Tolerance failure in test on arm64 architecture

[badshah400]

I package wannier90 rpms for openSUSE.

Today, when attempting to build — and submit packages for — version 4.0.2 of wannier90, on the aarch64/arm64 architecture only, we found the following relative tolerance issue (which looks minor to me!) causing ctest to fail; the same test runs just fine on x86_64, however.

[  454s] 76/97 Test #76: testw90_graphene_dis_proj_auto ................***Failed    2.14 sec
[  454s] Using executable: /home/abuild/rpmbuild/BUILD/wannier90-openmpi4-4.0.2-build/wannier90-4.0.2/build/test-suite/tests/../../wannier90.x.
[  454s] Test id: 05092026.
[  454s] Benchmark: default.
[  454s] 
[  454s] tests/testw90_graphene_dis_proj_auto - graphene.win: **FAILED**.
[  454s] omegaD
[  454s]     ERROR: relative error 5.20e-06 greater than 5.00e-06. (Test: 0.181238503.  Benchmark: 0.181239445.)
[  454s] All done. ERROR: only 0 out of 1 test passed.
[  454s] Failed test in:
[  454s] 	/home/abuild/rpmbuild/BUILD/wannier90-openmpi4-4.0.2-build/wannier90-4.0.2/build/test-suite/tests/testw90_graphene_dis_proj_auto

Thanks for the great library.

[jryates]

Due to various recent changes (e.g. CODATA) many of the tests differ from the benchmark by a small amount - less than the tolerance for failure. However, this means that another small deviation can cause an actual failure. This should be checked carefully - but we believe that a reasonable solution is to regenerate the benchmark set.