NSE for `trending:::fit.list()` is broken

[TimTaylor]

The NSE for trending:::fit.list() is broken when used programatically (i.e within functions). I think it is something to do with passing the correct environment around for evaluation but it's tricky.

library(trending)

# this works
x = rnorm(100, mean = 0)
y = rpois(n = 100, lambda = exp(1.5 + 0.5*x))
dummy <- data.frame(x = x, y = y)
poisson_model <- glm_model(y ~ x , family = poisson)

# this works
fit(poisson_model, data = dummy)
#> <trending_fit_tbl> 1 x 3
#>   result warnings errors
#>   <list> <list>   <list>
#> 1 <glm>  <NULL>   <NULL>

# this works
fun1 <- function(xx) fit(poisson_model, data = xx)
fun1(dummy)
#> <trending_fit_tbl> 1 x 3
#>   result warnings errors
#>   <list> <list>   <list>
#> 1 <glm>  <NULL>   <NULL>

# this works
fit(list(pm = poisson_model), data = dummy)
#> <trending_fit_tbl> 1 x 4
#>   model_name result       warnings     errors      
#>   <chr>      <named list> <named list> <named list>
#> 1 pm         <glm>        <NULL>       <NULL>

# this does not work
fun2 <- function(xx) fit(list(pm = poisson_model), data = xx)
(out <- fun2(dummy))
#> <trending_fit_tbl> 1 x 4
#>   model_name result       warnings     errors      
#>   <chr>      <named list> <named list> <named list>
#> 1 pm         <NULL>       <NULL>       <chr [1]>
out$errors
#> $pm
#> [1] "object 'xx' not found"

^{Created on 2023-04-14 with reprex v2.0.2}

[TimTaylor]

This means fixing reconhub/trendbreaker#56 is not trivial