Support for lazy tibbles

[mgirlich]

Lazy tibbles currently have two issues with eval_select():

1. They are not a vector so one has to simulate a "dummy" tibble with the appropriate column names, e.g with dtplyr:::simulate_vars().
2. The column types are not generally known which is why allow_predicates would make sense: Optionally suppress use of predicates #226

The second point is still a bit annoying when trying to do some general work in a generic, e.g. in tidyr::nest() unnamed dots are tried to be named

tidyr::nest <- function (.data, ..., .names_sep = NULL, .key = deprecated()) {
  cols <- enquos(...)
  if (any(names2(cols) == "")) {
    col_names <- names(tidyselect::eval_select(expr(c(...)), .data))
    ...
  }
  UseMethod("nest")
}

So, it would be necessary to check the type to figure out whether allow_predicate should be TRUE or FALSE. It might make sense if tidyselect() could handle this.

Also see tidyverse/tidyr#1134

[hadley]

Note that there are at least two lazy tibble variants, one in dbplyr and the other in dtplyr.

[hadley]

One possibility would be to provide a tidyselect_proxy generic. It could return list(data, supports_predicates).

[hadley]

Better to do tidyselect_proxy() + tidyselect_supports_predicates().