How to report multiple documents on extract

[mih]

I implement a dataset-level metadata extractor. I think I need to be able to report multiple, individual metadata records. In principle, one be able to build these records in a way that they can be reported in a nested fashion (thereby reporting just a single object). However, in my case I have no control over the nature of these documents, and they might be linked (or not) in different ways.

What is a desirable approach here?

• an arbitrary top-level key that maps onto an array?
• a JSON-LD style @graph top-level key (as a realization of the above)?
• something else?

Related: We might be talking about a lot of stuff to return. If I see things correctly, I need to load multiple standalone records into memory (many), report them via immediate_data as a single dict, such that they can be written out as JSON (again). I am yet to understand why meta-extract turns a single return value of type ExtractorResult into a result record, rather than dealing with result records directly. This would make the standard machinery of seemlessly switching between return values and generator yields applicable to metadata extractors too

[christian-monch]

With the realization that the principle approach requires fixing, I would opt for the {"@graph": [ <objects>]}-approach as a quick-fix.

[christian-monch]

@mih I didn't think about that yesterday afternoon, but another option would be to return a list, which contains the individual results, in immediate_data.