Numba error in clean install of bempp
Closed this issue · 6 comments
I am getting an error with numba (due to an update maybe?) . Installing to a clean conda environment and running the first cell of notebook:
/bempp-cl/notebooks/laplace/dirichlet_weak_imposition.ipynb
I get the following traceback:
AttributeError Traceback (most recent call last)
in
----> 1 import bempp.api
2 import numpy as np
~/.local/lib/python3.9/site-packages/bempp/api/init.py in
7
8 import pyopencl as _cl
----> 9 from bempp.api.utils import DefaultParameters
10 from bempp.api.utils.helpers import MemProfiler
11
~/.local/lib/python3.9/site-packages/bempp/api/utils/init.py in
3 from .which import which
4 from .parameters import DefaultParameters
----> 5 from .octree import Octree
6
~/.local/lib/python3.9/site-packages/bempp/api/utils/octree.py in
5
6
----> 7 @_numba.jitclass(
8 [
9 ("_lbound", _numba.float64[:]),
AttributeError: module 'numba' has no attribute 'jitclass'
Hi
I got the same problem while installation.
It seems that numba update this part, and jitclass is no longer supported.
https://numba.pydata.org/numba-doc/latest/user/jitclass.html
I managed to get it working by preinstalling numba == 0.48.0. if it helps.
Hi
But I got conflict:
- bempp-cl -> numba[version='>=0.51.2']
how do you install an early version?
I just reinstalled from scratch.
Hi. Thanks for the comments. It looks like jitclass has been moved to a different location. I will prepare a new bugfix release after the weekend. There are a couple of other bugfixes that should be released as well.
Fixed in Master