Should quos_auto_name() use quo_name()?

[lionel-]

And should expr_name() deparse any kind of object?

This is about suboptimal error messages like:

grouped_mean <- function(data, summary_var, ...) {
  summary_var <- enquo(summary_var)
  group_vars <- enquos(...)

  data %>%
    group_by(!!group_vars)
}

grouped_mean(mtcars, disp, cyl, am)
#> Error in mutate_impl(.data, dots) :
#>   Column `structure(list(~cyl, ~am), .Names = c("", ""), class = "quosures")` must be length 32 (the number of rows) or one, not 2

Caused by quos_auto_name() that uses expr_text()

x <- quos(cyl, am)
quos_auto_name(quos(!!x))
#> $`structure(list(~cyl, ~am), class = "quosures", .Names = c("", \n""))`
#> <quosure>
#>   expr: ^<S3: quosures>
#>   env:  empty

expr_text(x)
#> [1] "structure(list(~cyl, ~am), class = \"quosures\", .Names = c(\"\", \n\"\"))"

It doesn't seem to make sense to use a multi-line deparser to give default names to expressions? Then we should use quo_name() / expr_name(). However they don't support literal objects:

expr_name(x)
#> Error: `expr` must quote a symbol, scalar, or call

I think objects should be deparsed using the pillar sum type. Would that make sense? Then the error message above would become:

Column `<S3: quosures>` must be length 32 (the number of rows) or one, not 2

which makes more sense.

[lionel-]

This also looks suboptimal for the purpose of creating default column names:

complex_var <- quote(mean(height, na.rm = TRUE, trim = 0.2))
quo_name(complex_var)
#> [1] "mean(height, na.rm = TRUE, trim = 0.2)"

tibble::tibble(mean(1:10, na.rm = TRUE, trim = 0.2))
#> # A tibble: 1 x 1
#>   `mean(1:10, na.rm = TRUE, trim = 0.2)`
#>                                    <dbl>
#> 1                                    5.5

It probably makes sense to deparse the first argument and collapse the rest:

quo_name(complex_var)
#> [1] "mean(height, ...)"

Then might need to take the max number of arguments as an option, e.g. for lobstr::cst() and rlang::trace_back().

[lionel-]

Then might need to take the max number of arguments as an option, e.g. for lobstr::cst() and rlang::trace_back().

Or use a different _line variant, cf #636.