Closed stonebig closed 3 years ago
Thanks for opening this issue.
Another reason for getting rid of orjson is compatibility with jupyterlite, ipycanvas won't be installable on jupyterlite unless orjson is not a dependency.
The problem is orjson is super fast, so maybe we should not make it a hard dependency, and fallback on another JSON library if it's not found, but still use it if it's found.
Would rapidjson be a good replacement in your opinion?
I need something compatible with pypy3.7 on Windows.
Performance is not the concern.
It's all about the belief that to go x5 speed up on cpython, the recipe needed for pypy3 will have to be applied, so working on pypy3 compatibility is a bijection with working on cpython performance
I implemented a fallback to the Python standard json
library. I understand pypy's json library should be faster.
I've never looked into building a pypy package on conda-forge. Is it something you'd be able/willing to work on?
I'm not on conda. I'm on pypi.
could ipycanvas rely on something else than orjson when running on PyPy ?
orjson is not compatible with PyPy https://github.com/ijl/orjson#will-it-support-pypy