`nest()` generic might error for unnamed dots

[mgirlich]

Before calling the method corresponding to the class the nest() generic tries to name any unnamed dots. This doesn't work for lazy tibbles:

library(dtplyr)
library(tidyr)

dt <- lazy_dt(tibble(x = c(1, 2, 1), y = c("a", "a", "b")))
dt %>% nest(y)
#> Error: `x` must be a vector, not a <dtplyr_step_first/dtplyr_step> object.

^{Created on 2021-07-01 by the reprex package (v2.0.0)}

The issue is the following line

col_names <- names(tidyselect::eval_select(expr(c(...)), .data))

The question is how to handle this?

• Should each method implement this functionality itself? Probably not.
• Should nest() get the underlying tibble of .data? But what about predicates (column types)? See Optionally suppress use of predicates r-lib/tidyselect#226.
• Should eval_select() be able to handle this?

[DavisVaughan]

I have a feeling we can resolve this by avoiding eval_select() at this point altogether. I think we just need to do something like expr(c(!!!cols)), and we can use as_label() on that to create a decent message.

tidyr/R/nest.R

Lines 124 to 136 in 89b409d

	if (any(names2(cols) == "")) {
	col_names <- names(tidyselect::eval_select(expr(c(...)), .data))
	cols_expr <- expr(c(!!!syms(col_names)))
	.key <- if (missing(.key)) "data" else as.character(ensym(.key))
	cols <- quos(!!.key := !!cols_expr)
	warn(paste0(
	"All elements of `...` must be named.\n",
	"Did you want `", .key, " = ", expr_text(cols_expr), "`?"
	))
	return(nest(.data, !!!cols))
	}

I'm not sure that would fix the problem for dtplyr though. Eventually in the dtplyr method for nest() you will have to call eval_select() and you'll need access to the column names of the lazy data.table, and I don't think there is any way to access that right now Oh I see there is dtplyr:::simulate_vars() so maybe you can work with that an the eventual allow_predicate = FALSE in tidy select.