eugene yokota

For this to work, I think there needs to be `mvn install` equivalent feature for Coursier. Copy-pasting Coursier cache sometimes works, but likely not safe since you could end up...

I did implement retrieve feature for sbt Launcher in https://github.com/sbt/launcher/pull/86, so I wonder if that can be reused here actually.

There are Coursier fields called `extraProperties` and `forceProperties`. We should expose either of them to sbt via lm-coursier: https://github.com/coursier/coursier/blob/d9a0fcc1af4876bec7f19a18f2c93d808e06df8d/modules/core/shared/src/main/scala/coursier/core/Resolution.scala#L677 A workaround probably would look like using `cs` CLI to grab...

To a potential contributor: The relevant files would be similar to https://github.com/coursier/sbt-coursier/pull/205

Nice!

sbt uses lm-coursier-shaded_2.12 (https://repo1.maven.org/maven2/io/get-coursier/lm-coursier-shaded_2.12/). I don't think you can use the latest pre-releases unless you're willing to build and publish the artifact locally or something.

I am open to suggestions. `remoteCacheId` is a task, so you can choose whatever the implementation that makes sense to you. Currently I think I try to calculate the content...

I am guessing this is a duplicate of https://github.com/sbt/sbt/issues/5568. sbt currently does not bubble up `Runtime` scoped dependencies transitively.

@viveksjain Thanks for the report. For the sake of reproducibility, could you suggest exact line someone could make in the Spark code base that would reproduce this behavior please?

Inside the `target` directory, sbt maintains incremental state file also known as Analysis. The Analysis contains hopefully-machine-independent content hash. If I remember correctly, timeWrapBinaryStamps is a way of optimizing this...