How to use Arc<dyn PhysicalExpr> inside HashMap as key, or inside HashSet.

[mustafasrepo]

I think some of the mappings during projection or in other parts of the codebase, can be simplified. If we were to use Arc<dyn PhysicalExpr> inside a HashMap (e.g HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>>). However, I gave it a try, and couldn't do so.

Is there a way to do so? I have tried to use a wrapper struct inside HashMap such as HashMap<ExprWrapper, ExprWrapper>
where ExprWrapper is

struct ExprWrapper{
    expr: Arc<dyn PhysicalExpr>,
}

impl PartialEq for ExprWrapper {
    fn eq(&self, other: &ExprWrapper) -> bool {
        self.expr.eq(&other.expr)
    }
}

impl Eq for ExprWrapper {}

impl Hash for ExprWrapper {
    fn hash<H: Hasher>(&self, state: &mut H) {
        self.expr.hash(state);
    }
}

This suggests me that, there should be a way to use directly Arc<dyn PhysicalExpr>> inside HashMap. I wonder if there is anyone who has done similar to this before, and How can I do so?

Thanks in advance.

[alamb]

What is the problem with using HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>>

We implemented Hash for PhysicalExpr for just the purpose I thought 🤔

https://docs.rs/datafusion/latest/datafusion/physical_expr/trait.PhysicalExpr.html#impl-Hash-for-dyn+PhysicalExpr

[mustafasrepo]

Actually we can create the hashMap for the type HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>>. Below code works correctly.

    // Ok, no error
    #[test]
    fn test_hash_map_creation() -> Result<()> {
        let col_a = &Column::new("a", 0);
        let col_b = &Column::new("b", 1);
        let expr_a = Arc::new(col_a.clone()) as Arc<dyn PhysicalExpr>;
        let expr_b = Arc::new(col_b.clone()) as Arc<dyn PhysicalExpr>;
        let hash_map: HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>> = HashMap::new();
        Ok(())
    }

However, when we want to insert something to the hashmap it gives an error. Code below produces error the method insert exists for struct HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>>, but its trait bounds were not satisfied.

    // Gives the error "the method `insert` exists for struct `HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>>`, but its trait bounds were not satisfied";
    #[test]
    fn test_hash_map_insertion() -> Result<()> {
        let col_a = &Column::new("a", 0);
        let col_b = &Column::new("b", 1);
        let expr_a = Arc::new(col_a.clone()) as Arc<dyn PhysicalExpr>;
        let expr_b = Arc::new(col_b.clone()) as Arc<dyn PhysicalExpr>;
        let mut hash_map: HashMap<Arc<dyn PhysicalExpr>, Arc<dyn PhysicalExpr>> = HashMap::new();
        hash_map.insert(expr_a, expr_b);
        Ok(())
    }

[mustafasrepo]

It seems that when we use a type with unknown size inside Arc. Eq property is not satisfied. This prevents us to use this type as key in the HashMap. However, I am surprised this error doesn't arise during initialization