Support objects with buffer-protocol as an input to loads

Currently the input to orjson.loads must be a bytearray or a bytes object. It would be useful if possible to also support passing in other objects supporting the buffer protocol.

In my use case, I have a large bytearray that contains some serialized json at known offsets. If possible, I’d like to read this in without having to copy it out of the bytearray into a new bytearray that only contains the json. Example:

In [25]: header = b"some bytes before"

In [26]: footer = b"some bytes after"

In [27]: json = orjson.dumps({"key": "val"})

In [28]: data = bytearray(header + json + footer)  # a potentially large bytearray

In [29]: subset = memoryview(data)[len(header):len(header) + len(json)]  # a view on the json data, without a copy

In [30]: bytearray(subset)  # this makes a copy, would be nice to just pass `subset` to loads
Out[30]: bytearray(b'{"key":"val"}')

[ijl]

https://github.com/ijl/orjson/releases/tag/3.5.0

[ijl]

@mattsta thanks for the research, it’s helpful.