np.array copy=False vs copy=None: portable "copy only if needed" across NumPy 1.x and 2.x

The Portable Solution

To achieve "copy only if needed" behavior that is portable across both NumPy 1.x and 2.x, use np.asarray(). This function is the stable, cross-version standard for ensuring an input is an ndarray while avoiding a copy if the input is already an array of the requested dtype.

Behavioral Comparison

The copy parameter in np.array() underwent a significant semantic shift in NumPy 2.0. The following table outlines the differences in how these calls handle inputs that require a copy (e.g., a Python list or a dtype mismatch).

Call	NumPy 1.x Behavior	NumPy 2.x Behavior
np.asarray(x)	Copies only if needed	Copies only if needed
np.array(x, copy=False)	Copies only if needed	Raises ValueError if copy is required
np.array(x, copy=None)	TypeError (invalid arg)	Copies only if needed

Why np.asarray is the correct choice

• Consistency: np.asarray has maintained the "copy only if needed" contract across both major versions.
• Safety: Unlike np.array(copy=False) in 2.x, it will not crash when encountering a Python list or a different dtype; it will simply perform the necessary copy to satisfy the array requirement.
• Avoids Branching: Using np.asarray removes the need to check numpy.__version__, reducing maintenance overhead during the transition.

Verification Steps

You can verify this behavior in your environment with the following scoped tests:

import numpy as np

# Case 1: Already an array (Should NOT copy)
base = np.array([1, 2, 3])
result = np.asarray(base)
print(f"No copy: {np.may_share_memory(base, result)}")

# Case 2: Python list (MUST copy)
result_list = np.asarray([1, 2, 3])
# This succeeds in both 1.x and 2.x, whereas np.array(copy=False) fails in 2.x

Note on Dtype: If you specify a dtype in np.asarray(x, dtype=...), a copy will be forced if the input array's dtype does not match, regardless of the NumPy version.