Everyone's seen the "sticky" newtalk notification problem -- that ugly orange box hangs around for longer than it's meant to and can be hard to get rid of. I fixed one cause of this: the client cache. We've always had a system to invalidate the client cache when the newtalk status changes, but it was broken shortly before 1.4 was branched and nobody noticed. I'm pretty sure it's now fixed.
The changes were inapplicable to HEAD, since newtalk handling has changed completely. It will need to be tested separately.
-- Tim Starling
Tim:
as I wrote in [1]
"fully reestablished memcache-efficient user_newtalk table,but only for the newmessage marker."
Please consider to commit this bugzilla (patchlet #1) - I have told you this several times, but the collaboration were stoped from your and Brions side with saying that I shall concentrate on other issues. The patchlet #1 solves the problem you are referring to (at least in my reference implementations and implementatation of other people).
Tom
[1] http://bugzilla.wikipedia.org/show_bug.cgi?id=2014 published 26 April 2005
Tim Starling schrieb:
Everyone's seen the "sticky" newtalk notification problem -- that ugly orange box hangs around for longer than it's meant to and can be hard to get rid of. I fixed one cause of this: the client cache. We've always had a system to invalidate the client cache when the newtalk status changes, but it was broken shortly before 1.4 was branched and nobody noticed. I'm pretty sure it's now fixed.
The changes were inapplicable to HEAD, since newtalk handling has changed completely. It will need to be tested separately.
-- Tim Starling
Wikitech-l mailing list Wikitech-l@wikimedia.org http://mail.wikipedia.org/mailman/listinfo/wikitech-l
Thomas Gries wrote:
The patchlet #1 solves the problem you are referring to (at least in my reference implementations and implementatation of other people).
...
[1] http://bugzilla.wikipedia.org/show_bug.cgi?id=2014 published 26 April 2005
No it doesn't, it reintroduces the bug, which you fixed (albeit inefficiently) with a previous patch. That patch replaces the use of UserTalkUpdate with a call to User::saveNewtalk(). The latter function which was written by Evan for User.php 1.140, does not update the user_touched field, which is the source of this bug. UserTalkUpdate calls User::saveSettings(), which does update user_touched. Both versions suffer from concurrency and replication lag problems due to their use of $wgMemc->delete() instead of $wgMemc->set() to refresh the cache.
-- Tim Starling
wikitech-l@lists.wikimedia.org