My last question for the day: Do

docker_image

targets only support dependency inference on paths that "look like" the outputs of

pex_binary

targets? Or, can you directly put target addresses anywhere in the Dockerfile and have the dependency inference work? So if, for example I had a Dockerfile with:

FROM path/to/my/parent/docker/image:target

RUN pip install path/to/a/python_distribution:target

Would pants infer the dependency on both the parent docker image and the python package automatically, and inject the output paths for those targets into the build context automatically? I'm digging through the source code now but I honestly can't really figure out if it supports anything beyond

FROM target_address

and

COPY output_path_that_can_be_mapped_to_a_target

Let me check

I did not know it infers

FROM

. Is that really true as I cannot find this to be documented? What about

COPY --from=parent-image

?

It hasn’t found it’s way into the Docker section of the docs, but it is in the reference section for the

docker_image

target: https://www.pantsbuild.org/docs/reference-docker_image

Great point about the

COPY

! And in the

FROM

lines, will it only infer dependencies if you embed a target address there? Or will it detect an image URI that's in one of the image registries set up in your config, then search your repo for other

docker_image

targets that package and publish an image to the registry and image tag in the

FROM

line and infer a dependency on the matching

docker_image

target automatically?