[R-pkg-devel] Compiler warning using C versions of base R integrate function

[Tomas Kalibera]

On 11/21/19 7:25 PM, Devin Incerti wrote:
> Hi,
>
> I'm attempting to submit a very minor update to my package where I replaced
> a deprecated C++ function with an alternative. However, I'm receiving an
> unrelated warning "*array subscript -1 is outside array bounds of ‘double
> [52]’ [-Warray-bounds] *" on Debian during the CRAN pre-tests. I do not
> receive this warning on my local OS X, Ubuntu 16.04 (on Travis-CI), with
> Win-builder, or with R-hub builder. I also did not receive this warning on
> previous submissions to CRAN, but it does now show up on the most recent CRAN
> checks
> <http://s.bl-1.com/h/cH9zB21r?url=https://cran.r-project.org/web/checks/check_results_hesim.html>
> .
>
> The warning is related to the header file integrate.h
> <http://s.bl-1.com/h/cH9zB6Qt?url=https://github.com/hesim-dev/hesim/blob/master/inst/include/hesim/Rbase/integrate.h>,
> which is just integrate.c
> <http://s.bl-1.com/h/cH9zBCpw?url=https://github.com/wch/r-source/blob/5a156a0865362bb8381dcd69ac335f5174a4f60c/src/appl/integrate.c>
> from
> R source with function pointers replaced by C++ functors. These integration
> functions are needed so that the package can integrate at the C++ level.
>
> I don't know how to avoid this warning. Is it perhaps related to the
> particular version of GCC used during the CRAN checks. Any thoughts or
> suggestions would be much appreciated.
>
> Thank you!

The warning is generated by GCC when inlining the function rdqelg(), I 
can reproduce it with the integrate.c we have in R, but only if I add 
"inline" to rdqelg() definition (GCC 6 and newer - on Ubuntu but that 
should not matter). The inliner knows the callsite, so it knows that 
what is being passed is an array (and not an arbitrary pointer), and 
hence it emits the warning. It is not smart enough to know that the 
first element of the new "array" will (unless there is a bug) never be 
used. When not inlining and the function is compiled independently from 
its callsites, it assumes it is a pointer, and hence there is no warning.

The code is automatically converted from Fortran, maybe these pointer 
decrements in function prologues are to adjust to Fortran 1-based array 
indexing. If you must use this modified copy of integrate.c from R, 
perhaps a quick hack to get rid of the warning would be to avoid the 
"inline" keywords in declarations of functions like rdqelg(). In a 
perfect world, the code would be rewritten to more idiomatic C (not just 
to avoid these decrements). Or perhaps the Fortran original could be used.

Tomas

> ()
>
> Devin
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-package-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-package-devel