Re-remove CONNECTIVITY_CHANGE_DELAY and supporting functions.

Since optimistic addresses are useable upon kernel notification
there is no need for this extra connectivity delay.

---

This functionality was originally submitted in ag/572619.  Owing
to issues with bind()ing to optimistic addresses (see b/18609055)
this was reverted in ag/598673.

This reverts the revert.  :-)

Bug: 17769720
Change-Id: Ibee490b2af72050693b6bd748193f51e312ca527
This commit is contained in:
Erik Kline 2014-12-08 16:25:20 +09:00
commit b883659514

View file

@ -100,7 +100,7 @@ public class ConnectivityManager {
/**
* Identical to {@link #CONNECTIVITY_ACTION} broadcast, but sent without any
* applicable {@link Settings.Global#CONNECTIVITY_CHANGE_DELAY}.
* historic {@link Settings.Global#CONNECTIVITY_CHANGE_DELAY}.
*
* @hide
*/
@ -427,18 +427,6 @@ public class ConnectivityManager {
@Deprecated
public static final int DEFAULT_NETWORK_PREFERENCE = TYPE_WIFI;
/**
* Default value for {@link Settings.Global#CONNECTIVITY_CHANGE_DELAY} in
* milliseconds. This was introduced because IPv6 routes seem to take a
* moment to settle - trying network activity before the routes are adjusted
* can lead to packets using the wrong interface or having the wrong IP address.
* This delay is a bit crude, but in the future hopefully we will have kernel
* notifications letting us know when it's safe to use the new network.
*
* @hide
*/
public static final int CONNECTIVITY_CHANGE_DELAY_DEFAULT = 3000;
/**
* @hide
*/