Detect if there are .mod files for GCCcore installs

[branfosj]

If compiling Fortran then we should not do that with GCCcore due to the incompatibility in the .mod files generated. It would be helpful if we searched the installed software and warn if any are found.

Possibly:

• config option to turn on the check
• config option to control if this is a warning or error
• we will need ability to say that an easyconfig produces .mod files but that we are sure that they are not from Fortran or to allowing us to skip the check for an easyconfig

[branfosj]

From a check on my installed software:

• Clang - has .mod files - in the flang directory. I suspect that these are compiled by clang/flang and not an issue
• CPLEX - uses .mod for its own command files. So we will definitely need a 'this has non-fortran .mod files' flag.
• kim-api - has .mod files in lib64/kim-api/mod. I expect that this should be moved to GCC / intel-compilers as required.

[ocaisa]

I thought we could restrict the search only to directories being added to CPATH, if I understand things, that is how they are picked up by software...unfortunately I just found fortran-lang/fpm#441 (comment) which indicates that compilers behave differently when it comes to CPATH

[branfosj]

kim-api fixed in easybuilders/easybuild-easyconfigs#19399, easybuilders/easybuild-easyconfigs#19401, easybuilders/easybuild-easyconfigs#19402, easybuilders/easybuild-easyconfigs#19403, easybuilders/easybuild-easyconfigs#19404, and easybuilders/easybuild-easyconfigs#19405

[branfosj]

On generoso using

find */*GCCcore* -name *.mod > modfiles
cat modfiles | cut -d"/" -f7,8 | uniq

Compilers (AOCC/AOMP/Clang)

AOCC/2.3.0-GCCcore-10.2.0
AOCC/2.3.0-GCCcore-9.3.0
AOCC/3.0.0-GCCcore-10.2.0
AOCC/3.0.0-GCCcore-10.3.0
AOCC/3.1.0-GCCcore-10.2.0
AOCC/3.1.0-GCCcore-10.3.0
AOCC/3.1.0-GCCcore-11.2.0
AOCC/3.2.0-GCCcore-11.3.0
AOCC/4.0.0-GCCcore-11.3.0
AOCC/4.0.0-GCCcore-12.2.0
AOCC/4.0.0-GCCcore-12.3.0
AOMP/13.0-2-GCCcore-10.2.0
Clang/15.0.5-GCCcore-11.3.0
Clang/16.0.4-GCCcore-12.2.0
Clang/16.0.6-GCCcore-12.3.0

In PR14840

FACE/1.1.1-GCCcore-10.3.0
forbear/1.2.0-GCCcore-10.3.0

In PR15837

Fortran-YAML/20210420-GCCcore-11.2.0

Not Fortran module file

GeneMark-ET/4.65-GCCcore-10.2.0
GeneMark-ET/4.71-GCCcore-11.2.0
GeneMark-ET/4.71-GCCcore-11.3.0

Fixed in easybuilders/easybuild-easyconfigs#17221

HDF5/1.10.5-GCCcore-8.3.0-serial
HDF5/1.12.2-GCCcore-11.3.0-serial
HDF5/1.13.1-GCCcore-11.3.0-serial

Fixed, as per the previous comment

kim-api/2.2.1-GCCcore-10.2.0
kim-api/2.2.1-GCCcore-10.3.0
kim-api/2.3.0-GCCcore-11.2.0
kim-api/2.3.0-GCCcore-11.3.0
kim-api/2.3.0-GCCcore-12.2.0
kim-api/2.3.0-GCCcore-12.3.0

Fixed in easybuilders/easybuild-easyconfigs#19283

json-fortran/8.3.0-GCCcore-11.2.0
json-fortran/8.3.0-GCCcore-11.3.0
json-fortran/8.3.0-GCCcore-12.2.0

In PR16721

HEALPix/3.82-GCCcore-11.2.0

Fixed in previous PRs

mctc-lib/0.3.1-GCCcore-11.2.0
mctc-lib/0.3.1-GCCcore-11.3.0
mctc-lib/0.3.1-GCCcore-12.2.0
mstore/0.2.0-GCCcore-11.3.0
mstore/0.2.0-GCCcore-12.2.0

Hmm..

Python/3.9.5-GCCcore-10.3.0

File is in the tests

Python/3.9.5-GCCcore-10.3.0/lib/python3.9/site-packages/numpy/f2py/tests/src/module_data/mod.mod

But the wider question here is why that Python has numpy in it at all.

[boegel]

Keeping this open since we need to update a couple of easyconfigs to weed out false positives now that #4389 is merged...

[branfosj]

Keeping this open since we need to update a couple of easyconfigs to weed out false positives now that #4389 is merged...

• Clang, AOCC, AOMP, and CPLEX: disable check for .mod files in Clang, AOCC, AOMP, and CPLEX easyblocks easybuild-easyblocks#3067
• GeneMark-ET: bypass .mod file in GeneMark-ET easybuild-easyconfigs#19500

Are you able to scan your installs to see if there are anyother false postives?

[branfosj]

If there remains items left to do here then we'll catch them during regression testing for the release and fix them then.