[John-Cusack]

pydantic's EmailStr calls validate_email() with check_deliverability=False for every value it validates (networks.py), so a bulk import runs it once per record.

For a domain with no xn-- label, the idna.decode() and idna.encode() round trip at the end of validate_email_domain_name() returns it unchanged, and the character checks after it cannot fail on a domain that has already matched the hostname pattern and passed the length checks. This returns early in that case. Domains with Punycode labels take the existing path.

On the script below, validation goes from about 54 µs to 30 µs per address.

The test suite passes on Python 3.10 and 3.14 (on 3.12.3, test_domain_literal fails on main too), and results and error messages match main on about 52,000 addresses.

Benchmark

Ryzen 9 5950X, CPython 3.12.3, idna 3.15. Best of 21 runs over 2,000 addresses.

import random
import timeit

from email_validator import EmailNotValidError, validate_email

rng = random.Random(0)


def make(i):
    r = rng.random()
    if r < 0.2:
        return f"user{i}@bad_domain{i}.com"  # rejected
    if r < 0.3:
        return f"user{i}@münchen-{i % 50}.de"  # internationalized domain
    return f"first.last+{i}@mail{i % 200}.example.com"


addrs = [make(i) for i in range(2000)]


def run():
    for a in addrs:
        try:
            validate_email(a, check_deliverability=False)
        except EmailNotValidError:
            pass


run()
best = min(timeit.repeat(run, number=1, repeat=21))
print(f"{best / len(addrs) * 1e6:.1f} us per address")

[JoshData]

But why?