repair_call() uninformative error when `x` isn't a parsnip model

[EmilHvitfeldt]

The documentation for repair_call() says that x should be "A fitted parsnip model." But if you didn't see that (like I did) it produces errors when applied to non-parsnip fits.

I have two thoughts

1. Have repair_call() explicitly check if x is of class model_fit and error accordingly.
2. Expand the use of repair_call() to work with workflow objects, and possibly the raw engine fit.

Right now it is a little awkward because you need to call two extract_fit_*() functions

wf_fit %>%
  extract_fit_parsnip() %>%
  repair_call(mtcars) %>%
  extract_fit_engine()

library(tidymodels)

lm_spec <- linear_reg() %>%
  set_mode("regression") %>%
  set_engine("lm")

lm_wf <- workflow() %>% 
    add_formula(mpg ~ .) %>% 
  add_model(lm_spec)

wf_fit <- fit(lm_wf, data = mtcars)
parsnip_fit <- extract_fit_parsnip(wf_fit)
engine_fit <- extract_fit_engine(wf_fit)

repair_call(parsnip_fit, mtcars)
#> parsnip model object
#> 
#> Fit time:  4ms 
#> 
#> Call:
#> stats::lm(formula = ..y ~ ., data = mtcars)
#> 
#> Coefficients:
#> (Intercept)          cyl         disp           hp         drat           wt  
#>    12.30337     -0.11144      0.01334     -0.02148      0.78711     -3.71530  
#>        qsec           vs           am         gear         carb  
#>     0.82104      0.31776      2.52023      0.65541     -0.19942
repair_call(wf_fit, mtcars)
#> Error: No `call` object to modify.
repair_call(engine_fit, mtcars)
#> Error: No `call` object to modify.

^{Created on 2021-11-12 by the reprex package (v2.0.1)}

[juliasilge]

We talked a fair amount of extending this to a fitted workflow, and put some ideas in tidymodels/workflows#61, but came to the conclusion that it is not very doable.