Update docs for schema_id()

docs/2-implementing.md

@@ -10,22 +10,54 @@ permalink: /implementing/
 [Deriving `JsonSchema`]({{ site.baseurl }}{% link 1-deriving.md %}) is usually the easiest way to enable JSON schema generation for your types. But if you need more customisation, you can also implement `JsonSchema` manually. This trait has two associated functions which must be implemented, and one which can optionally be implemented:
 
 ## schema_name
+
 ```rust
 fn schema_name() -> String;
 ```
 
-This function returns the name of the type's schema, which frequently is just the name of the type itself. The schema name is used as the title for root schemas, and the key within the root's `definitions` property for subschemas.
+This function returns the human-readable friendly name of the type's schema, which frequently is just the name of the type itself. The schema name is used as the title for root schemas, and the key within the root's `definitions` property for subschemas.
+
+NB in a future version of schemars, it's likely that this function will be changed to return a `Cow<'static, str>`.
+
+## schema_id
 
-If two types return the same `schema_name`, then Schemars will consider them identical types. Because of this, if a type takes any generic type parameters, then its schema name should depend on the type arguments. For example, the imlementation of this function for `Vec<T> where T: JsonSchema` is:
 ```rust
-fn schema_name() -> String {
+fn schema_id() -> Cow<'static, str>;
-    format!("Array_of_{}", T::schema_name())
+```
+
+This function returns a unique identifier of the type's schema - if two types return the same `schema_id`, then Schemars will consider them identical types. Because of this, if a type takes any generic type parameters, then its ID should depend on the type arguments. For example, the implementation of this function for `Vec<T> where T: JsonSchema` is:
+
+```rust
+fn schema_id() -> Cow<'static, str> {
+    Cow::Owned(
+        format!("[{}]", T::schema_id()))
 }
 ```
 
-`BTreeSet<T>`, `LinkedList<T>`, and similar collection types also use that implementation, since they produce identical JSON schemas so they can be considered the same type.
+`&mut Vec<&T>`, `LinkedList<T>`, `Mutex<LinkedList<Arc<T>>>`, and similar collection types also use that implementation, since they produce identical JSON schemas so they can be considered the same type.
+
+For a type with no generic type arguments, a reasonable implementation of this function would be to return the type name including module path (in case there is a type with the same name in another module/crate), e.g.:
+
+```rust
+impl JsonSchema for NonGenericType {
+    fn schema_name() -> String {
+        // Exclude the module path to make the name in generated schemas clearer.
+        "NonGenericType".to_owned()
+    }
+
+    fn schema_id() -> Cow<'static, str> {
+        // Include the module, in case a type with the same name is in another module/crate
+        Cow::Borrowed(concat!(module_path!(), "::NonGenericType"))
+    }
+
+    fn json_schema(_gen: &mut SchemaGenerator) -> Schema {
+        todo!()
+    }
+}
+```
 
 ## json_schema
+
 ```rust
 fn json_schema(gen: &mut gen::SchemaGenerator) -> Schema;
 ```
@@ -35,6 +67,7 @@ This function creates the JSON schema itself. The `gen` argument can be used to
 `json_schema` should not return a `$ref` schema.
 
 ## is_referenceable (optional)
+
 ```rust
 fn is_referenceable() -> bool;
 ```