Support missing subfields in EDF+ header fields.

[marcoross]

Increase tolerance regarding EDF+ files not compliant with the EDF+ standard. At the same time, make the API more comfortable to use by accepting None or empty strings as unspecified subfields.

[hofaflo]

Great idea! I'd like to avoid return type unions with None wherever possible, as this makes the type annotations much less helpful. Can we instead return "X" for all non-existent fields? This would also allow to keep the __init__ methods as they are.

[cbrnr]

The first field of local_recording_information must always be Startdate, so I'd use this as a default instead of X.

[hofaflo]

The first field of local_recording_information must always be Startdate, so I'd use this as a default instead of X.

Accessing this field via edfio.Recording is currently not possible. Do you think it should be?

[cbrnr]

Accessing this field via edfio.Recording is currently not possible. Do you think it should be?

I saw it was printed in the repr though. This question is orthogonal to what the field's default value should be, right? I guess it should be accessible somewhere, not sure what the best place is.

[marcoross]

Great idea! I'd like to avoid return type unions with None wherever possible, as this makes the type annotations much less helpful. Can we instead return "X" for all non-existent fields? This would also allow to keep the __init__ methods as they are.

I changed the return types to str and return "X" in case of unspecified subfields.

[marcoross]

Accessing this field via edfio.Recording is currently not possible. Do you think it should be?

I saw it was printed in the repr though. This question is orthogonal to what the field's default value should be, right? I guess it should be accessible somewhere, not sure what the best place is.

I made the get_subfield(idx: int) method public, so now it can be accessed using this method with index 0.

[hofaflo]

I'd prefer to avoid allowing to pass None instead of "X" (or omitting the parameter) when instantiating a Patient or Recording. This would simplify the signatures and let us avoid some is None checks. Do you see a use case that requires to accept None?

[marcoross]

I made the requested change. I don't know, however, if this is still relevant given the ongoing discussion about strictness regarding the EDF+ standard and the possibility of having an "auto-fix" method.

[hofaflo]

I don't know, however, if this is still relevant given the ongoing discussion about strictness regarding the EDF+ standard and the possibility of having an "auto-fix" method.

I think it still makes sense to have this, as one might want to access header fields without trying to autofix stuff.

Note: I rebased and force-pushed to fix the position of the changelog entry.

[marcoross]

I made all requested changes in the last commit.