Use ujson as a drop in replacement for json library? about jupyterlab_server HOT 5 OPEN

The speed gains from ujson are good... but orjson, pysimdjson... the list goes on of other libraries that offer a different set of features/performance against increasing complexity... for example, ujson just got working binary wheels again for all supported platforms.

If anything, I'd imagine trying to find a library that wrapped (but did not depend on) all of these...

from jupyterlab_server.

I'd support any of the above provided they're installable as wheels and available on conda-forge.

from jupyterlab_server.

at least orjson says it may not work on 32-bit, which means jupyter presumably wouldn't work on my Raspberry pi anymore :(.

What about pypy support as well?

I think we could perhaps try to use something like ujson or orjson if we can import it, but fall back to the standard library json if it is not available.

from jupyterlab_server.

I think we could perhaps try to use something like ujson or orjson if we can import it, but fall back to the standard library json

Sure, that makes sense too.

from jupyterlab_server.

Can I ask what leaned you towards picking ujson or orjson over pysimdjson? It helps me improve the library and packaging.

Note pysimdjson has pretty wide compatibility, and is tested against linux/macosx/windows/freebsd (x86_64 and ARM) for py3.5 -> 3.9beta, as well as pypy3. It has binary ARM packages as well, so it will install quickly on your raspberry pi (3 or 4) as long as you're using one of the 64 bit OSes.

from jupyterlab_server.