storing non ndarray types as blocks

[CagtayFabry]

Is there a way to store any other types besides ndarray as a block in an ASDF file?

I am looking for a simple way to serialize (and tag) bytes, bytearray, BytesIO or similar without going through numpy

Could this be done in an extension library or would this need changes in asdf?

[eslavich]

Currently the code requires that anything written to a block be wrapped in ndarray, but I think it's possible to wrap bytes in ndarray without copying the memory:

In [21]: some_bytes = b'a' * 1024

In [22]: array = np.frombuffer(some_bytes, dtype=np.uint8)

In [23]: array.flags
Out[23]:
  C_CONTIGUOUS : True
  F_CONTIGUOUS : True
  OWNDATA : False
  WRITEABLE : False
  ALIGNED : True
  WRITEBACKIFCOPY : False
  UPDATEIFCOPY : False

Now that array can be assigned to a block using the BlockManager.find_or_create_block_for_array method.

On the way back I don't know how to retrieve the bytes except to call ndarray.tobytes, which does make a copy.

[CagtayFabry]

yes that is a possible workaround we used before

However I was looking for a possibility to explicitly differentiate from ndarray but in this case I will probably just wrap the numpy calls in a custom tag/object