Use withr to not change global state

[strengejacke]

Fixes easystats/easystats#388

[codecov]

Codecov Report

Merging #829 (26503f5) into main (919ac84) will not change coverage.
Report is 1 commits behind head on main.
The diff coverage is n/a.

❗ Current head 26503f5 differs from pull request most recent head 4914275. Consider uploading reports for the commit 4914275 to get more accurate results

@@           Coverage Diff           @@
##             main     #829   +/-   ##
=======================================
  Coverage   55.04%   55.04%           
=======================================
  Files         124      124           
  Lines       15388    15388           
=======================================
  Hits         8470     8470           
  Misses       6918     6918           

[strengejacke]

@IndrajeetPatil Can I add multiple test_that() code chunks into one withr::*() call?
Like:

withr::with_environment(
  new.env(),
  m <- lm(...)
  test_that("1", {
    ...
  })
  test_that("2", {
    ...
  })
  test_that("3", {
    ...
  })
)

[IndrajeetPatil]

Yes, but IINM you'd need to do something like this:

withr::with_environment(
  new.env(),
  code = {
    m <- lm(...)
    test_that("1", {
      ...
    })
    test_that("2", {
      ...
    })
    test_that("3", {
      ...
    })
  }
)

[strengejacke]

Cool! This should resolve our scoping issues, and probably also the random test order thing.

[IndrajeetPatil]

You could do that, or maybe a better approach is to create a new environment in each test and test the expectations within this scope.

For example, see https://github.com/r-lib/withr/blob/0c5254ebc74590e80cc0056303d74b049b943920/tests/testthat/test-namespace.R#L70-L87

[strengejacke]

You could do that, or maybe a better approach is to create a new environment in each test and test the expectations within this scope.

This is what I would normally do, but if you look at the test structure of insight, e.g. https://github.com/easystats/insight/blob/main/tests/testthat/test-lm.R, I would need to re-fit the model every time again for each test then (in those cases where we use <<-, which we want to avoid).

[IndrajeetPatil]

Maybe add back the removed "random test order" check?

[strengejacke]

I think that check should work once we addressed the scoping issues using withr, so I'll do that after the withr-work is done.