`expect_vector` with NULL terminates the test prematurely

[plietar]

With all testthat expect_* functions, the tests keep running even after a failed check.

However I've noticed that this is not the case if the first argument of expect_vector is NULL. Instead a vctrs_error_scalar_type condition is thrown and the test terminates immediately.

For example, the following prints the two failed assertions:

test_that("expect_true", {
  expect_true(FALSE)
  expect_true(FALSE)
})

Whereas the test below aborts after the first line:

test_that("expect_vector", {
  expect_vector(NULL)
  expect_vector(NULL)
})

<vctrs_error_scalar_type/vctrs_error/rlang_error/error/condition>
Error in `expect_vector(NULL)`: `NULL` must be a vector, not `NULL`.
Backtrace:
    ▆
 1. └─testthat::expect_vector(NULL) at ../../foo.R:7:3
 2.   ├─base::tryCatch(...)
 3.   │ └─base (local) tryCatchList(expr, classes, parentenv, handlers)
 4.   │   └─base (local) tryCatchOne(expr, names, parentenv, handlers[[1L]])
 5.   │     └─base (local) doTryCatch(return(expr), name, parentenv, handler)
 6.   └─vctrs::vec_assert(act$val, ptype = ptype, size = size, arg = act$lab)
 7.     └─vctrs:::stop_scalar_type(x, arg, call = call)
 8.       └─vctrs:::stop_vctrs(...)
 9.         └─rlang::abort(message, class = c(class, "vctrs_error"), ..., call = call)

Error:
! Test failed
Run `rlang::last_trace()` to see where the error occurred.

---

vec_assert is documented to throw either a vctrs_error_scalar_type or a vctrs_error_assert, but expect_vector only catches the latter. Catching both kinds should be enough to fix this behaviour I imagine.

[hadley]

I think you're generalising based on a limited set of examples, because of expect_ functions do error:

library(testthat)

expect_match(1)
#> Error in `expect_match()`:
#> ! `object` must be a character vector, not the number 1.
expect_setequal(mean, 1:10)
#> Error in `expect_setequal()`:
#> ! `object` must be a vector, not a function.

^{Created on 2025-08-29 with reprex v2.1.1}

expect_true() and expect_false() don't error, because they're fundamentally tests of equality so can allow any input.

[plietar]

I guess this is a bit of a philosophical distinction, but to me expect_match assumes the argument is a string and checks additional properties about it, whereas (in my mind at least), expect_vector is "well-typed" for any value and makes no prior assumption, and then checks that the value is in fact a vector. I think this is evidenced by the fact that expect_vector can be called with just one argument, where "vectorness" is the property being tested for.

Anyway this is obviously not a big issue and it makes little difference in practice. It just surprised me for a bit.

[hadley]

Oh hmmm, you make a good point.

[hadley]

Since you convinced me, do you want to do a PR to fix the problem? 😄

[plietar]

Sure, will do next week