support build stamp with docker_push

[mikedanese]

support build stamp with docker_push

[mattmoor]

docker_push supports make-style expansion for the components:

vardef("MY_TAG", "latest")

docker_push(
    name = "foo_push",
    image = ":foo_image",
    registry = "gcr.io",
    repository = "foo/bar/baz",
    tag = "$(MY_TAG)",
)

Or is that not what you want?

[mattmoor]

@damienmg How can I access this in skylark?

[mattmoor]

Ok, blocked by: bazelbuild/bazel#1054

[ixdy]

Similar / clarifying request: I'd like to be able to tag an image automatically using the git sha of the workspace. We use --workspace_status_command to set this for go rules, but there's currently no way to use that in docker rules.

Make-style expansion isn't ideal, since that requires adding --defines to the bazel command line.

[mattmoor]

In other places we allow @foo to load the value from a file. I wonder if this would be sufficient for your purposes?

[ixdy]

That might work. I implemented a hack (ixdy/bazel@3b29803) to do that a few months ago, and it's what the kubernetes bazel build is currently using, though it still feels pretty hacky

(I got preoccupied with other things and didn't follow up on some review feedback, and now things have moved around a bit so I'd need to rebase my change.)

The go rules got around waiting for Skylark support (bazelbuild/bazel#1054) by implementing stamping through reading the ctx.info_file and ctx.version_file directly.

Since IIRC most of the actual docker image/layer-building implementation happens in python, I wonder if we could do something similar and have the info/version files passed through to the scripts on stamp=1?
I haven't thought through all of the implications yet, or if the rules will work correctly with unexpanded variables in skylark.

[mikedanese]

Is this fixed?

[mattmoor]

docker_push doesn't support it, but docker_bundle does.