last() gives cryptic error message

[t-kalinowski]

Hello!

a small example:

library(dplyr)
n <- 9
x <- c(ifelse(sample(c(T,F), n, T), 1:n, NA ), NA)
df <-  data_frame(x)

last(na.omit(x)) # works

mutate(df, y = last(na.omit(x)))            # doesn't work
# Error: Unsupported vector type language

mutate(df, z = last(x[!is.na(x)]))          # doesn't work
# Error: Unsupported vector type language

mutate(df, y = x %>% na.omit() %>% last() ) # doesn't work
# Error in mutate_impl(.data, dots) : bad value


mutate(df, y = x %>% na.omit %>% last ) # works
last(x[!is.na(x)]) # works

mutate(df, z = last(x)) # "works"

I really love dplyr and am immensely grateful for all the hard work that goes into it!

I'm not savvy enough to even diagnose what's leading to these error messages, but I suspect this is not intended behavior.

[py-b]

Related to this ?
#1980

[krlmlr]

Thanks, confirmed. Would you like to contribute a testthat test?

This looks related to, but probably not the same as #1980.

[dholstius]

Similar behavior exposed via filter().

I would not have thought the two expressions below would yield different outcomes — since identical(6L, c(6L)) is TRUE — but they do:

> filter(mtcars, cyl == last(6L)) # FIXME: returns 0 rows

> filter(mtcars, cyl == last(c(6L))) # FIXME: throws "Unsupported vector type language"

In contrast to last, first yields the expected outcome:

# All of these yield the correct result
> filter(mtcars, cyl == first(6L))
> filter(mtcars, cyl == first(c(6L))) 
> filter(mtcars, cyl == 6L)

Feels like ... maybe an off-by-one in nth?

[dholstius]

Hi @krlmlr,

I'm not good at debugging Rcpp stuff — I only know how to insert stop() statements, load_all(), think, and repeat — but it seems that:

1. Supplying order_by = NULL makes the problem disappear. For example, filter(mtcars, cyl == last(c(6L), order_by = NULL)) produces the correct result, while filter(mtcars, cyl == last(c(6L))) throws the error noted by @t-kalinowski ; and

2. The "easy case" (nargs == 2, line 157 of hybrid_nth.cpp, switch(TYPEOF(data))) only handles INTSXP, REALSXP, STRSXP, and LGLSXP. So it falls through when TYPEOF(data) is language, which happens when order_by is not supplied.

Hope these observations are helpful.

[krlmlr]

This will be resolved by #2143, but that might take some time until it's ready.

[BrazilForever11]

Here is another workaround

mutate(df,y = tail(na.omit(x),1))