IssueTracker

Thanks for the report! It looks like the lint check in ParcelChecker.java has to be updated to account for the new behavior on sealed classes. Right now the checker just looks for @Parcelize annotations on the current class. It should also check whether there's a direct seelad superclasses that is annotated with @Parcelize.

It seems likely due to robolectric relying on targetSdkVersion to determine what SDK to load

https://github.com/robolectric/robolectric/blob/master/robolectric/src/main/java/org/robolectric/plugins/DefaultSdkPicker.java#L90

+hoisie@ as FYI

targetSdkVersion is the last fallback option when the sdk level is not configured using other ways. If the user has not conveyed at all which sdk version it should be using, it will use target sdk version.

See http://robolectric.org/configuring/ for the other ways to configure the SDK level, all of which will supersede the targetSdkVersion.

Is there a stack trace for this issue? It could be something else, maybe an issue with manifest parsing.

"If the user has not conveyed at all which sdk version it should be using" how is that done exactly when running robolectric from Gradle?

I did a little research: AGP suppose to call ProcessTestManifest to generate manifests for unit tests. This task calls Manifest merger with ManifestMerger2.MergeType.APPLICATION as a parameter. It's going with type PACKAGED_MANIFESTS. This makes targetSdk appears in merged manifest even if it was not defined (falling back to minSdk). Experimented with ProcessTestManifestTest.testUnitTestManifestPlaceholdersFromVariantApi

Some additional information like error or stack trace will be helpful.

I am not sure what trace to provide other than that targetSdkVersion in the manifest file should be there or robolectric will pick the wrong SDK version.

fix is in CL: https://android.googlesource.com/platform/tools/base/+/fee1c78dc91f3f54f4efb71a1256ddf9b0698d1c

Fix will be available in AGP 7.4 alpha 9