Skip to main content

Storage, notification, permission, and background compatibility

These areas evolve frequently because they affect privacy, battery, and user control. Model user intent and durable work first; then map it to platform-specific mechanisms.

ConcernCompatibility questions
PermissionsDeclared? Runtime request needed? User denied/don't-ask-again? One-time/auto-reset? Enterprise policy?
NotificationsChannel exists? Runtime permission? User/app/channel disabled? PendingIntent valid?
StorageApp-private, MediaStore, SAF, or shared media? URI permission persisted? File moved/deleted?
Background workImmediate user-visible or deferrable? Constraints? FGS type/permission/start eligibility? Exact timing justified?

State-machine example​

sealed interface NotificationReadiness {
    data object Ready : NotificationReadiness
    data object NeedsRuntimePermission : NotificationReadiness
    data object DisabledByUser : NotificationReadiness
    data object ChannelDisabled : NotificationReadiness
}

A Boolean hasPermission cannot explain the correct recovery UI. Query current state when acting rather than assuming yesterday's grant remains. WorkManager, foreground services, and alarms are not interchangeable fallbacks; each represents a different guarantee and policy.

For storage, persist URI permission only when the provider offers it and still handle revoked/missing content. Never turn a content URI into a guessed filesystem path. For notifications, creating a channel after posting is too late, and channel importance is user-controlled after creation.

Upgrade test scenarios​

  • Old app scheduled work, then OS/app target upgrades.
  • Permission was granted, auto-reset/revoked, then process restores.
  • User restores backup onto a device where a document URI is invalid.
  • Existing notification channel has user-modified importance.
  • Exact alarm/foreground-service eligibility changes while work is queued.

Practical mistakes​

  • ❌ Route every platform failure to “Grant permission.”
  • ✅ Explain whether the remedy is request, settings, channel settings, retry, or unsupported.
  • ❌ Fall back from failed WorkManager to an unrestricted service.
  • ✅ redesign around allowed platform guarantees.
  • ❌ Test only new installs.
  • ✅ test historical upgrades and persisted system references.

Sources​