left_join() over empty grouped data frame produces error in dplyr 0.7.5

[javierluraschi]

Found out that this works in dplyr 0.7.4:

library(dplyr)
x <- data_frame(name = c("a", "b"), type = c("integer", "string"))
x %>% group_by(name) %>%
    select(name) %>%
    setdiff(select(x, name)) %>%
    left_join(x, by = "name")

# A tibble: 0 x 2
# Groups:   [?]
# ... with 2 variables: name <chr>, type <chr>

but not in dplyr 0.7.5:

Error in grouped_df_impl(out, vars$alias[groups_in_alias], group_drop(x),  : 
  no variables to group by

We found and pushed a fix for sparklyr with sparklyr/sparklyr#1510, but reporting to consider fixing upstream.

[ejmacrae1]

Hi,

I have a similar issue; looks like it's related to yours. The below works in dplyr 0.7.4 but not in 0.7.5:

k <- data.frame(id=letters[1:5], col1=c("ha", "ha", NA, "ha", "ha"), stringsAsFactors=F)
k %>%
  group_by(id) %>%
  summarise(temp_col=first(col1[!is.na(col1)]))

Error is: "Error in summarise_impl(.data, dots) :
Column temp_col must be length 1 (a summary value), not 0"

While the expected output is:

   id temp_col
1  a       ha
2  b       ha
3  c     <NA>
4  d       ha
5  e       ha

[krlmlr]

Is it worth fixing this in v0.7.6?

[ejmacrae1]

In my opinion, yes (unless it's desired behaviour).

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/