compatibility.md

Python Compatibility Guide

This document outlines compatibility considerations and solutions for supporting different Python versions in the tensorcontainer project.

Supported Python Versions

tensorcontainer requires Python 3.9 or higher.

Compatibility Changes

Union Types and isinstance() Checks

Union types can now be handled using typing.get_args():

from typing import Union, get_args
import torch
from tensorcontainer import TensorContainer

TDCompatible = Union[torch.Tensor, TensorContainer]

# Use get_args to extract types from Union:
if isinstance(val, get_args(TDCompatible)):
    pass

Type Annotations

Note: The | operator for union types was introduced in Python 3.10:

# Python 3.10+ only:
def func(x: int | str) -> None:
    pass

For broader compatibility with Python 3.9+, use Union from typing:

from typing import Union

# Compatible with Python 3.9+:
def func(x: Union[int, str]) -> None:
    pass

General Compatibility Tips

Import Compatibility

Use from __future__ import annotations at the top of files to enable forward references and improve compatibility:

from __future__ import annotations

# This allows using string annotations that are evaluated later
def func(x: 'SomeClass') -> 'SomeClass':
    pass

Version-Specific Features

When using features that are only available in newer Python versions, use version checks:

import sys

if sys.version_info >= (3, 10):
    # Use Python 3.10+ features
    pass
else:
    # Fallback for Python 3.9
    pass

Version-specific features should be avoided whenever possible. Instead, use one of the above solutions.