Add test for reproducing panics on resize

See #380

Author: bikeshedder

tests/managed_resize.rs

@@ -0,0 +1,51 @@
+#![cfg(feature = "managed")]
+
+use std::convert::Infallible;
+
+use deadpool::managed::{self, Metrics, Object, RecycleResult};
+
+type Pool = managed::Pool<Manager, Object<Manager>>;
+
+struct Manager {}
+
+impl managed::Manager for Manager {
+    type Type = ();
+    type Error = Infallible;
+
+    async fn create(&self) -> Result<(), Infallible> {
+        Ok(())
+    }
+
+    async fn recycle(&self, _conn: &mut (), _: &Metrics) -> RecycleResult<Infallible> {
+        Ok(())
+    }
+}
+
+#[tokio::test]
+async fn test_grow_reuse_existing() {
+    // Shrink doesn't discard objects currently borrowed from the pool but
+    // keeps track of them so that repeatedly growing and shrinking will
+    // not cause excessive object creation. This logic used to contain a bug
+    // causing an overflow.
+    let mgr = Manager {};
+    let pool = Pool::builder(mgr).max_size(2).build().unwrap();
+    let obj1 = pool.get().await.unwrap();
+    let obj2 = pool.get().await.unwrap();
+    assert!(pool.status().size == 2);
+    assert!(pool.status().max_size == 2);
+    pool.resize(0);
+    // At this point the two objects are still tracked
+    assert!(pool.status().size == 2);
+    assert!(pool.status().max_size == 0);
+    pool.resize(1);
+    // Only one of the objects should be returned to the pool
+    assert!(pool.status().size == 2);
+    assert!(pool.status().max_size == 1);
+    drop(obj1);
+    // The first drop brings the size to 1.
+    assert!(pool.status().size == 1);
+    assert!(pool.status().max_size == 1);
+    drop(obj2);
+    assert!(pool.status().size == 1);
+    assert!(pool.status().max_size == 1);
+}