How are kerchunk Codecs registered at runtime?

[mpiannucci]

I am working to add kerchunk support to my experimental grib lirbary gribberish.

Everything works fine when I manually import the custom codec (which registers the codes with numcodecs the same way as the built in kerchunk codecs), but if I don't then I predictably get codec not found errors.

However, I notice that the built in kerchunk codecs have no such problems and a dataset will work fine even if you load it in without importing kerchunk. As far as I can tell the only place the registration happens is when the codecs.py module is imported at the kerchunk package root, but I am wondering if there is another step I am missing here? How is the GribCodec registered if I don't manually import kerchunk when opening the dataset?

Thanks for any pointers you can provide!

[mpiannucci]

Figured it out with the entry points duh!!