General updates | JW – Worked on topology interop PR yesterday. I’m in some sort of really weird git state where GH and my computer disagree on what’s in certain files. But I’ve mostly got that resolved and plan to merge the topology interop PR soon. JW – Two industry people have asked for bespokefit single file installer. So I’ll put that on the sprint planning board next week. MT – I actually started on that this morning in response to your DM. It’s looking promising. JW – There’s some complexity with optional deps - WE’ll want to bundle all of them, except that psi4 won’t work on mac (so let’s not make mac builds) psi4 on linux needs the default channel to be higher priority than conda-forge psi4 and ambertools cause the env solve to get really complex, but JHorton reports that there are valid combinations, it just takes a long time.
MT – Smaller tasks we discussed on wednesday: Roundtripping to/from OFFXMLs is kinda working, but it’s hackier than I’d like There’s a kinda hacky manually-replace-ASCII-special-characters solution It may be possible to make Pint output english full words MT – There’s two kinds of pint vs. openmm interop: String-on-disk conversion, and in-memory conversion. The in-memory conversion is robust. It’s just the string-on-disk interop that we’re dealing with here. It would be very constraining to require string-on-disk readability by openmm forever. JW – Seems like there’s three families of options: Sooner or later no longer supporting old-style OFFXMLs Always supporting OFFXML back-compability but not other formats Making openff-units and openmm units be convertible in both in-memory and string-on-disk representations
(General) – The SMIRNOFF spec is ambiguous about how units should be handled. We should resolve what this is, since it will affect external people who parse OFFXMLs.
Tried out Simon’s contextmanager idea, where we’d do the toolkit registry switch in create_openmm_system before calling Intercahnge.from_smirnoff
|