Selecting columns from a grouped_df with [ results in lost grouping

[wch]

m <- mtcars %>% group_by(cyl)

# Selecting rows keeps grouping
m[1:3, ]
# Source: local data frame [3 x 11]
# Groups: cyl
# 
#                mpg cyl disp  hp drat    wt  qsec vs am gear carb
# Mazda RX4     21.0   6  160 110 3.90 2.620 16.46  0  1    4    4
# Mazda RX4 Wag 21.0   6  160 110 3.90 2.875 17.02  0  1    4    4
# Datsun 710    22.8   4  108  93 3.85 2.320 18.61  1  1    4    1

# Selecting columns loses grouping
m[1:3, 1:3]
# Source: local data frame [3 x 3]
# Groups: 
# 
#                mpg cyl disp
# Mazda RX4     21.0   6  160
# Mazda RX4 Wag 21.0   6  160
# Datsun 710    22.8   4  108

# All attributes are lost
str(m[1:3, 1:3])
# Classes ‘grouped_df’, ‘tbl_df’, ‘tbl’ and 'data.frame':  3 obs. of  3 variables:
#  $ mpg : num  21 21 22.8
#  $ cyl : num  6 6 4
#  $ disp: num  160 160 108

This is especially problematic because R crashes if you use do() on this grouped_df which has no groups:

m[, 1:3] %>% do(mpg = mean(.$mpg))
# [segfault]

[romainfrancois]

I've put some protection in place. We now get an error:

> m <- mtcars %>% group_by(cyl)
> m[1:3,1:3] %>% do(mpg = mean(.$mpg))
Erreur : no variables to group by

[wch]

It might make sense to keep the groups of any grouping columns that are selected (cyl in this case), but drop the groups of any grouping columns that aren't selected. If no grouping columns are selected, you could drop the grouped_df class.

I realize that select() works differently -- it keeps the grouping columns, even if the user doesn't ask for them specifically.

[romainfrancois]

We'd have to write a [.grouped_df for this. Not sure we want to go there.

[hadley]

I think we should probably protect attributes in [ methods (we wouldn't encourage users to use these methods but they are useful for developers). I'll add the code