Support deploy environments (deploy_env)

[hmemcpy]

(subsumes #560, expands on bazelbuild/bazel#1402)

Bazel has an internal feature called deploy_env for java_binaries, which allows performing classpath subtractions for specified targets (see a comment by Ulf here).

We tested this feature (by cherrypicking the commit to expose it in java_binary) and it is suitable for our needs for specifying different deployment targets having different dependencies (something we can't use neverlink for).

As other people seemed to find it useful as well, I would like to propose adding support for this for scala_binaries. The classpath subtraction can be done in skylark using the code we already wrote, to support deployment scenarios explained in #560.

WDYT? cc @ittaiz

[johnynek]

Do we need to do anything in rules_scala? Why not use java_binary?

We have talked about getting rid of scala_binary or making it only a macro that wraps java_binary. Does his ticket inform that discussion?

[samschlegel]

It looks like support for deploy_env in java_binary has finally landed in master https://github.com/bazelbuild/bazel/commit/a92347e405ebe022a7f216541aaa46753e311563

[johnynek]

did that make it into 0.22?

[samschlegel]

Doesn't look like it

[thundergolfer]

For anyone else arriving here, deploy_env is on track to be released with 0.23 in late Feb 2019. https://github.com/bazelbuild/bazel/issues/6495#issuecomment-460932694

[stijndehaes]

Any updates on this? Would be useful for building fat jars for spark application. I do not want to include spark every time as this dependency is big and already exists on the spark cluster itself

[johnynek]

Did you try using deploy_env with java_binary?

You should be able to use java_binary with scala_library.

[johnynek]

You could also use jar jar to zap the classes you don’t want from the deploy jar:

https://github.com/johnynek/bazel_jar_jar

[ittaiz]

Stijn, What about using java_binary like Oscar suggested back then? I don’t remember why we didn’t want to use it but it does make sense

On Sat, 20 Jul 2019 at 0:45 P. Oscar Boykin notifications@github.com wrote:

You could also use jar jar to zap the classes you don’t want from the deploy jar:

https://github.com/johnynek/bazel_jar_jar

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/bazelbuild/rules_scala/issues/576?email_source=notifications&email_token=AAKQQF3PD3RDIT3EGI6MYNLQAIYW5A5CNFSM4FONJVC2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2M27JA#issuecomment-513388452, or mute the thread https://github.com/notifications/unsubscribe-auth/AAKQQF7R6LVHEZWCJKKEUNDQAIYW5ANCNFSM4FONJVCQ .

[thundergolfer]

What about using java_binary

This is what we do. It works. The main_class can be 'fake', and you put the provided deps in runtime_deps. You then can use this target like so:

deploy_env = [
   ""//tools/build/spark-cluster-runtime",
]

[johnynek]

The reason we made scala_binary is that it predates “java sandwich” so java could not depend on scala rules. Now that motivation is gone.