Bug in assert messages (1h)

[slamao]

Reproducible example:

adae <- tibble::tribble(
  ~AEENDTC, ~DTHDT, ~DCUTDT,
  "2020-12", "2020-12-06", "2020-12-24",
  "2020-11", "2020-12-06", "2020-12-24"
)

derive_vars_dtm(
  adae,
  dtc = AEENDTC,
  new_vars_prefix = "AEN",
  date_imputation = "last",
  time_imputation = "last",
  max_dates = list(DTHDT, DCUTDT)
)

Since the DTHDT and DCUTDT variables are characters (not dates), an error message based on on_failure(is_date) should have been shown: https://github.com/Roche-GSK/admiral/blob/devel/R/assertions.R#L498
However, the error message seems not to be correctly defined according to this picture (see the "bad error message").

[thomas-neitmann]

Thanks for spotting this!

[thomas-neitmann]

For some reason attr(res, "msg") returns a character vector of length 2 which causes the issue when calling stop().

[slamao]

I had a look and the problem seems to be as follows:
eval(call$arg, envir = env) in on_failure(is_date) evaluates to 2 element vector. If adae had 3 rows, then this would be a 3 element vector, because argument max_date (which is DTHDT) is a 3 element vector.
I think that a possible fix could be to update eval(call$arg, envir = env) by using paste with collapse argument (and maybe using head as well).

The reason that we are getting this error is because of the design of the is_date function:

assert_that(is_date(c('a', 'b'))) # shows bad error message - wrong
assert_that(is_timeunit(c('a', 'b'))) # shows error from assert_that: "Error: assert_that: length of assertion is not 1" - correct

is_date always returns length 1 vector, whereas is_timeunit returns vector of the same length as its input. When this length is > 1, then this is caught by assert_that. But for is_date, this will never happen (and then the error propagates into eval(call$arg, envir = env). Maybe we could also suggest this to the assertthat package authors ...