Slots

__slots__ is a Python class attribute that fundamentally changes how Python stores instance attributes.

Normal Python Classes (without __slots__)

By default, Python stores instance attributes in a dictionary called __dict__ for each object. This means: - You can add any attribute to an instance at runtime - Each instance has its own __dict__ dictionary - Memory overhead includes the dictionary structure

With __slots__

When you define __slots__, Python: 1. Eliminates the __dict__ - No dictionary is created for each instance 2. Pre-allocates fixed slots - Creates a fixed number of memory slots for the specified attributes 3. Restricts attribute assignment - You can only set attributes listed in __slots__

Concrete Example from the Codebase

Looking at AgentSpanData, it defines __slots__ = ("name", "handoffs", "tools", "output_type"). This means:

• Each AgentSpanData instance can only have these 4 attributes
• No __dict__ is created
• Memory is pre-allocated for exactly these 4 slots
• You cannot do instance.some_random_attribute = value - it will raise an AttributeError

Memory Impact

For classes like FunctionSpanData with __slots__ = ("name", "input", "output", "mcp_data"), each instance uses significantly less memory because it avoids the overhead of a dictionary. In a tracing system that might create thousands of span objects, this memory savings adds up quickly.

Notes

The trade-off is flexibility vs efficiency - you lose the ability to dynamically add attributes but gain substantial memory and performance benefits. This is why it's used extensively in the span data classes where the structure is well-defined and performance matters.