From: Christian Heller Date: Sat, 12 Mar 2016 20:02:46 +0000 (+0100) Subject: TCE: Set up scenario-specific AI. X-Git-Tag: tce~40 X-Git-Url: https://plomlompom.com/repos/%7B%7B%20web_path%20%7D%7D/decks/static/ledger?a=commitdiff_plain;h=0a0ed866ffc56cfdce529dc183412721f7e87871;p=plomrogue TCE: Set up scenario-specific AI. --- diff --git a/libplomrogue.c b/libplomrogue.c index 9f872e1..68d3431 100644 --- a/libplomrogue.c +++ b/libplomrogue.c @@ -566,6 +566,90 @@ extern uint8_t dijkstra_map() return 0; } + +/* 7DRL/TCE addition: movement cost map setting. */ +static uint8_t * TCE_move_cost_map = NULL; +extern uint8_t TCE_set_movement_cost_map(char * mem_map) +{ + uint32_t map_size = maplength * maplength; + free(TCE_move_cost_map); + TCE_move_cost_map = malloc(map_size * sizeof(uint8_t)); + uint32_t pos = 0; + for (; pos < map_size; pos++) + { + TCE_move_cost_map[pos] = 0; + } + if (!TCE_move_cost_map) + { + return 1; + } + for (pos = 0; pos < map_size; pos++) + { + switch(mem_map[pos]) { + case '0': + TCE_move_cost_map[pos] = 1; + break; + case '1': + TCE_move_cost_map[pos] = 2; + break; + case '2': + TCE_move_cost_map[pos] = 4; + break; + case '3': + TCE_move_cost_map[pos] = 3; + break; + case '4': + TCE_move_cost_map[pos] = 6; + break; + } + } + return 0; +} + + +/* 7DRL/TCE addition: Like dijkstra_map(), but with movement costs applied. */ +extern uint8_t TCE_dijkstra_map_with_movement_cost() +{ + if (!score_map || !TCE_move_cost_map) + { + return 1; + } + uint16_t max_score = UINT16_MAX - 1; + uint32_t map_size = maplength * maplength; + uint32_t pos; + uint16_t i_scans, neighbors[6], min_neighbor; + uint8_t scores_still_changing = 1; + uint8_t i_dirs; + for (i_scans = 0; scores_still_changing; i_scans++) + { + scores_still_changing = 0; + for (pos = 0; pos < map_size; pos++) + { + uint16_t score = score_map[pos]; + uint8_t mov_cost = TCE_move_cost_map[pos]; + if (mov_cost > 0 && score > i_scans) + { + get_neighbor_scores(pos, max_score, neighbors); + min_neighbor = max_score; + for (i_dirs = 0; i_dirs < 6; i_dirs++) + { + if (min_neighbor > neighbors[i_dirs]) + { + min_neighbor = neighbors[i_dirs]; + } + } + if (score_map[pos] > min_neighbor + mov_cost) + { + score_map[pos] = min_neighbor + mov_cost; + scores_still_changing = 1; + } + } + } + } + return 0; +} + + extern uint8_t zero_score_map_where_char_on_memdepthmap(char c, char * memdepthmap) { diff --git a/plugins/server/TheCrawlingEater.py b/plugins/server/TheCrawlingEater.py index 658d57d..1bd03d2 100644 --- a/plugins/server/TheCrawlingEater.py +++ b/plugins/server/TheCrawlingEater.py @@ -179,8 +179,7 @@ world_db["test_hole"] = test_hole def test_air(t): - if (world_db["wetmap"][t["pos"]] - ord("0")) \ - + (world_db["MAP"][t["pos"]] - ord("0")) > 5: + if world_db["terrain_fullness"](t["pos"]) > 5: world_db["die"](t, "You SUFFOCATE") return False return True @@ -191,6 +190,7 @@ def die(t, message): t["T_LIFEPOINTS"] = 0 if t == world_db["Things"][0]: t["fovmap"] = bytearray(b' ' * (world_db["MAP_LENGTH"] ** 2)) + t["T_MEMMAP"][t["pos"]] = ord("@") log(message) world_db["die"] = die @@ -259,7 +259,7 @@ def turn_over(): update_map_memory(t) if 0 == tid: return - ai(t) + world_db["ai"](t) if t["T_LIFEPOINTS"]: t["T_PROGRESS"] += 1 taid = [a for a in world_db["ThingActions"] @@ -387,6 +387,222 @@ def write_wetmap(): return write_map(visible_wetmap, world_db["MAP_LENGTH"]) +def command_ai(): + if world_db["WORLD_ACTIVE"]: + world_db["ai"](world_db["Things"][0]) + world_db["turn_over"]() + + +def