Multiprocessing returns Array, RawArray of wrong length python 3.10.4

[tendermonster]

Bug report

When initializing Array or RawArray of type int, wrong length of the arrays is returned

Reproducible example:

from multiprocessing import RawArray, RawValue, Array, Value
a = Array('d',12)
X = np.frombuffer(a.get_obj())
print(X) # [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]

a = Array('i',12)
X = np.frombuffer(a.get_obj())
print(X) # [0. 0. 0. 0. 0. 0.]

a = RawArray('d',12)
X = np.frombuffer(a)
print(X) # [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]

a = RawArray('i',12)
X = np.frombuffer(a)
print(X) # [0. 0. 0. 0. 0. 0.]

Your environment

working with vscode
conda-forge python 3.10.4
using ubuntu uname-a: 5.11.0-46-generic #51~20.04.1-Ubuntu SMP Fri Jan 7 06:51:40 UTC 2022

• CPython versions tested on: conda-forge python 3.10.4
• Operating system and architecture: using ubuntu uname-a: 5.11.0-46-generic 51~20.04.1-Ubuntu

[mdboom]

The buffer interface doesn't retain type information about the items in the array, so using np.frombuffer defaults to double (8 bytes per item), even when the type in the array is i (4 bytes per item) since converting to a buffer throws that information out.

I think you want to use np.asarray(a) instead to get the behaviour you want -- it will correctly look at the type of the array (using the memoryview API), and it likewise avoids the memory copy.

[tendermonster]

Thanks for quick response