aboutsummaryrefslogtreecommitdiff
path: root/src/emerge.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/emerge.cpp')
-rw-r--r--src/emerge.cpp778
1 files changed, 472 insertions, 306 deletions
diff --git a/src/emerge.cpp b/src/emerge.cpp
index d6bda731a..93e8f2b30 100644
--- a/src/emerge.cpp
+++ b/src/emerge.cpp
@@ -20,77 +20,102 @@ with this program; if not, write to the Free Software Foundation, Inc.,
#include "emerge.h"
-#include "server.h"
+
#include <iostream>
#include <queue>
-#include "jthread/jevent.h"
-#include "map.h"
-#include "environment.h"
+
#include "util/container.h"
#include "util/thread.h"
-#include "constants.h"
-#include "voxel.h"
+#include "threading/event.h"
+
#include "config.h"
-#include "mapblock.h"
-#include "serverobject.h"
-#include "settings.h"
-#include "scripting_game.h"
-#include "profiler.h"
+#include "constants.h"
+#include "environment.h"
#include "log.h"
-#include "nodedef.h"
-#include "mg_biome.h"
-#include "mg_ore.h"
-#include "mg_decoration.h"
-#include "mg_schematic.h"
+#include "map.h"
+#include "mapblock.h"
+#include "mapgen_flat.h"
+#include "mapgen_fractal.h"
#include "mapgen_v5.h"
#include "mapgen_v6.h"
#include "mapgen_v7.h"
+#include "mapgen_valleys.h"
#include "mapgen_singlenode.h"
+#include "mg_biome.h"
+#include "mg_ore.h"
+#include "mg_decoration.h"
+#include "mg_schematic.h"
+#include "nodedef.h"
+#include "profiler.h"
+#include "scripting_game.h"
+#include "server.h"
+#include "serverobject.h"
+#include "settings.h"
+#include "voxel.h"
+
struct MapgenDesc {
const char *name;
MapgenFactory *factory;
+ bool is_user_visible;
};
-MapgenDesc reg_mapgens[] = {
- {"v5", new MapgenFactoryV5},
- {"v6", new MapgenFactoryV6},
- {"v7", new MapgenFactoryV7},
- {"singlenode", new MapgenFactorySinglenode},
-};
-
-class EmergeThread : public JThread
-{
+class EmergeThread : public Thread {
public:
- Server *m_server;
- ServerMap *map;
- EmergeManager *emerge;
- Mapgen *mapgen;
bool enable_mapgen_debug_info;
int id;
- Event qevent;
- std::queue<v3s16> blockqueue;
-
- EmergeThread(Server *server, int ethreadid):
- JThread(),
- m_server(server),
- map(NULL),
- emerge(NULL),
- mapgen(NULL),
- enable_mapgen_debug_info(false),
- id(ethreadid)
- {
- }
+ EmergeThread(Server *server, int ethreadid);
+ ~EmergeThread();
+
+ void *run();
+ void signal();
+
+ // Requires queue mutex held
+ bool pushBlock(v3s16 pos);
+
+ void cancelPendingItems();
+
+ static void runCompletionCallbacks(
+ v3s16 pos, EmergeAction action,
+ const EmergeCallbackList &callbacks);
+
+private:
+ Server *m_server;
+ ServerMap *m_map;
+ EmergeManager *m_emerge;
+ Mapgen *m_mapgen;
+
+ Event m_queue_event;
+ std::queue<v3s16> m_block_queue;
+
+ bool popBlockEmerge(v3s16 *pos, BlockEmergeData *bedata);
+
+ EmergeAction getBlockOrStartGen(
+ v3s16 pos, bool allow_gen, MapBlock **block, BlockMakeData *data);
+ MapBlock *finishGen(v3s16 pos, BlockMakeData *bmdata,
+ std::map<v3s16, MapBlock *> *modified_blocks);
- void *Thread();
- bool popBlockEmerge(v3s16 *pos, u8 *flags);
- bool getBlockOrStartGen(v3s16 p, MapBlock **b,
- BlockMakeData *data, bool allow_generate);
+ friend class EmergeManager;
};
+////
+//// Built-in mapgens
+////
+
+MapgenDesc g_reg_mapgens[] = {
+ {"v5", new MapgenFactoryV5, true},
+ {"v6", new MapgenFactoryV6, true},
+ {"v7", new MapgenFactoryV7, true},
+ {"flat", new MapgenFactoryFlat, true},
+ {"fractal", new MapgenFactoryFractal, true},
+ {"valleys", new MapgenFactoryValleys, true},
+ {"singlenode", new MapgenFactorySinglenode, false},
+};
-/////////////////////////////// Emerge Manager ////////////////////////////////
+////
+//// EmergeManager
+////
EmergeManager::EmergeManager(IGameDef *gamedef)
{
@@ -104,34 +129,34 @@ EmergeManager::EmergeManager(IGameDef *gamedef)
// Note that accesses to this variable are not synchronized.
