It is not a matter of v2.1 requires lower python than 3.12 and v2.0 dot not, and more that python changed some components such as casting in v3.12, which is incompatible with all of pyRDDLGym versions. So, in version 2.1 we added the restriction of using a lower version of python.

I'm thinking that at some point we will have to open compatibility with 3.12 and later python versions, regardless of whether numpy will fix that on their end.

Please check the latest PR. I am not sure I remember exactly what the problems were but one of them with casting was that numpy can_cast was causing a crash on latest version. I have fixed this particular issue, and the code seems to run correctly in a new python 3.12 environment.

It is not a matter of v2.1 requires lower python than 3.12 and v2.0 dot not, and more that python changed some components such as casting in v3.12, which is incompatible with all of pyRDDLGym versions. So, in version 2.1 we added the restriction of using a lower version of python.

We probably do not use the full extent of pyRDDLGym, but in the ci/cd process of our project scikit-decide, we do use pyRDDLGym on python 3.12 environments without any issues (this is still in a PR for now).

I believe the numpy casting checks were added sometime after v 2.0 was released, so is it possible you were testing on pyrddlgym 2.0? 2.1 does not work with python 3.12 on my end in a fresh environment.

Yes i was speaking about 2.0 (since pip installing 2.1 does not work because of the python<3.12 requirement)