Transition to exec configuration for `bazel test --run_under=... --platforms=...`

[tpudlik]

Description of the bug:

bazel test --run_under=:test_runner --platforms=:device is a natural way to run on-device tests with Bazel: you point --run_under to your test runner (which flashes tests to the hardware), and --platforms to the target device platform.

But it's awkward in practice because the target that --run_under points to is built for the target platform, even though it will execute on the exec platform. So you need to wrap your runner into a platform_data:

platform_data(
  name = "run_under_wrapper",
  target = ":test_runner",
  platform = "@local_config_platform//:host",
)

@katre @gregestren

Which category does this issue belong to?

Configurability

What's the simplest, easiest way to reproduce this bug? Please provide a minimal example if possible.

No response

Which operating system are you running Bazel on?

No response

What is the output of bazel info release?

No response

If bazel info release returns development version or (@non-git), tell us how you built Bazel.

No response

What's the output of git remote get-url origin; git rev-parse HEAD ?

No response

Is this a regression? If yes, please try to identify the Bazel commit where the bug was introduced.

No.

Have you found anything relevant by searching the web?

This is similar to the case of bazel run --run_under --platforms, discussed here.

Any other information, logs, or outputs that you want to share?

No response

[katre]

It seems like the right answer here is that, for the command bazel test --run_under=:runner --platforms=:platform :test, we should have the following:

1. :runner should be built for the test execution platform (whatever that is), probably the host platform but maybe some form of remote exec platform.
2. :test should be built for the platform specified by --platforms
3. If these aren't compatible, it's up to the runner to report the error in a useful fashion.

Is that a reasonable summary?

[katre]

I suspect that bazel run will also need similar handling. A quick glance at the code makes it seem like these are handled very differently.

[tpudlik]

Yes, that summary sounds right to me.

[tpudlik]

It turns out that the platform_data workaround I mentioned is somewhat inconvenient because of bazelbuild/rules_platform#11: if the binary target you wanted to invoke with --run_under had an args attribute, that's lost when the binary is wrapped by platform_data. In fact it's not clear (at least to me) how to implement a rule that forwards the args through a platform transition.

[gregestren]

What happens if :runner is only target-compatible with a Mac platform?

Should that change the whole test to run on a Mac execution platform?

Stated differently, should exec platform resolution be based on :runner's compatibility settings?

[tpudlik]

I'm not sure how this should interact with exec platform resolution: I spend all my time in the simple world where the exec and host platforms are one and the same! If my host platform is Linux and the :runner is only target-compatible with a Mac platform, I'd just expect an error stating that.