// This is because the *only* thread ever starting or stopping
// EmergeThreads should be the ServerThread.
- this->threads_active = false;
+ this->m_threads_active = false;
- mapgen_debug_info = g_settings->getBool("enable_mapgen_debug_info");
+ enable_mapgen_debug_info = g_settings->getBool("enable_mapgen_debug_info");
- // if unspecified, leave a proc for the main thread and one for
+ // If unspecified, leave a proc for the main thread and one for
// some other misc thread
s16 nthreads = 0;
if (!g_settings->getS16NoEx("num_emerge_threads", nthreads))
- nthreads = porting::getNumberOfProcessors() - 2;
+ nthreads = Thread::getNumberOfProcessors() - 2;
if (nthreads < 1)
nthreads = 1;
- qlimit_total = g_settings->getU16("emergequeue_limit_total");
- if (!g_settings->getU16NoEx("emergequeue_limit_diskonly", qlimit_diskonly))
- qlimit_diskonly = nthreads * 5 + 1;
- if (!g_settings->getU16NoEx("emergequeue_limit_generate", qlimit_generate))
- qlimit_generate = nthreads + 1;
+ m_qlimit_total = g_settings->getU16("emergequeue_limit_total");
+ if (!g_settings->getU16NoEx("emergequeue_limit_diskonly", m_qlimit_diskonly))
+ m_qlimit_diskonly = nthreads * 5 + 1;
+ if (!g_settings->getU16NoEx("emergequeue_limit_generate", m_qlimit_generate))
+ m_qlimit_generate = nthreads + 1;
// don't trust user input for something very important like this
- if (qlimit_total < 1)
- qlimit_total = 1;
- if (qlimit_diskonly < 1)
- qlimit_diskonly = 1;
- if (qlimit_generate < 1)
- qlimit_generate = 1;
+ if (m_qlimit_total < 1)
+ m_qlimit_total = 1;
+ if (m_qlimit_diskonly < 1)
+ m_qlimit_diskonly = 1;
+ if (m_qlimit_generate < 1)
+ m_qlimit_generate = 1;
for (s16 i = 0; i < nthreads; i++)
- emergethread.push_back(new EmergeThread((Server *) gamedef, i));
+ m_threads.push_back(new EmergeThread((Server *)gamedef, i));
infostream << "EmergeManager: using " << nthreads << " threads" << std::endl;
}
@@ -139,27 +164,25 @@ EmergeManager::EmergeManager(IGameDef *gamedef)
EmergeManager::~EmergeManager()
{
- for (u32 i = 0; i != emergethread.size(); i++) {
- if (threads_active) {
- emergethread[i]->Stop();
- emergethread[i]->qevent.signal();
- emergethread[i]->Wait();
+ for (u32 i = 0; i != m_threads.size(); i++) {
+ EmergeThread *thread = m_threads[i];
+
+ if (m_threads_active) {
+ thread->stop();
+ thread->signal();
+ thread->wait();
}
- delete emergethread[i];
- delete mapgen[i];
+
+ delete thread;
+ delete m_mapgens[i];
}
- emergethread.clear();
- mapgen.clear();
delete biomemgr;
delete oremgr;
delete decomgr;
delete schemmgr;
- if (params.sparams) {
- delete params.sparams;
- params.sparams = NULL;
- }
+ delete params.sparams;
}
@@ -171,33 +194,37 @@ void EmergeManager::loadMapgenParams()
void EmergeManager::initMapgens()
{
- if (mapgen.size())
+ if (m_mapgens.size())
return;
+ MapgenFactory *mgfactory = getMapgenFactory(params.mg_name);
+ if (!mgfactory) {
+ errorstream << "EmergeManager: mapgen " << params.mg_name <<
