like the correct traits here either, since they would require cloning too.
This means if you want something like Seq which supports other types, you will need to create your own type. The trait implementations are not that complicated.