[R-pkg-devel] Reviving IndependenceTests and LAPACK WARNINGs

[Ivan Krylov]

В Fri, 19 Sep 2025 06:36:12 +0000
Pierre Lafaye de Micheaux via R-package-devel
<r-package-devel using r-project.org> пишет:

> I was able to resolve most of the issues, but I still have 2 WARNINGs
>  and 1 NOTE, such as:
> 
>     Warning: GNU Extension: Nonstandard type declaration COMPLEX*16

These only happen on Win-Builder due to extra checks being enabled
there using the _R_CHECK_WALL_FORTRAN_=TRUE environment variable;
normally R CMD check ignores the warnings about COMPLEX*16 or DOUBLE
COMPLEX, including during CRAN submission checks. See, for example, a
recent upload of the 'QZ' package: the compiler warnings are present
[1], but the check status is OK [2].

WRE 1.6.5 expresses a weak preference for DOUBLE COMPLEX over
COMPLEX*16. If you're willing to make the old Fortran code closer to
Fortran 90 instead of Fortran 77 with extensions, replacing COMPLEX*16
with COMPLEX(KIND(0.0D0)) (and removing 10 spaces to the right to avoid
hitting the line length limit) seems to work too.

The C++ code at https://github.com/lafaye/IndependenceTests has a
couple of more serious problems:

 * The 'new[]' allocation at src/cubature1.cpp:429 is freed using
   'delete' instead of 'delete[]' at src/cubature1.cpp:462, which
   crashes the process for me.
 * The allocation at src/dependogram.cpp:635 seems to be leaked.

Hopefully the July 12 version of the package no longer contains them.

-- 
Best regards,
Ivan

[1]
https://win-builder.r-project.org/incoming_pretest/QZ_0.2-4_20250413_040405/Windows/00install.out

[2]
https://win-builder.r-project.org/incoming_pretest/QZ_0.2-4_20250413_040405/Windows/00check.log