TODO - notes from ai.cc comments

I am stripping these from the code, will put them somewhere else. This is a temp spot. ` /* TO DO // i just noticed there is Object::NewTree and Object::NewPine... I thought only baby pines exist? check and see if there is an unused baby non-pine tree graphic!

something is up with idle_serfs check, the underlying function seems correct, it only checks for StateIdleInStock, but knights might have issues, and geologists are having issues. Maybe two separate problems? If so, go back and un-do the extra step that was added to knight calculations to avoid what I thought was a problem with idle serf detection
a significant amount of time is spent on expand_borders call, which is spent doing many score_area calls around all occupied_military_buildings. There are probably excessive redundant calls because every dir is checked from every military building (straight line to where it hits border) INSTEAD - try walking the border and trying every so many positions? what about circular borders? could get stuck NO, better yet just give up on finding borders in a given direction after 10 or so tiles, because huts at edges of territory borders should be ~8 or less tiles from border
seeing some new roads being connected inefficiently... like the pathfinding went the wrong (non-optimal) way and didn't find the clearly better way. Saw this twice in one game when creating geologist roads on mountains, they ended up being connected to a non-optimal flag instead of the nearest flag. Maybe change these to use nearest only?
toolmaker is making too many tools, hogging all the iron, even though many tools exist. I think this is related to warehouse/stock parallel economy see TOOLMAKER.save file for example
I think knight shuffling issue remains, likely need to add a check for if all knights have reached their destination before changing value again DONE - see if this fixes it
silly enhancement idea, make transporter serfs become lazier the longer it has been since they transported something. Make them eventually tap their feet, sit down, sleep. this could be purely visual or could cause them to take longer to wake up and get back to work when materials come
update_building_counts is being called far to often. Because it requires a mutex lock it slows down the many functions that use it. Maybe better to simply call it sparingly? or, find a way to make it not require mutex lock
modify do_manage_knight_occupation_levels so that on game load it doesn't change knight occupation! it assumes a new game was started and so sets to zero check to see if this actually still a problem
when building roads, consider a path THROUGH another path where a flag could be built and continuing, instead of only considering ending the road at a new flag
have AI overlay status text indicate when planks_crit reached, as it makes you wonder why the AI is doing nothing. better yet, so sliders or text indicator showing general AI resources status for critical resources in terms of min/med/max
bug confirmation - when the stuck serf detector boots a serf that is on its way to occupy a building, the serf returns to the castle but the building will never be occupied! saw this early in a game where a fisherman was on way to a new fisherman hut, got stuck, was booted, returned to castle, but never was sent back to hut!
seems like knights are shuffling in and out of the castle too often. maybe the min/max occupation/staffing settings are flipping too often? this might be fixed now - nov01 2020
minor issue - when mines are placed early, sometimes their flags happen to fall on existing roads and cause them to be unexpectedly connected. Maybe check this before building the mine and... disfavor it somehow without totally eliminating that spot? Maybe only if this is the 2nd or 3rd mine of same type?
make the direction that pathfinding starts random? I saw an issue where AI would build a building in a tight area, fail to connect it, burn it, try another spot, and keep trying until it exhausted every spot, BUT it could have build a road. I think it was always trying to go one direction and hitting obstacle (large lake in this case) DONE - updated build_best_road to do this - test it out
oct21 2020 - changing building affinity road connection logic. Currently only COMPLETED buildings will be used. I think it is better to use incomplete buildings if available, or maybe "started but not yet complete" buildings DIDNT ACTUALLY MAKE THIS CHANGE YET, it looks like single affinity uses different and likely older code than dual affinity make them both use the same logic and update both at once
attack scoring should probably be different than expansion scoring. When expand_towards list has a lot of needs, enemy target scores are very high, but when expand_towards list is small enemy target scores are low. Maybe ignore things that are easy to find elsewhere (fields, hills, trees, stone) and only focus on buildings, gold, iron? what about desperate situations where stones, hills, nowhere else to be found? some kind of "cannot expand borders by building huts" or "cannot obtain goals by building huts" to trigger desperate attacks??
when doing mid/high resource start where planks are initially high and no sawmill is needed yet, when a sawmill is eventually needed its placement the placement of lumberjacks near it is inefficient. This is actually a general problem with higher resource starts. It might be better to build all buildings in order regardless of current resources OR work on the rebuild_all_roads idea? Or maybe rebuild all roads near castle??
