Constraints: what `T` can do

Operations need permissions

Inside a generic method, T starts with no abilities — you can't call a.CompareTo(b) without promising T supports comparison. Constraints are that promise:

where T : IComparable<T>   // T knows how to compare to itself
where T : class            // T is a reference type
where T : struct           // T is a value type (non-nullable)
where T : new()            // T has a parameterless constructor
where T : SomeBase         // T is (or derives from) SomeBase

With where T : IComparable<T>, the compiler unlocks CompareTo on every T — and rejects type arguments that lack it. Constraints both restrict callers and enable the method body.

The honest rule

Add a constraint only when the body needs that capability. An empty where T : class "for safety" restricts callers for no reason. Start from the operations the algorithm needs; the constraint list writes itself.