Add openmmdl as MDAKit

[talagayev]

Addition of OpenMMDL to MDAKits:

Package for Interface to OpenMM for easy setup of molecular dynamic simulations of protein-ligand complexes and the analysis thereof, available at:

https://anaconda.org/conda-forge/openmmdl

[orbeckst]

The main issue is that one set of tests for py 3.12 fails with an installation issue for mdtraj

Could not solve for environment specs
The following package could not be installed
└─ openmmdl is installable and it requires
   └─ mdtraj 1.9.7.*  with the potential options
      ├─ mdtraj 1.9.7 would require
      │  └─ python >=3.10,<3.11.0a0 , which can be installed;
      ├─ mdtraj 1.9.7 would require
      │  └─ python >=3.11,<3.12.0a0 , which can be installed;
      ├─ mdtraj 1.9.7 would require
      │  └─ python >=3.7,<3.8.0a0 , which can be installed;
      ├─ mdtraj 1.9.7 would require
      │  └─ python >=3.8,<3.9.0a0 , which can be installed;
      └─ mdtraj 1.9.7 would require
         └─ python >=3.9,<3.10.0a0 , which can be installed.

I am not really sure what's going on so I am bouncing this back to you to look into and report back...

UPDATE: Looking more closely into the "develop" test run, which supposedly passed, I can now see that this run did not actually install the package:

Run if [[ develop == "develop" ]]; then
  if [[ develop == "develop" ]]; then
    type="src"
  else
    type="install"
  fi
  
  install=$(python utils/get_install.py --itype ${type} --mdakit openmmdl)
  echo "install=${install}"
  eval ${install}
  shell: /usr/bin/bash -l {0}
  env:
    PYVER: 
    INPUT_RUN_POST: true
    CONDA: /usr/share/miniconda3
    CONDA_PKGS_DIR: /home/runner/conda_pkgs_dir
install=

Similarly, the "run tests" also come back empty and report success

Run tests=$(python utils/get_runtests.py --mdakit openmmdl --runtype develop)
  tests=$(python utils/get_runtests.py --mdakit openmmdl --runtype develop)
  echo "tests: ${tests}"
  eval ${tests}
  shell: /usr/bin/bash -l {0}
  env:
    PYVER: 
    INPUT_RUN_POST: true
    CONDA: /usr/share/miniconda3
    CONDA_PKGS_DIR: /home/runner/conda_pkgs_dir
tests: 

This means that you also need instructions how to install from source (and my hunch is that the testing command that I already mentioned also needs updating).

[talagayev]

@orbeckst hey Oliver,

Hm I can take a look into mdtraj to adjust the mdtraj version, since the code mainly uses mdtraj only for one function, which ist image_molecules(), else I can just replace it with an mdanalysis function to do the same :)

[talagayev]

Ah ambertools also complains about the imcompatability with the python version, so that currently only python 3.10 is the python version covering all of the packages. I will then update the packages and test them to be compatible with python 3.11 and python 3.12, make a release and get then the compatable release version on conda-forge so that it works with all 3 python versions.

[IAlibay]

@lilyminium @fiona-naughton can either of you assign yourselves to shepherd this PR?

[orbeckst]

✅ The openmmmdl (latest) installs and runs the tests successfully.
❌ The openmmmdl (develop) does not properly install, see e.g. https://github.com/MDAnalysis/MDAKits/actions/runs/10688427502/job/29628268563?pr=161#step:7:18 — the following steps all complete seemingly successfully

The develop tests should also run properly with an installed version of the software.

[talagayev]

@orbeckst If I understand correctly the tests are successfull with the latest version of mdanalysis, but errors appear with the devlopment version of mdanalysis, so I can check if they run locally by using the development mdanalysis package locally and looking if it installs and runs the tests successuflly?

[orbeckst]

When you look at the GH actions report at the line that I linked, you see that

install=$(python utils/get_install.py --itype ${type} --mdakit openmmdl)

gives an empty install variable.

I would assume that this is because there's an issue with the metadata.yml file and utils/get_install.py is getting confused somehow. That's where I would start.

[IAlibay]

Apologies folks, I haven't been keeping up with this.

Looks like the develop failure is because there's a missing src_install entry, which is a requirement for mdakit registration.

