import xlb fails on Google Colab

[astefanz]

With xlb-0.2.1, in a Google Colab instance, the following:

!pip install xlb
import xlb

throws this exception:

ImportError                               Traceback (most recent call last)
[<ipython-input-2-bc1869afc235>](https://localhost:8080/#) in <cell line: 0>()
----> 1 import xlb

6 frames
[/usr/local/lib/python3.11/dist-packages/numpy/_core/strings.py](https://localhost:8080/#) in <module>
     12 from numpy._core.multiarray import _vec_string
     13 from numpy._core.overrides import set_module
---> 14 from numpy._core.umath import (
     15     isalpha,
     16     isdigit,

ImportError: cannot import name '_center' from 'numpy._core.umath' (/usr/local/lib/python3.11/dist-packages/numpy/_core/umath.py)

On both a CPU and A100 instance, with both !pip install xlb and !pip install "xlb[tpu]".

As a workaround, without XLB installed, install dependencies, clone XLB and import xlb as local code:

!pip install ruff warp-lang trimesh numpy-stl vtk pyvista
!git clone https://github.com/autodesk/XLB
!ln -s ./XLB/xlb ./xlb
import xlb

[astefanz]

pip throws this error message when executing !pip install xlb on a fresh Colab instance:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
numba 0.60.0 requires numpy<2.1,>=1.22, but you have numpy 2.2.4 which is incompatible.
tensorflow 2.18.0 requires numpy<2.1.0,>=1.26.0, but you have numpy 2.2.4 which is incompatible.

[mehdiataei]

Colab instances have their own environment with pre-installed libraries such as TensorFlow, which may conflict with the requirements of XLB. To resolve this error, you need to uninstall the conflicting libraries and install the correct version of NumPy.

[astefanz]

Turns out the correct NumPy version (2.2.4) is installed correctly with !pip install xlb, but Colab does not reload it. Restarting the session after installing XLB fixes this, and import xlb runs without error.

Reloading programatically with importlib does not work. The session must be manually restarted or closed with os._exit(0).