Source: numpy Version: 1:2.3.4+ds-3 Severity: normal X-Debbugs-Cc: [email protected] User: [email protected] Usertags: powerpc Control: forwarded -1 https://github.com/numpy/numpy/issues/30257 Control: affects -1 src:fenics-dolfinx
Debian tests of dolfinx on powerpc have exposed an irregularity in ppc ctypes https://buildd.debian.org/status/fetch.php?pkg=fenics-dolfinx&arch=powerpc&ver=1%3A0.10.0.post2-1exp2&stamp=1763496611&raw=0 /usr/lib/python3/dist-packages/numpy/ctypeslib.py:382: in _ctype_from_dtype_scalar raise NotImplementedError( E NotImplementedError: Converting dtype('int32') to a ctypes type The failing line numpy/numpy/ctypeslib/_ctypeslib.py Line 385 in a251f2a ctype = _scalar_type_map[dtype_native] is simply trying to access a ctype for dtype int32. The problem can be reproduced manually. The strange thing is that inspecting the _scalar_type_map dict (on ppc), it clearly shows an entry dtype('int32'): <class 'ctypes.c_long'> import numpy as np import platform print("testing architecture", platform.machine()) a = np.array([1], dtype=np.int32) dtype_native = a.dtype.newbyteorder('=') print("'np.int32' is", dtype_native) print("ctypeslib _scalar_type_map is", np.ctypeslib._scalar_type_map) print(f"extract ctype for {dtype_native}:") ctype = np.ctypeslib._scalar_type_map[dtype_native] print("got ctype", ctype)
