Proposed new feature or change:

When summing large integer arrays,np.sum can silently overflow depending on the dtype and system architecture.
For example, summing anint32 array on a 32-bit system can produce incorrect results without any warning.

Reproduction

importnumpyasnparray=np.arange(10**8+1,dtype=np.int32)np.sum(array)# May overflow depending on the environment

On 64-bit systems (or when usingint64), this gives the correct result, but on 32-bit systems or withint32 arrays, it silently wraps around due to overflow.

Proposed behavior

It would be helpful if np.sum provided one of the following features:

• Automatic dtype promotion — promote smaller integer types (e.g.int32 →int64) to avoid overflow, or
• Optional warning parameter — e.g.warn_on_overflow=True, which would raise or log a warning when overflow is detected.

This would makenp.sum more consistent with Python’s built-insum, which automatically handles large integers, and could reduce confusion when switching between platforms.

Motivation

While this behavior is consistent with NumPy’s current design, it might be unintuitive for users who expect consistent numerical results across different architectures.
Providing a warning or an optional type promotion could make this behavior easier to understand and improve usability.