initial_rowno in expand_cat_json() is wrong if by=F, JSON empty, and na.rm = T

[berg-michael]

Hi,

Not sure if this is a bug or intended behavior. The JSON in the second row of the input datatable is empty. If I expand the JSON with by = F and na.rm = T, the initial_rowno variable for rows 3 and 4 of the output is 2, when it should be 3. If I set na.rm = F, it becomes 3.

Obviously this issue can be avoided entirely by setting na.rm = F. Maybe it should be obvious to me why this behavior occurs, but it confused me so I thought I'd bring it up.

Thanks for all your work on this package, by the way!

patterns <- data.table::data.table(state_fips = c(1,2,3),
                                   cat_origin = c('{"a": "2", "b": "3"}',
                                                  '{}',
                                                  '{"a": "4", "b": "5"}'))
> patterns
   state_fips           cat_origin
1:          1 {"a": "2", "b": "3"}
2:          2                   {}
3:          3 {"a": "4", "b": "5"}
> 
expand_cat_json(
  patterns,
  'cat_origin',
  'index',
  by = F,
  na.rm = T
)
   initial_rowno cat_origin index
1:             1          2     a
2:             1          3     b
3:             2          4     a
4:             2          5     b
expand_cat_json(
  patterns,
  'cat_origin',
  'index',
  by = F,
  na.rm = F
)
   initial_rowno cat_origin index
1:             1          2     a
2:             1          3     b
3:             3          4     a
4:             3          5     b