TESTS: expect_true(inherits(plan(), 'sequential')) -> expect_true(identical(plan(), oplan))

[HenrikBengtsson]

Hi, I'm running reverse package dependency checks on future. I do this with the default settings as well as with env var R_FUTURE_PLAN=multisession, which will force the default future plan to be multisession rather than sequential. This will further validate that all packages work when running in parallel, e.g. all global variables are properly exported, etc.

This results in errors when checking your package because some of your tests assert at the end that sequential is still set. You can reproduce this yourself by running:

export R_FUTURE_PLAN=multisession
R CMD check codalm_0.1.0.tar.gz
...
* checking tests ... ERROR
  Running ‘testthat.R’
Running the tests in ‘tests/testthat.R’ failed.
Complete output:
  > library(testthat)
  > library(codalm)
  > 
  > test_check("codalm")
  ── 1. Failure: bootstrap CI works with sequential evaluation (@test-codalm_ci.R#
  inherits(plan(), "sequential") isn't true.
  
  ── 2. Failure: bootstrap CI works with multisession evaluation (@test-codalm_ci.
  inherits(plan(), "sequential") isn't true.
  
  ── 3. Failure: independence test works with sequential evaluation (@test-indepen
  inherits(plan(), "sequential") isn't true.
  
  ── 4. Failure: independence test works with multisession evaluation (@test-indep
  inherits(plan(), "sequential") isn't true.
  
  ══ testthat results  ═══════════════════════════════════════════════════════════
  [ OK: 22 | SKIPPED: 0 | WARNINGS: 1 | FAILED: 4 ]
  1. Failure: bootstrap CI works with sequential evaluation (@test-codalm_ci.R#20) 
  2. Failure: bootstrap CI works with multisession evaluation (@test-codalm_ci.R#42) 
  3. Failure: independence test works with sequential evaluation (@test-independence_test.R#9) 
  4. Failure: independence test works with multisession evaluation (@test-independence_test.R#19) 
  
  Error: testthat unit tests failed
  Execution halted

Could you please update to check something like:

oplan <- plan()
...
expect_true(identical(plan(), oplan))

[HenrikBengtsson]

Actually, it should be:

oplan <- plan("list")
...
expect_true(identical(plan(), oplan))

[jfiksel]

Hi Henrik, thank you for taking the time to let us know about this issue. However, even after installing the version of the future package on your GitHub (version 1.19.1), just running

library(future)
oplan <- plan("list")
identical(plan(), oplan)

returns FALSE, while running

library(future)
oplan <- plan()
identical(plan(), oplan)

returns TRUE. So it seems like your most recent comment would cause an error in the testing script, but let me know if this seems wrong.

[HenrikBengtsson]

My bad (again), should be:

library(future)
oplan <- plan("list")
identical(plan("list"), oplan)

[jfiksel]

Excellent, this has been fixed with commit ac0eaa9. Thanks for your help!

[HenrikBengtsson]

Almost. Right now you're basically testing identical(plan("list"), plan("list")), no different than testing identical(1, 1). You need to record:

oplan <- plan("list")

at the beginning, before you call your functions that set plan() internally. You can test that it works by setting the environment variable R_FUTURE_PLAN=multisession before running your tests.

[jfiksel]

Ah, got it. Ok this should now actually be fixed with 8b48c71