Using complete with a subset of levels silently drops observations

[gowerc]

Based upon the description of complete its purpose is to add missing observations for specific combinations of variable values. I believe therefore it is not very intuitive for it to silently drop observations if certain levels are not present, for example:

library(tidyr)
library(dplyr)

iris %>% 
    as_data_frame() %>% 
    mutate( Species = as.character(Species)) %>% 
    complete( Species = c("versicolor", "virginica"))

This code silently drops all observations with Species = setosa

I would propose either changing the function to use a full_join() to ensure the original data is preserved or at the very least adding a warning/message so the user is aware that data has been lost.

I am happy to make a PR if people agree this is a beneficial change.

[hadley]

I think the right way to handle this is to union() the supplied value with the existing values. A PR would be greatly appreciated 😄

[Ryo-N7]

Going to work on this for tidy-dev-day.
@batpigandme please add a tidy-dev-day label.

[batpigandme]

@Ryo-N7 All set 👍