asdf.load and asdf.dump

[braingram]

dump and load (and the related dumps and loads) is the (perhaps informal) standard for serialization/deserialization in python. This API is supported by:

• pickle
• json
• yaml (doesn't support dumps and loads)
• toml

I propose that we add support for this API at the top level module to align asdf with other serialization packages in python.

This will have the added benefit of simplifying test code and examples.

asdf.dump example

For example writing a dict/tree to a file at the moment can be done with:

asdf.AsdfFile({"foo": 1}).write_to("foo.asdf")

this could be replaced with:

asdf.dump({"foo": 1}, "foo.asdf")

Non-lazy asdf.load

For load I propose that we set the default for lazy_load to False as it would otherwise cause issues for array data:

# assuming "foo.asdf" has an array at "arr"
with asdf.open("foo.asdf") as af:
    tree = af.tree
tree["arr"][:]  # results in an exception because the file is closed

So load could by used as follows:

tree = asdf.load("foo.asdf")
tree["arr"][:]  # access to array data here results in no exception because it was not lazy loaded

Supported "file" argument

To detail a few supported options I suggest that we allow asdf.load and asdf.dump to handle:

• strings as filenames
• Paths as filenames
• file-like objects
• fsspec file objects (to support http, s3, etc)

Additional arguments

We will also want to consider how to handle additional keyword arguments as both asdf.open and AsdfFile (and AsdfFile.write_to) support a number of useful options that we can map to asdf.load and asdf.dump.

[braingram]

Added in #1930