search

MRChemSoft / mrchem

MultiResolution Chemistry
GNU Lesser General Public License v3.0
27 stars 21 forks source link

HF_molecular_gradient_LDA test fails in Fedora rawhide #457

Closed susilehtola closed 1 year ago

susilehtola commented 1 year ago

The development branch of Fedora, a.k.a rawhide, has a new version of Python, which requires rebuilding packages.

The rebuild of mrchem 1.1.2 fails due to a failure in the HF_molecular_gradient_LDA test both on x86_64 and i686

The builds are against mrcpp 1.4.2.

susilehtola commented 1 year ago

So... I think there's something really odd going on here. The MRChem energies are way off. MRChem gives a total energy around -83.314 Eh, while the HF molecule should have a total energy around -99 or -100 Eh. What is going on here?!

stigrj commented 1 year ago

Hi @susilehtola, due to my lateness it seems your build logs are no longer available, do you have links to a more recent attempt?

susilehtola commented 1 year ago

@stigrj yes,

https://kojipkgs.fedoraproject.org//work/tasks/1440/104791440/root.log https://kojipkgs.fedoraproject.org//work/tasks/1440/104791440/build.log

stigrj commented 1 year ago

Thanks, reproduced on my (non-fedora) system with Python-3.12

stigrj commented 1 year ago

Actually, the error is caused by a failure in the parsing of a docstring in the python input validator. For some reason :math: is not recognized in the string anymore in Python-3.12, which throws an error that is completely unrelated to the test case (the weird HF energy is due to poor core Hamiltonian orbitals used for computing the gradient in this particular test).

Fix coming up

stigrj commented 1 year ago

Patched MRChem-v1.1.3 is now available