Performance issue with version 0.14.5 compared to 0.13.4

[PetitLepton]

Hi, we are using pandera to validate on the fly data that we then serve through an API. Yesterday, we switched from version 0.13.4 to 0.14.5 which, unexpectedly, lead to a significant increase in the latency of our endpoints. After a little bit of digging, we figured out that there was a difference of behavior when using Series[str]. The following example, with pandas version 1.4.4,

import pandas
import pandera
import time
import numpy

class Model(pandera.SchemaModel):
    string: pandera.typing.Series[str]  # pandera.typing.Series[pandera.STRING]


@pandera.check_types
def identity(data: pandera.typing.DataFrame[Model]) -> pandera.typing.DataFrame[Model]:
    return data


if __name__ == "__main__":
    data = pandas.DataFrame(
        data={
            "string": 1_000_000 * ["a"],
        }
    )#.assign(string=lambda df: df["string"].astype("string[python]"))

    times = []

    for _ in range(10):

        start = time.time()
        _ = data.pipe(identity)
        end = time.time()
        times.append(end - start)

    print(numpy.array(times).mean(), numpy.array(times).std())

leads to an average computation time, numpy.array(times).mean(), four times larger in version 0.14.5 compared to 0.13.4 (on my local machine, x86_64).

If we convert the series into Python strings and use pandera.STRING, the performance of 0.13.4 and 0.14.5 are similar.

What would be your advice? Should we switch to Python strings? Is it an expected behaviour?

Thanks in advance for your help!