Is there a way to access workspace files during feature execution?

[nikaro]

I am trying something like this:

{
  "name": "exercice2-go",
  "image": "mcr.microsoft.com/devcontainers/go:1-1.21-bullseye",
  "features": {
    "ghcr.io/nikaro/features/ansible:1": {
      "dependencies": "-r ${containerWorkspaceFolder}/deploy/requirements.txt"
    }
  }
}

The feature uses the dependencies argument like this:

if [ -n "${DEPENDENCIES:-}" ]; then
  # ensure packages are separated by spaces
  DEPENDENCIES=$(printf '%s' "$DEPENDENCIES" | tr ',' ' ')
  "${VENV_PATH}/bin/pip" install $DEPENDENCIES
fi

And it fails with this error:

ERROR: Could not open requirements file: [Errno 2] No such file or directory: '/workspaces/Exercice_2/deploy/requirements.txt'

Is it possible to do something like that in some way?

[samruddhikhandale]

Yes, you can access workspace files during the execution of a Dev Container Feature. However, the timing of when the Feature scripts are run in relation to when the workspace files are mounted can affect this.

In a Dev Container, the workspace files are mounted into the container at the path specified by the workspaceMount property in devcontainer.json (or at /workspaces/<folder-name> by default if workspaceMount is not specified).

Feature scripts are run after the Docker image is built but before the workspace files are mounted. This means that if you try to access workspace files in a feature script, they might not be available yet.

To work around this, you can use a postCreateCommand or postStartCommand in devcontainer.json to run commands after the workspace files are mounted. These commands are run every time the container is created or started, respectively. I believe, you could also make use of Feature's lifecycle hooks.

Looping in @joshspicer, feel free to correct my understanding if necessary.

[nikaro]

Thank for your answer!

Using postCreateCommand (or similar) would defeat the purpose of the feature, because otherwise i could do everything i want through postCreateCommand with arbitrary commands.

AFAIK, feature's lifecycle hooks cannot access features arguments (dependencies in my example)?

[joshspicer]

That's true that lifecycle hooks don't have access to Feature arguments - that sounds like a helpful feature request, and would be useful in this scenario.

One way I can think to do this today is similar to how the git-lfs Feature works, writing the lifecycle hook script in the install.sh and injecting the relevant Feature's options. That script is then run as a postCreate hook once the filesystem is mounted.

Note the usage of both EOF and 'EOF' to control when variables are substituted or not (https://unix.stackexchange.com/questions/138418/passing-a-variable-to-a-bash-script-that-uses-eof-and-considers-the-variable-a)

[joshspicer]

Using postCreateCommand (or similar) would defeat the purpose of the feature, because otherwise i could do everything i want through postCreateCommand with arbitrary commands.

Also, I don't think this is defeating the purpose of the Feature. You still need to install ansible, which is a natural thing to do in the Feature's install.sh script. Tying the lifecycle hook to the Feature also makes this functionality more portable (people can just add your Feature instead of copy/pasting config into their repo!)

[nikaro]

Yes, the postCreateCommand at the feature's lifecycle hook would do the trick, but not at the .devcontainer.json level (or at least it would feel like a hack/workaround to me).

I will try the trick of the git-lfs feature, i think it should work for my use case.

[nikaro]

I confirm that it worked. Thank you @samruddhikhandale and @joshspicer for your help 🙏

Here is the feature for the record: https://github.com/nikaro/features/tree/main/src/ansible

[joshspicer]

Thanks for sharing!