Python SDK / Runtime: Suggestion for improving Environments, non-cloudevent types

[rabejens]

The current Python SDK has a number of shortcomings. It uses an outdated version of Python and it is cumbersome to add your own environment. Not many people use bare-bones Python today.

I suggest the following enhancements:

• Specify Python version and packages to use with Conda or requirements.txt
  • The buildpack should detect presence of an environment.y(a)?ml and create an env for it
  • Or, if only requirements.txt and an optional pip.conf is there, it could use an environment variable like BP_PYTHON_VERSION to specify the Python to use and build the environment from requirements.txt
    • environment.yml supports using pip by specifying a pip section:

      dependencies:
        - pip
        - pip:
          - --some-pip-option
          - --some-other-pip-option
          - some-package
          - some-other-package>41.0,<43.0

    • With this, the builder could create a temporary environment.yml from requirements.txt and also use conda / mamba (which is faster) to build the env
  • The env can then either be compressed and uncompressed to the same location on the target container, or one could use conda-pack but that might cause some problems.
  • The builder should also implicitly add dependencies to additional libraries needed when running the func (i.e., HTTP server, CloudEvents library)
• Ease the use of plain JSON (non-cloudevents)
  • Python has a concept called dataclasses that can easily be serialized to/from JSON

[lkingland]

Thank you for these suggestions. An update to the Python middleware is currently in progress, so thank you for your helpful suggestions.

[github-actions]

This issue is stale because it has been open for 90 days with no
activity. It will automatically close after 30 more days of
inactivity. Reopen the issue with /reopen. Mark the issue as
fresh by adding the comment /remove-lifecycle stale.