englue-ing NULL yields uninformative error

[hadley]

library(tidyverse)

histogram <- function(df, var, binwidth = NULL) {
  label <- rlang::englue("A histogram of {{var}} with binwidth {binwidth}")
  
  df |> 
    ggplot(aes({{ var }})) + 
    geom_histogram(binwidth = binwidth) + 
    labs(title = label)
}

diamonds |> histogram(carat)
#> Error in `rlang::list2()`:
#> ! The LHS of `:=` must be a string or a symbol

#> Backtrace:
#>     ▆
#>  1. ├─global histogram(diamonds, carat)
#>  2. │ ├─rlang::englue("A histogram of {{var}} with binwidth {binwidth}")
#>  3. │ │ └─rlang::inject(rlang::list2(!!expr), caller_env())
#>  4. │ └─rlang::list2(...)
#>  5. └─rlang::abort(message = message)

^{Created on 2022-09-26 with reprex v2.0.2}

[lionel-]

This is tricky because this behaviour is from recycling rules of glue():

unclass(glue::glue("foo {NULL}"))
#> character(0)

We could detect a change in size and throw an error about sizes of interpolated elements causing recycling.

[hadley]

Oh yeah: tidyverse/glue#246

[lionel-]

@hadley With the recent refactoring of englue() the error is gone:

fn <- function(x) {
  englue("{{ x }} {NULL}")
}

fn(foo)
#> character(0)

But I'm thinking we should enforce a size 1 return from englue()?

[hadley]

Yeah, I think that's a good idea.