is.atomic(NULL) now returns FALSE on R-devel

[yutannihilation]

I found the CI with R-devel runner started to fail with a real error, not the timeout. Actually, the CRAN check now fails with the same error:

     ══ Failed tests ════════════════════════════════════════════════════════════════
     ── Failure ('test-aes-calculated.R:50:3'): make_labels() deprases mappings properly ──
     make_labels(aes(x = NULL)) (`actual`) not identical to list(x = "x") (`expected`).
    
     `actual$x`: "NULL"
     `expected$x`: "x"
     ── Error ('test-aes.R:29:3'): aes_q() & aes_string() preserve explicit NULLs ───
     Error in `FUN(X[[i]], ...)`: Aesthetic must be a one-sided formula, call, name, or constant.
     Backtrace:
     ▆
     1. ├─base::suppressWarnings(expect_equal(aes_q(NULL), aes(NULL))) at test-aes.R:29:2
     2. │ └─base::withCallingHandlers(...)
     3. ├─testthat::expect_equal(aes_q(NULL), aes(NULL))
     4. │ └─testthat::quasi_label(enquo(object), label, arg = "object")
     5. │ └─rlang::eval_bare(expr, quo_get_env(quo))
     6. └─ggplot2::aes_q(NULL)
     7. └─base::lapply(mapping, as_quosure_aes)
     8. └─ggplot2 (local) FUN(X[[i]], ...)
     9. └─cli::cli_abort("Aesthetic must be a one-sided formula, call, name, or constant.")
     10. └─rlang::abort(...)
     ── Failure ('test-aes.R:102:3'): labelling doesn't cause error if aesthetic is NULL ──
     p$labels$x (`actual`) not identical to "x" (`expected`).
    
     `actual`: "NULL"
     `expected`: "x" 

I'm not sure, but I'm suspecting this breaking change NEWS is somehow related.

is.atomic(NULL) now returns FALSE, as NULL is not an atomic vector. Strict back compatibility would replace is.atomic(foo) by (is.null(foo) || is.atomic(foo)) but should happen sparingly only.

This seems commit:

wch/r-source@c97dbd1

[teunbrand]

Thanks for posting this! Thomas was informed that this would change soon in R-devel, so that we can release a hotfix to patch this (see #5437 (comment)). Now that it is implemented, we can start testing solutions.

[yutannihilation]

Oh, sorry, this was the very reason for the patch release! Sorry, I overlooked the comment (and the announcement on the R-devel mailing list). Then, I'm closing this issue in favor of it.