take

@jorisvandenbossche - I'm having a hard time trying to replicate np.array(ser, copy=False) to return an error using pandas latest release (2.2) or 2.3.x branch with Numpy v2.0 or Numpy>v2.0. I'm trying to replicate it to use it for debugging, Thanks

Try this example with latest main:

In [1]: ser = pd.Series(["a", "b"], dtype="category")

In [2]: np.array(ser, copy=False)
...
ValueError: Unable to avoid copy while creating an array as requested.

You need to use a dtype that cannot be converted zero-copy to numpy, such as category I used above (if you would use integers, for example, that will not error).

And also you need latest main (or 2.3.x), this is not yet included in a released version.

Can't seem to replicate on my end, on both main and 2.3.x. Running:

import numpy as np
import pandas as pd

print(f"NumPy version: {np.__version__}")
print(f"Pandas version: {pd.__version__}")

ser = pd.Series(["a", "b"], dtype="category")
x = np.array(ser, copy=False)
print(x)

Output:

(pandas-dev) kev@mac pandas % python test.py
+ /Users/kev/.pyenv/versions/3.10.14/bin/ninja
[1/1] Generating write_version_file with a custom command
NumPy version: 1.26.4
Pandas version: 3.0.0.dev0+1580.g68d9dcab5b
['a' 'b']

I'm running on macOS 15.1.1

Ah, you need numpy > 2.0

Thanks, I'm now able to replicate it on my end. Working on the PR 🔧

Great!

@jorisvandenbossche - Does this warning message look good to you?

Numpy>=2.0 changed copy keyword's behavior, making copy=False raise an error when a zero-copy numpy array is not possible

I would still add something like "pandas will follow that behaviour starting with pandas 3.0" and "this conversion to numpy requires a copy, but 'copy=False' was passed. This will start raise an error in the future. Use np.asarray(..) instead."