2021_11_10 Thompson/Wagner Check in

General updates

• Roadmap review happened just before this meeting.

PRs that need reviews

• • MT fixed root cause in an interchange branch. Will merge, cut release, and then tag PR as ready for review.

• Open

  

  • MT – I’ll contact OE to determine which of their RCs/betas/dev builds should be used for external testing

  • MT – Should we test ALL upstream rcs at once? Or one jub per upstream?

    • JW – Let’s do all of them at once, and revisit this if it ends up being super fragile

    • MT – Agree.

• • (Discussed some API stuff, JW left some comments on PR to keep from introducing performance regression. Determined that the duck-typing inputs for hill-formula method lead to complexity/secretly required dependencies/fragile code)

Other topics

• Sander crashes in examples

  • MT – Should be fixed, see above

• Toolkit stuff

• Fragmenter stuff

• Bespokefit stuff

• Miniforge issues – Fixes needed?

  • MT – I’ll make a new mdtraj release. Tried making a new release from the most recent and that didn’t fix it. Once there’s a new tarball out there, it should run on M1 macs

  • MT – I think AmberTools will also need an M1 release.

  • JW – Do we have some way to find other packages that will need M1 support?

    • MT – GHA may offer M1 runners. Mike Henry getting ARM builds running may overlap with this? If we can get an M1 runner we should be able to investigate this more closely.

    • MT – Not sure why the conda failures couldn’t describe which packages were missing. I have a script to iterate through conda yaml and try installing each dep which could pointedly address this.

  • JW – Can we mix M1 and 64-bit (rosetta) packages?

    • MT – Not sure

• Protein ligand workflow performance

  • JW – I’m working on performance improvements but ended up running tech support monday and tuesday (my development days). Improvements will be

    • Fast find-identical-molecules (should cut down parameter assignment time for thousands of waters)

      • MT – Make sure that everything that’s needed for reasonable performance is either under the hood of the OFFTK API that I’m using, that new API points are available, or that reasonable performance does somehow get into interchange.

    • (Maybe) Trust OpenMM’s residue names to narrow down substructure searching in Molecule.from_pdb

    •