🤔 I understand the issue with masking the base package, but I'm not sure I like arrow_array() and arrow_table() for these factories, esp that other factories, e.g. record_batch() or chunked_array() presumably stay the same ?

vctrs used to be more important earlier in the development, specifically when array() took ... and vctrs::vec_c() would collapse the ... into one vector. So yeah it's totally fine to move it to Suggests, perhaps we can even get rid of it for the tests.

We can totally just not use crayon at all.

Fair point, I worried about that too. My preference would be to namespace everything (like sparklyr), e.g. arrow_record_batch, rather than masking the base functions, but that was a bigger, more disruptive change so I wanted to try this. Happy to try that out if you're open to it--I think it also raises some design questions about how we want the R developer interface to Arrow to look and feel, which is a good thing to think about anyway.

Yeah, maybe it's worth have a broader discussion about the interface as a whole. These functions array(), chunked_array(), table(), and record_batch() are factories that make objects, maybe they should be e.g. :

new_Array(), new_ChunkedArray(), new_Table(), new_RecordBatch()

Maybe the R6 classes should be made more user visible and then they would be e.g.

Array$new(), ChunkedArray$new(), Table$new(), RecordBatch$new() ...

In any case, we should be consistent in what we do, and not just prefixing with arrow_ for the cases where we do mask another function.

Closing for now; we can continue the discussion on Jira/elsewhere and revisit.