get_dir_to_target(t, target): + + from server.utils import rand, libpr, c_pointer_to_bytearray + from server.config.world_data import symbols_passable + + def zero_score_map_where_char_on_memdepthmap(c): + map = c_pointer_to_bytearray(t["T_MEMDEPTHMAP"]) + if libpr.zero_score_map_where_char_on_memdepthmap(c, map): + raise RuntimeError("No score map allocated for " + "zero_score_map_where_char_on_memdepthmap().") + + def set_map_score(pos, score): + test = libpr.set_map_score(pos, score) + if test: + raise RuntimeError("No score map allocated for set_map_score().") + + def set_movement_cost_map(): + memmap = c_pointer_to_bytearray(t["T_MEMMAP"]) + if libpr.TCE_set_movement_cost_map(memmap): + raise RuntimeError("No movement cost map allocated for " + "set_movement_cost_map().") + + def seeing_thing(): + def exists(gen): + try: + next(gen) + except StopIteration: + return False + return True + mapsize = world_db["MAP_LENGTH"] ** 2 + if target == "food" and t["T_MEMMAP"]: + return exists(pos for pos in range(mapsize) + if ord("2") < t["T_MEMMAP"][pos] < ord("5")) + elif target == "fluid_certain" and t["fovmap"]: + return exists(pos for pos in range(mapsize) + if t["fovmap"] == ord("v") + if world_db["MAP"][pos] == ord("0") + if world_db["wetmap"][pos] > ord("0")) + elif target == "fluid_potential" and t["T_MEMMAP"] and t["fovmap"]: + return exists(pos for pos in range(mapsize) + if t["T_MEMMAP"][pos] == ord("0") + if t["fovmap"] != ord("v")) + elif target == "space" and t["T_MEMMAP"] and t["fovmap"]: + return exists(pos for pos in range(mapsize) + if ord("0") <= t["T_MEMMAP"][pos] <= ord("2") + if (t["fovmap"] != ord("v") + or world_db["terrain_fullness"](pos) < 5)) + return False + + def init_score_map(): + test = libpr.init_score_map() + set_movement_cost_map() + mapsize = world_db["MAP_LENGTH"] ** 2 + if test: + raise RuntimeError("Malloc error in init_score_map().") + if target == "food" and t["T_MEMMAP"]: + [set_map_score(pos, 0) for pos in range(mapsize) + if ord("2") < t["T_MEMMAP"][pos] < ord("5")] + elif target == "fluid_certain" and t["fovmap"]: + [set_map_score(pos, 0) for pos in range(mapsize) + if t["fovmap"] == ord("v") + if world_db["MAP"][pos] == ord("0") + if world_db["wetmap"][pos] > ord("0")] + elif target == "fluid_potential" and t["T_MEMMAP"] and t["fovmap"]: + [set_map_score(pos, 0) for pos in range(mapsize) + if t["T_MEMMAP"][pos] == ord("0") + if t["fovmap"] != ord("v")] + elif target == "space" and t["T_MEMMAP"] and t["fovmap"]: + [set_map_score(pos, 0) for pos in range(mapsize) + if ord("0") <= t["T_MEMMAP"][pos] <= ord("2") + if (t["fovmap"] != ord("v") + or world_db["terrain_fullness"](pos) < 5)] + elif target == "search": + zero_score_map_where_char_on_memdepthmap(mem_depth_c[0]) + + def rand_target_dir(neighbors, cmp, dirs): + candidates = [] + n_candidates = 0 + for i in range(len(dirs)): + if cmp == neighbors[i]: + candidates.append(dirs[i]) + n_candidates += 1 + return candidates[rand.next() % n_candidates] if n_candidates else 0 + + def get_neighbor_scores(dirs, eye_pos): + scores = [] + if libpr.ready_neighbor_scores(eye_pos): + raise RuntimeError("No score map allocated for " + + "ready_neighbor_scores.