eval_pull()

[hadley]

Equivalent of vars_pull() with new interface; works like eval_select() but must select exactly one columns. (And negative values count backwards from RHS)

[lionel-]

It probably should be stricter than vars_pull() (see #251), which entails some design work and probably a dedicated interpreter as in eval_select().

[hadley]

I'd be tempted to say that eval_pull() should be exactly eval_select() with the added restriction that it must return a single column. That would allow syntax that's not that useful, but a single tidyselect syntax would be easier to explain and to implement.

OTOH the different meanings of df |> select(-1) and df |> pull(-1) is going to make this at least a little tricky.

[lionel-]

Random thought, it seems like it might be useful for interactive usage to allow helpers like starts_with() and matches() or predicate functions in pull(). This would always select the first matching element and throw an error if none.