[C++] Expose `{Array,...}FromJSON` as public APIs

[pitrou]

Describe the enhancement requested

The various XXXFromJSON functions have been initially written to ease writing our own unit tests, but they could actually be generally helpful for users to experiment with Arrow, write examples, etc. Perhaps we could expose them as public APIs?

We should just be careful to document that they are not meant for high-performance data ingestion.

Component(s)

C++

[pitrou]

What do you think @amoeba @zanmato1984 @bkietz ? This came to mind while reviewing #45411

(also FYI @EnricoMi )

[bkietz]

I think it'd be invaluable as an aid for documentation; most of the example input/outputs for the compute kernels for example use an informal JSON-ish format, illustrating that terse array definitions are useful in explanations.

Related issues:

• [C++] Expose Arrow's *FromJson, Assertion and Random generator helper functions #33538
• [C++] Implement Array to JSON function #24864

[zanmato1984]

+1 on making them public. Could be very useful in demoing things using arrow.

We should just be careful to document that they are not meant for high-performance data ingestion.

Maybe we can export them in libarrow_testing to avoid any performance expectations?

[pitrou]

Maybe we can export them in libarrow_testing to avoid any performance expectations?

I'm not sure we expose libarrow_testing for public consumption. I would be lukewarm about that, since it contains a lot of disparate stuff, some of which we might not want to encourage reliance on.

[zanmato1984]

I'm not sure we expose libarrow_testing for public consumption. I would be lukewarm about that, since it contains a lot of disparate stuff, some of which we might not want to encourage reliance on.

Agreed. Another practical problem about libarrow_testing just came to me as well: we have multiple levels of data containers spread in multiple sub-libraries, e.g. Array/RecordBatch/Table in arrow core and ExecBatch in compute. We better export *FromJSON in the library where * belongs to, rather than exporting all FromJSON in one place where dependencies would become a mess.

[amoeba]

I created #46180 to move just the helpers in json_simple to make it as easy to review as possible.