[tomkinstinch]

Python environment headaches have been largely solved for me thanks to Miniconda[1]. Not only do the environments isolate dependencies, they can easily use different versions of Python, and can include arbitrary binary packages too. It helps that for my work in biology an extensive number of packages are available from the bioconda[2] channel (with many non-bio packages from conda-forge[3]). Environments can be described via environment files, allowing them to be transferred to collaborators, managed with source control, or included with publications to support reproducible science. If you like virtualenvs, you may want to give Miniconda a try. One current limitation is that environment files cannot specify a source channel for specific packages--they're installed from channels based on the global channel precedence.

1. https://conda.io/miniconda.html

2. https://bioconda.github.io

3. https://conda-forge.org

[icebraining]

I used Conda recently and it was OK, but it's annoying that it can't install stuff directly from PyPI. Unless you're lucky, you still have to use Pip inside the Conda environment to install some packages.

[makmanalp]

A sparkle of hope is that the conda people recently announced in a github ticket that this would be supported as a first class citizen soon.

[jhbadger]

But it's annoying how many packages like qiime make it basically a requirement to use Anaconda/miniconda. I really don't want to pollute my computer with multiple installations of even things like R just to run a package.

[pwang]

What if the right answer to the packaging nightmare that plagues languages with native extensions, is to actually have multiple isolated copies of the interpreter, each with well-known build chain metadata and sane, isolated locations for shared libraries?