How is use_first_valid_of different from dplyr's coalesce?

[rgknight]

Would be helpful to have an explanation in the help file, if it is indeed different.

https://github.com/hadley/dplyr/blob/master/R/coalesce.R

[sfirke]

Great catch. It's not that different :-( I didn't know about dplyr::coalesce.

I compared them and think they are close enough that use_first_valid_of should be deprecated and phased out, with a message to use dplyr::coalesce instead. There are two features in use_first_valid_of that are missing in coalesce:

1. If there are any type mismatches in the input vectors, the output is coerced to a character vector (with a message). In coalesce, the operation fails with a warning. Stylistic difference, and okay with me.
2. Factors are coerced to character in use_first_valid_of, while coalesce fails:

> x <- factor(c("a", NA))
> y <- factor(c("b", "c"))
> use_first_valid_of(x, y)
[1] "a" "c"
Warning message:
In use_first_valid_of(x, y) :
  Input vectors are of class 'factor' and will be coerced to class 'character' before combining
> coalesce(x, y)
[1] a    <NA>
Levels: a
Warning message:
In `[<-.factor`(`*tmp*`, i, value = 2L) :
  invalid factor level, NA generated

Course of action:

• Deprecate use_first_valid_of with message referring to coalesce
• Post dplyr issue identifying issue 2 above; acceptable if coalesce can't handle factors but in that case it should fail with an error message, not return the current output, IMO.

[rgknight]

I haven't looked too closely at convert_to_NA, but it may also be a candidate for deprecation with a pointer to dplyr's na_if?

[rgknight]

I submitted an issue on dplyr for this tidyverse/dplyr#2254

[sfirke]

Updating the vignette to remove the deprecated convert_to_NA and I kind of want to put it back after reading this discussion: tidyverse/dplyr#1972...