CI on main branch failing after #2044 merge

[praveksharma]

The new CI test to test cuPQC builds introduced by #2044 is failing on the main branch. This problem is occuring because CMake can not find NVCC, it first occured when working on #2044; although it was resolved and CI was green before the merge that test is failing again now. I shall work on getting this fixed.

[baentsch]

The error is a simple one: There's a hard-coded compiler path:

liboqs/.github/workflows/linux.yml

Line 199 in 6a16ac6

run: mkdir build && cd build && cuPQC_DIR=/cupqc/cupqc/cupqc-pkg-0.2.0/cmake/ CUDACXX=/usr/local/cuda-12.6/bin/nvcc cmake -GNinja -DOQS_USE_CUPQC=ON -DCMAKE_CUDA_ARCHITECTURES=80 .. && cmake -LA -N ..

that's out of sync with the CI image.

But in looking at the issue, I noticed that the default CI image (openquantumsafe/ci-ubuntu-latest:latest) grew to > 14GB. This IMO is out of sync with our goal of being "environmentally friendly": Just the load of this huge image takes 1-2 minutes, i.e., 30% to 70% of the complete CI runtime of several tests (also making me wonder why there's apparently no caching?).

May I thus suggest splitting out a (arbitrarily large) separate image for "thorough" testing (say once a week or only at release) and revert the main "workhorse image" to something more efficient? In other words, as part of the fix, may I suggest placing testing for cuPQC somewhere else in the CI setup (and use a separate image)?