[R-pkg-devel] Issue with R Package on CRAN - OpenMP and clang17

[Ivan Krylov]

В Thu, 2 Nov 2023 10:38:56 +0100 (CET)
Romain Pierlot <romain.pierlot using u-bordeaux.fr> пишет:

> Does a compiler bug imply that the issue doesn't come from from the
> code?

Yes. I'm sure (corrections welcome!) that it's not an error to use a
Fortran variable before an OpenMP parallel block that also uses it for
a reduction operation.

One more OpenMP-related problem that may still bite you is having to
limit the number of threads. Since there's currently no centralised way
of regulating all kinds of parallelism used in a given R session [1],
you will have to use no more than two threads by default [2], although
you are always welcome to start more if the user explicitly passes such
an option to one of your functions.

(I'm trying to help with this [3], but it may take a long while to
implement properly and even longer to ensure that this is the right
solution from both technical and social viewpoints.)

-- 
Best regards,
Ivan

[1] https://stat.ethz.ch/pipermail/r-package-devel/2023q3/009513.html

[2] https://cran.r-project.org/web/packages/policies.html

[3] https://stat.ethz.ch/pipermail/r-package-devel/2023q4/009769.html