Memory leaks on dedicated servers with long uptimes.

[Frank Sapone]

This bug exists in the original source but since servers weren't auto-rotating maps and there were very little users connecting to trigger changelevels you would rarely see it.

As soon as an endmatch is triggered then the map rotates to the next one memory keeps climbing. The issue became obvious after I implemented sv_auto_rotate_map cvar. I verified it in 1.2 by opening a ded server and changing the map manually via the console. Using changelevel (which does not unload/reload DLLs) exhibits the same behaviour.

[Frank Sapone]

It happens on the client side as well.

[Frank Sapone]

Okay this has now mostly been fixed within reason.

Any HUGE memory leaks have now been resolved thanks to me setting up valgrind in a linux VM. There's a few very very minor memory leaks between invoking "map <mapname>" but they are very small. You'd have to load thousands of maps for it to really make a dent in your memory usage.

Client-side memory leaks are also resolved now too. I have sucessfully played the game upto e1m6a so far with no issues and my servers now have sv_auto_rotate_map 1 set to on and memory usage is in sane limits.