count() retains grouping, but should not

[krlmlr]

I believe the implementation of group_vars.tbl_lazy() needs a tweak.

library(tidyverse)
library(dbplyr)
#> 
#> Attaching package: 'dbplyr'
#> The following objects are masked from 'package:dplyr':
#> 
#>     ident, sql

memdb_frame(a = 1:3, b = 1:3) %>%
  count(a, b) %>%
  group_vars()
#> [1] "a"

tibble(a = 1:3, b = 1:3) %>%
  count(a, b) %>%
  group_vars()
#> character(0)

^{Created on 2019-08-13 by the reprex package (v0.3.0)}

[hadley]

Hmmm, I think the fix should be to provide a count method as below, but count isn't a generic?

count.tbl_lazy <- function(x, ..., wt = NULL, sort = FALSE, name = NULL) {
  if (!missing(...)) {
    out <- group_by(x, ..., .add = TRUE, .drop = .drop)
  } else {
    out <- x
  }
  out <- dplyr::tally(out, wt = !!enquo(wt), sort = sort, name = name)
  out <- group_by(out, !!!group_vars(x))
  out
}

[krlmlr]

Would it be too bad to add the group_by(out, !!!group_vars(x)) to dplyr::count() without introducing a generic?

[hadley]

@krlmlr we keep patching count() and it keeps breaking in surprising ways.