lost Groups cause dplyr functions to crash R

[statsandwich]

Losing Groups in a grouped_df, and subsequently using dplyr functions, causes R to crash.

If I mix non-dplyr functions like subset on grouped_dfs, the Groups are lost. When the groups are lost, any use of filter(), select(), mutate(), summarise() causes R to crash.

require( dplyr)
DF <- data.frame( C1 = 1:2, C2 = 1:2) %.% group_by(C1)

DF
Source: local data frame [2 x 2]
Groups: C1

C1 C2
1 1 1
2 2 2

** Using dplyr functions on the following grouped_df will crash R. Notice, no Groups:

DF[, names(DF)] ### instead of using select()
Source: local data frame [2 x 2]
Groups:

C1 C2
1 1 1
2 2 2

subset( DF, C1 == 1) ### instead of using filter()
Source: local data frame [1 x 2]
Groups:

C1 C2
1 1 1

This crash sneaks up over and over in converting old functions to dplyr.

[romainfrancois]

So what happens is that R's regular [ keeps the class but drops all other attributes instead of setting the class to just data.frame.

> load_all()
Loading dplyr
>
> DF <- data.frame( C1 = 1:2, C2 = 1:2) %.% group_by(C1)
>
> d <- DF[ , names(DF) ]
> str(d)
Classes ‘grouped_df’, ‘tbl_df’, ‘tbl’ and 'data.frame': 2 obs. of  2 variables:
 $ C1: int  1 2
 $ C2: int  1 2
>
> summarise(d, x = mean(C2) )

 *** caught segfault ***
address 0x0, cause 'memory not mapped'

Traceback:
 1: .Call("dplyr_summarise_impl", PACKAGE = "dplyr", df, args, env)
 2: summarise_impl(.data, named_dots(...), environment())
 3: summarise.tbl_df(d, x = mean(C2))
 4: summarise(d, x = mean(C2))

The internal code is now more strict at deciding that an object is a grouped data frame, instead of just relying on the class.

[statsandwich]

Cool - makes sense. Thanks for the explanation.