Add BUILD_UUID to app bundle manifests

[simonbowring]

Goal

Add a Build UUID to app bundle manifests.

The Build UUID currently only gets added when generating APKs directly without using app bundles.

Changeset

• Return a path to the bundle manifest when generating an app bundle rather than the APK manifest
• Add the Build UUID to the bundle manifest rather than the merged manifest when an app bundle is being generated
• Use the Gradle task graph to determine whether an assemble task (generate APK) is being run or bundle task (generate app bundle)
• When manually uploading a mapping file, determine which manifest file to read based on which one has a Build UUID.
  • Incidentally, both a merged manifest and bundle manifest get generated for assemble and bundle tasks, but we only put the build UUID into one of them.

Tests

Made sure that the mazerunner tests pass.

Manual tests:

• Check that an app built with the bundleDebug Gradle task contains the Build UUID in the manifest in the app bundle
  • Checked with AGP versions:
    • 3.2.1
    • 3.3.0
    • 3.3.2
• Check that an app built with the assembleDebug Gradle task contains the Build UUID in the manifest in the APK
  • Checked with AGP versions:
    • 3.0.1 (pre-app bundles)
    • 3.2.1
    • 3.3.0
    • 3.3.2
• Check that manually uploading a mapping file after building a bundle then the build UUID is read from the bundle manifest
• Check that manually uploading a mapping file after building an APK then the build UUID is read from the merged manifest

[fractalwrench]

I left one comment inline, otherwise I believe this should work for the common case. We should verify in an example app running AGP 3.3.0, and one running AGP 3.0.1, which won't have App Bundles. We should also test whether this fix works with DexGuard - I'm not sure that it will as currently implemented.

This approach is not ideal in the long-term as it effectively doubles the amount of IO each task needs to perform, regardless of whether App Bundles are actually being used. If other manifest locations are added to the build folder in future, then this will become even more of a problem. Also I think this approach is likely to make it harder to make this task incremental, which has been scheduled in the past.

If we can verify that the plugin works for the above scenarios then I'm happy to approve as a short-term solution, with the caveat that we should revisit the design at a later date.

[simonbowring]

I've made a change to the approach so that either the merged manifest, the bundle manifest or both are written to depending on which tasks are in the Gradle task graph for each variant.

This has also been tested with an app running AGP 3.3.0, and one running AGP 3.0.1

Tasks left to do:

• Test with DexGuard
• Get mazerunner tests working

[fractalwrench]

I believe we've recently changed this to throw an exception in #151

[simonbowring]

I haven't changed this block of code, only the line numbers have changed. https://github.com/bugsnag/bugsnag-android-gradle-plugin/blob/master/src/main/groovy/com/bugsnag/android/gradle/BugsnagVariantOutputTask.groovy#L74

Maybe you mean this which changed in #151?
https://github.com/bugsnag/bugsnag-android-gradle-plugin/blob/master/src/main/groovy/com/bugsnag/android/gradle/BugsnagMultiPartUploadTask.groovy#L43

[fractalwrench]

I believe the mazerunner tests are failing as the fixture's gradle wrapper has been updated, which seems to alter the order in which requests are sent. Setting it back to the previous version passes the scenarios, and the failing scenarios appear to be in the wrong order when inspecting the maze_output directory. I don't think this is an issue in a real world scenario and is just something that needs fixing in our tests.

[simonbowring]

I've reverted the gradle and AGP version changes in the maze fixture.

[fractalwrench]

I'm happy to approve this as a short-term fix with the caveat that this has been tested to work with DexGuard. We should create a ticket to address previous concerns I've raised about this general approach at a later date.

[simonbowring]

I've tested this with DexGuard, deobfuscation is working.