unnest, unchop, unpack

[moodymudskipper]

the tidyverse also has unnest_longer, unnest_wider, unnest_auto, unnest_legacy, but we might not need this complexity.

Taking inspiration from the margin argument from apply :

• tb1[1 ~ foo] expands horizontally (similar to unchop() and unnest_longer())
• tb1[2 ~ foo] expands vertically (similar to unpack and unnest_wider)
• tb1[3 ~ foo] expands in both directions (similar to unnest(), unnest_legacy(), or unnest_auto()

We can remove created unnested columns :

• tb1[((1)) ~ foo]
• tb1[((2)) ~ foo]
• tb1[((3)) ~ foo]

These arguments could be labelled, with one (usually 1st case) or several column names.

By convention we place the n ~ BEFORE the label, so we'll have tb1[1 ~ foo := ...], this is needed because the expansions occur in the end, of the [.tb call.

I'm not completely comfortable with this decision however, as :

• technically expanding horizontally could be done through a function call that would be spliced.
• horizontal expansion doesn't have to happen at the end of the call, it is more useful to do it right away, but more consistent to do it in the end, so what do we do ?

We could just not have ~2 and 3 and just use ~ ... for 2~ ... (vertical expansion), and use splicing to do the horizontal expansion, but not being able to delete the source column will be frustrating.

Another way is to use -- on the rhs to delete columns. but the equivalent of ((3)) ~ foo becomes ~+--foo ! Which technically is still more compact that unnest, and we might get used to the syntax, but still a bit intimidating.

[moodymudskipper]

Maybe if we want to keep a column that we want to unnest we just make a copy right before ?

tb[keep = nested, ~+nested]

Now it doesn't seem to me we want to keep it so often, and it makes things much easier.

[moodymudskipper]

should ~+ and +~ be synonyms ?

Maybe not, they'd be the same for data frame elements but a vector would behave differently depending on what's done first, so both should be supported