Why do I no longer use Anaconda？

This post discusses the pros and cons of Anaconda and why I no longer use it.

Today, I finally made up my mind to remove Anaconda, which was taking up more than ten GiB of space. Anaconda played a significant role in my journey of learning and using Python, and I’m grateful for its existence.

However, I’ve now abandoned Anaconda and hope that readers studying computer science will also avoid using it. There are three reasons for this:

Independent Compilers/Dynamic Libraries from the System

The Python versions that come with Linux systems are CPython, compiled by C/C++ compilers. Therefore, if you use the system’s compiler to build a pybind11 project, the resulting module can be directly recognized by the system’s Python.

Anaconda’s Python, on the other hand, is different: to ensure consistency and ease of use, Anaconda uses pre-built binaries and corresponding versions of dynamic libraries, independent of the system’s compilers and dynamic libraries. As a result, the interoperability between Python in Anaconda’s virtual environment and the system’s compilers & dynamic libraries is poor. This caused significant issues in a project I worked on that used pybind11 (we needed to export C++ interfaces to Python).

It’s undeniable that this "separate ecosystem" approach makes Anaconda work out of the box, but the cost is also substantial.

PATH Pollution

As mentioned above, Anaconda comes with its own binaries and dynamic libraries, and it includes far more than just Python—tools like OpenSSL, mpic{c|xx}, and many others are also included. These tools make it convenient for Anaconda to maintain its own runtime environment, but when conda adds its bin to the PATH, it pollutes the system’s PATH, overriding critical files like the system’s OpenSSL and causing extremely difficult-to-debug mysterious bugs.

Huge Space Consumption

The reason is also mentioned above. To maintain its own environment, Anaconda needs to install various packages and tools. Before I deleted Anaconda, it was taking up 11.4 GiB of space on my system.

What to Use Instead?

I’m now using poetry. If you need manual multi-version dependency management, consider pyenv with the pyenv-virtualenv plugin.