I read about Pypy, that should be a faster python. My questions are: Is it possible to use it as a Ipykernel? If Yes, is it worth it? Is there any limitation that make it incompatible with recent jupyter versions?
At present the whole stack is installable in a pypy environment, and some packages have pypy in their continuous integration setup. As the bulk of core jupyter packages are pure python, it is likely these will continue to work indefinitely.
Outside of jupyter, many domain-specific packages with compiled components may not publish pypy wheels, or test against pypy, which can create additional headaches for the user.
As for reproducibly installing pypy: unfortunately, conda-forge had to drop PyPy, leaving that whole stack on the python 3.9 syntax and standard library, which will be EOL in a few months. The best course today for a reproducible pypy environment (particularly important for non-mainstream architecture choices) would be uv (which can be installed from conda-forge).
2 Likes
I’ve build a rudimentary pypy kernel a while back.
Pushed it to github for you: here
This will only work with pypy in environments PATH (and only tested on Linux).
1 Like