group_by with a grouping variable with missing data causes R to crash

[jeromyanglim]

It seems that group_by with missing data on the grouping variable causes R to crash

library(dplyr)
library(MASS)

names(survey) <- tolower(names(survey))
group_by(na.omit(survey), m.i) #works fine
group_by(survey), m.i) # causes r to crash

 *** caught segfault ***
address 0xfffffffd0214f288, cause 'memory not mapped'

Traceback:
 1: .Call("dplyr_grouped_df_impl", PACKAGE = "dplyr", data, symbols,     drop)
 2: grouped_df_impl(data, unname(vars), drop)
 3: grouped_df(x, value)
 4: regroup.data.frame(x, new_groups)
 5: regroup(x, new_groups)
 6: group_by(survey, m.i)

Possible actions:
1: abort (with core dump, if enabled)
2: normal R exit
3: exit R without saving workspace
4: exit R saving workspace

[hadley]

Ooops, I thought we'd tested that case already.

[romainfrancois]

More precisely, this is because m.i is a factor and FactorVisitor does not take that case into account.
https://github.com/hadley/dplyr/blob/master/inst/include/dplyr/VectorVisitorImpl.h#L153

What should happen ? Should we have one group related to all NA from m.i ?

[hadley]

Right, NAs should just turn into another group (which by convention, would be the last group)