275 lines
10 KiB
C++
275 lines
10 KiB
C++
#include "plannedbuilding.h"
|
|
#include "buildingplan.h"
|
|
|
|
#include "Debug.h"
|
|
|
|
#include "modules/Items.h"
|
|
#include "modules/Job.h"
|
|
#include "modules/Materials.h"
|
|
|
|
#include "df/building_design.h"
|
|
#include "df/item.h"
|
|
#include "df/job.h"
|
|
#include "df/world.h"
|
|
|
|
#include <unordered_map>
|
|
|
|
using std::map;
|
|
using std::string;
|
|
using std::unordered_map;
|
|
|
|
namespace DFHack {
|
|
DBG_EXTERN(buildingplan, cycle);
|
|
}
|
|
|
|
using namespace DFHack;
|
|
|
|
struct BadFlags {
|
|
uint32_t whole;
|
|
|
|
BadFlags() {
|
|
df::item_flags flags;
|
|
#define F(x) flags.bits.x = true;
|
|
F(dump); F(forbid); F(garbage_collect);
|
|
F(hostile); F(on_fire); F(rotten); F(trader);
|
|
F(in_building); F(construction); F(in_job);
|
|
F(owned); F(in_chest); F(removed); F(encased);
|
|
F(spider_web);
|
|
#undef F
|
|
whole = flags.whole;
|
|
}
|
|
};
|
|
|
|
bool itemPassesScreen(df::item * item) {
|
|
static const BadFlags bad_flags;
|
|
return !(item->flags.whole & bad_flags.whole)
|
|
&& !item->isAssignedToStockpile();
|
|
}
|
|
|
|
bool matchesFilters(df::item * item, df::job_item * job_item) {
|
|
// check the properties that are not checked by Job::isSuitableItem()
|
|
if (job_item->item_type > -1 && job_item->item_type != item->getType())
|
|
return false;
|
|
|
|
if (job_item->item_subtype > -1 &&
|
|
job_item->item_subtype != item->getSubtype())
|
|
return false;
|
|
|
|
if (job_item->flags2.bits.building_material && !item->isBuildMat())
|
|
return false;
|
|
|
|
if (job_item->metal_ore > -1 && !item->isMetalOre(job_item->metal_ore))
|
|
return false;
|
|
|
|
if (job_item->has_tool_use > df::tool_uses::NONE
|
|
&& !item->hasToolUse(job_item->has_tool_use))
|
|
return false;
|
|
|
|
return Job::isSuitableItem(
|
|
job_item, item->getType(), item->getSubtype())
|
|
&& Job::isSuitableMaterial(
|
|
job_item, item->getMaterial(), item->getMaterialIndex(),
|
|
item->getType());
|
|
}
|
|
|
|
bool isJobReady(color_ostream &out, const std::vector<df::job_item *> &jitems) {
|
|
int needed_items = 0;
|
|
for (auto job_item : jitems) { needed_items += job_item->quantity; }
|
|
if (needed_items) {
|
|
DEBUG(cycle,out).print("building needs %d more item(s)\n", needed_items);
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
static bool job_item_idx_lt(df::job_item_ref *a, df::job_item_ref *b) {
|
|
// we want the items in the opposite order of the filters
|
|
return a->job_item_idx > b->job_item_idx;
|
|
}
|
|
|
|
// this function does not remove the job_items since their quantity fields are
|
|
// now all at 0, so there is no risk of having extra items attached. we don't
|
|
// remove them to keep the "finalize with buildingplan active" path as similar
|
|
// as possible to the "finalize with buildingplan disabled" path.
|
|
void finalizeBuilding(color_ostream &out, df::building *bld) {
|
|
DEBUG(cycle,out).print("finalizing building %d\n", bld->id);
|
|
auto job = bld->jobs[0];
|
|
|
|
// sort the items so they get added to the structure in the correct order
|
|
std::sort(job->items.begin(), job->items.end(), job_item_idx_lt);
|
|
|
|
// derive the material properties of the building and job from the first
|
|
// applicable item. if any boulders are involved, it makes the whole
|
|
// structure "rough".
