[Bug] importing "Litestar" somehow makes "Faker" Not Random anymore

[saltcable]

• Faker version: 19.1.0
• Litestar version: 2.0.0b2
• OS: Python 3.10.5 (tags/v3.10.5:f377153, Jun 6 2022, 16:14:13) [MSC v.1929 64 bit (AMD64)] on win32

Faker is working fine but when Litestar is imported, it's not random anymore. Same result/output every time.

Steps to reproduce

1. code (lab.py)

import faker
import litestar  # <<=== the problem

fake = faker.Faker()
print(fake.md5())

2. run few time

python lab.py
python lab.py
python lab.py

Expected behavior (by removing import litestar)

Random output all the time
7442b46046a7c2d7ca080483a0d45ad4
a87b76986a9afbb9da63dde9ed4d2495
6924bc922dd1db07440ec3298c42afb8

Actual behavior (same environment)

Same every time. for all the functions
9fc4f6348d3311fa3376be5aa395877a
9fc4f6348d3311fa3376be5aa395877a
9fc4f6348d3311fa3376be5aa395877a

[saltcable]

joke2k/faker#1889

[Goldziher]

Litestar uses polyfactory to generate openapi examples. It seeds the random seed for random to ensure that the examples are deterministic.

You can set a different random seed on faker after importing litestar to override this.

Why does this happen? The reason is that faker no longer supports seeding a per faker random seed but rather requiires a global random seed.

[saltcable]

Already closed the issue! I did some tests with other libraries and found that they work well with Faker. Here's an example code snippet using different libraries:

import faker

import starlette
import fastapi
import flask
import sanic
import quart

fake = faker.Faker()

print(fake.md5())

The outputs :

d71a751d94055529478e28f567c50d9c
5da22a60148536ebf46774c249a5f15a
ba1c26b24bb94047a6d18239c73429d8

It seems that other libraries and Faker handle the seeds correctly. However, Litestar is somehow "messing" with the seeds.

The Faker maintainer responded with the following comment:

Given the litestar maintainer comment in the linked issue, what would be your desired solution for faker? I do not see faker ensuring compatibility with each third-party package which might "mess" with the seeding automatically.

Quick question: I attempted to disable the openapi feature by using openapi_config=None, but it didn't have the desired effect. Is there any way to disable Litestar's seed modification, even if I don't intend to use the openapi functionality?

[Goldziher]

Already closed the issue! I did some tests with other libraries and found that they work well with Faker. Here's an example code snippet using different libraries:

Well this is a pretty useless exercise. I already explained what you're seeing. The issue is closed because this is not a bug. I can convert it to a discussion instead.

import faker

import starlette
import fastapi
import flask
import sanic
import quart

fake = faker.Faker()

print(fake.md5())

The outputs :

d71a751d94055529478e28f567c50d9c
5da22a60148536ebf46774c249a5f15a
ba1c26b24bb94047a6d18239c73429d8

It seems that other libraries and Faker handle the seeds correctly. However, Litestar is somehow "messing" with the seeds.

The Faker maintainer responded with the following comment:

Given the litestar maintainer comment in the linked issue, what would be your desired solution for faker? I do not see faker ensuring compatibility with each third-party package which might "mess" with the seeding automatically.

Quick question: I attempted to disable the openapi feature by using openapi_config=None, but it didn't have the desired effect. Is there any way to disable Litestar's seed modification, even if I don't intend to use the openapi functionality?

Not at present. We could consider adding this but it's a very marginal use case, since you can simply set a seed on your own and even abstract this stuff into a fixture.

I need to understand why exactly this is a big issue for you?

As to the other question - if faker supports seed random per faker instance it will prevent this issue. The problem is the use of a global namespace.