Can we relax the requirement for inputs data frames to have at least one row?

[ddsjoberg]

I think with the new ARD framework, we may be able to get away with removing the requirement.

See note here: #2006

[ddsjoberg]

I did a small test where I commented out the check that the data frame passed to tbl_summary() has at least one row, and this is the table generated. I think we can fairly easy relax this requirement in many of the functions.

devtools::load_all(path = "/Users/sjobergd/Documents/GitHub/gtsummary")
#> ℹ Loading gtsummary
tbl_summary(trial[0,], include = grade) |> 
  as_kable()

Characteristic	N = 0
Grade
I	0 (NA%)
II	0 (NA%)
III	0 (NA%)

^{Created on 2025-02-16 with reprex v2.1.1}

[Melkiades]

Many users often want to have a null report when the data has 0 rows. In {chevron} there is an example of that. In general, I would allow something like the above example (i.e. keeping the expected structure with empty values). I would rely on the users to call the null report if they want. Wdyt?

We need to test the trial[0,] case also in sorting and other post-processing functions that may break if there are no rows in the data (if the structure is preserved as in the above, I think it should be quite solid, but I would test it)