[talagayev]

@IAlibay hm shouldn't be a problem to add the src_install entry, the only problem is that I would not be able to have all of the packages being installable with pip due to using plip for protein-ligand interaction analysis, which uses openbabel that has the problem with the wheel building, so the installation with pip would miss this package. I will then add the src_install entry now :)

[talagayev]

I added the src_install entry and will then move the packages that are possible to be installed with pip tomorrow to the pyproject.toml in the OpenMMDL repository, with the exception of plip, thanks @IAlibay for helping with the PR:)

[IAlibay]

@IAlibay hm shouldn't be a problem to add the src_install entry, the only problem is that I would not be able to have all of the packages being installable with pip due to using plip for protein-ligand interaction analysis, which uses openbabel that has the problem with the wheel building, so the installation with pip would miss this package. I will then add the src_install entry now :)

You can have a separate mamba install call before the pip install, i.e. it's a list so you can just add more things as necessary - I believe a few other mdakits do this (I'll try to find an example later if I get a chance)

[talagayev]

@IAlibay hm shouldn't be a problem to add the src_install entry, the only problem is that I would not be able to have all of the packages being installable with pip due to using plip for protein-ligand interaction analysis, which uses openbabel that has the problem with the wheel building, so the installation with pip would miss this package. I will then add the src_install entry now :)

You can have a separate mamba install call before the pip install, i.e. it's a list so you can just add more things as necessary - I believe a few other mdakits do this (I'll try to find an example later if I get a chance)

Ah, then it should be fine, I will then go to sleep now and take a look tomorrow, since if it is possible to first run the mamba install then it shouldn't be a problem :)

[talagayev]

@IAlibay okay mdahole2, lipyds and openmm-mdanalysis-reporter all do it the way you mentioned, by git cloning the repository, then installing through mamba install and then going into the cloned folder and then using pip install . I will see if it works now with mamba install and directly afterwards pip install from github without the git cloning and cd into the folder

[orbeckst]

@talagayev please ping me when you've made progress and think it's ready (i.e., everything installs and passes tests) or if there are questions.

[talagayev]

@orbeckst I think everything is ready and should be now running correctly including the installation with src_install and all the pytests :)

[orbeckst]

I need some clarification on what happens in the CI develop during mdakit installation:

Successfully built openmmdl
Installing collected packages: openmmdl
  Attempting uninstall: openmmdl
    Found existing installation: openmmdl 0.0.1
    Uninstalling openmmdl-0.0.1:
      Successfully uninstalled openmmdl-0.0.1
Successfully installed openmmdl-0.9.2.1+3.g2217545.dirty

The conda-forge package is at version 0.9.2.1

+ openmmdl                              0.9.2.1  pyhd8ed1ab_0                       conda-forge      85MB

and it is being installed to get all the deps.

Then the pip package is built and installed. But why does pip uninstall version 0.0.1 ??

As mentioned in my comments, it would be cleaner to explicitly list dependencies for a source installation.

[orbeckst]

Please see comments!

(I wrote these days ago but apparently forgot to submit review, sorry for the delay.)

[orbeckst]

mdanalysis may not be necessary because I think we install it ourselves — but I doubt it hurts.

Do you have any specific recommendations @IAlibay @fiona-naughton @lilyminium ?

[talagayev]

also a similar case is for versioningit and setuptools which I have in the pyproject.toml so they are installed there in theory, but I thought if I list the remaining packages I can also add them in the mamba install step so that all of the packages necessary for the package are present

[talagayev]

I will keep for now the mdanalysis>=2.3.0 to see what the others say if it should be removed or retained :)

[orbeckst]

ok

[IAlibay]

I think it makes sense to have MDAnalysis here, in the end this will go in the docs as "this is how you install this".

I have moved pytest things into test_dependencies though, because as far as I'm aware they aren't necessary for the standard operations of the code.

[talagayev]

Please see comments!

(I wrote these days ago but apparently forgot to submit review, sorry for the delay.)

will do, all good :)

[orbeckst]

Thank you for addressing all my comments. Looks ready to me (modulo comments from others).

[orbeckst]

ok

[orbeckst]

@IAlibay I believe your specific comment #161 (review) was addressed but would you mind having a quick look through? Thanks.

[orbeckst]

@IAlibay if you want to have a look please do over the next few days, otherwise I'll dismiss.

[IAlibay]

I've made one direct change on the dependencies but otherwise lgtm!

[IAlibay]

I think it makes sense to have MDAnalysis here, in the end this will go in the docs as "this is how you install this".

I have moved pytest things into test_dependencies though, because as far as I'm aware they aren't necessary for the standard operations of the code.

[talagayev]

Nice, thank you @orbeckst and @IAlibay for accepting and helping with this MDAKits entry :)