Implement .null argument

[echasnovski]

This PR add .null argument to glue() and glue_data(). As it was suggested in comment, this fixes #84.

[jimhester]

Thanks for working on this!

However the PR is not currently quite the behavior we want.

If .null = NULL it should work like paste0() does with NULL inputs, e.g. they are silently dropped.

paste0("foo", NULL, "bar")
#> [1] "foobar"

If .null = character() it should work how glue does currently, e.g. return a 0 length vector.

I think we have also decided we are going to change the default behavior of glue() to the NULL case, as it makes doing things like glue::glue("foo{bar}", bar = if(x == 1) "y") nicer.

[echasnovski]

Added described behavior but didn't change the default value to NULL, because I am not sure that you want to do this now.

This introduces one peculiar edge case. What should be the output of glue("{NULL}", .null = NULL)? The output of paste0(NULL) is character(), but in this case I would expect the output to be "" because "NULL is inside quotes".

Currently I decided to stick with ".null = NULL emulates paste0()" because the "NULL inside quotes results into """ can be achieved with glue("{NULL}", .null = "").

[echasnovski]

Also a question about a code I stumbled here. Is there a need for length(unnamed_args) < sum(!named) test? I removed this check, and no test got broken.

Git history shows that it was altered for two reasons: in bcfaf72 to account for lazy loading and in 9cb52dd to fix when last expression is NULL. However, lapply() returns output with the same length as input and which(!named) should have the same length as sum(!named) (as named seems to always be logical vector). Am I missing something here?

[jimhester]

It seems the test cases for that PR were not added when it was committed manually.

https://github.com/tidyverse/glue/pull/46/files#diff-dfe1d2c4b81146f1b57d333a16c3436162380561ea424fb6141aa71c978ee97aR245-R246

Could you add them and verify removing the conditional does not break the behavior?

[echasnovski]

Yep, they are present. Removing that check doesn't break anything, both on master and in this branch (as long as .null = character() is default).

I'll remove this check here.

[jimhester]

Thanks for working on this!