Some tests failing with devtools::check(), not with devtools::test()

[dimfalk]

Hi everyone,

I'm really not sure whether this is the right place to ask or if this is an issue at all - since I'm currently working on tests for my very first package, I totally might have messed up somewhere unconsciously - but let me try to describe the problem first:

As already addressed in the topic, some of my tests are failing with devtools::check() at the moment. Having read some material on this topic, e.g. at community.rstudio.com and hadley/r-pkgs#483, I checked the mentioned usual suspects like dependencies in DESCRIPTION and namespace issues, but was not able to resolve this yet, unfortunately.

However, this is the repo I am currently working on.

Everything was fine via devtools::check() before a recent commit, where I relocated some of the function calls to be tested from setup.R to the relevant test_that() blocks because it felt like it was the wrong place to run the code (I did not want to repeat these calls at first) and then I picked up the recommendation to "inline everything inside test_that() calls" at r-pkgs.org.

All tests (7) located in test-calc_designstorm.R currently fail with basically the same error:

── Error ('test-calc_designstorm.R:3'): Output class is as expected. ───────────
<vctrs_error_scalar_type/vctrs_error/rlang_error/error/condition>
Error in `vec_size()`: `x` must be a vector, not a <sfc_POLYGON/sfc> object.

Backtrace:
     ▆
  1. ├─kostra2010R::calc_designstorm(kostra_ref, d = 60, tn = 20, type = "EulerII") at test-calc_designstorm.R:3:2
  2. │ ├─dplyr::filter(tiles, INDEX_RC == attr(x, "id"))
  3. │ └─dplyr:::filter.data.frame(tiles, INDEX_RC == attr(x, "id"))
  4. │   ├─dplyr::dplyr_row_slice(.data, loc, preserve = .preserve)
  5. │   └─dplyr:::dplyr_row_slice.data.frame(.data, loc, preserve = .preserve)
  6. │     ├─dplyr::dplyr_reconstruct(vec_slice(data, i), data)
  7. │     │ └─dplyr:::dplyr_new_data_frame(data)
  8. │     │   ├─row.names %||% .row_names_info(x, type = 0L)
  9. │     │   └─base::.row_names_info(x, type = 0L)
 10. │     └─vctrs::vec_slice(data, i)
 11. └─vctrs:::stop_scalar_type(`<fn>`(`<s_POLYGO>`), "x", `<fn>`(vec_size()))
 12.   └─vctrs:::stop_vctrs(...)
 13.     └─rlang::abort(message, class = c(class, "vctrs_error"), ..., call = vctrs_error_call(call))

What I assume to be the potential source for this error is this code block where I filter an sf obejct with dplyr by attributes. Probably the way I call the lazy loaded data in line 131 causes this to fail. However, no error messages pop up with other tests where I also access some of the other datasets. But maybe the way I load my "reference datasets" in setup.R is also not best-practice.

I was also able to find a "workaround" - restoring the state before mentioned commit - and found out that all tests mysteriously pass when I call get_stats() before, either in setup.R or in even in the first test_that() block (c.f. here). Comment this line again to reproduce failing tests.

I can't wrap my head around this since nothing relevant happens by calling get_stats() from the perspective of calc_designstorm(), at least from my naive point of view. Also, since I thought the individual containers are to be regarded as independent environments (?), why does a call in the first container influences all the others resulting in passing the test?

Thank you very much in advance!

[dimfalk]

Fixed by adding import(sf) to NAMESPACE - e.g. via @import sf roxygen tag.