Sketch implementation of new separate functions

[hadley]

TODO:

• Better function names
• extra and fill need prefixes
• Better way to get bad rows? — No actual demand, no maybe in the future.
• Should list2df() be baked into unnest_wider() or provided in stringr? — Doesn't really matter; leave in tidyr for this PR.
• Warnings for separate_wider_fixed() & separate_wider_regex()
• Need to pass along names_repair in *_wider
• Check inputs
• Add tests
• Extract out repeated code to map_unpack() + map_unchop()?
• Don't use named groups to avoid problems with special var names
• separate_longer() that extracts group from multiple matches
• Warn if separate_wider_regex() uses parens (i.e. check number of columns in match)
• str_separate_wider_fixed should be able to create arbitrary number of columns too
• Make into generics? — Maybe in the future.
• Need way to diagnose/iterate with separate_wider_regex()
• Think about how to incorporate into names_pattern etc in pivot_longer()
• Update error handling — argument checking in top level functions, so no longer needed
• Same spec for pivot_longer(names_pattern)
• Use sep instead of delim
• Some way to suppress warnings in separate_regex_wider
• diagnose = TRUE
• Add news bullet
• Update superseded functions
• Don't include NA values in problems

[DavisVaughan]

Mainly lots of small things. Overall looks really nice

I had a tough time knowing how to review the "progressively relax the matches" part of str_separate_wider_regex() because I'm not super familiar with regex

[mine-cetinkaya-rundel]

I think the PR looks great. I like the naming of the functions and the arguments. I've added some comments throughout but they're mostly minor.

One thing that I think would be beneficial is a table like https://github.com/tidyverse/tidyups/blob/9fbe5d48ec9c8dbf3bcd5db5995d905fba9532a7/002-tidyr-stringr.Rmd#L108-L114. I wonder if that's more of a vignette-thing than a function docs thing, but I think it would be good to expose a version of that with the new functions somewhere in the tidyverse docs.

A minor / potentially nitpicky comment is about styling of function arguments in documentation. For example, we should use "code text" (backticks) for the names_repair options in the pack.R documentation (

tidyr/R/pack.R

Lines 91 to 94 in f6b9509

	#' * "minimal": no name repair or checks, beyond basic existence,
	#' * "unique": make sure names are unique and not empty,
	#' * "check_unique": (the default), no name repair, but check they are unique,
	#' * "universal": make the names unique and syntactic

), that then gets used in various other places, e.g., "minimal". I think it's also worthwhile to make a similar change to the extra and fill options for separate() as well. (Note: I checked other similar arguments and .groups for summarize() doesn't use code text but method for geom_smooth() does. I think it's good to consistently distinguish character strings that are function arguments with styling.)

[mine-cetinkaya-rundel]

I think all of these examples would be easier to follow if the dfs are defined with tribble() instead of tibble(), but I realize that takes space...

[mine-cetinkaya-rundel]

This is a general comment about this sort of statement, I think it makes more sense above the description, as the first thing in the documentation for the function. I often find myself regretting having spent mental energy deciphering the description of the function just to find out it's been superseded/deprecated.

[hadley]

Interesting; I put them below in the belief that most people would want to know what a function does before knowing what replaces it. I can move them.