davidcellis / snakerun

Tool to launch python scripts in environments with their dependencies

** Archived after the acceptance of PEP-723 **

I will make a new module to support the PEP-723 semantics rather than continuing this one.

---

The format used for this is still subject to change.

This is currently an experiment with using the PEP-722 style dependency format and is only lightly tested.

SnakeRun

Launch python scripts with specific python versions and dependencies.

How to use

Inside your python script specify requirements in a PEP-722 style format.

eg:

# Script Dependencies:
#     prefab_classes>=0.9.3
#     requests

Optionally specify python version with a non-standard # x-requires-python: >=3.10.

Run with snakerun myscript.py Clear caches with snakerun --clear-cache

Logic

In order to save time on launching, environments are cached based upon the specification, not the environment. An environment will only be reused if the specification matches.

If there is no match then a new venv will be generated. If a python version is given it will be checked against the version used to launch snakerun first. If that is sufficient it will be used to create the venv.

Otherwise:

• On Windows:
  • snakerun will ask the py launcher for available python versions if a sufficient version is found that will be used, otherwise an error will be raised. (On windows this will only look for 3.X, not 3.X.Y)
• On Linux/MacOS:
  • snakerun will look for pyenv environments. If a sufficient version is found that will be used, otherwise an error will be raised.

If dependencies are specified they will be installed into the new venv and then the script will be launched.

If no python version or dependencies are given, the version of python running snakerun.py will launch the script without creating a venv.

Performance considerations

As a tool that is intended to start, find an environment matching and launch as quickly as possible in the fast case, many modules that would otherwise make development easier are avoided in the code that will be loaded in this case.

Most of the available argument parsers and the modules they depend on add a significant proportion of runtime so the current goal is to avoid them if possible.

• re would be necessary for most current commandline parsers but adds 60% to the launch time (on my development machine)
  • This is actually still used when pip installed as part of the entrypoint script that is generated. There is a script included to replace that entrypoint script if so desired.
  • Note: on windows re gets used as part of the zipapp format anyway
• argparse would be useful to allow for the addition of command line arguments but nearly doubles the launch time.
• subprocess is the new intended way to launch processes, but os.spawnv is used in the fast path because importing subprocess doubles the launch time
• packaging is useful to compare specifications but its use is delayed until after the initial comparison because importing packaging.specifiers would make launching take 4x longer.

All of these are used when it is clear that a new env will have to be constructed.