New features for players:
* Loading for multiplayer. Any save could be used for multiplayer.
* Restart for multiplayer. All clients will restart together.
* Loading from single save.
* Hotseat mixed with network game. Multiple players per client.
* Now connection to server could be cancelled.
* Return to menu on disconnections instead of crashes.
* Restoring of last selected map, save or campaign on next run.
TLDR on important changes in engine code:
* UI: work with server separated from UI
* UI: all explitic blitting replaced with IntObject's
* UI: all new code use smart pointers instead of DISPOSE
* Gameplay always start through lobby controlled by server.
* Threads receiving netpacks now shared for lobby and gameplay.
* Campaigns: heroes for crossover now serialized as JsonNode.
* Indirect spell effects loading
* Json serializer improvements
* spell->canBeCastAt do not allow useless cast for any spell
* Added proxy caster class for spell-created obstacles
* Handle damage from spell-created obstacles inside mechanics
* Experimental GameState integration/regression tests
* Ignore mod settings and load only "vcmi" mod when running tests
* fixed https://bugs.vcmi.eu/view.php?id=2765 (with tests)
* Huge improvements of BattleAI regarding spell casts
* AI can cast almost any combat spell except TELEPORT, SACRIFICE and obstacle placement spells.
* Possible fix for https://bugs.vcmi.eu/view.php?id=1811
* CStack factored out to several classes
* [Battle] Allowed RETURN_AFTER_STRIKE effect on server side to be optional
* [Battle] Allowed BattleAction have multiple destinations
* [Spells] Converted limit|immunity to target condition
* [Spells] Use partial configuration reload for backward compatibility handling
* [Tests] Started tests for CUnitState
* Partial fixes of fire shield effect
* [Battle] Do HP calculations in 64 bits
* [BattleAI] Use threading for spell cast evaluation
* [BattleAI] Made AI be able to evaluate modified turn order (on hypothetical battle state)
* Implemented https://bugs.vcmi.eu/view.php?id=2811
* plug rare freeze when hypnotized unit shots vertically
* Correctly apply ONLY_MELEE_FIGHT / ONLY_DISTANCE_FIGHT for unit damage, attack & defense
* [BattleAI] Try to not waste a cast if battle is actually won already
* Extended JsonSerializeFormat API
* fixed https://bugs.vcmi.eu/view.php?id=2847
* Any unit effect can be now chained (not only damage like Chain Lightning)
** only damage effect for now actually uses "chainFactor"
* Possible quick fix for https://bugs.vcmi.eu/view.php?id=2860
* new configuration option 'general.saveRandomMaps'
* maps being saved to 'userCachePath/RandomMaps'
* no deletion of old random maps
* map filename generated based on template name and random seed
Now server-side code should always use CRandomGenerator::getDefault which is serialized in GH.
CGameState::getRandomGenerator should be only used from GS code and CPackForClient-based applyGs.
Long time ago it's was used without prefix to make future switch from boost to std version easier.
I discusses this with Ivan and decide to drop these using from Global.h now.
This change wouldn't break anything because there was already code with prefix for each of three cases.
Now pathfinder take into account different bonuses for different tuns. So if you only have FLYING_MOVEMENT bonus from Fly spell for one turn then pathfinder will only let you use air layer within one turn only.
That work for cost calculations too. Let's say you have two bonuses:
- FLYING_MOVEMENT with 20% penalty for next 2 turns
- FLYING_MOVEMENT with 40% penalty for 5 turns
Now pathfinder using correct penalty for each turn so movements in air layer going to be more expensive on 3-5 turns.
Suggested by @DjWarmonger as better alternative from performance standpoint while struct still more organized than bunch of variables.
Other reason of change it's that in future we may need non-boolean options, e.g for patrol movement and some new pathfinder usages.
There no need in dedicated function for embark / disembark and other conditions belong there perfectly.
This function is going to check what movements are plain impossible so we don't have to bother check their cost.
Each kind of teleporter have own function that determine if it's should or shouldn't be used.
For now Monolith with bidirectional channels and Subterranean Gate are united.
