Software-specific branching in several places should be avoided. Branching must be enveloped somehow.

[comcon1]

I have already seen software-specific branching in several places. At the moment the organization makes adding new software or file formats for an old software very tedious. We should think how to resolve it. In general, the good thing is to make application general - software independent via drivers, which are software specific. So, a driver - middle-layer API, should take care of all software-specific manipulations producing a standardized output, from which the following steps should be done. Then, if software specific modifications are needed, they should be done in one place, instead of introducing modifications to many different files throughout the project.

Originally posted by @pbuslaev in #201 (comment)

[comcon1]

Continue citing @pbuslaev

At the moment the organization makes adding new software or file formats for an old software very tedious. We should think how to resolve it. In general, the good thing is to make application general - software independent via drivers, which are software specific. So, a driver - middle-layer API, should take care of all software-specific manipulations producing a standardized output, from which the following steps should be done. Then, if software specific modifications are needed, they should be done in one place, instead of introducing modifications to many different files throughout the project.

We should agree here, how we ask people to provide topology, and how we will use the topology. Now, NMRlipids uses TPR for GROMACS and doesn't use topology files from other engines, which is quite strange because, I see that MDanalysis actually does allow using various formats:
https://docs.mdanalysis.org/1.1.1/documentation_pages/topology/init.html

... including for example PSF. We also allow PSF in namd_dict (see databank_defs.py )

    "TOP": {
        "REQUIRED": False,
        "TYPE": "file",
        "EXTENSION": ("psf"),
    },

But then we... just don't use it and use PDB instead!