since adding multiple economies, does it still make sense to stop sending geologists once XX conditions met?
toolmaker not working right since parallel economies implemented possibly fixed now? check seems okay as of oct22 2020 wait, no it is still not right nov03 2020
additional no transporter bug! Saw no transporter on road, never was detected. After some digging and testing with savegames, determined that the flags at either end of the road say a transporter is there (so my check doesn't detect it), but there is no serf object with the map coordinates and job that match any point on that road. By changing some random IdleInStock serf to being a Transporter with coordinates (pos) of a point on the road - in the savegame where it was broken, it fixes the problem. To detect this issue, could run a check for any road that does not actually have a serf transporter where it thinks it does DONE
building too many geologists, saw 13 on a low resource start! this is mostly fixed by hard capping total geologists at 4, sometimes a few more are created but not more than ~8
expand_borders is finding occupied_military_buildings - redundant? have it use update_buildings one?
IT IS CONFUSING THAT stocks_pos / stock_buildings USES FLAG-POS INSTEAD OF BUILDING-POS, CHANGE IT??
check to see if this bug/issue with no-transporter detection affects warehouse/stocks in addition to castle // it seems the castle shows as having a path Up-Left into it... // I wonder if any other buildings do also? warehouses? if (flag->get_position() == castle_flag_pos) continue;
possible huge performance improvement... change all the Game::update_XXX functions to use a loop that can be safely invalidated (somehow?) so the entire call doesn't have to be mutex locked
idea - like pos_add_spirally, create a pos_add_cone with a direction. To find positions in a "pie slice" radiating outward from center use this to place faraway things that would require a massive spiral
BUG - by allowing roads to be built to flags that have no paths if the flag is target_pos, it allows buildings to never be connected to road system, but only to their affinity buildings. Change this to only allow connecting to disconnected if doing rebuild_all_roads
BUG - freeserf bug? saw an occupied sawmill not receive any wood to saw despite ample logs being logged nearby, they are all placed in castle and none were coming out of castle. halts economy add a generic timer like "if building hasn't produced anything X time destroy it" ???? saw this AGAIN right afteR?? next game wtf?
rebuild all roads still not working right. The roads appear to be rebuilt fine, but even after lost serf chaos eventually calls down, materials are still not being transported on certain roads/flags. Not sure why, probably related to missing transporter bug?
maybe add a "compare two roads" function? or is that redundant
the nearby_flags search doesn't seem to reliably be ending early (within 6 tiles if any flag found), at least when watching rebuild_all_roads Look into it
for finding faraway things... maybe use get_straightline_dist to compare rather than giant spirals?
work on rebuild_all_roads... might be an acceptable solution to inefficient AI roads
avoid building Large civilian buildings near borders with enemies
clean up ai.h or create another for functions needed by other code, keep all the private AI stuff from having to be included?
try to connect warehouse flags to all nearby flags
really do need a CONNECTED building count after all, for mines's free-miner/tools check. Maybe just do dynamically instead of keeping a record??
prioritize AI borders expansion BEFORE building Production economy. De-prioritize expansion once running low on idle knights
send_geologist bug... I saw AI sending geologists to a dead-end flag built next to a big desert with no mountains nearby. is it possible that desert tiles are accidently included in part of the geologist find terrain check????
idea - burn any building that hasn't acquired an owner/occupant serf after so many ticks?
PUT AN UNFINISHED BUILDINGS / MAX CHECK at the beginning of each do_build function!!! or in the main AI loop but that is ugly
add if ptr != nullptr checks to all "foreach Serf/Building/Flag/whatever" where a copy is fetched inside mutex locking and then iterated over
idea - add a "started building" check like building_count and completed_building_count to avoid starting multiple mines if one is already connected and begun contruction? this keeps coming up as a want
idea - avoid/discourage building road UP-Right, THEN right from castle flag. This prevents roads from being built UP-LEFT/UP-RIGHT from castle
rebuild_all_roads not working right at all, missing roads, wrong connections, etc. look into it if still needed
idea - consider making sawmill (or stonecutter??) the secondary affinity for ALL buildings?
implement Road cache, have a job run every so often, or maybe continuously in a separate thread, that traces all roads in realm, caches them
use this to detect roads that are not connected to the castle
use the cache to feed the "build better roads" process, and if a cached road is no longer the same, just skip it and continue, or invalidate the cache? */

tlongstretch / freeserf-with-AI-plus

TODO - notes from ai.cc comments #41