[R-pkg-devel] Has something changed regarding chol()?

[Krzysztof Bartoszek]

Hello!
I am trying to update my mvSLOUCH R package and am getting it rejected due to failures on r-devel-linux-x86_64-debian-clang and r-devel-linux-x86_64-debian-gcc . The error is thrown when running the examples : Error in chol.default( Some matrix) :  the leading minor of order 1 is not positive definite.
The same error can be seen in the current version of mvSLOUCH on CRAN: https://cran.r-project.org/web/checks/check_results_mvSLOUCH.html but only for the two Debian flavors not the other ones. I understand what the error means and in principle can get around it by using chol()'s pivot option, i.e. I did this the first time the submission failed, but then it fails with same error on the next call to chol(). But before I go and set pivot=TRUE in all calls to chol() I wanted to ask whether you know what change has taken place regarding chol() and the Debian flavor? All these examples previously ran without error  (set.seed() is run first-so everything should be the same) and moreover the problem is only on the Debian flavor, not on the others. In the new, being rejection submission, I have also used in set.seed(..., kind = "Mersenne-Twister", normal.kind = "Inversion") and ran RNGversion("3.6.1") just in case there were differences here between flavors.

I would be grateful for some pointers.
Best wishes
Krzysztof Bartoszek

Sent with ProtonMail Secure Email.
	[[alternative HTML version deleted]]