qcint should have a different soversion than libcint

[susilehtola]

See pyscf/pyscf#1855

[chillenb]

Perhaps the SONAME of qcint can be changed so that pyscf built against libcint cannot dynamically load qcint?

[susilehtola]

Perhaps the SONAME of qcint can be changed so that pyscf built against libcint cannot dynamically load qcint?

Yes, definitely! This would be the correct solution.

[chillenb]

I have an idea related to this @sunqm

What do you think about cmakedefining SIMDD at build time? After all, the simd width is set when the library is built and should not be changed afterwards.

This would help to ensure that the correctness of an application linking qcint does not depend on the compile flags used to build it.

[susilehtola]

@chillenb please file other tickets for unrelated issues

[susilehtola]

It appears that the issue is that CMake doesn't allow you to change the soname.

The alternative is then to define a different soversion in qcint vs libcint, which would achieve the same end result. For example, from now on define all libcint soversions to be even, 2n and all qcint to be odd, 2n+1.

[chillenb]

@chillenb please file other tickets for unrelated issues

I did do this later on, but didn't link it here: #21

Mentioned it here because I felt it might help with some of your Fedora packaging work by improving ABI consistency (ensuring that the ABI does not depend on the compile flags)