How to install from source ("develop")?

[orbeckst]

What's the best way to install MDAnalysis from source, in particular the develop branch after you forked the mdanalysis repo:

git clone https://github.com/your-user-name/mdanalysis.git
cd mdanalysis
git remote add upstream https://github.com/MDAnalysis/mdanalysis

The installing develop instructions in the User Guide are outdated (issue MDAnalysis/UserGuide#368). What's the best/cleanest way to install?

1. using conda-forge packages (with mamba/conda)
2. using pip packages (pip and virtualenv ... and your compiler toolchain)

[orbeckst]

For a quick conda/mamba environment one can use

mamba create -n mda312dev python=3.12
mamba install -n mda312dev --only-deps mdanalysis mdanalysistests
mamba activate mda312dev

which makes use of mamba --only-deps mdanalysis mdanalysistests to pull the dependencies for the published mdanalysis and mdanalysistests conda-forge packages.

It has the disadvantage that these dependencies may be different in the current source.... but it's quick.

[IAlibay]

because of the disadvantage you list, doing this is a really bad idea - especially with the current state of the develop branch

[IAlibay]

Since we don't offer an environment file, the best approach is to just manually list all the deps to install imho.

We could offer an environment file, and I looked into this a while ago, but it requires an action to check our CI vs what we store.

Note: Before someone says "let's switch completely to environment files" - the source of truth is our CI deps build action, we don't use environment files because they are too inflexible, it really won't work, I tried.

[RMeli]

What about parsing the dips build action (it's a YAML file after all)? See MDAnalysis/UserGuide#368 (comment) for a quick stab at it.

[IAlibay]

If we want an env file then we can have one and create the necessary tooling to verify it (that's what I meant by checking what we store, it's fine to me, but it's extra tooling and every time I make extra tooling I get complaints about extra tooling).

[orbeckst]

Let's start simple with a working environment file, update the UG, and then see if we need CI to check that this env file agrees with what we use in CI.

(This was the conclusion from a discussion at the MDAKits meeting today.)

[orbeckst]

And once we have the env file in the rep, we can add a proper answer here (which will be mirrored in the UG.)

[marinegor]

@orbeckst I'm not sure if this was your initial request, but I think you can try following syntax of pip to install directly from git(hub):

# install default package
python -m pip install 'mdanalysis @ git+https://github.com/MDAnalysis/mdanalysis.git@develop#subdirectory=package'

# install a particular tag -- note the single quotes!
python -m pip install 'mdanalysis[analysis] @ git+https://github.com/MDAnalysis/mdanalysis.git@develop#subdirectory=package'

Note the single quotes in the latter option -- won't work without them (will install default branch). And this is for bash shell, not sure if it works for others without hard quoting (though I think you can safely quote the git+... part).

I usually do that in a virtual environment, which works perfectly fine, thanks to the dev team.

Also, for the sake of huge speed gain, I recommend doing this with uv, which supports exactly the same syntax (uv pip install ...), and also installs whole [analysis] tag in 1m30s with a clean cache / 2s (pip installs [analysis] in 4 minutes with clean cache, and like 40 seconds with cached dependencies, i.e. 20 times slower).

[orbeckst]

That's a nice answer for pip installation without the cloned local repository.