`flutter build "apk" succeeds without libraries` from `native_assets_without_cbuild_assemble_test.dart` is slow and flaky

[fluttergithubbot]

The post-submit test builder Linux tool_integration_tests_6_6 had a flaky ratio 6.25% for the past (up to) 100 commits, which is above our 2.00% threshold.

One recent flaky example for a same commit: https://ci.chromium.org/ui/p/flutter/builders/prod/Linux%20tool_integration_tests_6_6/6
Commit: 461ad0d

Flaky builds:
https://ci.chromium.org/ui/p/flutter/builders/prod/Linux%20tool_integration_tests_6_6/6

Recent test runs:
https://flutter-dashboard.appspot.com/#/build?taskFilter=Linux%20tool_integration_tests_6_6

Please follow https://github.com/flutter/flutter/blob/master/docs/infra/Reducing-Test-Flakiness.md#fixing-flaky-tests to fix the flakiness and enable the test back after validating the fix (internal dashboard to validate: go/flutter_test_flakiness).

[bkonyi]

There's only been 36 runs on this configuration, so the reported flaky rate may be slightly higher than the actual rate.

It looks like flutter build "apk" succeeds without libraries in native_assets_without_cbuild_assemble_test.dart is failing due to a timeout and an unexpected exit code from Gradle.

I'm assigning this to @matanlurey (who recently rewrote this test) and @dcharkes (who owns native asset functionality) to investigate further.

[matanlurey]

I think a no-brainer move is to just disable the apk variant of the tests for now, the actual functionality being tested by this is specific for macOS, and while it would be great to have more coverage I wouldn't even know where to start in terms of debugging why this specific test configuration times out (maybe hidden Java heap issues?)

[matanlurey]

I also wonder if this is similar to @andrewkolos' other investigations (i.e. #157847) where it appears that flutter build apk can just take wildly different amounts of time or #152604 where OOMs are the cause.

[matanlurey]

Interestingly enough android_plugin_example_app_build_test.dart which otherwise is doing something very similar, does not appear to have this problem. The only difference I can see is the AGP version being used?