Support Build Arg to Override Default requirements.txt

[derks]

The standard for Python+PIP is a requirements.txt obviously, however in my use case for a project there are officially no required dependencies, but you can optionally install extensions that might have external dependencies. Therefore you end up with an official requirements.txt that is empty, but also a requirements-dev.txt that has all the optional requirements for development only.

It would be a minor change to set this as a build time arg, and not introduce any backward compatibility issues.

ARG REQUIREMENTS_FILE=requirements.txt
ONBUILD COPY $REQUIREMENTS_FILE /usr/src/app/

And then us outliers could utilize it:

$ docker build --build-arg REQUIREMENTS_FILE=requirements_dev.txt ...

Or more realistically, this will be from a development focused docker-compose.yml:

version: "3"
services:
    myapp:
        image: myapp:dev
        build:
            context: .
            args:
                REQUIREMENTS_FILE=requirements-dev.txt
        volumes:
            - .:/usr/src/app

[yosifkit]

Unfortunately this change would have not have the intended effect for the end user since the ARG would be processed by our builder. When the user does FROM python:onbuild and builds, only the ONBUILD lines get triggered in the dependent image.

We are currently working toward removing onbuild images from the official images: docker-library/official-images#2076, since they are easily replaced with a more transparent Dockerfile and are inadequate beyond basic use cases as is the case you ran into with this issue.

So we recommend not using a Dockerfile of FROM python:onbuild and to instead be explicit:

FROM python:3.6

WORKDIR /usr/src/app

COPY requirements.txt /usr/src/app/
RUN pip install --no-cache-dir -r requirements.txt

COPY . /usr/src/app

This way you can change what needs to be copied in with or instead of requirements.txt and even make it FROM python:3.6-alpine and have more control over the size of the image by adding and removing build dependencies within the layer that they are used.

[derks]

@yosifkit thank you for the feedback, and for clarifying that for me. I will move forward with your suggestions, thanks.