[ra@hamburger ~]$ ll /usr/local/games/quake3/ioquake3.x86_64
-rwxr-xr-x. 1 ra users 2607969 Sep 11 19:27 /usr/local/games/quake3/ioquake3.x86_64

Not exactly current but Sep 11 is close enough.   Also, Im on Fedora 19.

I started quake, clicked team arena in the menu and the game switched over.  Then I just clicked single play and enter arena and I was playing.
I can verify that my binds did not carry over :)

check out http://www.ra.is/missionpack/

That JBravo in those screenshots is me, not a bot.   Are you using some other method to trigger the behaviour you describe?

Sorry, I spoke to soon.  Once I choose a CTF type game I saw what you see.
It seems that the join menu in TA is broken.  If you pull down the console and join red or blue all works fine.

Using the Q3 start server menu the player also does not join a team in CTF.

Setting the team from the UI is time based [1], [2]. The command is dropped at the top of ClientCommand in g_cmds.c [3].

Apparently ioq3 is taking too long to get player to connected state in game VM.

[1] https://github.com/ioquake/ioq3/blob/master/code/q3_ui/ui_startserver.c#L818
[2] https://github.com/ioquake/ioq3/blob/master/code/ui/ui_main.c#L3068
[3] https://github.com/ioquake/ioq3/blob/master/code/game/g_cmds.c#L1648

Created attachment 3387 [details]
allow early team command

As pointed out to me by Ensiform, the issue is caused by a new check added to ClientCommand "|| ent->client->pers.connected != CON_CONNECTED" in commit 56f16e. Using the original id qagame.qvm works.

Adding a special case for allowing early team command for local client fixes joining team. Disabled running early ClientBegin as it may be able to cause problems? Anyone have comments on the patch before it's committed?

In that the original change was a security change, it's probably worth thinking carefully about the ramifications of removing the check. I'm not saying I fully understand what's going on here, just that exceptions to security/safety code need to be thoroughly thought through.

@Zach, When I searched the code last night to find the problem, I came to the same check in line ui_main.c#L3068 you posted.

Why is that security check needed? What was the idea behind adding it? Because a security check that breaks the game, is too much security :)

This check fixed a few bugs iirc.
Though I can't remember the whole list of problems:

1. I think disconnected clients sometimes have broken the configstrings ('ghost'  
   clients with malformated strings).
2. Under some circumstances game crashes (after map_restart?) if kicked bots 
   (because of wrong function 'BotReadSessionData') or related to 1?.
3. Maybe related bug: "Under certain circumstances, already "disconnected" 
   clients can disconnect"?
   http://svn.icculus.org/tremulous?view=rev&revision=2009
4. gameplay bugs like the flag disappearing? (reported by dev/humancontroller)
5. ...

To be honest I can't remember what exactly was the biggest problem without the check, but I'm definitly sure that game.qvm can crash without the check (the already disconnected client is still there as 'ghost').

So, yes it is a security fix. The added check in g_cmd.c fixed the bug(s), though technically I don't understand why!
--------
It seems that the current problem of not joining a team must be OS specific. I neither have problems on WinXP nor on Win7!
Why is this a problem for Non-Windows OS?

Thanks

Thanks for the info on the nature of the problems that were fixed with the added checks. :) What I don't understand yet is, how can something for disconnecting clients affect connecting clients?

I only have linux machines to try it on. Anyone with a mac?

Created attachment 3575 [details]
allow early team command v2

Well, no-one seems to have thought about this patch for a while, so I decided to think about it.

It seems safe to me.  The only behavior changed is when a local client requests a team change when it's technically not connected.  The UDP spoofing protection in Netchan_Process() in net_chan.c should prevent any spoofed messages from triggering this code path, and if that's not reliable, all bets are already off.

The only change I would make is to re-enable the initial connected check and add an exception for a local client, to avoid always calling the trap_Argv() syscall.

Changes attached.

g_teamAutoJoin "1"

g_teamAutoJoin isn't a great solution, because of two things.

First, I really don't want to override user cvars.

Second, it can lead to joining the wrong team.  Try starting a CTF game in Quake 3 (through single player...skirmish) with g_teamAutoJoin 1 and using default settings, except change your team to red.  You'll magically join blue when the game starts.

As well, I tried making the g_teamAutoJoin check in g_session.c G_InitSessionData() line 110 to (g_teamAutoJoin.integer || client->pers.localClient) and that didn't work, I'm guessing localClient isn't set at that point.

Fixed in https://github.com/ioquake/ioq3/commit/4506ebd5d7690b8677aaa598c173dcc71f013426.