()") + for i in range(len(dirs)): + scores.append(libpr.get_neighbor_score(i)) + return scores + + def get_dir_from_neighbors(): + import math + dir_to_target = False + dirs = "edcxsw" + eye_pos = t["pos"] + neighbors = get_neighbor_scores(dirs, eye_pos) + minmax_start = 65535 - 1 + minmax_neighbor = minmax_start + for i in range(len(dirs)): + if minmax_neighbor > neighbors[i]: + minmax_neighbor = neighbors[i] + if minmax_neighbor != minmax_start: + dir_to_target = rand_target_dir(neighbors, minmax_neighbor, dirs) + return dir_to_target, minmax_neighbor + + dir_to_target = False + mem_depth_c = b' ' + run_i = 9 + 1 if "search" == target else 1 + minmax_neighbor = 0 + while run_i and not dir_to_target and \ + ("search" == target or seeing_thing()): + run_i -= 1 + init_score_map() + mem_depth_c = b'9' if b' ' == mem_depth_c \ + else bytes([mem_depth_c[0] - 1]) + if libpr.TCE_dijkstra_map_with_movement_cost(): + raise RuntimeError("No score map allocated for dijkstra_map().") + dir_to_target, minmax_neighbor = get_dir_from_neighbors() + libpr.free_score_map() + if dir_to_target and str == type(dir_to_target): + action = "move" + from server.utils import mv_yx_in_dir_legal + move_result = mv_yx_in_dir_legal(dir_to_target, t["T_POSY"], + t["T_POSX"]) + if 1 != move_result[0]: + return False, 0 + pos = (move_result[1] * world_db["MAP_LENGTH"]) + move_result[2] + if world_db["MAP"][pos] > ord("2"): + action = "eat" + t["T_COMMAND"] = [taid for taid in world_db["ThingActions"] + if world_db["ThingActions"][taid]["TA_NAME"] + == action][0] + t["T_ARGUMENT"] = ord(dir_to_target) + return dir_to_target, minmax_neighbor +world_db["get_dir_to_target"] = get_dir_to_target + + +def terrain_fullness(pos): + return (world_db["MAP"][pos] - ord("0")) + \ + (world_db["wetmap"][pos] - ord("0")) +world_db["terrain_fullness"] = terrain_fullness + + +def ai(t): + + if t["T_LIFEPOINTS"] == 0: + return + + def standing_on_fluid(t): + if world_db["MAP"][t["pos"]] == ord("0") and \ + world_db["wetmap"][t["pos"]] > ord("0"): + return True + else: + return False + + def thing_action_id(name): + return [taid for taid in world_db["ThingActions"] + if world_db["ThingActions"][taid] + ["TA_NAME"] == name][0] + + t["T_COMMAND"] = thing_action_id("wait") + needs = { + "safe_pee": (world_db["terrain_fullness"](t["pos"]) * t["T_BLADDER"]) / 4, + "safe_drop": (world_db["terrain_fullness"](t["pos"]) * t["T_BOWEL"]) / 4, + "food": 33 - t["T_STOMACH"], + "fluid_certain": 33 - t["T_KIDNEY"], + "fluid_potential": 32 - t["T_KIDNEY"], + "search": 1, + } + from operator import itemgetter + needs = sorted(needs.items(), key=itemgetter(1,0)) + needs.reverse() + for need in needs: + if need[1] > 0: + if need[0] in {"fluid_certain", "fluid_potential"}: + if standing_on_fluid(t): + t["T_COMMAND"] = thing_action_id("drink") + return + elif t["T_BLADDER"] > 0 and \ + world_db["MAP"][t["pos"]] == ord("0"): + t["T_COMMAND"] = thing_action_id("pee") + return + elif need[0] in {"safe_pee", "safe_drop"}: + action_name = need[0][len("safe_"):] + if world_db["terrain_fullness"](t["pos"]) < 4: + t["T_COMMAND"] = thing_action_id(action_name) + return + else: + test = world_db["get_dir_to_target"](t, "space") + if test[0]: + if (not test[1] < 5) and \ + world_db["terrain_fullness"](t["pos"]) < 5: + t["T_COMMAND"] = thing_action_id(action_name) + return + if t["T_STOMACH"] < 32 and \ + world_db["get_dir_to_target"](t, "food")[0]: + return + continue + if world_db["get_dir_to_target"](t, need[0])[0]: + return + elif t["T_STOMACH"] < 32 and \ + need[0] in {"fluid_certain", "fluid_potential"} and \ + world_db["get_dir_to_target"](t, "food")[0]: + return +world_db["ai"] = ai + + from server.config.io import io_db io_db["worldstate_write_order"] += [["T_STOMACH", "player_int"]] io_db["worldstate_write_order"] += [["T_KIDNEY", "player_int"]]