|
|
bool rough = false;
|
|
for (auto attached_item : job->items) {
|
|
df::item *item = attached_item->item;
|
|
rough = rough || item->getType() == df::item_type::BOULDER;
|
|
if (bld->mat_type == -1) {
|
|
bld->mat_type = item->getMaterial();
|
|
job->mat_type = bld->mat_type;
|
|
}
|
|
if (bld->mat_index == -1) {
|
|
bld->mat_index = item->getMaterialIndex();
|
|
job->mat_index = bld->mat_index;
|
|
}
|
|
}
|
|
|
|
if (bld->needsDesign()) {
|
|
auto act = (df::building_actual *)bld;
|
|
if (!act->design)
|
|
act->design = new df::building_design();
|
|
act->design->flags.bits.rough = rough;
|
|
}
|
|
|
|
// we're good to go!
|
|
job->flags.bits.suspend = false;
|
|
Job::checkBuildingsNow();
|
|
}
|
|
|
|
static df::building * popInvalidTasks(color_ostream &out, Bucket &task_queue,
|
|
unordered_map<int32_t, PlannedBuilding> &planned_buildings) {
|
|
while (!task_queue.empty()) {
|
|
auto & task = task_queue.front();
|
|
auto id = task.first;
|
|
if (planned_buildings.count(id) > 0) {
|
|
auto bld = planned_buildings.at(id).getBuildingIfValidOrRemoveIfNot(out);
|
|
if (bld && bld->jobs[0]->job_items[task.second]->quantity)
|
|
return bld;
|
|
}
|
|
DEBUG(cycle,out).print("discarding invalid task: bld=%d, job_item_idx=%d\n", id, task.second);
|
|
task_queue.pop_front();
|
|
}
|
|
return NULL;
|
|
}
|
|
|
|
static void doVector(color_ostream &out, df::job_item_vector_id vector_id,
|
|
map<string, Bucket> &buckets,
|
|
unordered_map<int32_t, PlannedBuilding> &planned_buildings) {
|
|
auto other_id = ENUM_ATTR(job_item_vector_id, other, vector_id);
|
|
auto item_vector = df::global::world->items.other[other_id];
|
|
DEBUG(cycle,out).print("matching %zu item(s) in vector %s against %zu filter bucket(s)\n",
|
|
item_vector.size(),
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
buckets.size());
|
|
for (auto item_it = item_vector.rbegin();
|
|
item_it != item_vector.rend();
|
|
++item_it) {
|
|
auto item = *item_it;
|
|
if (!itemPassesScreen(item))
|
|
continue;
|
|
for (auto bucket_it = buckets.begin(); bucket_it != buckets.end(); ) {
|
|
auto & task_queue = bucket_it->second;
|
|
auto bld = popInvalidTasks(out, task_queue, planned_buildings);
|
|
if (!bld) {
|
|
DEBUG(cycle,out).print("removing empty bucket: %s/%s; %zu bucket(s) left\n",
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
bucket_it->first.c_str(),
|
|
buckets.size() - 1);
|
|
bucket_it = buckets.erase(bucket_it);
|
|
continue;
|
|
}
|
|
auto & task = task_queue.front();
|
|
auto id = task.first;
|
|
auto job = bld->jobs[0];
|
|
auto filter_idx = task.second;
|
|
if (matchesFilters(item, job->job_items[filter_idx])
|
|
&& Job::attachJobItem(job, item,
|
|
df::job_item_ref::Hauled, filter_idx))
|
|
{
|
|
MaterialInfo material;
|
|
material.decode(item);
|
|
ItemTypeInfo item_type;
|
|
item_type.decode(item);
|
|
DEBUG(cycle,out).print("attached %s %s to filter %d for %s(%d): %s/%s\n",
|
|
material.toString().c_str(),
|
|
item_type.toString().c_str(),
|
|
filter_idx,
|
|
ENUM_KEY_STR(building_type, bld->getType()).c_str(),
|
|
id,
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
bucket_it->first.c_str());
|
|
// keep quantity aligned with the actual number of remaining
|
|
// items so if buildingplan is turned off, the building will
|
|
// be completed with the correct number of items.
