[R-pkg-devel] Compilation used the "non-portable flag"

[Iñaki Ucar]

On Wed, 15 May 2019 at 13:16, Dirk Eddelbuettel <edd using debian.org> wrote:
>
>
> On 15 May 2019 at 11:49, Gábor Csárdi wrote:
> | Thinking about this a bit more, I think this might be a bug in R CMD
> | check, actually.
> |
> | I understand that these flags are not portable, and yes, they should
> | be reported if they are specified in the packages's Makevars* files.
> | But in this case they are coming from the Ubuntu system's Makevars
> | setup, and surely, they do work there. I would even say that it is
> | actually good practice to include them in the system setup.
> |
> | I can probably provide a patch for this.
>
> See discussions on r-sig-debian -- in essence I had the same thouht and this
> has been taken care of now in R 3.6.0-2.  We add this to Renviron.site
>
> ## edd May 2019
> ##
> ## Debian (and its derivatives) impose a set of compiler flags to prevent some known
> ## security issues with compiled code.  These flags then become defaults for R as well
> ## (see eg /etc/R/Makeconf) but nevertheless confuse R as warnings get triggers.  Users,
> ## on the other hand, are often stumped about these.  So with this variable we declare
> ## these options as known for the local checks on the machine this file is on.
> ##
> ## See Section 8 of the R Internals manual for many more options to customize tests.
> _R_CHECK_COMPILATION_FLAGS_KNOWN_='-Wformat -Werror=format-security -Wdate-time'
>
>
> It may eventually move to Renviron now that Kurt kindly fixed an Renviron parser
> buglet with `=` assignments in r-devel
>
> So please add the environment variable now on all Debian and Ubuntu flavours,
> and/or upgrade to 3.6.0-2.

Fedora adds these flags too, plus -D_GLIBCXX_ASSERTIONS since Fedora 28, see:

https://stat.ethz.ch/pipermail/r-package-devel/2018q2/002754.html

This last one is particularly difficult to conceal, because the
resulting code includes calls to abort, for which R complains.

Iñaki