NULL in Mutate as result of expression

[aaronjg]

If a column is set to NULL in mutate it is deleted. However, this does not work if mutate is returned as a result of an expression. It would also be nice if you could use NULL in summarize to exclude a column as well, but maybe that should go into a new feature request?

> df <- data.frame(a=5,b=6) 
> df %>% mutate(a= NULL)
  b
1 6
> z <- NULL
> df %>% mutate(a=z)
Error: Unsupported type NILSXP for column "a

Previous discussion of the NULL feature.
#462
8104dd6

[krlmlr]

Thanks. This is by design, in particular because for a grouped mutate() it would be unclear what to do if the mutator returns NULL for some groups only. For summarize() we don't seem to need this, because the result includes only the grouping variables and those that are explicitly created.

[hadley]

Don't we have another open issue about this?

[krlmlr]

We decided to improve error messages only, but apparently we did this only for the ungrouped case... #2187

[krlmlr]

I'm now getting:

library(dplyr)
z <- NULL
data.frame(a=5,b=6) %>% group_by(b) %>% mutate(a=z)
#> Error in mutate_impl(.data, dots): Column `a` is of unsupported type NULL

Same for ungrouped data.

@hadley: Do we need to add a special-case for the wording of the error message here?

[hadley]

The error message is ok for now. I'd rather make it possible to remove the column by setting it to NULL.

[krlmlr]

We don't remove because it's unclear what to do in a grouped call when it's removed only for some groups:

tibble(a = 1:3, b = 4:6) %>%
  group_by(a) %>%
  mutate(b = if (a < 2) NULL else b)

[hadley]

Right, but that's mostly an artificial example, and could presumably be handled in the same way as

tibble(a = 1:3, b = 4:6) %>%
  group_by(a) %>%
  mutate(b = if (a < 2) 1 else "b")

[aaronjg]

But similar issues can happen with differing group lengths, and in that case, an error is thrown only when the lengths are invalid:

> tibble(a = 1:3, b = 4:6) %>% group_by(a) %>% mutate(b = if (a < 2) rep(1,b) else b)
Error in mutate_impl(.data, dots) : 
   Column `b` must be length 1 (the group size), not 4