[cppyy] cast to long long pointer result in unsigned long long pointer

[AlkaidCheng]

Check duplicate issues.

• Checked for duplicates

Description

Casting an int64 numpy array into a signed long long pointer gives an unsigned long long pointer instead.

Reproducer

import cppyy
import cppyy.ll
import numpy as np

arr = np.array([-1, 0, 1], dtype=np.int64)
view = cppyy.ll.cast[f"long long*"](arr.ctypes.data)
# prints 18446744073709551615 instead of -1
print(view[0])
# prints 'L' instead of 'l'
print(view.typecode)

ROOT version

cppyy==3.5.0

Installation method

conda (bundled with root)

Operating system

Linux

Additional context

No response

[siliataider]

Hi @AlkaidCheng thanks for your report!

I cannot reproduce this neither with ROOT 6.38.02 bundled cppyy (Fedora, Linux x86_64) nor standalone cppyy==3.5.0 via conda.

Both give typecode q and value -1 correctly.

Could you provide more details about your environment:

• Linux distribution and version
• Architecture
• Python version
• Is this reproducible in a fresh conda environment with only cppyy installed?

[AlkaidCheng]

Hi @siliataider , it seems to be an OS dependent issue. The one that failed is with SUSE Linux. Specifically:

Operating System: SLES 15.6
    Architecture: x86_64

This is the OS used by perlmutter.

I also tested on Ubuntu (Ubuntu 22.04), which gives the same result as you stated.

[guitargeek]

Thanks! What are the versions of Python, ROOT, and NumPy?

[AlkaidCheng]

Hi @guitargeek , the relevant software versions are:

ROOT version: 6.38.00
cppyy version: 3.5.0
numpy version: 1.26.4

[guitargeek]

Thanks! The Python version is also relevant, please share that too 🙂

NumPy 1 to 2 was a big change, and we don't actively test and support NumPy 1 anymore. Probably that's it! Can you not install NumPy 2? NumPy 2 was released in June 2024, so by now it should be easy to get it.

[AlkaidCheng]

I think I had numpy 2.x before but recently installed some specific tensorflow version which forces it to downgrade. But anyway, the bug still remain when I upgrade numpy version:

python version: 3.11.14 | packaged by conda-forge | (main, Jan 26 2026, 23:48:32) [GCC 14.3.0]
ROOT version: 6.38.00
cppyy version: 3.5.0
numpy version: 2.4.2

[guitargeek]

How did you determine the cppyy version, did you install cppyy and ROOT at the same time? ROOT also ships cppyy internally, so if you also install the standalone cppyy package in the same environment things will go wrong.

[AlkaidCheng]

I use cppyy.__version__ and it is shipped with ROOT via conda install.

[AlkaidCheng]

It also feels strange to me that the type code for unsigned long long is L in SUSE Linux (not 100% sure it is because of OS at the point), and Q in Ubuntu Linux.

[siliataider]

Hi!

I tested with ROOT 6.38.02 on openSUSE Leap 15.6 (installed via conda) and cannot reproduce the issue, I get -1 and typecode q as expected.

There were fixes in 6.38.02 that could be relevant here, could you try upgrading to 6.38.02 (which is now available on conda) and check if the issue is still present?

Note that we're testing and adding an openSUSE Leap 15.6 image to our CI, and the reproducer passes correctly there.

[AlkaidCheng]

Thanks for the suggestion @siliataider ! Indeed, with ROOT 6.38.02, the problem seems to be fixed. I will close the issue then.