+ " not registered; falling back to " << DEFAULT_MAPGEN << std::endl;
+
+ params.mg_name = DEFAULT_MAPGEN;
+
+ mgfactory = getMapgenFactory(params.mg_name);
+ FATAL_ERROR_IF(mgfactory == NULL, "Couldn't use any mapgen!");
+ }
+
if (!params.sparams) {
- params.sparams = createMapgenParams(params.mg_name);
- if (!params.sparams) {
- params.mg_name = DEFAULT_MAPGEN;
- params.sparams = createMapgenParams(params.mg_name);
- assert(params.sparams);
- }
+ params.sparams = mgfactory->createMapgenParams();
params.sparams->readParams(g_settings);
}
- // Create the mapgens
- for (u32 i = 0; i != emergethread.size(); i++) {
- Mapgen *mg = createMapgen(params.mg_name, i, &params);
- assert(mg);
- mapgen.push_back(mg);
+ for (u32 i = 0; i != m_threads.size(); i++) {
+ Mapgen *mg = mgfactory->createMapgen(i, &params, this);
+ m_mapgens.push_back(mg);
}
}
Mapgen *EmergeManager::getCurrentMapgen()
{
- for (u32 i = 0; i != emergethread.size(); i++) {
- if (emergethread[i]->IsSameThread())
- return emergethread[i]->mapgen;
+ for (u32 i = 0; i != m_threads.size(); i++) {
+ if (m_threads[i]->isCurrentThread())
+ return m_threads[i]->m_mapgen;
}
return NULL;
@@ -206,329 +233,486 @@ Mapgen *EmergeManager::getCurrentMapgen()
void EmergeManager::startThreads()
{
- if (threads_active)
+ if (m_threads_active)
return;
- for (u32 i = 0; i != emergethread.size(); i++)
- emergethread[i]->Start();
+ for (u32 i = 0; i != m_threads.size(); i++)
+ m_threads[i]->start();
- threads_active = true;
+ m_threads_active = true;
}
void EmergeManager::stopThreads()
{
- if (!threads_active)
+ if (!m_threads_active)
return;
// Request thread stop in parallel
- for (u32 i = 0; i != emergethread.size(); i++) {
- emergethread[i]->Stop();
- emergethread[i]->qevent.signal();
+ for (u32 i = 0; i != m_threads.size(); i++) {
+ m_threads[i]->stop();
+ m_threads[i]->signal();
}
// Then do the waiting for each
- for (u32 i = 0; i != emergethread.size(); i++)
- emergethread[i]->Wait();
+ for (u32 i = 0; i != m_threads.size(); i++)
+ m_threads[i]->wait();
- threads_active = false;
+ m_threads_active = false;
}
-bool EmergeManager::enqueueBlockEmerge(u16 peer_id, v3s16 p, bool allow_generate)
+bool EmergeManager::isRunning()
{
- std::map<v3s16, BlockEmergeData *>::const_iterator iter;
- BlockEmergeData *bedata;
- u16 count;
- u8 flags = 0;
- int idx = 0;
+ return m_threads_active;
+}
+
+bool EmergeManager::enqueueBlockEmerge(
+ u16 peer_id,
+ v3s16 blockpos,
+ bool allow_generate,
+ bool ignore_queue_limits)
+{
+ u16 flags = 0;
if (allow_generate)
- flags |= BLOCK_EMERGE_ALLOWGEN;
+ flags |= BLOCK_EMERGE_ALLOW_GEN;
+ if (ignore_queue_limits)
+ flags |= BLOCK_EMERGE_FORCE_QUEUE;
- {
- JMutexAutoLock queuelock(queuemutex);
+ return enqueueBlockEmergeEx(blockpos, peer_id, flags, NULL, NULL);
+}
- count = blocks_enqueued.size();
- if (count >= qlimit_total)
- return false;
- count = peer_queue_count[peer_id];
