This happens because edges cease processing when the delta between their zones is small enough. This is normally a fine optimization, but when it results in a room at 4kPa with a window open to space, it breaks imurshuns.
Two main changes to solve this problem without too much cpu cost:
1) Stop edges from sleeping if one side is a hard vacuum. This ensures that a zone doesn't freeze at a low-but-non-zero pressure when touching hard vacuum.
2) Prevent #1 from causing the edge to stay alive for ages while pressures asymptotically approach zero as they are repeatedly equalized but only half is dumped to space. (Would happen if ZoneA---ZoneB---Space arrangement exists) by detecting when the total amount of air left is small enough that it would normally sleep anyway, and just setting it to zero.
The end outcome is that behavior is mostly the same as before, except when zones have an open path to unsimulated space, they will reach equilibrium at zero instead of semi-random lowish values.
New turf proc: assume_gas(). Maps to air.adjust_gas_temp().
Lots of optimizations to processing, fire, lighting, HasEntered() and
more.
Zones now process all fire data and existance in one go, fire objects
only handle spreading.
Most code has been ported straight so some of it mightn't be ideally
structured for the new gas_mixtures.
Signed-off-by: Mloc-Argent <colmohici@gmail.com>
