Proposal: Ability to drop to an interactive shell via Earthfile

[dchw]

Introduce a new command that triggers the interactive debugger as a step. Once the shell session is complete; aggregate all changes made in the container and continue execution (including the modifications).

Here are the options I have thought of for syntax:

my-target:
    FROM alpine:latest
    INTERACTIVE SHELL
   ...

my-target:
    FROM alpine:latest
    DEBUG
    ...

Use-cases:

• Explore problematic containers that otherwise build fine with Earthly
• Debug at any state
• Allow manual modification for development and/or testing while creating an Earth target
• Enables a repl-like development cycle

Alternatives:

• Right now, to get similar effects, you would need to insert a failing command (I usually use RUN breakhere); and then include the interactive debugging mode CLI argument. This is cumbersome and annoying; plus it does not allow for continuation after the fact (since something has already failed).
• Build the image and explore it post-build using tools like dive. This lets you see the state of the filesystem at a given point; but its not easy to run commands at a given layer. Additionally; we are encumbered more by the lack of good layer descriptions in our built images right now.

[vladaionescu]

Another possible syntax could be

RUN --interactive /bin/sh

Then you could also run arbitrary programs directly. Like RUN --interactive python or RUN --interactive node.

Possible extension: in interactive mode, also allow bind-mounting any local folder. This could enable watch mode use-case for front-end developers.

[leighst]

Was this added ?