- u16 qlimit_peer = allow_generate ? qlimit_generate : qlimit_diskonly;
- if (count >= qlimit_peer)
+bool EmergeManager::enqueueBlockEmergeEx(
+ v3s16 blockpos,
+ u16 peer_id,
+ u16 flags,
+ EmergeCompletionCallback callback,
+ void *callback_param)
+{
+ EmergeThread *thread = NULL;
+ bool entry_already_exists = false;
+
+ {
+ MutexAutoLock queuelock(m_queue_mutex);
+
+ if (!pushBlockEmergeData(blockpos, peer_id, flags,
+ callback, callback_param, &entry_already_exists))
return false;
- iter = blocks_enqueued.find(p);
- if (iter != blocks_enqueued.end()) {
- bedata = iter->second;
- bedata->flags |= flags;
+ if (entry_already_exists)
return true;
- }
- bedata = new BlockEmergeData;
- bedata->flags = flags;
- bedata->peer_requested = peer_id;
- blocks_enqueued.insert(std::make_pair(p, bedata));
+ thread = getOptimalThread();
+ thread->pushBlock(blockpos);
+ }
- peer_queue_count[peer_id] = count + 1;
+ thread->signal();
- // insert into the EmergeThread queue with the least items
- int lowestitems = emergethread[0]->blockqueue.size();
- for (u32 i = 1; i != emergethread.size(); i++) {
- int nitems = emergethread[i]->blockqueue.size();
- if (nitems < lowestitems) {
- idx = i;
- lowestitems = nitems;
- }
- }
+ return true;
+}
- emergethread[idx]->blockqueue.push(p);
+
+//
+// Mapgen-related helper functions
+//
+
+v3s16 EmergeManager::getContainingChunk(v3s16 blockpos)
+{
+ return getContainingChunk(blockpos, params.chunksize);
+}
+
+
+v3s16 EmergeManager::getContainingChunk(v3s16 blockpos, s16 chunksize)
+{
+ s16 coff = -chunksize / 2;
+ v3s16 chunk_offset(coff, coff, coff);
+
+ return getContainerPos(blockpos - chunk_offset, chunksize)
+ * chunksize + chunk_offset;
+}
+
+
+int EmergeManager::getSpawnLevelAtPoint(v2s16 p)
+{
+ if (m_mapgens.size() == 0 || !m_mapgens[0]) {
+ errorstream << "EmergeManager: getSpawnLevelAtPoint() called"
+ " before mapgen init" << std::endl;
+ return 0;
}
- emergethread[idx]->qevent.signal();
- return true;
+ return m_mapgens[0]->getSpawnLevelAtPoint(p);
}
int EmergeManager::getGroundLevelAtPoint(v2s16 p)
{
- if (mapgen.size() == 0 || !mapgen[0]) {
+ if (m_mapgens.size() == 0 || !m_mapgens[0]) {
errorstream << "EmergeManager: getGroundLevelAtPoint() called"
- " before mapgen initialized" << std::endl;
+ " before mapgen init" << std::endl;
return 0;
}
- return mapgen[0]->getGroundLevelAtPoint(p);
+ return m_mapgens[0]->getGroundLevelAtPoint(p);
}
bool EmergeManager::isBlockUnderground(v3s16 blockpos)
{
- /*
+#if 0
v2s16 p = v2s16((blockpos.X * MAP_BLOCKSIZE) + MAP_BLOCKSIZE / 2,
(blockpos.Y * MAP_BLOCKSIZE) + MAP_BLOCKSIZE / 2);
int ground_level = getGroundLevelAtPoint(p);
return blockpos.Y * (MAP_BLOCKSIZE + 1) <= min(water_level, ground_level);
- */
+#endif
- //yuck, but then again, should i bother being accurate?
- //the height of the nodes in a single block is quite variable
+ // Use a simple heuristic; the above method is wildly inaccurate anyway.
