TypeError 'numpy._DTypeMeta' object is not subscriptable

[lukegwoolley]

Happens on both latest numpy version and numpy 1.21.5

Error sits in this line:
File /local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.10/site-packages/highspy/highs.py:21
19 # backwards typing support information for HighspyArray
20 if sys.version_info >= (3, 9):
---> 21 ndarray_object_type = np.ndarray[Any, np.dtype[np.object_]]

[jajhall]

Any thoughts on how to fix this @mathgeekcoder ?

[mathgeekcoder]

Oh interesting. Python typing support seems a little troublesome. Since I can't reproduce, @lukegwoolley, can you please try to modify your highs.py, and see if this works for you?

# backwards typing support information for HighspyArray
if sys.version_info >= (3, 9):
    from typing import TypeAlias
    obj_type: TypeAlias = 'np.dtype[np.object_]'
    ndarray_object_type = np.ndarray[Any, obj_type]
else:
    ndarray_object_type = np.ndarray

If that works, I can make a new PR with some relevant fixes.

BTW: You can import the TypeAlias with the imports at the top of highs.py instead, I included it in the code snippet for easier copy/paste.

[lukegwoolley]

Hi @mathgeekcoder

I had a play around with the code and I think the issue is the sys.version issue is wrong.

The below code works for me in python 3.10 and 3.12 but I wasnt able to test for 3.11.

# backwards typing support information for HighspyArray
if sys.version_info >= (3, 11):
    ndarray_object_type = np.ndarray[Any, np.dtype[np.object_]]
else:
    ndarray_object_type = np.ndarray

[mathgeekcoder]

Thanks @lukegwoolley. Did you happen to test the code I sent? Can you confirm if it didn't work?

Updating that version check just disables the numpy typing for your python 3.10. According to the docs, this typing should be supported in 3.10.

Also, do you know if there's anything special about your environment? I'm wondering since this error wasn't picked up in our CI build - which also tests the python wheel on many different versions.

[lukegwoolley]

It errors out here - it might be due to the numpy version then as unfortunately databricks is where I am running it and it uses a very old numpy and pandas version.

numpy is 1.21.5

The previous version of highspy 1.7.3 works completely fine btw