Recommend mambaforge + mamba in the installation instructions and contributing guides

[seisman]

In the installation instructions, we recommend using Anaconda:

We recommend using the Anaconda Python distribution to ensure you have all dependencies installed and the conda package manager is available. Installing Anaconda does not require administrative rights to your computer and doesn’t interfere with any other Python installations on your system.

Here is a comparison between Anaconda and miniconda: https://stackoverflow.com/a/45421527.

From my own experience, Anaconda packages too many packages by default, which causes slow installation, large disk spaces and potential package conflicts.

I wonder if we should recommend miniconda instead.

[maxrjones]

Yes, I think we should recommend miniconda or micromamba.

IMO we should also recommend mamba for contributors in https://www.pygmt.org/latest/contributing.html#setting-up-your-environment, since the conda solver can get tedious with the number of development dependencies that we have.

[seisman]

IMO we should also recommend mamba for contributors in pygmt.org/latest/contributing.html#setting-up-your-environment, since the conda solver can get tedious with the number of development dependencies that we have.

I agree.

Yes, I think we should recommend miniconda or micromamba.

micromamba is not a good option, because (https://mamba.readthedocs.io/en/latest/user_guide/micromamba.html):

Still at early stage, it’s main usage is in continous integration pipelines: since it’s a single executable, it reduces dramatically bandwidth usage and provide fast operations.

[weiji14]

Related thread at https://discourse.pangeo.io/t/mamba-mature-enough/3154/4 on how mamba is pretty stable already. Agree that micromamba isn't a good option (mostly because you'd have to call commands using micromamba install instead of mamba install), but I'd prefer mambaforge instead of miniconda as recommended by https://mamba.readthedocs.io/en/stable/installation.html#fresh-install.

[seisman]

Related thread at discourse.pangeo.io/t/mamba-mature-enough/3154/4 on how mamba is pretty stable already. Agree that micromamba isn't a good option (mostly because you'd have to call commands using micromamba install instead of mamba install), but I'd prefer mambaforge instead of miniconda as recommended by mamba.readthedocs.io/en/stable/installation.html#fresh-install.

Sounds great. mambaforge + mamba is the way to go.

[weiji14]

Kinda unrelated, but I just found out about https://prefix.dev/blog/online_environment_solving today and was playing with solving a PyGMT development environment (conda-lock.yml file) at https://prefix.dev/envs/weiji14/pygmt. This is today's resolved environment at https://prefix.dev/envs/weiji14/pygmt/285fy5jawbs0, installable using micromamba create -f https://prefix.dev/envs/weiji14/pygmt/285fy5jawbs0/conda-lock.yml -n pygmt:

It's still a very young project built by the mamba folks, but it might be useful for your own projects if you ever need to quickly test out solving an environment without messing with your local environment, or do cross-platform solves. They also do versioning of generated conda-lock.yml, which might be useful for reproducibility.

P.S. Couldn't get linux-aarch64 nor linux-ppc64le to solve for some reason, but that's for another issue.