return blockpos.Y * (MAP_BLOCKSIZE + 1) <= params.water_level;
}
-void EmergeManager::getMapgenNames(std::list<const char *> &mgnames)
+void EmergeManager::getMapgenNames(
+ std::vector<const char *> *mgnames, bool include_hidden)
{
- for (u32 i = 0; i != ARRLEN(reg_mapgens); i++)
- mgnames.push_back(reg_mapgens[i].name);
+ for (u32 i = 0; i != ARRLEN(g_reg_mapgens); i++) {
+ if (include_hidden || g_reg_mapgens[i].is_user_visible)
+ mgnames->push_back(g_reg_mapgens[i].name);
+ }
}
-Mapgen *EmergeManager::createMapgen(const std::string &mgname, int mgid,
- MapgenParams *mgparams)
+MapgenFactory *EmergeManager::getMapgenFactory(const std::string &mgname)
{
- u32 i;
- for (i = 0; i != ARRLEN(reg_mapgens) && mgname != reg_mapgens[i].name; i++);
- if (i == ARRLEN(reg_mapgens)) {
- errorstream << "EmergeManager; mapgen " << mgname <<
- " not registered" << std::endl;
- return NULL;
+ for (u32 i = 0; i != ARRLEN(g_reg_mapgens); i++) {
+ if (mgname == g_reg_mapgens[i].name)
+ return g_reg_mapgens[i].factory;
}
- MapgenFactory *mgfactory = reg_mapgens[i].factory;
- return mgfactory->createMapgen(mgid, mgparams, this);
+ return NULL;
}
-MapgenSpecificParams *EmergeManager::createMapgenParams(const std::string &mgname)
+bool EmergeManager::pushBlockEmergeData(
+ v3s16 pos,
+ u16 peer_requested,
+ u16 flags,
+ EmergeCompletionCallback callback,
+ void *callback_param,
+ bool *entry_already_exists)
{
- u32 i;
- for (i = 0; i < ARRLEN(reg_mapgens) && mgname != reg_mapgens[i].name; i++);
- if (i == ARRLEN(reg_mapgens)) {
- errorstream << "EmergeManager: Mapgen " << mgname <<
- " not registered" << std::endl;
- return NULL;
+ u16 &count_peer = m_peer_queue_count[peer_requested];
+
+ if ((flags & BLOCK_EMERGE_FORCE_QUEUE) == 0) {
+ if (m_blocks_enqueued.size() >= m_qlimit_total)
+ return false;
+
+ if (peer_requested != PEER_ID_INEXISTENT) {
+ u16 qlimit_peer = (flags & BLOCK_EMERGE_ALLOW_GEN) ?
+ m_qlimit_generate : m_qlimit_diskonly;
+ if (count_peer >= qlimit_peer)
+ return false;
+ }
}
- MapgenFactory *mgfactory = reg_mapgens[i].factory;
- return mgfactory->createMapgenParams();
-}
+ std::pair<std::map<v3s16, BlockEmergeData>::iterator, bool> findres;
+ findres = m_blocks_enqueued.insert(std::make_pair(pos, BlockEmergeData()));
+
+ BlockEmergeData &bedata = findres.first->second;
+ *entry_already_exists = !findres.second;
+ if (callback)
+ bedata.callbacks.push_back(std::make_pair(callback, callback_param));
-////////////////////////////// Emerge Thread //////////////////////////////////
+ if (*entry_already_exists) {
+ bedata.flags |= flags;
+ } else {
+ bedata.flags = flags;
+ bedata.peer_requested = peer_requested;
+
+ count_peer++;
+ }
+
+ return true;
+}
-bool EmergeThread::popBlockEmerge(v3s16 *pos, u8 *flags)
+
+bool EmergeManager::popBlockEmergeData(
+ v3s16 pos,
+ BlockEmergeData *bedata)
{
- std::map<v3s16, BlockEmergeData *>::iterator iter;
- JMutexAutoLock queuelock(emerge->queuemutex);
+ std::map<v3s16, BlockEmergeData>::iterator it;
+ std::map<u16, u16>::iterator it2;
+
+ it = m_blocks_enqueued.find(pos);
+ if (it == m_blocks_enqueued.end())
+ return false;
- if (blockqueue.empty())
+ *bedata = it->second;
+
+ it2 = m_peer_queue_count.find(bedata->peer_requested);
+ if (it2 == m_peer_queue_count.end())
return false;
- v3s16 p = blockqueue.front();
- blockqueue.pop();
- *pos = p;
+ u16 &count_peer = it2->second;
+ assert(count_peer != 0);
+ count_peer--;
+
+ m_blocks_enqueued.erase(it);
+
+ return true;
+}
+
+
+EmergeThread *EmergeManager::getOptimalThread()
+{
+ size_t nthreads = m_threads.size();
+
+ FATAL_ERROR_IF(nthreads == 0, "No emerge threads!");
+
+ size_t index = 0;
+ size_t nitems_lowest = m_threads[0]->m_block_queue.size();
+
+ for (size_t i = 1; i < nthreads; i++) {
+ size_t nitems = m_threads[i]->m_block_queue.size();
+ if (nitems < nitems_lowest) {
+ index = i;
+ nitems_lowest = nitems;
+ }
+ }
+
+ return m_threads[index];
+}
+
+
+////
+//// EmergeThread
+////
+
+EmergeThread::EmergeThread(Server *server, int ethreadid) :
+ enable_mapgen_debug_info(false),
+ id(ethreadid),
+ m_server(server),
+ m_map(NULL),
+ m_emerge(NULL),
+ m_mapgen(NULL)
+{
+ m_name = "Emerge-" + itos(ethreadid);
+}
+
+
+EmergeThread::~EmergeThread()
+{
+ //cancelPendingItems();
+}
+
+
+void EmergeThread::signal()
+{
+ m_queue_event.signal();
+}
+
+
+bool EmergeThread::pushBlock(v3s16 pos)
+{
+ m_block_queue.push(pos);
+ return true;
+}
+
+
+void EmergeThread::cancelPendingItems()
+{
+ MutexAutoLock queuelock(m_emerge->m_queue_mutex);
+
+ while (!m_block_queue.empty()) {
+ BlockEmergeData bedata;
+ v3s16 pos;
+
+ pos = m_block_queue.front();
+ m_block_queue.pop();
+
+ m_emerge->popBlockEmergeData(pos, &bedata);
- iter = emerge->blocks_enqueued.find(p);
- if (iter == emerge->blocks_enqueued.end())
- return false; //uh oh, queue and map out of sync!!
