Fixed a rare case in which a town-defender could become an indirect part of a troop-delivery-task by doing a sub-task like capturing a shipyard that takes less than 1 turn but leaves him out in the open.
Fixed that armyInvolvement was only filled in when the action involved a target to interact with rather than just a tile.
Since armyInvolvement was used for scoring actions such as retreating towards the closest town, this caused the AI to never retreat to their towns when they were supposed to.
These can happen when an enemy spawns ontop of an AI-hero. If the action would win, it wouldn't be executed anyways. So now AI does the next best thing instead, which likely what it wanted to do anyways.
AI will no longer skip turns with heroes that are waiting for a delivery that takes more than one turn. Instead they will do something until their delivery is close enough to get it at the same turn.
Fixed an issue where a bunch of heroes all tried to do the same tasks:
Tasks that involve no fighting will now always be performed by the closest eligible hero while all other heroes look for something else to do.
Just killing stuff even if there is no apparent reason now also is considered for the mere purpose of gaining XP.
This also helps the non-cheating AI to keep attacking enemies when they can't see anything worth exploring behind them.
Final goal (of multiple PR's) is to remove all remaining pointers from
serializeable game state, and replace them with either identifiers or
with shared/unique pointers.
CGTownInstance::town and CGHeroInstance::type members have been removed.
Now this data is computed dynamically using subID member.
VLC entity of a town can now be accessed via following methods:
- getFactionID() returns ID of a faction
- getFaction() returns pointer to a faction
- getTown() returns pointer to a town
VLC entity of a hero can now be accessed via following methods:
- getHeroTypeID() returns ID of a hero
- getHeroClassID() returns ID of a hero class
- getHeroType() returns pointer to a hero
- getHeroClass() returns pointer to a hero class
AI is more careful when gathering stuff near enemies.
The wasted movement-points are no longer considered when calculating which own city to fall back to when there's nothing better to do.
Fix an issue that caused AI to take their hero's attributes into consideration twice when calculating how much army they think they'll lose.
Fixed an issue where offensive defending didn't take into consideration whether the hero would actually be strong enough to beat the enemy hero it was trying to dispatch.
A recent fix made it so that towns that weren't supposed to be defended are now no longer defended.
This caused scouts with minimal army to also go after them in addition to the main-hero.
Problem is when two heroes go for the same town it's a massive waste of movement-points. So for the time being only main-heroes will go for faraway captures.
Better solution would be to memorize who was sent to attack what on the same turn and filter out tasks going for the same target.
When attacking non-neutral towns that are far away, the AI now considers whether their attack would arrive in the same week. If it wouldn't, it means there's a high risk that newly bought troops might flip around who is stronger. So they now refrain from sending a hero towards an enemy town that is too far away.
The AI should no longer chase enemy heroes that are not reachable in the same turn, when there's other options as this behavior was quite exploitable.
The AI should now take their overall strength into account when deciding whether to attack or not.
Previously it would attack as long as their assumed army-loss was at most 25%.
Now that is 50% times the ratio of their power compared to the total power of everyone.
Manarecoveryreward now uses float instead of unsigned int in order to avoid extremely high instead of negative scores when the hero has more mana than his mana-limit for example due to mana-vortex.
Moved upgrading armies to a lower priority tier as otherwise the AI would go back to their cities all the time even though there were plenty of other things to do.
Improved exploration logic by putting different kinds of exploration to different priority-tiers.
Looking at the other side of a portal has high priority, visiting an observatory has medium priority and scouting by visiting nearby tiles has low priority.
