Consider adding support for #[serde(default = $expr)]

[erickt]

There is a bit of a dance that happens when deserializing a struct, and one of the fields is missing. serde_codegen effectively does:

• Was the field attributed with #[serde(default)]? If so, set the value Default::default(), assuming that the field type actually implements Default.
• Otherwise, call the Deserializer's missing_field() method. If the format has some built-in support for a value for a missing field, such as JSON considering a missing value as a null value, try to deserialize from that value, or return an error.

This unfortunately means that as a consumer, you may have to newtype yourself an implementation of Default. Instead, it might be better if we optionally allowed for #[serde(default = $expr)], where if the value is not present, we evaluate some expression. This would cut down on some of the boilerplate necessary for using serde_codegen.

Open question: If we do implement this, is this expression being evaluated in the same scope as Deserialize::deserialize? On one level it'd be potentially interesting that it'd be able to access one of the other values to implement itself, but I consider the serde_codegen generated bodies to not be a stable interface. So I'm leaning towards us putting the $expr into a sub-function to keep it from accessing the internal deserialization state.

cc #198, which has a similar idea for specifying alternative serializers/deserializers.

[erickt]

This was implemented in #238!