[Bug]: Nondeterministic Hypothesis data generation

[jacksondm33]

Bug Explanation

Some test data generated via Ivy's helper strategies is nondeterministic (between runs) when using the derandomize option with Hypothesis. Specifically, this appears to be caused by the conversion of sets of string values to lists/tuples, which results in a nondeterministic ordering. This behavior is ultimately caused by Python's hashing randomization:

By default, the hash() values of str and bytes objects are “salted” with an unpredictable random value. Although they remain constant within an individual Python process, they are not predictable between repeated invocations of Python.

This most notably occurs in the get_dtypes helper function, which causes most Ivy and frontend API tests to be nondeterministic. A straightforward fix would be to always sort any list/tuple converted from a set that is used in a composite strategy. Otherwise, deterministic tests would need to be run with the PYTHONHASHSEED environment variable set.

Steps to Reproduce Bug

1. Run a (generally) failing Ivy test using the Hypothesis derandomize option
2. Run the same test again, and it will likely fail with a different example

Environment

Arch Linux using Emacs and venv

Ivy Version

commit 0437358

Backend

• NumPy
• TensorFlow
• PyTorch
• JAX

Device

No response