When building/installing Python dependencies in Docker/CI for multiple Python versions, keep the environment deterministic and aligned across interpreters: use the same dependency constraint/lock inputs everywhere and ensure the packaging toolchain is consistent per interpreter. Apply this by: - Always pass the shared constraints/lock file to pip installs (for every supported Python/PyPy build). - Before installing requirements, upgrade `pip`, `setuptools`, and `wheel` using the target interpreter. - Keep configuration formatting consistent (e.g., Docker `ENV KEY=VALUE` style) to avoid lint/tooling drift. Example (adapted): ```dockerfile RUN pyenv exec python3.11 -m pip install --upgrade pip setuptools wheel \ && --mount=type=cache,target=/home/$CELERY_USER/.cache/pip \ pyenv exec python3.11 -m pip install -r requirements/test.txt \ --build-constraint requirements/constraints.txt # Prefer consistent ENV syntax ENV PYTHONUNBUFFERED=1 ENV PYTHONIOENCODING=UTF-8 ```