Image too large; split extras to separate tag?

[lutzky]

From #38 (comment):

I'm considering reverting this change, as fava-portfolio-returns depends on beangrow, which in turn depends on scipy. As the results of that dependency, the fava docker image is ballooning from ~60MiB to ~150MiB (with the uncompressed image to ~500MiB when inspected using https://github.com/wagoodman/dive).

Maybe it's better to build with this dependency with vxxx-extras and latest-extras labels, so that people can choose between a minimum setup and a full-featured setup. But I'm not sure how to do that the best with a combination of the Dockerfile and the actions workflow....

I like it! Here's what I'm thinking: We can use build variables for this.

• In the github workflow, add a matrix for two build_variant options - normal and extras.
• In the Build and push Docker image step:
  • Set the tags to conditionally add -extras as appropriate.
  • Add build-args: [BUILD_VARIANT=${{ matrix.build_variant }}]
• Split requirements.txt so some of it is in requirements-extra.txt.
• In Dockerfile, add ARG BUILD_VARIANT. Only perform the extra pip3 install steps for the extra variant.

What do you think?

[yegle]

Oh thanks for the tip on the matrix! I'll give it a go.