Implement support for dockerignore and containerignore

[sgaist]

Currently repo2docker creates a context object that includes the whole content of the repository it builds an image for. Thus it includes folders like .git which is usually something that has no interest in the final image, can take quite a lot of space and most importantly, kills the caching of that layer.

This patch adds support for reading dockerignore and containerignore files that are used to ensure only the relevant data are used to build the image.

By default it also excludes the .git folder if neither of these files are provided.

It makes use of docker-py's own file discovery mechanism to ensure coherence and avoid re-inventing a working wheel.

Note: containerignore comes from podman that reads either file. See the podman-build documentation.

Fixes #937
Related to #268 and #269 final discussions.

[welcome]

Thanks for submitting your first pull request! You are awesome! 🤗

If you haven't done so already, check out Jupyter's Code of Conduct. Also, please make sure you followed the pull request template, as this will help us review your contribution more quickly.

You can meet the other Jovyans by joining our Discourse forum. There is also a intro thread there where you can stop by and say Hi! 👋

Welcome to the Jupyter community! 🎉

[sgaist]

I just thought about something to discuss based on the default "ignore .git" behaviour introduced by this patch.

Since the goal of repo2docker is to provide a ready to use work environment with only the relevant pieces from a repo, should we have a configurable list of paths that we could ignore by default so people don't need to introduce another configuration file ?

I am currently mainly thinking of git and forge specific files and folders like .github, .gitlab-ci.yml, etc.

[manics]

I think we need to keep .git since it provides a link back to the source repo if users are in a terminal of have a Jupyter git extension installed. It's also good to have it since repo2docker is about reproducibility.

[sgaist]

@manics I follow you on the reproducibility side but I thought that the following labels were there for that:

LABEL repo2docker.ref="<commit hash>"
LABEL repo2docker.repo="git@gitlab.example.org/mygroup/myprojectgit"
LABEL repo2docker.version="2022.10.0"

?

I just saw that repo2docker.ref could contain HEAD which does not provide a useful information for reproducibility. The information would only be retrievable if one I explicitly runs the image to go into the git clone or based on the image generated name. May I suggest that HEAD should be replaced by the corresponding hash at image build time ?

In case we keep the .git folder by default, what are you thinking about my suggestion of a new argument that can be used with a list of files/folders to ignore ?

[consideRatio]

I rebased this PR by mistake, sorry for the noise!

[sgaist]

@consideRatio no worries !

On that matter, something I am unsure of, it it globally preferred to rebase or to merge main back to the branch ? I am used to do the former to have a "semi-linear" history as they say in GitLab's configuration. I thought I have read something about it in the contributor documentation but I can't find it anymore.

[consideRatio]

On that matter, something I am unsure of, it it globally preferred to rebase or to merge main back to the branch ?

There is clear no consensus, but I think of two situations. When a PR is accepted an merged/rebased/squashed, and how to handle PR development before that.

• We almost always use the merge button when merging PRs in jupyterhub org to retain commit authorship as I understand it.
• I appreciate rebasing my PRs before merging them to have an easier to follow git history, but I dislike if rebasing makes a reviewer uncertain what is already reviewed once and what are new changes etc. Due to this, I'm often rebasing PRs before review efforts are put in, but not after.

[sgaist]

Thanks @consideRatio, sounds good to me.

I haven't yet had a really big review on GitHub were I could see that rebasing did break the process for reviewers. Some times I have encountered policies that would be "we use merge from main" when updating a pull request but that is all. If it could help shape a consensus for the Jupyter project, we can use one of my contribution to do some testing in that regard.

[sgaist]

@manics a small ping to get your opinion an my reproducibility related questions above as well as .git handling.

[sgaist]

Having re-read #269, the ignore .git by default would be "out of spec" with regard to what docker does so I will remove it.

I'll open tickets for the other points as they are not directly related to this patch.

[welcome]

Congrats on your first merged pull request in this project! 🎉

Thank you for contributing, we are very proud of you! ❤️