Developer environment Dockerfile for ide

[bhack]

I want to add a Vscode devcontainer entry for tensorflow/addons in microsoft/vscode-dev-containers#14.
Something like the already existing Tensorflow-examples repo.

But it is not clear to me what to put in a complete developer environment Dockerfile receipts for Addons cause the current docker images receipts are fragmented in stages (Test, pre-commit, build etc..) as I see in https://github.com/tensorflow/addons/tree/master/tools/docker

[gabrieldemarmiesse]

I believe that gpu_tests.Dockerfile has all the necessary tools for developement. It's what we actually tell people to use in the CONTRIBUTING.md to compile interactively. For the code format, no need to have the tools installed locally. bash tools/pre-commit.sh works in pretty much all situations, even on Windows.

[bhack]

I suppose that we need IDE python and cpp extensions for the addons right?

[gabrieldemarmiesse]

I'm not sure I understand your question. Could you expand on this?

[bhack]

I meant developer need to have both python and cpp support in the IDE right?
Also have you arleady a published gpu_tests.Dockerfile image on Dockerhub where I can start FROM?
If not I need to copy tools/install_deps/tensorflow.txt requirements.txt and tools/install_deps/finish_bazel_install.sh to replicate the same image build.

[seanpmorgan]

@bhack per your comment on the Discord issue, would you like to join our gitter if you think thats a better discussion forum?
https://gitter.im/tensorflow/sig-addons

We've avoided pushing docker images to hub since we have no access to the tensorflow account and creating more accounts which we have to share across a community SIG has challenges. Not to say it can't be done we could use GitHub secrets and push images through CI.

[bhack]

An initial stub is at microsoft/vscode-dev-containers#258.
Probably we want some other kind of stuffs in the container cause the formatting is in the IDE but the LSP (languge server protocol) daemon run in the (local/remote) container.

[bhack]

I moved the PR here at #1309 cause it is directly ready for devs with Vscode at checkout.

[seanpmorgan]

@karmel as discussed in our monthly meeting we would like to be able to publish our own development docker image for this issue and other reasons. #1309 may have more details on this particular request.

Would you be able to get us in touch with someone who would know if it's possible to publish under tensorflow namespace on DockerHub. That would be preferable to how we're currently publishing images under one of our own accounts. Perhaps commiting a Dockerfile to the tensorflow/tensorflow repo that gets picked up? That may be problematic since we need code from this repo, but would help to know what our options are