[R-pkg-devel] Puzzling error from win-builder

I have a puzzling error from win-builder for a package.

My package is https://github.com/jonathan-g/datafsm, also 
https://cran.r-project.org/web/packages/datafsm. I am preparing a 
bug-fix release to CRAN to fix an error that CRAN recently identified in 
the previous release of this package (see 
https://cran.r-project.org/web/checks/check_results_datafsm.html).

The package passes R CMD check with no errors or warnings, and one note 
(harmless) on win-builder for R-devel 
(https://win-builder.r-project.org/8mp9LN00YVw4/00check.log) and 
R-oldrelease 
(https://win-builder.r-project.org/1K21X458WtYt/00check.log), but for 
R-release, win-builder has two errors in running tests and building a 
vignette (https://win-builder.r-project.org/exK86Fkyir2s/00check.log), 
all of which say the CRAN caret package is not available for Windows x64:

* checking for unstated dependencies in 'tests' ... OK
* checking tests ...
** running tests for arch 'i386' ... [7s] OK
?? Running 'testthat.R' [6s]
** running tests for arch 'x64' ... [6s] ERROR
?? Running 'testthat.R' [6s]
Running the tests in 'tests/testthat.R' failed.
Complete output:
?? > library(testthat)
?? > library(datafsm)
?? >
?? > test_check("datafsm")
?? == Failed tests 
================================================================
?? -- Error (test_mainfunc.R:7:9): evolve_model() returns correct type 
of object --
?? Error: DLL 'caret' not found: maybe not installed for this 
architecture?
?? Backtrace:
?????? x
??? 1. +-datafsm::evolve_model(cdata, cv = FALSE) test_mainfunc.R:7:8
??? 2. \-base::loadNamespace(x)
??? 3.?? \-base::library.dynam(lib, package, package.lib)
?? -- Error (test_mainfunc.R:14:3): evolve_model() returns warnings 
and errors ----
?? Error: DLL 'caret' not found: maybe not installed for this 
architecture?
?? Backtrace:
?????? x
??? 1. +-testthat::expect_warning(evolve_model(cdata, cv = FALSE), 
"did not supply a data.frame") test_mainfunc.R:14:2
??? 2. | \-testthat:::quasi_capture(enquo(object), label, 
capture_warnings)
??? 3. |?? +-testthat:::.capture(...)
??? 4. |?? | \-base::withCallingHandlers(...)
??? 5. |?? \-rlang::eval_bare(quo_get_expr(.quo), quo_get_env(.quo))
??? 6. +-datafsm::evolve_model(cdata, cv = FALSE)
??? 7. \-base::loadNamespace(x)
??? 8.?? \-base::library.dynam(lib, package, package.lib)
?? ?? [ FAIL 2 | WARN 0 | SKIP 0 | PASS 3 ]
?? Error: Test failures
?? Execution halted
* checking for unstated dependencies in vignettes ... OK
* checking package vignettes in 'inst/doc' ... OK
* checking re-building of vignette outputs ... [10s] WARNING
Error(s) in re-building vignettes:
--- re-building 'datafsm_introduction.Rmd' using rmarkdown
Quitting from lines 173-175 (datafsm_introduction.Rmd)
Error: processing vignette 'datafsm_introduction.Rmd' failed with 
diagnostics:
DLL 'caret' not found: maybe not installed for this architecture?
--- failed re-building 'datafsm_introduction.Rmd'

--- re-building 'FRD_vignette.Rmd' using rmarkdown_notangle
--- finished re-building 'FRD_vignette.Rmd'

SUMMARY: processing the following file failed:
?? 'datafsm_introduction.Rmd'

Error: Vignette re-building failed.
Execution halted

My package declares the "caret" package as an import in DESCRIPTION, so 
I am puzzled what this error from win-builder means and how to fix it. 
If R CMD check got this far, and installed the package for testing 
without reporting missing dependencies, why is it suddenly unable to use 
the caret package for testing the package, but only for the x64 
architecture? Is this something I am doing wrong, or is it a bug in 
win-builder?

Is anyone else finding problems with win-builder reporting errors about 
CRAN packages not being available?

For what it's worth, the package passes R CMD check with no errors or 
warnings or notes under GitHub actions (Windows, MacOS, and Ubuntu), on 
rhub (Debian, Fedora, Ubuntu, MacOS, and Windows except for R-devel 
under Windows, where R CMD check fails with a message about BioConductor 
not having binaries for R 4.2, even though my package doesn't use any 
BioConductor packages), and on my local Windows and Ubuntu boxes, so 
this seems to be a problem specific to R-release under win-builder.

If anyone has any suggestions for how I can fix this problem and submit 
the new release of my package to CRAN, I would be grateful.