How to make Python repl startup 100x faster

Python packages are so large in size that they would quickly fill up Replit's 2 GiB temporary disk. This required a lengthy installation process every time, and packages with many dependencies like TensorFlow or Torch were unusable due to insufficient disk space.

To solve this, Replit released a new nix-based Python template. This template supports standard virtual environments and introduces an innovative Content-Addressable cache mechanism.

The new cache operates at the level of individual files rather than entire packages. It manages a pool of files based on each file's SHA-256 hash value, and instead of extracting files directly during package installation, it creates a symbolic link to files that already exist in the pool.

This approach dramatically reduces disk usage while allowing large-scale packages to run smoothly. Additionally, since the necessary files are already prepared in the cache, Python repl startup speed has increased dramatically.