Collection of nvfortran problems in our code

The following tarball contains all our unit tests that fail with the nvfortran 23.1 (but do work with ifort, gfortran and nagfor). If you unpack, and do make, then you can check the individual tests with
./whizard_check --check testXX (with XX from 01 to 44)
whizard-230404-2235.tar.gz (1.9 MB)
{note that originally in this file the reference files for the tests were missing, now they are there, and I even reduced the code a tiny bit)

I will try to break this up into individual cases. The reference file for the corresponding tests is in the folder ref-output, deviating files will go into err-output. Many tests however seem to have problems with segmentation faults.

Thanks Juergen,

This will take me a bit to get through these to determine the issues and then report them to engineering.

The package doesn’t include the reference files so I downloaded Whizard 3.1.2 from https://whizard.hepforge.org/ and used the files from “share/tests/unit_tests/ref-output”

Some of the tests are failing due to “libtool”, “omega_QED.opt”, etc. being missing so I’ll ignore those for now Looks like the code forces a segv in these cases.

Are there any particular tests I should focus on first?

-Mat

Hi Mat,
yes, indeed, there are a lot of things that have the same root cause, and some don’t work for you without the setup on my computer. I exchanged the tarball above, which should now contain also the reference files. I will reduce the cases further. I put the next stage of reduction here:
whizard-repro_v2.tar.gz (1.1 MB)
Cheers,
Juergen

Here is a first separate test case, where the allocation of a polymorphic entity from an abstract base entity leads to a segmentation faul, and a tarball (v3) where I removed that individual test case.
test_alloc_polymorph.f90 (4.1 KB)

And because I can only put one link in a post, here comes the reduced reproducer container:
whizard-repro_v3.tar.gz (1.1 MB)

I isolated another issue which has to do with allocatable character objects, here:

There is another isolated single test from this test suite reported here:

Another spin-off from this is here:

Matt, this is now the tarball with 5 or 6 remaining cases, I will stop it for now. If you like, please contunue:
whizard_v4.tar.gz (886.9 KB)
The tests are labelled test_02/03/04/05/06/10/12/15.

I removed all the separate issues from the tarball, which has shrunk now quite a bit. There are only 5 different tests left, test02-06. The rest were apparently duplicates.
whizard_v5.tar.gz (258.6 KB)