+ runCompletionCallbacks(pos, EMERGE_CANCELLED, bedata.callbacks);
+ }
+}
- BlockEmergeData *bedata = iter->second;
- *flags = bedata->flags;
- emerge->peer_queue_count[bedata->peer_requested]--;
+void EmergeThread::runCompletionCallbacks(
+ v3s16 pos,
+ EmergeAction action,
+ const EmergeCallbackList &callbacks)
+{
+ for (size_t i = 0; i != callbacks.size(); i++) {
+ EmergeCompletionCallback callback;
+ void *param;
+
+ callback = callbacks[i].first;
+ param = callbacks[i].second;
+
+ callback(pos, action, param);
+ }
+}
+
+
+bool EmergeThread::popBlockEmerge(v3s16 *pos, BlockEmergeData *bedata)
+{
+ MutexAutoLock queuelock(m_emerge->m_queue_mutex);
+
+ if (m_block_queue.empty())
+ return false;
+
+ *pos = m_block_queue.front();
+ m_block_queue.pop();
- delete bedata;
- emerge->blocks_enqueued.erase(iter);
+ m_emerge->popBlockEmergeData(*pos, bedata);
return true;
}
-bool EmergeThread::getBlockOrStartGen(v3s16 p, MapBlock **b,
- BlockMakeData *data, bool allow_gen)
+EmergeAction EmergeThread::getBlockOrStartGen(
+ v3s16 pos, bool allow_gen, MapBlock **block, BlockMakeData *bmdata)
{
- v2s16 p2d(p.X, p.Z);
- //envlock: usually takes <=1ms, sometimes 90ms or ~400ms to acquire
- JMutexAutoLock envlock(m_server->m_env_mutex);
-
- // Load sector if it isn't loaded
- if (map->getSectorNoGenerateNoEx(p2d) == NULL)
- map->loadSectorMeta(p2d);
-
- // Attempt to load block
- MapBlock *block = map->getBlockNoCreateNoEx(p);
- if (!block || block->isDummy() || !block->isGenerated()) {
- EMERGE_DBG_OUT("not in memory, attempting to load from disk");
- block = map->loadBlock(p);
- if (block && block->isGenerated())
- map->prepareBlock(block);
+ MutexAutoLock envlock(m_server->m_env_mutex);
+
+ // 1). Attempt to fetch block from memory
+ *block = m_map->getBlockNoCreateNoEx(pos);
+ if (*block && !(*block)->isDummy() && (*block)->isGenerated())
+ return EMERGE_FROM_MEMORY;
+
+ // 2). Attempt to load block from disk
+ *block = m_map->loadBlock(pos);
+ if (*block && (*block)->isGenerated())
+ return EMERGE_FROM_DISK;
+
+ // 3). Attempt to start generation
+ if (allow_gen && m_map->initBlockMake(pos, bmdata))
+ return EMERGE_GENERATED;
+
+ // All attempts failed; cancel this block emerge
+ return EMERGE_CANCELLED;
+}
+
+
+MapBlock *EmergeThread::finishGen(v3s16 pos, BlockMakeData *bmdata,
+ std::map<v3s16, MapBlock *> *modified_blocks)
+{
+ MutexAutoLock envlock(m_server->m_env_mutex);
+ ScopeProfiler sp(g_profiler,
+ "EmergeThread: after Mapgen::makeChunk", SPT_AVG);
+
+ /*
+ Perform post-processing on blocks (invalidate lighting, queue liquid
+ transforms, etc.) to finish block make
+ */
