Add new "python-no-pyc" test

[tianon]

This test validates that the Python images do not contain any pre-compiled .pyc or .pyo files in /usr/local (or /opt, for good measure).

Ideally this would also include all of /usr (not just /usr/local) but then it would also have to filter out results that are from Debian packages (python's default variants are FROM buildpack-deps which includes a fair amount of Python packages).

See docker-library/python#207

(The goal is to simplify how we avoid .pyc and .pyo files in the images but with a safety net to validate that we don't regress.)

[tianon]

$ bashbrew list --uniq --arch-filter python pypy | xargs ./test/run.sh -t python-no-pyc
testing python:3.11.0a5-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.11.0a5-slim-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.11.0a5-buster
	'python-no-pyc' [1/1]...passed
testing python:3.11.0a5-slim-buster
	'python-no-pyc' [1/1]...passed
testing python:3.11.0a5-alpine3.15
	'python-no-pyc' [1/1]...passed
testing python:3.11.0a5-alpine3.14
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-slim-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-buster
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-slim-buster
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-alpine3.15
	'python-no-pyc' [1/1]...passed
testing python:3.10.2-alpine3.14
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-slim-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-buster
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-slim-buster
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-alpine3.15
	'python-no-pyc' [1/1]...passed
testing python:3.9.10-alpine3.14
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-slim-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-buster
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-slim-buster
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-alpine3.15
	'python-no-pyc' [1/1]...passed
testing python:3.8.12-alpine3.14
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-slim-bullseye
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-buster
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-slim-buster
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-alpine3.15
	'python-no-pyc' [1/1]...passed
testing python:3.7.12-alpine3.14
	'python-no-pyc' [1/1]...passed
testing pypy:3.8-7.3.7-bullseye
	'python-no-pyc' [1/1]...passed
testing pypy:3.8-7.3.7-slim
	'python-no-pyc' [1/1]...passed
testing pypy:3.8-7.3.7-buster
	'python-no-pyc' [1/1]...passed
testing pypy:3.8-7.3.7-slim-buster
	'python-no-pyc' [1/1]...passed
testing pypy:3.7-7.3.7-bullseye
	'python-no-pyc' [1/1]...passed
testing pypy:3.7-7.3.7-slim
	'python-no-pyc' [1/1]...passed
testing pypy:3.7-7.3.7-buster
	'python-no-pyc' [1/1]...passed
testing pypy:3.7-7.3.7-slim-buster
	'python-no-pyc' [1/1]...passed
testing pypy:2.7-7.3.6-bullseye
	'python-no-pyc' [1/1]...passed
testing pypy:2.7-7.3.6-slim
	'python-no-pyc' [1/1]...passed
testing pypy:2.7-7.3.6-buster
	'python-no-pyc' [1/1]...passed
testing pypy:2.7-7.3.6-slim-buster
	'python-no-pyc' [1/1]...passed

[tianon]

I wrote this because it's a little bit complicated, but on second thought I think we should probably change the message to say it's explicitly not a goal of the Windows images.

[imomaliev]

Hi, could you please help me understand this part

rem N/A: Windows-based Python images may or may not contain ".pyc" and/or ".pyo" files because they are not a significant contributor to the overall image size

I never used Windows's docker images and see that they are quite large, I assume the main contributor to their size is the base Windows image. After a bit of research, I see that there is also mcr.microsoft.com/windows/nanoserver image and using it as a base in possible future images could require fighting for every byte as in Linux based images. Is similar test checking py[co] for Windows would be hard to implement/maintain or this would be just premature optimization and is not worth the effort?

[tianon]

Yeah, the main size contributor there is definitely the base, and I think the .py[co] files come with the base install of pre-built artifacts from Python. Even if we were to implement docker-library/python#169, I'm not convinced it would make enough of a difference to be worth the hassle.

The flip side of that is that implementing a check for that in pure batch scripting is not going to be a very fun time. 🙈