Updating to 2.1.8 breaks non-android builds if ANDROID_HOME variable is set in the system but android SDK isn't configured

[hb-man]

After updating to 2.1.8 I noticed a new transitive dep on rules_android in lock file.
And than I run into following error on a non-android build of hirschgarten

	File "/private/var/tmp/_bazel_Danil.Tereshchenko/a1a6972b46838be5e55c38c576f4e9d8/external/rules_android+/rules/android_sdk_repository/rule.bzl", line 114, column 13, in _android_sdk_repository_impl
		fail("No Android SDK apis found in the Android SDK at %s. Please install APIs from the Android SDK Manager." % android_sdk_path)
Error in fail: No Android SDK apis found in the Android SDK at dummy. Please install APIs from the Android SDK Manager.
ERROR: Analysis of target '//ijwb:ijwb_bazel_zip' failed; build aborted: No Android SDK apis found in the Android SDK at android_sdk_folder. Please install APIs from the Android SDK Manager.
INFO: Elapsed time: 12.611s, Critical Path: 0.01s
INFO: 1 process: 1 internal.
ERROR: Build did NOT complete successfully

turned out, now if you build even non-android code, but you have ANDROID_HOME (possibly ANDROID_SDK_HOME as well) variable in your system withotu actual android SDK configured - it will fail, because it tries to fetch rules_android

Reproduction:

1. git clone https://github.com/JetBrains/hirschgarten
2. bazel build --action_env=ANDROID_HOME=dummy --action_env=ANDROID_SDK_HOME=dummy //tools/format/...
3. bazel build --action_env=ANDROID_HOME= --action_env=ANDROID_SDK_HOME= //tools/format/...

in step 2 you'll see that it errors out, in step three it builds. Build doesn't depend on android.
same behaviour can be reproduced in https://github.com/bazelbuild/intellij - it also uses rules_kotlin

[restingbull]

$ bazel query --implicit_deps 'let alldeps = deps(//tools/format/...) in filter(".*kotlin.*", $alldeps) union filter(".*android.*", $alldeps)'
INFO: Empty results

I'm not clear where rules_kotlin figures into this behaviour. Can you produce a smaller repo example?

cc @ahumesky -- is this expected behaviour if ANDROID_HOME is anywhere in the environment?

[restingbull]

bazel aquery --implicit_deps '//tools/format:ktfmt' | grep android

Also returns an empty set. Though, the java actions do include Environment: [ANDROID_HOME=, ANDROID_SDK_HOME=, PATH=/bin:/usr/bin:/usr/local/bin] -- so linkage may be coming from inside bazel.

[restingbull]

It is worth noting that it only fails if the ANDROID_HOME/SDK_HOME is set to something invalid.

edit

Ah, here we are. If ANDROID_HOME is empty, it creates a dummy repo.

I suspect the android_sdk_repository is an implicit dependency of the java rules (somewhere) if it exists.

[JetBrainsBazelBot]

It is worth noting that it only fails if the ANDROID_HOME/SDK_HOME is set to something invalid.

on our CI we have ANDROID_HOME/SDK_HOME set up to valid paths, but android SDK itself is not fully configured

here is a minimal example repo, containing only dependency to rules_kotlin
example.zip

1. bazel build --action_env=ANDROID_HOME=dummy --action_env=ANDROID_SDK_HOME=dummy //...
2. bazel build --action_env=ANDROID_HOME= --action_env=ANDROID_SDK_HOME= //...

[laureanobrs]

Also, another scenario where this can happen is when using RBE. Example:

1. On the GitHub Actions runners, the Android SDK is configured and the ANDROID_HOME/SDK_HOME variables are set.
2. When running bazel build on that runner, the SDK is detected, but the execution is sent to the RBE remote executors (which, in this case, do not have the Android SDK installed).
3. The build fails.

In my case, as a workaround, I overwrote the ANDROID_HOME/SDK_HOME variables by setting them to empty on the GHA runner before running the build.