HanatoK

Hi @giacomofiorin and @jhenin! Aside from the atoms in `cartesian`, are the atoms in an atom group expected be aranged in the same order as their insertion order? In other...

> See `cvm::atom_group::create_sorted_ids():` you would certainly have to update `sorted_atoms_ids` and `sorted_atoms_ids_map` after changing the atom order. I'm thinking on the rmsd and eigenvector components in particular. The value of...

> If it helps, we could also remove the inlined coordinates feature altogether. I don't think I have used it much myself in 20 years or so, i.e. from before...

> > Do you mean the `refPositions`? I have tried @jhenin's test case and [HanatoK@a66fe46](https://github.com/HanatoK/colvars/commit/a66fe4655a5a26acfe74d55412499d40a889ecd8) indeed breaks it. In my opinion, the atoms in the atom group are better to...

Commit https://github.com/Colvars/colvars/pull/788/commits/5eea84f9a213ce2883eddd91d261626cd04179c0 adds support of SOA atom groups in the NAMD CudaGlobalMaster interface.

Maybe I need to remove the ifdef guards in https://github.com/Colvars/colvars/blob/789029131bca128ff36017d4bdd0d98aaec495aa/src/colvar.cpp#L933-L935

> Same here on both points! > A compile-time comparable version number would be helpful to have, even if the new preprocessor code is slightly more complex than the current...

> But you are changing the conversion from string to integer. If I build a test code using your header and the following lines: > > std::cout proxy->get_version_from_string(COLVARS_VERSION) std::cout >...

After some thoughts I feel there is no way to make explicit gradients working correctly with reusable components, so I will disable it in this PR.

### The problem of `calc_gradients()` Colvars was not designed with [automatic differentiation (AD)](https://en.wikipedia.org/wiki/Automatic_differentiation) in mind. At first glance, it seems to perform the forward AD because for each CVC, the...