pivot_longer() regression with tibble 3.0.1

[krlmlr]

From tidyverse/tibble#766, CC @mvazquezs.

Is the previous behavior correct? Do we need a better error message?

tibble 2.1.3

library(tidyverse)

tibble(id = 1, x_1 = 1, x_2 = 2, y_1 = 1, y_2 = 2) %>%
  pivot_longer(
    cols = -id,
    names_to = c("x", "x_dil"),
    names_pattern = "(.+)_(.+)",
    values_to = c("mfi", "dil")
  )
#> # A tibble: 4 x 5
#>      id x     x_dil   mfi   dil
#>   <dbl> <chr> <chr> <dbl> <dbl>
#> 1     1 x     1         1    NA
#> 2     1 x     2        NA     2
#> 3     1 y     1         1    NA
#> 4     1 y     2        NA     2

^{Created on 2020-04-29 by the reprex package (v0.3.0)}

tibble 3.0.1

library(tidyverse)

tibble(id = 1, x_1 = 1, x_2 = 2, y_1 = 1, y_2 = 2) %>%
  pivot_longer(
    cols = -id,
    names_to = c("x", "x_dil"),
    names_pattern = "(.+)_(.+)",
    values_to = c("mfi", "dil")
  )
#> Error: Assigned data `values_to` must be compatible with existing data.
#> x Existing data has 4 rows.
#> x Assigned data has 2 rows.
#> ℹ Only vectors of size 1 are recycled.

^{Created on 2020-04-29 by the reprex package (v0.3.0)}

[hadley]

Almost certainly a bug. But should be simple fix.

[hadley]

Looking at the code, I don't think there's any suggestion that values_to should be a vector, so I think this can just be a better error.

[hadley]

I think the call should be more like this:

library(tidyr)

tibble(id = 1, x_mfi = 1, x_dil = 2, y_mfi = 1, y_dil = 2) %>%
  pivot_longer(
    cols = -id,
    names_to = c("x", ".value"),
    names_pattern = "(.+)_(.+)"
  )
#> # A tibble: 2 x 4
#>      id x       mfi   dil
#>   <dbl> <chr> <dbl> <dbl>
#> 1     1 x         1     2
#> 2     1 y         1     2

^{Created on 2020-05-04 by the reprex package (v0.3.0)}

[mvazquezs]

The special .values name works fine and simplifies the code a lot. Thanks so much @hadley and @krlmlr for your help.