SystemError setting ID custom property to a list containing NumPy scalars #117656

New Issue

Thomas Barlow · 2024-01-30T03:23:25+01:00

Thomas Barlow commented

2024-01-30 03:23:25 +01:00

System Information
Operating system: Windows-10-10.0.19045-SP0 64 Bits
Graphics card: NVIDIA GeForce GTX 1070 Ti/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 546.33

Blender Version
Broken: version: 4.1.0 Alpha, branch: main, commit date: 2024-01-29 00:53, hash: dab651537471
Worked: (newest version of Blender that worked as expected)

Short description of error
Trying to set an id prop array to a list containing numpy scalars (or another sequence that is not considered a compatible buffer) raises a SystemError when run from the Python Console (or a delayed TypeError when run from the Text Editor and another line checks for an exception being set afterward):

Error: Python: TypeError: object of type 'numpy.int32' has no len()

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "D:\Downloads\blender-4.1.0-alpha+main.dab651537471-windows.amd64-release\4.1\scripts\modules\console_python.py", line 149, in execute
    with (
SystemError: <function _TempModuleOverride.__exit__ at 0x000001DC9D302D40> returned a result with an exception set

On a debug build, an assertion instead fails in Python itself:

Assertion failed: !PyErr_Occurred(), file C:\db\build\S\VS1564D\build\python\src\external_python\Objects\typeobject.c, line 3823

Exact steps for others to reproduce the error
Run in the Python Console:

import numpy as np
bpy.context.scene['test_id_prop'] = list(np.array([0]))

or (because id prop arrays do not support 64-bit int and the array will not be considered a compatible buffer)

import numpy as np
bpy.context.scene['test_id_prop'] = np.array([0], dtype=np.int64)

**System Information** Operating system: Windows-10-10.0.19045-SP0 64 Bits Graphics card: NVIDIA GeForce GTX 1070 Ti/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 546.33 **Blender Version** Broken: version: 4.1.0 Alpha, branch: main, commit date: 2024-01-29 00:53, hash: `dab651537471` Worked: (newest version of Blender that worked as expected) **Short description of error** Trying to set an id prop array to a list containing numpy scalars (or another sequence that is not considered a compatible buffer) raises a `SystemError` when run from the Python Console (or a delayed TypeError when run from the Text Editor and another line checks for an exception being set afterward): ``` Error: Python: TypeError: object of type 'numpy.int32' has no len() The above exception was the direct cause of the following exception: Traceback (most recent call last): File "D:\Downloads\blender-4.1.0-alpha+main.dab651537471-windows.amd64-release\4.1\scripts\modules\console_python.py", line 149, in execute with ( SystemError: <function _TempModuleOverride.__exit__ at 0x000001DC9D302D40> returned a result with an exception set ``` On a debug build, an assertion instead fails in Python itself: ``` Assertion failed: !PyErr_Occurred(), file C:\db\build\S\VS1564D\build\python\src\external_python\Objects\typeobject.c, line 3823 ``` **Exact steps for others to reproduce the error** Run in the Python Console: ``` import numpy as np bpy.context.scene['test_id_prop'] = list(np.array([0])) ``` or (because id prop arrays do not support 64-bit int and the array will not be considered a compatible buffer) ``` import numpy as np bpy.context.scene['test_id_prop'] = np.array([0], dtype=np.int64) ```

Thomas Barlow added the

labels 2024-01-30 03:23:26 +01:00

Pratik Borhade commented

2024-01-30 07:56:02 +01:00

Thanks for the report. Can confirm. Seems to work with other dtypes (int_, float_, int32)

Pratik Borhade added

and removed

labels 2024-01-30 07:56:15 +01:00

Thomas Barlow referenced this issue

2024-04-26 15:19:51 +02:00

`TypeError: object of type 'numpy.int32' has no len()` assigning ID's custom attribute #121072

Andrej commented

2024-04-26 16:02:03 +02:00

A note that setting numpy array of numpy np.int32 does work without errors though (C.object['np_data'] = np.array([3,4,5])), error occurs only when np.int32 are in default python list.

A note that setting numpy array of numpy np.int32 does work without errors though (`C.object['np_data'] = np.array([3,4,5])`), error occurs only when np.int32 are in default python list.

Thomas Barlow commented

2024-04-26 17:23:54 +02:00

A note that setting numpy array of numpy np.int32 does work without errors though (C.object['np_data'] = np.array([3,4,5])), error occurs only when np.int32 are in default python list.

The np.int32 array works because it runs a different code path. Numpy arrays implement the Buffer Protocol and the code responsible for reading the elements in the array sees this and that the format (element type) of the buffer is compatible and then reads the elements of the array directly from the array's memory.

Python lists don't implement the Buffer Protocol, so the code has to iterate the list similar to a for loop written in Python and somewhere in that iteration the error occurs.

I think it may relate to either how numpy scalars can be accessed like read-only zero-dimensional arrays:

zero_dim_array = np.array(6, dtype=np.int32)
# 6
print(zero_dim_array[()])

scalar = np.int32(12)
# 12
print(scalar[()])

or how numpy scalars also implement the buffer protocol:

x = np.int32(12)
# Would error if `x` was not a buffer
mv = memoryview(x)

with the code then incorrectly assuming that having one of these properties means that __len__() should be implemented.

I'm guessing it maybe tries to read each numpy scalar as a sequence rather than as a long in e3894f0a07/source/blender/python/generic/idprop_py_api.cc (L678) since np.int32 is not a subclass of Python int so I think will fail PyLong_Check()

> A note that setting numpy array of numpy np.int32 does work without errors though (`C.object['np_data'] = np.array([3,4,5])`), error occurs only when np.int32 are in default python list. The np.int32 array works because it runs a different code path. Numpy arrays implement the Buffer Protocol and the code responsible for reading the elements in the array sees this and that the format (element type) of the buffer is compatible and then reads the elements of the array directly from the array's memory. Python lists don't implement the Buffer Protocol, so the code has to iterate the list similar to a for loop written in Python and somewhere in that iteration the error occurs. I think it may relate to either how numpy scalars can be accessed like read-only zero-dimensional arrays: ```py zero_dim_array = np.array(6, dtype=np.int32) # 6 print(zero_dim_array[()]) scalar = np.int32(12) # 12 print(scalar[()]) ``` or how numpy scalars also implement the buffer protocol: ```py x = np.int32(12) # Would error if `x` was not a buffer mv = memoryview(x) ``` with the code then incorrectly assuming that having one of these properties means that `__len__()` should be implemented. I'm guessing it maybe tries to read each numpy scalar as a sequence rather than as a long in https://projects.blender.org/blender/blender/src/commit/e3894f0a0708d1619dbfce3fb081f12bcee3f0c3/source/blender/python/generic/idprop_py_api.cc#L678 since `np.int32` is not a subclass of Python `int` so I think will fail `PyLong_Check()`

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

SystemError setting ID custom property to a list containing NumPy scalars #117656