Indeed, Jackson2Tokenizer.tokenize has a boolean flag tokenizeArrays that determines whether to flatten the array or not, and there are tests that parse the array both ways. However, we always pass true from AbstractJackson2Dedcoder and there is no way to change that.

It should be an easy change then, just a matter of how to decide when to set the flag. We could perhaps make it conditional on whether the media type is known as a streaming type. We have similar checks on the encoder side.

On second though, NDJSON implies a stream, and that should be streamed rather than collected to a List. So I'm not sure we can do anything transparently for your case. At best, we could provide some way to control it through a protected method.

Could you explain a little better your case and why you are collecting NDJSON stream to a list?

I'm not aggregating multiple NDJSON lines into a single list. Rather, each line represents a distinct item, which could be structured as a JSON array itself. This maintains the NDJSON format as a continuous stream of data.

In my use case, I need to handle multiple objects within the same NDJSON line as a single transaction. Therefore, I need the framework to automatically group these objects in the same way they are structured within an NDJSON format. For example:

[{"name": "Alice", "age": 25, "city": "San Francisco"}]
[{"name": "John", "age": 30, "city": "New York"},{"name": "Bob", "age": 35, "city": "Chicago"}]
[{"name": "Eve", "age": 28, "city": "Los Angeles"}]

I reviewed the NDJSON specification and did not find any explicit statement that prohibits a line from containing a JSON array instead of a single object. However, I wouldn't be surprised if this is indeed the case. Therefore, my request might deviate from the specification and may be considered invalid.

Okay, I see now. We can decide then whether to set tokenize based on the target type. If the element type is an array or collection, then false (don't flatten), or true otherwise.