Notify about major session changes events.

ash/session_state_delegate.h

Index: ash/session_state_delegate.h
diff --git a/ash/session_state_delegate.h b/ash/session_state_delegate.h
index f0e7f22f42005c677abf234299ce8a4955a7c278..9aea5619c553ab0fe3b795730cf0d5c251cd99ab 100644
--- a/ash/session_state_delegate.h
+++ b/ash/session_state_delegate.h
@@ -45,6 +45,27 @@ class ASH_EXPORT SessionStateDelegate {
     CYCLE_TO_PREVIOUS_USER,  // Cycle to the previous user.
   };
 
+  // Defines session state i.e. whether session is running or not and
+  // whether user session is blocked by things like lock screen and
+  // multi-profile login.
+  enum SessionState {
+    STATE_UNKNOWN = 0,

[Daniel Erat, 2014/04/10 16:30:32]

please don't add an "unknown" state; this means that every class using this enum
will need to handle it. more about this in the delegate

[Nikita (slow), 2014/04/10 17:22:51]

Done.

+
+    // When primary user login UI is shown i.e. after boot or sign out,
+    // no active user session exists yet.
+    STATE_LOGIN_PRIMARY,
+
+    // Inside user session, no blocking UI (login/lock) is shown.
+    STATE_SESSION,
+
+    // User session is active, currently is locked.
+    STATE_LOCK,

[stevenjb, 2014/04/10 17:00:06]

I'm not sure I understand this state. If this is synonymous with screen lock I
might suggest removing this state and checking screen lock directly where
necessary. Otherwise please elaborate in the comment.

[Nikita (slow), 2014/04/10 17:22:51]

Done.

+
+    // When secondary user login UI is shown i.e. other users are
+    // already logged in and is currently adding another user to the session.
+    STATE_LOGIN_SECONDARY,

[stevenjb, 2014/04/10 17:00:06]

While I hate to second guess a change in an area I'm only peripherally familiar
with, I'm also uncomfortable with adding yet another state, even in the "short"
term.

I assume there are future plans for GetSessionState(), since this CL doesn't
appear to actually use it outside of tests, but it looks to me like
LOGIN_SECONDARY is the only new "state" and we might be better off with adding
something like LOGIN_UI_STATE_{NONE|PRIMARY|SECONDARY}, rather than combining
states from different sources here? The sheer number of observers added to
SessionStateDelegateChromeos suggests to me that we are trying to do too much
with a single state; I think it will be OK to push some of that logic to the UI
code.

If we go this route then we should name these SESSION_STATE_ since, as mentioned
we already have so many login / session related states we want to reduce
confusion as much as possible.

[Nikita (slow), 2014/04/10 17:22:51]

I've dropped STATE_LOCK for now.

I would prefer to merge LOGIN_STATE into SessionStateDelegate.

[Nikita (slow), 2014/04/10 17:29:27]

Since adding this logic to ShellObserver::OnLoginStateChanged() doesn't seem
like a right place.

Next I would really like move OnLoginStateChanged() from there.

[stevenjb, 2014/04/10 17:47:55]

I agree. That's another one where we conflate "LoginState" and screen lock
state. It would be nice to eliminate ash::user::LoginStatus and use LoginState +
a separate screen lock state in the UI directly.

+  };
+
   virtual ~SessionStateDelegate() {};
 
   // Returns the browser context for the user given by |index|.
@@ -89,6 +110,9 @@ class ASH_EXPORT SessionStateDelegate {
   // session.
   virtual bool IsUserSessionBlocked() const = 0;
 
+  // Returns current session state.
+  virtual SessionState GetSessionState() const = 0;
+
   // Gets the displayed name for the user with the given |index|.
   // Note that |index| can at maximum be |NumberOfLoggedInUsers() - 1|.
   virtual const base::string16 GetUserDisplayName(