Failing package test on single-core machines

[HenrikBengtsson]

Hi, I'm running conservative revdep checks on parallelly and future, where I check if they work on single-core machines, which might be the case on cloud CI services, free RStudio Cloud accounts, etc. I use export R_PARALLELLY_AVAILABLE_CORES=1 to emulate this. Doing this, I spotted greta failing with:

* checking tests ...

  Running ‘spelling.R’
  Running ‘testthat.R’
 ERROR
Running the tests in ‘tests/testthat.R’ failed.
Last 50 lines of output:
  x Initialising python and checking dependencies, this may take a moment. ... ...
  
  i Initialising python and checking dependencies, this may take a moment.
  x Initialising python and checking dependencies, this may take a moment. ... ...
  
...
   4. │     ├─testthat .capture(...)
   5. │     │ └─base::withCallingHandlers(...)
   6. │     └─rlang::eval_bare(quo_get_expr(.quo), quo_get_env(.quo))
   7. └─greta:::check_future_plan()
  
  [ FAIL 1 | WARN 0 | SKIP 251 | PASS 4 ]
  Deleting unused snapshots:
  • greta-sitrep.md
  Error: Test failures
  Execution halted

Source: https://github.com/HenrikBengtsson/parallelly/blob/71e60044d1b6daf83f66f424ff380a004147155b/revdep/R_PARALLELLY_AVAILABLECORES_SYSTEM%3D1/problems.md#greta

I haven't tracked this down further, but it could be an assumption that plan(multisession) will produce multisession futures, but if availableCores() == 1, then they'll become sequential futures.

[njtierney]

Thanks for this, @HenrikBengtsson ! I'm not really sure how to track this test down, do you know which test in particular fails?

[HenrikBengtsson]

Sorry, I don't see more then the above either*.

You should be able to reproduce it by setting environment variable:

R_PARALLELLY_AVAILABLE_CORES=1

before loading parallelly the first time, i.e. before running the tests.

(*) This is actually the main reason why I still don't use modern unit test frameworks like testthat, RUnit, tidytest myself. With them, R CMD check will only report on one giant test. With individual tests/*.R test scripts, you see exactly which test scripts passed and which failed. But there's hope; markvanderloo/tinytest#78

[njtierney]

I've tried to create a GH action to add this environment variable to R CMD check, this felt like a somewhat decent way to check that across multiple systems? Specifically, added this code to the R-CMD-check.yml

      - name: Check on single core machine
        if: runner.os != 'Windows'
        env:
          R_PARALLELLY_AVAILABLE_CORES: 1
        run: rcmdcheck::rcmdcheck(args = c("--no-manual", "--as-cran", "--no-multiarch"))
        shell: Rscript {0}

[HenrikBengtsson]

That looks correct to me.

[njtierney]

OK awesome - I think I'm going to close this issue with PR #538 then, as all tests are passing now :)