Re: [PATCH proxmox 01/20] router: introduce shared state

["Lukas Wagner" <l.wagner@proxmox.com>]

On Fri Aug 21, 2026 at 3:59 PM CEST, Robert Obkircher wrote:
>> +use anyhow::{Error, bail};
>> +
>> +/// Registry of state values, keyed by their type.
>> +///
>> +/// It holds at most one value per type. Values are registered with
>> +/// [`register`](SharedStateRegistry::register) before the registry is handed over to the API
>> +/// environment, from where handlers can access them through
>> +/// [`RpcEnvironment::shared_state`](crate::RpcEnvironment::shared_state). This allows passing
>> +/// application context to handlers without resorting to globals.
>> +#[derive(Default)]
>> +pub struct SharedStateRegistry {
>> +    map: HashMap<TypeId, Box<dyn Any + Send + Sync>>,
>> +}
>
> We could get rid of some indirection by replacing the map with a
> Box<&dyn SharedState>. Probably not worth it, though.

Yeah, the map lookups happen once per API request, so I don't think the
indirection should be an issue here -- performance-wise at least.

The current approach has in my opinion the advantage that it keeps the
type-id magic contained within proxmox-router. With the trait-based
approach, if I understand it correctly, the application code would need
to perform the cast from dyn Any. Now, given that this is handled behind
the scenes by the API macro, it should not matter that much, probably.

>
> trait SharedState { fn lookup(&self, id: TypeId) -> &dyn Any; }
>
> struct PdmState { a: A, b: B }
>
> impl SharedState for PdmState {
>     fn lookup(&self, id: TypeId) -> &dyn Any {
>         if id == TypeId::of::<A>() { &self.a }
>         else if id == TypeId::of::<B>() { &self.b }
>         else { unreachable!() }
>     }
> }
>
>
>> +
>> +impl SharedStateRegistry {
>> +    /// Get a clone of the registered value of type `T`, if there is one.
>> +    pub fn lookup<T: 'static + Send + Sync + Clone>(&self) -> Option<T> {
>> +        self.map
>> +            .get(&TypeId::of::<T>())
>> +            .and_then(|s| s.downcast_ref())
>> +            .cloned()
> This could .expect("value must have correct type for key").


Makes sense, as a failed cast is clearly a bug here. Thanks!