+ m_map->finishBlockMake(bmdata, modified_blocks);
+
+ MapBlock *block = m_map->getBlockNoCreateNoEx(pos);
+ if (!block) {
+ errorstream << "EmergeThread::finishGen: Couldn't grab block we "
+ "just generated: " << PP(pos) << std::endl;
+ return NULL;
}
- // If could not load and allowed to generate,
- // start generation inside this same envlock
- if (allow_gen && (block == NULL || !block->isGenerated())) {
- EMERGE_DBG_OUT("generating");
- *b = block;
- return map->initBlockMake(data, p);
+ v3s16 minp = bmdata->blockpos_min * MAP_BLOCKSIZE;
+ v3s16 maxp = bmdata->blockpos_max * MAP_BLOCKSIZE +
+ v3s16(1,1,1) * (MAP_BLOCKSIZE - 1);
+
+ // Ignore map edit events, they will not need to be sent
+ // to anybody because the block hasn't been sent to anybody
+ MapEditEventAreaIgnorer ign(
+ &m_server->m_ignore_map_edit_events_area,
+ VoxelArea(minp, maxp));
+
+ /*
+ Run Lua on_generated callbacks
+ */
+ try {
+ m_server->getScriptIface()->environment_OnGenerated(
+ minp, maxp, m_mapgen->blockseed);
+ } catch (LuaError &e) {
+ m_server->setAsyncFatalError("Lua: " + std::string(e.what()));
}
- *b = block;
- return false;
+ EMERGE_DBG_OUT("ended up with: " << analyze_block(block));
+
+ /*
+ Activate the block
+ */
+ m_server->m_env->activateBlock(block, 0);
+
+ return block;
}
-void *EmergeThread::Thread()
+void *EmergeThread::run()
{
- ThreadStarted();
- log_register_thread("EmergeThread" + itos(id));
- DSTACK(__FUNCTION_NAME);
+ DSTACK(FUNCTION_NAME);
BEGIN_DEBUG_EXCEPTION_HANDLER
- v3s16 last_tried_pos(-32768,-32768,-32768); // For error output
- v3s16 p;
- u8 flags = 0;
-
- map = (ServerMap *)&(m_server->m_env->getMap());
- emerge = m_server->m_emerge;
- mapgen = emerge->mapgen[id];
- enable_mapgen_debug_info = emerge->mapgen_debug_info;
+ v3s16 pos;
- porting::setThreadName("EmergeThread");
+ m_map = (ServerMap *)&(m_server->m_env->getMap());
+ m_emerge = m_server->m_emerge;
+ m_mapgen = m_emerge->m_mapgens[id];
+ enable_mapgen_debug_info = m_emerge->enable_mapgen_debug_info;
- while (!StopRequested())
try {
- if (!popBlockEmerge(&p, &flags)) {
- qevent.wait();
+ while (!stopRequested()) {
+ std::map<v3s16, MapBlock *> modified_blocks;
+ BlockEmergeData bedata;
+ BlockMakeData bmdata;
+ EmergeAction action;
+ MapBlock *block;
+
+ if (!popBlockEmerge(&pos, &bedata)) {
+ m_queue_event.wait();
continue;
}
- last_tried_pos = p;
- if (blockpos_over_limit(p))
+ if (blockpos_over_limit(pos))
continue;
- bool allow_generate = flags & BLOCK_EMERGE_ALLOWGEN;
- EMERGE_DBG_OUT("p=" PP(p) " allow_generate=" << allow_generate);
-
- /*
- Try to fetch block from memory or disk.
- If not found and asked to generate, initialize generator.
