Hi! I'm autofix.ci, a bot that automatically fixes trivial issues such as code formatting in pull requests.

I would like to apply some automated changes to this pull request, but it looks like I don't have the necessary permissions to do so. To get this pull request into a mergeable state, please do one of the following two things:

1. Allow edits by maintainers for your pull request, and then re-trigger CI (for example by pushing a new commit).
2. Manually fix the issues identified for your pull request (see the GitHub Actions output for details on what I would like to change).

with the R8 classpath fix I'm very close to replicating the same test apk as gradle

This PR is far from functional, just posting if someone knows a few tips on the best way to exclude the dependency tree of the app module from the androidTest one

Each JavaModule has a coursierProject representing the internal coursier representation of the dependencies and constraints. Those get collected in transitiveCoursierProjects before resolution starts. Maybe, you can hook in there, to make the app module dep tree intransitive or something like that.

This PR is far from functional, just posting if someone knows a few tips on the best way to exclude the dependency tree of the app module from the androidTest one

Each JavaModule has a coursierProject representing the internal coursier representation of the dependencies and constraints. Those get collected in transitiveCoursierProjects before resolution starts. Maybe, you can hook in there, to make the app module dep tree intransitive or something like that.

thanks, I have it somewhat functional now. The problem is this as the steps in the packaging flow goes (just thinking out loud):

1. The dependency tree is needed for compilation
2. when packaging the test apk, the dependencies that were used for the base apk need to be excluded. This is somewhat problematic to do on a jar by jar basis (I started with this approach) because the versions don't always match between what the test deps pull and what the base module pulls

This extends beyond the jars, anything like android resources, native libs, meta info files. So it seems on a first glance that this transitiveCoursierProjects can be useful, if the androidTest compileClasspath has the base apk deps so that we can compile the test code , but not package it.

Thanks for the tip, now that I have 2 complex examples working, I can make this less hacky!

EDIT: Hm, I think after introducing the androidR8Jar I can use that in the android compile classpath and have no module dependency between the 2

For the case of jet-chat I've even managed to exclude redundant things that are included in the gradle test apk

Mill with this PR

Gradle

EDIT: Hm, I think after introducing the androidR8Jar I can use that in the android compile classpath and have no module dependency between the 2

The problem is that I still have to compute the dependency exclusions, because the test deps might bring the same dependencies as in base module/apk and even worse , different versions.

I've done some cleanup, avoiding duplication by overriding the resolvedRunMvnDeps.

I've tried a few other methods but it starts to get a bit complicated to play with coursier + manage the kotlin friend paths, I ended up having even more duplication.

I think with the latest commit, the implementation is fairly clean, but I'm open of course to further comments and suggestions to try .

In terms of technical things and results, this PR allows mill to produce tiny test apks.

Some result comparisons:

With the newest changes, there's a problem with androidtodo, so still not ready but will update accordingly

EDIT: I've added a small workaround for now for androidtodo and also kept backwards compatibility if the new AndroidR8InstrumentedTests is not extended

In addition to these size improvements, there's further improvement of setting up mill demonstrated here with a lot of duplicated dependencies not being needed in android test . Also the r8 error to warning mapping is no longer needed

If one wishes to reproduce these results, this branch should be built locally and then start an emulator and run mill Jetchat.app.androidTest .

I need to figure out how to fix mima due to the added bom argument in a few methods, but apart from that , I'm open to review and comments