GROMACS

• Whenever a PDB file is loaded into Gromacs, atom names of the form NABC (where N is a digit and the name is at least 4 chars) are transformed into ABCN for storage in t_atoms.atomname. The original atom name from the PDB file is stored into t_atoms.pdbinfo.atomnm, but this version is not trimmed of whitespace. It could be easier to work with if it was trimmed, but I don't know if the whitespace is significant to some current users of this information.
• Whenever a PDB file is written from Gromacs, a reverse transformation takes place, no matter what is the source of the original atom names (so taking a PDB input with atom name HG13 and writing that out produces 3HG1).
• Selections (and also, e.g., make_ndx) only use t_atoms.atomname for matching the atom names. So g_select and make_ndx both work the same.
• I assume that the reason for these transformations is somehow related to pdb2gmx, but I agree that the behavior of g_select (and also make_ndx) is not very intuitive.

1. Make name selections match t_atoms.pdbinfo.atomnm if it is present, otherwise use t_atoms.atomname. This resolves the unintuitive behavior in this change, but introduces another one: depending on the input format, the same selection may select different atoms. So if one just takes a tpr and a pdb file produced by the same invocation of mdrun (or a pdb file used as input to grompp, and the produced tpr file), the selection syntax will be different.
2. Make name selections match either t_atoms.pdbinfo.atomnm (if present) or t_atoms.atomname. Can also be quite unintuitive, as HG23 can then match both HG23 and 3HG2 in the input.
3. Make name selections always match against t_atoms.atomname, but also try applying the reverse PDB transformation. The unintuitive behavior from the second point is even stronger here.
4. Add a separate pdbname selection keyword that will match only t_atoms.pdbinfo.atomnm. It can either fall back to t_atoms.atomname or give an error for non-PDB input. In particular without the fallback, this keeps the selection syntax more predictable.
5. Try to keep the user-supplied atom names also when dealing with PDB files. There is probably some rationale (which I don't know) for this translation, so this may introduce other problems.
6. Keep the current behavior, but just document it. At least the current behavior is consistent between g_select and make_ndx, and also has the nice property that the same selection produces the same output for different input formats that were produced by Gromacs conversions.

Any of the alternatives except the fifth should be straightforward to implement.

Hmm--I would think 4 + 6 would be good--if we document the behavior and then have a pdbname selection keyword to allow matching the pdb convention. We can leave the whitespace convention alone for the moment, although it might be nice to allow matches to non-whitespace pdbname if that is straightforward.