- */
- BlockMakeData data;
- MapBlock *block = NULL;
- std::map<v3s16, MapBlock *> modified_blocks;
+ bool allow_gen = bedata.flags & BLOCK_EMERGE_ALLOW_GEN;
+ EMERGE_DBG_OUT("pos=" PP(pos) " allow_gen=" << allow_gen);
- if (getBlockOrStartGen(p, &block, &data, allow_generate) && mapgen) {
+ action = getBlockOrStartGen(pos, allow_gen, &block, &bmdata);
+ if (action == EMERGE_GENERATED) {
{
- ScopeProfiler sp(g_profiler, "EmergeThread: Mapgen::makeChunk", SPT_AVG);
+ ScopeProfiler sp(g_profiler,
+ "EmergeThread: Mapgen::makeChunk", SPT_AVG);
TimeTaker t("mapgen::make_block()");
- mapgen->makeChunk(&data);
+ m_mapgen->makeChunk(&bmdata);
if (enable_mapgen_debug_info == false)
t.stop(true); // Hide output
}
- {
- //envlock: usually 0ms, but can take either 30 or 400ms to acquire
- JMutexAutoLock envlock(m_server->m_env_mutex);
- ScopeProfiler sp(g_profiler, "EmergeThread: after "
- "Mapgen::makeChunk (envlock)", SPT_AVG);
-
- map->finishBlockMake(&data, modified_blocks);
-
- block = map->getBlockNoCreateNoEx(p);
- if (block) {
- /*
- Do some post-generate stuff
- */
- v3s16 minp = data.blockpos_min * MAP_BLOCKSIZE;
- v3s16 maxp = data.blockpos_max * MAP_BLOCKSIZE +
- v3s16(1,1,1) * (MAP_BLOCKSIZE - 1);
-
- // Ignore map edit events, they will not need to be sent
- // to anybody because the block hasn't been sent to anybody
- MapEditEventAreaIgnorer
- ign(&m_server->m_ignore_map_edit_events_area,
- VoxelArea(minp, maxp));
- try { // takes about 90ms with -O1 on an e3-1230v2
- m_server->getScriptIface()->environment_OnGenerated(
- minp, maxp, mapgen->blockseed);
- } catch (LuaError &e) {
- m_server->setAsyncFatalError("Lua: " + std::string(e.what()));
- }
-
- EMERGE_DBG_OUT("ended up with: " << analyze_block(block));
-
- m_server->m_env->activateBlock(block, 0);
- }
- }
+ block = finishGen(pos, &bmdata, &modified_blocks);
}
- /*
- Set sent status of modified blocks on clients
- */
- // Add the originally fetched block to the modified list
+ runCompletionCallbacks(pos, action, bedata.callbacks);
+
if (block)
- modified_blocks[p] = block;
+ modified_blocks[pos] = block;
- if (modified_blocks.size() > 0) {
+ if (modified_blocks.size() > 0)
m_server->SetBlocksNotSent(modified_blocks);
- }
}
- catch (VersionMismatchException &e) {
+ } catch (VersionMismatchException &e) {
std::ostringstream err;
- err << "World data version mismatch in MapBlock " << PP(last_tried_pos) << std::endl
+ err << "World data version mismatch in MapBlock " << PP(pos) << std::endl
<< "----" << std::endl
<< "\"" << e.what() << "\"" << std::endl
<< "See debug.txt." << std::endl
<< "World probably saved by a newer version of " PROJECT_NAME_C "."
<< std::endl;
m_server->setAsyncFatalError(err.str());
- }
- catch (SerializationError &e) {
+ } catch (SerializationError &e) {
std::ostringstream err;
- err << "Invalid data in MapBlock " << PP(last_tried_pos) << std::endl
+ err << "Invalid data in MapBlock " << PP(pos) << std::endl
<< "----" << std::endl
<< "\"" << e.what() << "\"" << std::endl
<< "See debug.txt." << std::endl
@@ -537,24 +721,6 @@ void *EmergeThread::Thread()
m_server->setAsyncFatalError(err.str());
}
- {
- JMutexAutoLock queuelock(emerge->queuemutex);
- while (!blockqueue.empty())
- {
- v3s16 p = blockqueue.front();
- blockqueue.pop();
-
- std::map<v3s16, BlockEmergeData *>::iterator iter;
- iter = emerge->blocks_enqueued.find(p);
- if (iter == emerge->blocks_enqueued.end())
- continue; //uh oh, queue and map out of sync!!
-
- BlockEmergeData *bedata = iter->second;
- delete bedata;
- }
- }
-
- END_DEBUG_EXCEPTION_HANDLER(errorstream)
- log_deregister_thread();
+ END_DEBUG_EXCEPTION_HANDLER
return NULL;
}