Error in get(i, parent.frame()) : object 'weights' not found

[fg-sempre]

Error description

When I try to use non-standard evaluation with box the following error occurs.

R/mod.R

box::use(
  blme,
  rlang,
)

#' @export
fit_w <- function(tbl, resp, w_form) {
  form <- paste0(resp, " ~ disp + (1|cyl)")

  blme$blmer(
    form,
    weights = !!as.name(w_form),
    data = tbl
  ) |>
    rlang$quo() |>
    rlang$eval_tidy()
}

main.R

box::use(R / mod)

mod$fit_w(mtcars, "mpg", "hp")

Error in get(i, parent.frame()) : object 'weights' not found

The same happens with lme4::lmer. When I replace box::use(R/mod) with source("R/mod.R") everything works fine.

R version

platform       x86_64-pc-linux-gnu         
arch           x86_64                      
os             linux-gnu                   
system         x86_64, linux-gnu           
status                                     
major          4                           
minor          3.3                         
year           2024                        
month          02                          
day            29                          
svn rev        86002                       
language       R                           
version.string R version 4.3.3 (2024-02-29)
nickname       Angel Food Cake

‘box’ version

'1.2.0'

[klmr]

This is a bug in ‘lme4’, not in ‘box’.

In fact, the following code exhibits the same bug and doesn’t use ‘box’ at all (nor ‘rlang’, or any kind of NSE except for the formula):

e <- new.env(parent = baseenv())

local(
  fit_w <- function(tbl) {
    form <- mpg ~ disp + (1 | cyl)
    lme4::lFormula(form, data = tbl, weights = tbl$hp)
  },
  envir = e
)

e$fit_w(mtcars)

lFormula() (which is called inside blmer()) evaluates (essentially) get("weights", parent.frame()), which is nonsense. It only works by pure coincidence when executed in the global environment (instead of a custom environment), since there’s a function stats::weights, which lFormula() is (erroneously!) finding. — In other words, the strictness of ‘box’ caught a bug in ‘lme4’.

Until this is fixed in ‘lme4’ you can work around the bug by creating a local variable weights with the desired value inside fit_w.

[fg-sempre]

Thanks. box is very good, indeed.