Multiple tests in test-run_now.R are failing; we didn't catch this because they're disabled on ci.
In test-run_now.R, there are tests that test what happens if a callback:
throws an error while run_now is called
is interrupted while run_now is called
Tests in the second category apparently don't work on CRAN or CI environments. But tests in the first should be fine, and they've started failing since this commit in Rcpp (between 1.0.9 and 1.0.10, in 2022):
The problem is that our invoke_wrapped() logic assumes that Rcpp's unwind protect feature is not in use, and the default became to use it. With unwind protect, the right thing to do is almost to just call invoke() directly and not use invoke_wrapped() at all. I say almost, because this works great for errors but not for interrupts--the interrupts somehow aren't caught by tryCatch(..., interrupt = ...) so the tests abruptly end.
Multiple tests in
test-run_now.Rare failing; we didn't catch this because they're disabled on ci.In
test-run_now.R, there are tests that test what happens if a callback:run_nowis calledrun_nowis calledTests in the second category apparently don't work on CRAN or CI environments. But tests in the first should be fine, and they've started failing since this commit in Rcpp (between 1.0.9 and 1.0.10, in 2022):
https://github.com/RcppCore/Rcpp/commit/d389a8a9540690fdc1839507abe95fdfdee8acd7
The problem is that our
invoke_wrapped()logic assumes that Rcpp's unwind protect feature is not in use, and the default became to use it. With unwind protect, the right thing to do is almost to just callinvoke()directly and not useinvoke_wrapped()at all. I say almost, because this works great for errors but not for interrupts--the interrupts somehow aren't caught bytryCatch(..., interrupt = ...)so the tests abruptly end.