slice_* function issue with data masking

[bairdj]

There is an issue when slice_* calls are translated to SQL when the order_by parameter uses .data$ notation rather than a bare symbol name. The generated SQL erroneously prefixes .data to the order by clause, which causes the query to fail.

# Expected behaviour (using local df)
mtcars %>%
  group_by(.data$cyl) %>%
  slice_max(.data$mpg)

# Returns 1 row per group, ordered by MPG

# Observed behaviour (fails)
rem_mtcars <- copy_to(connection, mtcars)
rem_mtcars %>%
  group_by(.data$cyl) %>%
  slice_max(.data$mpg)

# This works
rem_mtcars %>%
  group_by(.data$cyl) %>%
  slice_max(mpg)

An example of the generated SQL is

ROW_NUMBER() OVER (PARTITION BY "cyl" ORDER BY ".data"."mpg" DESC)

The translation works fine when using arrange, it is only for the window functions I have observed the issue.

I've tested this with Redshift and SQLite with the same output.