Email from Uwe Ligges (5 Oct 2019) suggests package check time exceeded 10 minutes (13 min on r-devel-windows). I have correspondingly reduced numbers of tests run, observing a local reduction to < 50% of former test time.
Other than that, and as stated on the immediately prior submission, I finally managed to reproduce the previous UBSCAN and valgrind errors observed by Brian Ripley on 20 Sept 2019. This submission definitely fixes, with both AddressSanitizer and valgrind on r-devel returning only the by-now usual loss of 2-3000 bytes due to TBB code bundled in RcppParallel (see below). I humbly apologise for any inconvenience which may have arisen during my previously unsuccessful attempts to resolve this issue.
This submissions generates the following NOTES on some systems:
- Two possibly invalid URLs: one for a valid vignette reference, due to jstor preventing automated queries; and one for the DOI for the paper associated with this pacakge, which is also valid (https://doi.org/10.32866/6945, with DOI simply given as 'doi = "10.32866/6945"').
- "GNU make is a SystemRequirements", which is unavoidable because of the need to remove compiled object files in src sub-directories.
Other than the above, this submission generates no additional notes, and no warnings on:
- Ubuntu 16.04 (on
travis-ci): R-release, R-devel - win-builder (R-release, R-devel, R-oldrelease)
Testing with "valgrind --tool=memcheck --leak-check=full" still reveals one potential memory leak of around 2,000 bytes, which must at this stage be considered an enduring issue. This is due to RcppParallel and not code within this submission; see RcppCore/RcppParallel#81