view TODO @ 11719:109ee3bfeac5

[gaim-migrate @ 14010] SF Patch #1333770 from corfe83 "Many times in gaim we use the function g_slist_remove(list,node->data) to remove an element from a GSList. If we already have the pointer to the node we want to delete, it is faster to send it the pointer to the node to delete rather than the data of the node (we can do this by calling g_slist_delete_link(list,node)). This change was made while looking at glib's documentation and the code in glib's gslist.c. This is because as the remove/delete function traverses each node in the list, it doesn't need to spend an extra memory access to retrieve the data for each element in the node it is traversing and then compare, it can simply compare the pointer. In my tests outside of gaim, this makes a big difference if the node you are deleting is at a high index in the list. However, even if you're deleting the first node, it about breaks even. So, I've found each case in gaim where we are calling g_slist_remove, and we already have the pointer to the appropriate node to delete (this is often the case when we're doing a for or while loop on a GSList). I've then replaced it with the appropriate call to g_slist_delete_link. I, however, didn't do this in situations where we are explicitly removing the first element in the list, because in those situations it is an unnecessary change. There should be no difference in behavior, but just in case I've tried running it with valgrind, which reports the same number of memory leaks after my patch as before my patch. Of course, I can't guarantee that my normal behavior on gaim is hitting all the functions I've changed, but in general testing it Works For Me (tm)." As with the last patch, this one may not have a practical performance impact (or maybe it does, I have no idea), but it's not worse for any case. Given two ways of doing things where one is always at least as fast and may be faster under some cases, I like to prefer that faster way. This doesn't make the code any uglier, so I'm applying. committer: Tailor Script <[e-mail]>

* conversation api {{{
	* need to handle: {{{
		* unnamed chats
		* chats with the same name as a buddy
			* this should be working now, but it needs to be kept in mind when (re-)designing
		* jabber Messages (one off messages vrs a conversation with a backlog)
		* multiple conversations with a single buddy (jabber, with thread tag)
		* bugs
	}}}
	* notes: {{{
		* (12:58:39) Robot101: user list always|auto|never
		* topic only where supported by the protocol
		* how do we log this? {{{
			* rlaager's (NEW) proposal: {{{
				* each conversation has 1 or more logs associated with it.
				* when someone joins, a new log is opened
					* file based loggers should try to use hard-links in a smart way
					  to avoid duplicating data on disk as much as possible...
					* this needs to be designed in such a way that a database logger
					  can handle it w/o duplication...
					* TO SOLVE THESE CONCERNS, a link and/or copy function callback
					  will likely need to be added
			}}}
		}}}
	}}}
}}}
* buddy list {{{
	* we have not committed any gobjectification, so this is not a blocker for 2.0.0 until we do.
		* if we do, it is only a blocker to the extent that gaim needs to run.  That is, we do not
		  *necessarily* have to gobjectifiy everything as long as we preserve functionality.
		* That being said, my understanding of the signals mess says that it is probly more work to
		  try to do this in stages than to start it or not start it.
	* This would require some sort of signal aggregation, as gobject signals are unique to the instance.
		* one way to do this would be to emit the signals from a non-descript buddy object.  this would
		  most closely mimic the current functionality
		* a second way would be to actually try to aggregate them in some way. I really don't know how
		  possible this is, but it would allow us to do things like set a pounce on everyone in a group
		  or define sounds on a per-group basis.
		* we could look at replacing the UI ops with signals/call backs
	* bugs
		* buddy shows as group on add until gaim restart
		* buddy shows as online when offline
		* buddy shows as offline when online
		* wrong buddy given priority
		* so on and so forth
}}}
* build targets {{{
  * this is not a blocker
  * we need build targets for libgaim, we need to test them, and make sure they work.
  * we ought to use our own build targets to build the executable itself.
  * due to the limitations of cvs, this cannot accompany moving files to other directories at this time.
}}}
* status {{{
	* Sean's segment of the UI needs to be finished.  We need to have support for the (new) default case
	  of one global status at all times.  we currently do not, unless that happens to be "online/present"
	* Tim's modifications to Sean's ui should be included to allow exceptions. refer to gaim-devel
	  archives for this.
}}}
* account editor {{{
	* this is not a blocker for 2.0.0
	* account editor is not intuitive, users do not find it.
	* Luke: my temptation is to get rid of this entirely, in favor of deryni's account menu. {{{
		* at this point, tools->accounts is only used for add/delete account, modify account, and
		  enable account.
		* most users do not have the 15+ accounts that some gaim developers do. a menu scales well for
		  anything from 2 to n, n "small" accounts.
		* this would allow ready access to the buddy icon stuff, and account actions could go here
		* status is handled as per the new api, status stuff need not go here beyond enabled.
		* the account modify dialog is already too big.  this would let us split it, for instance
		  buddy icon need not be in it this way.  Similarly, alias need not be in it.
			* splitting the account modify dialog to tabs seems to work nicely.  Still, I think
			  that the menu would be nicer.
	}}}
}}}
* Privacy {{{
	* this is not a blocker for 2.0.0
	* Privacy sucks. it doesn't handle many of the protocols in a way that users understand. notably
	  msn, but also yahoo, jabber, and icq.
	* We want to sync to the server as much as possible, entirely local-only privacy is not an option
	  except on protocols like irc that have no server-side.
	* We want to be as consistent across protocols as possible.  this will require renaming things on
	  some protocols.
	* Bleeter suggests that we create a privacy.xml, remove privacy from blist.xml.  He suggests
	  that each protocol define certain capabilities & defaults, with accounts.xml holding exceptions
	  to the defaults.
}}}
* Perl {{{
	* Block for 2.0.0 or remove perl:
		* Summer of Code seems to have largely solved this.
		* Extended testing and resolving the inevitable bugs remains.
		* Test each call to make sure it actually works
		* Make it work with G_MODULE_BIND_LOCAL
}}}
* Prefs {{{
	* this blocks for 2.0.0
	* Prefs cannot stay as-is.  the dialog is far too wide and not at all usable.
	* The biggest problem is that each new plugin creates horizontal space. {{{
		* I do not see it as a solution to remove the posability of plugin preferences.
		* In the past, we had a separate plugin management dialog.  People never found it,
		  and were often surprised to learn that gaim had plugins at all.  I am unsure that
		  people find the current plugin page of preferences any more frequently, but I
		  *suspect* that it is the case.  This leads to a conundrum, how do plugins
		  display preferences?
		}}}
	* Currently the window is, at my font size, 1129x505 (or should that be 505x1129?).  It *should* fit
	  in 800x600 at worst, I'm unsure that 480x640 is a reasonable goal.  still, this leaves us with
	  something either considerably wider or considerably taller than we are currently using (on any
	  given pane, the tabs force the width, not the contents).  Further, taking "Message Text" as an
	  example, it has 3 preferences and a text area, each in its own category (the text area sharing a
	  category with 1 preference). obvious waste of space here.  All 3 could clearly be uncategorised
	  without loss of meaning, categories only make sense for groups of preferences.  It may even be
	  possible to combine this with "Conversations" entirely.
}}}