Create single-file installers

[mattwthompson]

No description provided.

[codecov]

Codecov Report

Merging #161 (01f63fc) into main (cb5aa10) will decrease coverage by 0.89%.
The diff coverage is n/a.

[mattwthompson]

The last run successfully built an installer and installed against it, but I haven't yet got tests running and passing. I'm assuming no OpenEye license is needed to run tests.

[mattwthompson]

I have it all the way to some tests passing, but some are still failing. Could you have a look at the failing tests and see what I'm missing @jthorton? I can add the OpenEye license in if it's required, but I don't know if that would cover all test failures.

The other option would be to not run tests based on the installer - I'd prefer keeping them in at all possible.

[jthorton]

@mattwthompson Thanks for putting this together, and sure happy to help but not sure where I should be looking for tests? in the single file installer CI I see it failing before getting to the tests or do I need to download the installer and run the tests locally?

[mattwthompson]

I screwed something up in the last commit, here are some logs from earlier showing failing tests: https://gist.github.com/mattwthompson/0903635d42866f6826069edcc18c69cd

[j-wags]

Ah, one things that I'm seeing in my current testing as well is the "can't assign fractional bond orders" one - Users need EITHER OpenEye OR (AmberTools + RDKit) to run fragmentation. The bespokefit test env and conda package only install the openff-toolkit-base package, which doesn't come with AmberTools.

Since the vast majority of users won't have access to OpenEye we should make sure the SFIs bundle AmberTools.

Separately, I asked OE for permission to bundle their toolkits in our SFIs and they did not approve it. So let's also make sure not to bundle OETK in the SFIs.

[mattwthompson]

Updated openforcefield/toolkit-installer-constructor@5e072df and restarting now

[mattwthompson]

Most recent failures - PDB loading, QCArchive offline https://gist.github.com/mattwthompson/f39e6bc4a89275a2f4339e6578c9460b

[mattwthompson]

Some tests are failing because they require OpenEye behavior - I'll open a PR to skip these tests in that case.

[mattwthompson]

The single-file installers are being built (with AmberTools installed - no OpenEye license or software packaged) and seem to not only exist but also pass (most) tests). Most of the failures seem to be a hard OpenEye dependency, which I attempted to fix in #170. There's one test failing in that PR but otherwise is good, I think.

[mattwthompson]

Okay, #170 is merged in and hopefully the tests pass here now.

[mattwthompson]

Tests are passing; I'll do some final touch-ups, merge, and upload the installers to the latest release. As this does not touch functionality of the software itself, I don't think waiting for reviews is warranted, though feedback is always the case on a non-draft PR.

Please note that these are XTB + AmberTools (+ RDKit) bundles; adding a Psi4 bundle to this array is a separate task and, as noted above, none will be made with OpenEye Toolkits bundled in. I've updated the file names to reflect this, using the pattern

name: ${{ matrix.os }}_py${{ matrix.python-version }}_xtb.sh

[j-wags]

Fantastic. Thanks a million, @mattwthompson!! (I agree a review wasn't needed on this, I just happened to be checking my email when the notification came in)