|
|
--job->job_items[filter_idx]->quantity;
|
|
task_queue.pop_front();
|
|
if (isJobReady(out, job->job_items)) {
|
|
finalizeBuilding(out, bld);
|
|
planned_buildings.at(id).remove(out);
|
|
}
|
|
if (task_queue.empty()) {
|
|
DEBUG(cycle,out).print(
|
|
"removing empty item bucket: %s/%s; %zu left\n",
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
bucket_it->first.c_str(),
|
|
buckets.size() - 1);
|
|
buckets.erase(bucket_it);
|
|
}
|
|
// we found a home for this item; no need to look further
|
|
break;
|
|
}
|
|
++bucket_it;
|
|
}
|
|
if (buckets.empty())
|
|
break;
|
|
}
|
|
}
|
|
|
|
struct VectorsToScanLast {
|
|
std::vector<df::job_item_vector_id> vectors;
|
|
VectorsToScanLast() {
|
|
// order is important here. we want to match boulders before wood and
|
|
// everything before bars. blocks are not listed here since we'll have
|
|
// already scanned them when we did the first pass through the buckets.
|
|
vectors.push_back(df::job_item_vector_id::BOULDER);
|
|
vectors.push_back(df::job_item_vector_id::WOOD);
|
|
vectors.push_back(df::job_item_vector_id::BAR);
|
|
}
|
|
};
|
|
|
|
void buildingplan_cycle(color_ostream &out, Tasks &tasks,
|
|
unordered_map<int32_t, PlannedBuilding> &planned_buildings) {
|
|
static const VectorsToScanLast vectors_to_scan_last;
|
|
|
|
DEBUG(cycle,out).print(
|
|
"running buildingplan cycle for %zu registered buildings\n",
|
|
planned_buildings.size());
|
|
|
|
for (auto it = tasks.begin(); it != tasks.end(); ) {
|
|
auto vector_id = it->first;
|
|
// we could make this a set, but it's only three elements
|
|
if (std::find(vectors_to_scan_last.vectors.begin(),
|
|
vectors_to_scan_last.vectors.end(),
|
|
vector_id) != vectors_to_scan_last.vectors.end()) {
|
|
++it;
|
|
continue;
|
|
}
|
|
|
|
auto & buckets = it->second;
|
|
doVector(out, vector_id, buckets, planned_buildings);
|
|
if (buckets.empty()) {
|
|
DEBUG(cycle,out).print("removing empty vector: %s; %zu vector(s) left\n",
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
tasks.size() - 1);
|
|
it = tasks.erase(it);
|
|
}
|
|
else
|
|
++it;
|
|
}
|
|
for (auto vector_id : vectors_to_scan_last.vectors) {
|
|
if (tasks.count(vector_id) == 0)
|
|
continue;
|
|
auto & buckets = tasks[vector_id];
|
|
doVector(out, vector_id, buckets, planned_buildings);
|
|
if (buckets.empty()) {
|
|
DEBUG(cycle,out).print("removing empty vector: %s; %zu vector(s) left\n",
|
|
ENUM_KEY_STR(job_item_vector_id, vector_id).c_str(),
|
|
tasks.size() - 1);
|
|
tasks.erase(vector_id);
|
|
}
|
|
}
|
|
DEBUG(cycle,out).print("cycle done; %zu registered building(s) left\n",
|
|
planned_buildings.size());
|
|
}
|