How do I transform special values into Option::None when using Serde to deserialize?

I'm parsing data into:

struct Data {
    field1: Option<f32>,
    field2: Option<u64>,
    // more ...
}

The problem is that my input data format formats what would be a None in Rust as "n/a".

How do tell Serde that an Option<T> should be None for the specific string n/a, as opposed to an error? We can assume that this doesn't apply to a String.

This isn't the same question as How to deserialize "NaN" as `nan` with serde_json? because that's creating an f32 from a special value whereas my question is creating an Option<Anything> from a special value. It's also not How to transform fields during deserialization using Serde? as that still concerns a specific type.

Solution

You can write your own deserialization function that handles this case:

use serde::de::Deserializer;
use serde::Deserialize;

// custom deserializer function
fn deserialize_maybe_nan<'de, D, T: Deserialize<'de>>(
    deserializer: D,
) -> Result<Option<T>, D::Error>
where
    D: Deserializer<'de>,
{
    // we define a local enum type inside of the function
    // because it is untagged, serde will deserialize as the first variant
    // that it can
    #[derive(Deserialize)]
    #[serde(untagged)]
    enum MaybeNA<U> {
        // if it can be parsed as Option<T>, it will be
        Value(Option<U>),
        // otherwise try parsing as a string
        NAString(String),
    }

    // deserialize into local enum
    let value: MaybeNA<T> = Deserialize::deserialize(deserializer)?;
    match value {
        // if parsed as T or None, return that
        MaybeNA::Value(value) => Ok(value),

        // otherwise, if value is string an "n/a", return None
        // (and fail if it is any other string)
        MaybeNA::NAString(string) => {
            if string == "n/a" {
                Ok(None)
            } else {
                Err(serde::de::Error::custom("Unexpected string"))
            }
        }
    }
}

Then you can mark your fields with #[serde(default, deserialize_with = "deserialize_maybe_nan")] to use this function instead of the default function:

#[derive(Deserialize)]
struct Data {
    #[serde(default, deserialize_with = "deserialize_maybe_nan")]
    field1: Option<f32>,
    #[serde(default, deserialize_with = "deserialize_maybe_nan")]
    field2: Option<u64>,
    // more ...
}

Working playground example