select() silently refuses to drop grouping variable

[jennybc]

Maybe this is just documentation, but a warning would be nice. I know I can ungroup() and I shall.

x <- data.frame(y = 1:3, z = LETTERS[3:1])
x %>%
  group_by(z) %>% 
  select(-z)
#> Source: local data frame [3 x 2]
#> Groups: z [3]
#> 
#>        z     y
#>   (fctr) (int)
#> 1      C     1
#> 2      B     2
#> 3      A     3

Same issue on stack overflow here and here.

[hadley]

What do you expect to happen?

a) fail with a warning
b) remove z and hence remove the grouping
c) remove z and somehow keep the grouping

[jennybc]

I think a) or b) is good. For me, the biggest thing is the wake up call that a warning provides.

[hadley]

I'm leaning towards making this warning. Something along the lines of "Automatically adding z to variable list to preserve grouping".

Would you expect the same warning if you did:

x %>%
  group_by(z) %>% 
  select(x)

[jennybc]

Yes I would expect and appreciate that warning.

In real usage, the group_by() can be much farther away and I often forget that I've grouped, once the groups have served their purpose. Maybe I should always ungroup(), if no immediate reason to keep grouping.

[krlmlr]

Going a bit further: What are the use cases for more than one verb between group_by() and ungroup()? Perhaps summarize() and do() should always ungroup?

[hadley]

@krlmlr I take advantage of it quite a bit, plus I think it's too late to change now.

[DieselAnalytics]

@jennybc thanks for posting this issue. I thought this was an bug but now I know to use "ungroup". @hadley thanks for developing the package!