[R-pkg-devel] flang-new-18 complains: incompatiple object shapes

[Göran Broström]

More specifically: I submitted a new version of eha to CRAN, but:
"... does not pass the incoming checks automatically,..."

I found warnings in

https://win-builder.r-project.org/incoming_pretest/eha_2.11.5_20240919_123016/Debian/00install.out
-----
/getsums.f:65:16: warning: Reference to the procedure 'daxpy' has an 
implicit interface that is distinct from another reference: incompatible 
dummy argument #5: incompatible dummy data object shapes
                  call daxpy(j2, tmp2, zma, 1, szz(index), 1)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
./getsums.f:51:10: previous reference to 'daxpy'
            call daxpy(k, epy, zma, 1, sz, 1)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
--------
The arguments 'szz(index)' and 'sz' are compared and found to be 
'incompatible'. szz and sz are both double vectors, szz is the upper 
triangle of a symmetric covariance matrix, stored as a vector of length 
n * (n + 1) / 2 (space saving, this is F77 code written some 40 years ago).

This is the point of view of a compiler called 'flang-new-18' (Debian). 
However, gfortran and myself disagree, and I am interested in hearing a 
suggestion of what measures I should take.

It is of course easy (but tedious) to cut the Gordian knot by explicitly 
writing out the loops that defines 'daxpy', but is it necessary?

Thanks, Göran