Conda is not just for Python.

Conda

Conda is a cross-platform, language-agnostic binary package manager. Shame for me 😟, it's there for years, but only recently I've realized how awesome it is! Maybe Anaconda's daunting fat distribution scares away people like me. Anyhow, with MiniConda you can manage virtual environment of your *nix packages. If you works on app level development (not OS related), you can easily make almost every standard linux dev package virtual. This is super useful if you're not sudoer or in big tech giants (which usually has their own package mangers) but want to do something big or you don't like heavy rocket-launcher solutions (Docker/VirtualMachine). Actually, inside conda's x86_64-conda_*-linux-gnu/sysroot, you can build your own *nix kingdom or miniverse.

A miniverse, from Rick and Morty
A miniverse, from Rick and Morty

Use Conda for Everything

I've found a lot packages in anaconda shopping mall: zsh, clang, cmake, libtool, autoconfig, nodejs, ruby, openjdk... all linux toolchains and packages you could image. Vendor like conda-forge makes a lot of recipes, got to check it out. Next time when you want to brew/apt/yum install, you should first check if you can do it in conda elegantly. The good things is that you don't need to messing around with folders like /usr/bin etc. and you can easily revert to your previous install revisions. Excited by this finding, I've migrated all my old install scripts just to use conda so that my environment is more independent.

What? no recent neovim?

In anaconda, I can only find python package instead of native neovim app I need. There are some native neovim build but mostly outdated. One of the greatest things about conda ecosystem is: if some package's missing, you can build it from source and then share it on Anaconda store. Everyone else benefits by saving a lot of build/release time. Build conda packages is so easy. Always remember conda-forge is one of the most resourceful channels. After some searching, you could write your first build script for neovim quickly. Then you just do:

conda build nvim -c conda-forge

Later install this package locally. In my case, --use-local doesn't work, so I used this one.

conda install -c file://${CONDA_PREFIX}/conda-bld/<PACKAGENAME>

When build is working, you could upload it for example:

anaconda login # you need: conda install anaconda-client
anaconda upload ${CONDA_PREFIX}/conda-bld/linux-64/nvim-0.3.1-0.tar.bz2

Of course, you can do better using CI like Travis to automate releasing of all platforms. You can also push your recipe to conda-forge for publish like here or here

So now everyone who wants neovim in conda could do this in Linux/OSX, 🙌: 

conda install -y -c daizeng1984 nvim

Revisit My Previous Projects

If you worked with Python before, you must remember the nightmare with Python's native dependencies e.g. BLAS, OpenLdap, Cyrus-Sasl etc.. One of my python neovim plugin also has hard-to-get dependency. This sucks and frustrates a lot since we are advertised Python is very cross-platformy. Now with conda, you have more confidence to install all dependencies using merely conda install. There are still some Python packages (pip) or outlaws that don't respect conda, and blindly complains about e.g. missing lib's headers. In this case, we need keep the build folder (--no-clean), add lib and include locations in setup.cfg and reinstall.

For example, my neovim plugin Snip-n-Paste, requires pygobject which usually took quite some time to get it right, but now it's simply:

Another example is installing the servers for LanguageClient e.g. cquery.

Cons?

You got me, not perfect at everything, just for open *nix packages. You cannot install OS dependent app e.g. mac store app etc.. There are also some boundaries if you have some outlaw (mostly native code related) apps, e.g. maplot. In these case, you're still limited to brew install, but maybe think about: do you really need these not very portable bad guys?

conda might as well have issues with versions consistency. You need to choose the reliable channel. But again, when messed up, you can always revert back. To minimize these issues, I'd stick to conda-forge channel for all dependencies and when necessary, build my own. Therefore, for my conda experience so far, so good.

Of course, conda is not alone. For *nix environment, Nix as another package manager claims to be as cool as it and it seems to gain a lot of attention recently. However, no matter your choice, it's good to make your development environment as independent as possible.

...