auto_inline not working as expected

[nden]

There must be a subtle bug in auto_inline - it seems the behavior depends on previously run commands. Is something setting a state?

f = AsdfFile({'data': np.arange(6)})
f.write_to('test.asdf') # correctly writes the data inline

f.write_to('test.asdf', auto_inline=3) # still writes it inline

# force it to write to a block
f.write_to('test.asdf', all_array_storage='internal') 

f.write_to('test.asdf', auto_inline=3) # writes to a block as expected

f.write_to('test.asdf', auto_inline=10) # writes it inline

f.write_to('test.asdf', auto_inline=3) # still writes it inline

[drdavella]

Thanks for the report. I've noticed other places where write_to modifies the state of the AsdfFile object, which is not okay, in my opinion. I wish there was a better way to enforce this in Python (I'm thinking of const methods in C++, which can't update the associated object).

A related discussion came up in #597 regarding the use/implementation of inline_threshold. I think I'm going to have to rework that API and somehow combine it with auto_inline since they serve the same purpose. Since the inline_threshold feature is still in an unreleased version of ASDF, we have no concerns about making further API changes.

[drdavella]

This is still a real issue, even after #610. Unfortunately, however, it's one that is unlikely to be resolved without a major redesign. I'm considering some major API changes for 3.0, which would hopefully address this kind of issue.