Improve pivot_longer(names_pattern) syntax

[hadley]

To match new separate_wider_regex()

[DavisVaughan]

Does that look something like

Previously:

who %>% pivot_longer(
  cols = new_sp_m014:newrel_f65,
  names_to = c("diagnosis", "gender", "age"),
  names_pattern = "new_?(.*)_(.)(.*)",
  values_to = "count"
)

New: (where it makes names_to unnecessary?)

who %>% pivot_longer(
  cols = new_sp_m014:newrel_f65,
  names_pattern = c("new_?", diagnosis = ".*", "_", gender = ".", age = ".*"),
  values_to = "count"
)

It looks like maybe we could also use features from separate_wider_delim() and separate_wider_position()?

• separate_wider_position() could probably use the same idea as names_pattern above, but applied to names_sep when it is supplied as numeric positions
• separate_wider_delim() is basically already supported through a string names_sep, but our current internal usage of str_separate() doesn't allow stringr modifiers like stringr::regex() so maybe we'd be able to relax that? This one would still need names_to to be specified, like how separate_wider_delim() has names.

[DavisVaughan]

I think the toughest part is deciding when we are doing old separate/extract() like behavior vs using the new separate_wider_*() style behavior.

Possibly if we have a named character vector for names_pattern we use the new behavior and if we have an unnamed character vector we use the old behavior? Feels a little risky but might work. We could soft deprecate the unnamed character vector behavior

[DavisVaughan]

Other tough part is that we probably want to use the algorithm of str_separate_wider_regex() and friends, but we don't want their error messages because they refer to too_few = and those separate_wider_*() specific args

[DavisVaughan]

Could also introduce 3 new arguments, names_delim, names_patterns, and names_positions that align with separate_wider_*() and soft deprecate names_sep and names_pattern