zincware / znjson Goto Github PK

To avoid issues when e.g. de-serializing a dict {"Path": "this is some string"} which should return a dict and not a Path object,
it would be possible to serialize all objects to the dict {<type>: val} which would yield{"dict": {"Path": "this is some string"}} and therefore can not be confused with {"Path": "this is some string"}.

It is currently possible to register two converters with the same representation: str which can be avoided by checking for uniqueness in the register command..

A numpy (and maybe also pandas) converter could also convert to lists instead of binary.
This could be useful for small arrays.

Currently if _type is there it will try to deserialize and fail. Sometimes it could be useful to only serialize a specific type.

see https://github.com/zincware/ZnTrack/blob/4551776c4397a1fc12b82829cbf3ee292faeea55/zntrack/fields/zn/__init__.py#L316-L320
where we only would like to serialize the converters.

Currently you have to do:

znjson.config.register(CustomConverter)
json.dump(data, cls=znjson.ZnEncoder)

but maybe you only want to use a specific converter which could look something like:

json.dump(data, cls=znjson.ZnEncoder.from_converter(CustomConverter, allow_default=True)) # this could be a list of converters
# or
json.dump(data, cls=CustomConverter) # this would allow only one converter

Currently large numpy arrays will be stored uncompressed and can take a lot of space.

provide patched zndraw.json module and a magic patch

The package https://github.com/jsonpickle/jsonpickle provides similar functionality