#5397)

Introduces a new SessionDataService that is bound to by each process and
used to deliver session-ids.

Co-authored-by: jrothfeder <[email protected]>

Created a SessionLifecycleService class which contains all the
service-side code, and a SessionLifecycleClient object which handles the
connection to the server and the sending and receiving of events.

Avoid changing the api surface. Having `val`s at the file level causes
Kotlin to generate a public class with the filenameKt which adds a
public class to the public api surface: "The public api surface has
changed for the subproject firebase-sessions: error: Added class
com.google.firebase.sessions.SessionDatastoreKt [AddedClass]". So we
have to put them in a class or companion object.

…ance` so it can be accessed from a bound service. (#5407)

Co-authored-by: jrothfeder <[email protected]>

Make `SessionGenerator` injectable. Also made `WallClock` an object to
avoid needing to pass it around. Moved `collectEvents` out of
SessionGenerator since it will need to be handled in the service.

Co-authored-by: jrothfeder <[email protected]>

…ave vimean superpowers. (#5413)

Co-authored-by: jrothfeder <[email protected]>

Using the SessionGenerator to generate new sessions and create
SessionDetails instead of directly creating UUID session ids in the
service.

Also updated the boundClients queue to be a LinkedBlockingQueue to avoid
ConcurrentModificationExceptions if a client binds while a message is
being handled by the server.

Made SessionsSettings injectable. Also removed SessionMaintainer.

Co-authored-by: jrothfeder <[email protected]>

This will make it safe to call `Firebase.app` any time from anywhere in
this SDK.

#5427)

… the FirelogPublisher handle the full SessionEvent creation since
that's only relevant to Firelog.

#5431)

… the MainLooper which is the application main thread and so could
interfere with things there. Also fixes issue with missing import.

#5434)

…and use the same Messenger instance for every client. This makes sure
that all state manipulation happens on the same looper, so we can get
rid of the LinkedBlockingQueue or worry about volatile booleans.

#5436)

…tor.

Co-authored-by: Mila <[email protected]>
Co-authored-by: Rodrigo Lazo <[email protected]>
Co-authored-by: Greg Sakakihara <[email protected]>
Co-authored-by: David Motsonashvili <[email protected]>
Co-authored-by: David Motsonashvili <[email protected]>
Co-authored-by: Vinay Guthal <[email protected]>
Co-authored-by: cherylEnkidu <[email protected]>
Co-authored-by: Mark Duckworth <[email protected]>
Co-authored-by: Ehsan Nasiri <[email protected]>
Co-authored-by: Rosalyn Tan <[email protected]>
Co-authored-by: themiswang <[email protected]>

#5442)

…e of new log lines.

#5443)

…acks based on the sdk enablement in SessionsSettings.

#5452)

…irebaseSessionsDependencies` can be accessed directly.

---------

Co-authored-by: jrothfeder <[email protected]>

…ks (#5447)

Co-authored-by: Jamie Rothfeder <[email protected]>

#5461)

…e and managed by the Registrar in order to improve testability in the
SessionLifecycleService.

Adds fake implementations for the newly added interfaces and creates a
fake registrar installing our fake implementations.

#5468)

… lifecycle events. We can't have the handler thread reacting to
lifecycle events because:

* There's no guarantee that an activity is ever started on a process,
but we still want session ids pushed there (eg. service-only process)
* There can be more than one activity per process and the first time one
of those activities is stopped, the handlertthread will be stopped and
so we'll have no thread to process the callback messages.

Updated to use MainLooper

#5465)

…mplementations

#5472)

…t, and pulls the service binding into a different class to improve
testability of the client.

Keep track of pending foreground before settings fetched to catch the
first session in cold start with no settings cache.

#5481)

… processes and crash behaviour with the SDK.