Reputation: 14097
I'm experimenting with Rust and am trying to write a wrapper for Elasticsearch queries in Rust. There's a query which I have implemented and it works correctly, however I truly dislike the way I did it with json!
macro.
fn main() {
let actual = Query {
field: "field_name".into(),
values: vec![1, 2, 3],
boost: Some(2),
name: Some("query_name".into()),
};
let expected = serde_json::json!({
"terms": {
"field_name": [1, 2, 3],
"boost": 2,
"_name": "query_name"
}
});
let actual_str = serde_json::to_string(&actual).unwrap();
let expected_str = serde_json::to_string(&expected).unwrap();
assert_eq!(actual_str, expected_str);
}
#[derive(Debug)]
struct Query {
field: String,
values: Vec<i32>,
boost: Option<i32>,
name: Option<String>,
}
impl serde::Serialize for Query {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
let field = self.field.as_str();
let values = self.values.as_slice();
let value = match (&self.boost, &self.name) {
(None, None) => serde_json::json!({ field: values }),
(None, Some(name)) => serde_json::json!({ field: values, "_name": name }),
(Some(boost), None) => serde_json::json!({ field: values, "boost": boost }),
(Some(boost), Some(name)) => {
serde_json::json!({ field: values, "boost": boost, "_name": name })
}
};
serde_json::json!({ "terms": value }).serialize(serializer)
}
}
I would like to know how I could implement such a serializer using serde's built in traits, such as SerializeStruct
, SerializeMap
, etc. Basically I'd like to avoid using json macro or creating intermediate data structures.
Upvotes: 2
Views: 2859
Reputation: 58805
If you make your data structures match the JSON a bit more, then you could do the entire serialization with a couple of #[serde]
annotations, which is far more maintainable, and unlikely to be slower than a manual implementation. (Edit: the dynamically named field actually makes that tricky, so a custom implementation of Serialize
might actually be best in this case).
Serde serialization works by delegating to other implementations of Serialize
, so it's difficult to create a custom nested map from a flat structure in a single implementation.
You can create a tempory structure for the nesting like this:
// Need to import SerializeMap to call its methods
use serde::{Serialize, ser::SerializeMap};
// temporary wrapper for serializing Query as the inner object
struct InnerQuery<'a>(&'a Query);
impl<'a> serde::Serialize for InnerQuery<'a> {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
let query = self.0;
// We don't know the length of the map at this point, so it's None
let mut map = serializer.serialize_map(None)?;
map.serialize_entry(&query.field, &query.values)?;
if let Some(boost) = &query.boost {
map.serialize_entry("boost", boost)?;
}
if let Some(name) = &query.name {
map.serialize_entry("_name", name)?;
}
map.end()
}
}
This will give you the inner part:
{
"field_name": [1, 2, 3],
"boost": 2,
"_name": "query_name"
}
Then implement the serialization for the main struct something like:
impl serde::Serialize for Query {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
let mut map = serializer.serialize_map(Some(1))?;
map.serialize_entry("terms", &InnerQuery(&self))?;
map.end()
}
}
Note the the order of the keys is not deterministic so your test will need to be changed. One way is to parse it back to a serde_json::Value
which will compare for equality as you expect:
let actual_str = serde_json::to_string(&actual).unwrap();
let actual: serde_json::Value = serde_json::from_str(&actual_str).unwrap();
assert_eq!(actual, expected);
Upvotes: 3
Reputation: 26757
If you want to avoid as much as possible manual implementation, I think the following solution is not bad. Your requirement is quite odd so I would advice to just avoid this kind of requirement.
My solution is reusable for other structure, CustomName
can be use in a lot of context, you just need to flat it using serde attribute.
The following also only compare json Value
, this is better cause json is unordered.
use serde::ser::SerializeMap; // 1.0.120
struct Query {
field: String,
values: Vec<i32>,
boost: Option<i32>,
name: Option<String>,
}
fn main() {
let actual = Query {
field: "field_name".into(),
values: vec![1, 2, 3],
boost: Some(2),
name: Some("query_name".into()),
};
let expected = serde_json::json!({
"terms": {
"field_name": [1, 2, 3],
"boost": 2,
"_name": "query_name"
}
});
let result = serde_json::to_value(&actual).unwrap();
assert_eq!(result, expected);
}
// the following concern serrialize implemenation
struct CustomName<'a, 'b> {
field: &'a str,
values: &'b [i32],
}
#[derive(serde::Serialize)]
struct SerializeQuery<'a, 'b, 'c, 'd> {
boost: Option<&'a i32>,
#[serde(rename = "_name")]
name: Option<&'b str>,
#[serde(flatten)]
custom_name: CustomName<'c, 'd>,
}
#[derive(serde::Serialize)]
struct Terms<'a, 'b, 'c, 'd> {
terms: SerializeQuery<'a, 'b, 'c, 'd>,
}
impl<'a> From<&'a Query> for Terms<'a, 'a, 'a, 'a> {
fn from(query: &'a Query) -> Self {
Self {
terms: SerializeQuery {
boost: query.boost.as_ref(),
name: query.name.as_deref(),
custom_name: CustomName {
field: &query.field,
values: &query.values,
},
},
}
}
}
impl serde::Serialize for Query {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
Terms::from(self).serialize(serializer)
}
}
impl<'a, 'b> serde::Serialize for CustomName<'a, 'b> {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
let mut map = serializer.serialize_map(Some(1))?;
map.serialize_entry(self.field, self.values)?;
map.end()
}
}
As you can see, we don't do much in the serialize implemtation, I don't think we can get rid of using serialize_map
cause field name is dynamic.
Upvotes: 1