[Compatibility Issue] np.trapz removed in NumPy 2.4.0 — use np.trapezoid instead

[xyf5432]

Affected Files

• ivy/functional/backends/numpy/elementwise.py:749
• ivy/functional/backends/numpy/experimental/statistical.py:567

Current Code

# ivy/functional/backends/numpy/elementwise.py:749
def trapz(y, /, *, x=None, dx=1.0, axis=-1, out=None):
    return np.trapz(y, x=x, dx=dx, axis=axis)

# ivy/functional/backends/numpy/experimental/statistical.py:567
integral = np.trapz(y, t)

Root Cause

numpy.trapz was deprecated in NumPy 2.0 (2023-08-18) and fully removed in NumPy 2.4.0 (2025-12-20). In NumPy >= 2.4.0, any call to np.trapz() raises:

AttributeError: module 'numpy' has no attribute 'trapz'

Ivy defines its own trapz API across multiple backends (torch, jax, tensorflow, numpy). The numpy backend directly delegates to np.trapz, which will break.

Impact

• Severity: High — hard crash at runtime on NumPy >= 2.4.0
• The numpy backend trapz() is called via ivy.trapz() → backend dispatch → np.trapz()
• Also affects the numpy frontend: ivy.functional.frontends.numpy.trapz() → ivy.trapz() → numpy backend → np.trapz()
• The experimental igamma_cal function in statistical.py is also affected

Solution

# ivy/functional/backends/numpy/elementwise.py:749
# Before:
return np.trapz(y, x=x, dx=dx, axis=axis)
# After:
return np.trapezoid(y, x=x, dx=dx, axis=axis)

# ivy/functional/backends/numpy/experimental/statistical.py:567
# Before:
integral = np.trapz(y, t)
# After:
integral = np.trapezoid(y, t)

trapezoid has the same signature: trapezoid(y, x=None, dx=1.0, axis=-1).

For NumPy < 2.0 backward compatibility:

try:
    from numpy import trapezoid
except ImportError:
    from numpy import trapz as trapezoid

References

• NumPy 2.4.0 release notes — trapz removed
• NumPy 2.0 deprecation: https://numpy.org/doc/stable/reference/generated/numpy.trapz.html
• Similar fixes in other downstream projects: gpytorch#2695, NeuroKit#1155