Support setting appendix on Dockerfiles too

[yuvipanda]

Until now, 'appendix' was ignored for repos built with
their own Dockerfile. This fixes that, and sets the appendix
on repos built with Dockerfiles too.

I want to set a REPO_URL env var for all images built with
repo2docker-action
(jupyterhub/repo2docker-action#86)
and this is needed for that.

[yuvipanda]

https://github.com/jupyterhub/binderhub/blob/e04a8c3cec71ae8a1bf984da8623ecf8d81398bb/binderhub/binderspawner_mixin.py#L102 is how BINDER_REPO_URL is set. Instead, I just want to set REPO_URL in repo2docker itself.

[consideRatio]

My understanding made explicit:

• repo2docker can be run with an --appendix flag, but when a Dockerfile was created, it wasn't honored.
• REPO_URL is the proposed name of a new environment variable to solve a problem.

It is my understand that a discussion about an env called REPO_URL would be a new environment variable. If that is correct, I'd very much like that REPO_URL instead is named REPO2DOCKER_REPO_URL or similar. Environment variables with a prefix related to the source code that defines them is a major help to avoid confusion, not only the the project that defines them but in other projects where the confusion otherwise propagates. I'm thinking for example on the amount of questions jupyterhub/z2jh/binderhub could get about REPO_URL that really was about repo2docker if we would leave out a REPO2DOCKER prefix.

[yuvipanda]

@consideRatio ah sure re: naming of the env var. I'll open a separate PR to deal with that, this PR doesn't touch that.

[consideRatio]

This LGTM!

[consideRatio]

I labelled this as a bug because it wasn't documented that --appendix would only work for non-Dockerfile situations.

[manics]

Note this was discussed in a previous PR #1097 (comment)

[betatim]

I still like the logic/argument that manics linked to. This means "no appendix for customer Dockerfile based builds is a feature, not a bug".

[consideRatio]

Ah hmmm...

I haven't formed a clear opinion if it is a feature or a bug, but right now it is an undocumented exception as I understand it - which i consider a documentation bug at least.

For now I'll leave a blank vote with regards to making it supported or explicitly fail/emit warning, and a strong vote for any of those options is better than undocumented and silently ignored situation we currently have. Ping me for further input, I feel I'd need to onboard myself a bit further to be able to think clearly what I consider to make sense about supporting it or not.

[manics]

I've made an alternative proposal in jupyterhub/repo2docker-action#86 (comment)

[yuvipanda]

@manics I like that :)

[yuvipanda]

Going to close this as this was considered and rejected. I do agree with @consideRatio this should be documented though.