Performance impact of Nested with many vs List of Nested.

[mspiller]

We noticed that Nested is slower than List of Nested.
Not sure why this happens but should they be the same?

import timeit

import marshmallow
from marshmallow import fields
from marshmallow import validate


class PropertySchema(marshmallow.Schema):
    name = fields.String(validate=[validate.Regexp('^[0-9a-zA-Z -]{1,100}$')], required=True)
    value = fields.String(validate=[validate.Regexp('^[ -~]{0,255}$')], required=True)


class TrackInteractionActionSchema(marshmallow.Schema):
    properties = fields.Nested(PropertySchema, many=True, required=False)

class TrackInteractionActionSchemaNew(marshmallow.Schema):
    properties = fields.List(fields.Nested(PropertySchema), required=False)


def load_schema(schema: marshmallow.Schema, data: dict) -> None:
    try:
        schema.load(data)
    except validate.ValidationError as err:
        pass


if __name__ == '__main__':
    data = {
        "properties": [
            {"name": "x!0", "value": "v"} for x in range(20000)
        ]
    }

    a = TrackInteractionActionSchema()
    print(timeit.timeit(lambda: load_schema(a, data), number=10))
    b = TrackInteractionActionSchemaNew()
    print(timeit.timeit(lambda: load_schema(b, data), number=10))

It is couple of times slower:

7.725316457916051
1.1829904590267688

[deckar01]

I am going to refer to the runtime as a/b%.

On an Apple M1 I am seeing ~72% in cpython 3.13.2 and ~104% in pypy 3.11.11. The nested many schema has ~66% as many deserialize calls, and function invocation is a signification source of overhead in cpython. PyPy matches what I would expect from reading the code, basically identical logic with a little bit of extra code in nested many.

You are reporting ~653%. Nested many might have more branching, because it has some extra features, but that is quite a performance penalty. @mspiller What python version and processor are you using?