|
|
@ -9,6 +9,7 @@
|
|
|
|
#include "modules/Items.h"
|
|
|
|
#include "modules/Items.h"
|
|
|
|
#include "modules/Job.h"
|
|
|
|
#include "modules/Job.h"
|
|
|
|
#include "modules/Translation.h"
|
|
|
|
#include "modules/Translation.h"
|
|
|
|
|
|
|
|
#include "modules/Random.h"
|
|
|
|
|
|
|
|
|
|
|
|
#include "DataDefs.h"
|
|
|
|
#include "DataDefs.h"
|
|
|
|
#include "df/d_init.h"
|
|
|
|
#include "df/d_init.h"
|
|
|
@ -40,6 +41,8 @@ using df::global::created_item_subtype;
|
|
|
|
using df::global::created_item_mattype;
|
|
|
|
using df::global::created_item_mattype;
|
|
|
|
using df::global::created_item_matindex;
|
|
|
|
using df::global::created_item_matindex;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Random::MersenneRNG rng;
|
|
|
|
|
|
|
|
|
|
|
|
bool isUnitMoodable (df::unit *unit)
|
|
|
|
bool isUnitMoodable (df::unit *unit)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if (!Units::isCitizen(unit))
|
|
|
|
if (!Units::isCitizen(unit))
|
|
|
@ -107,7 +110,7 @@ df::job_skill getMoodSkill (df::unit *unit)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (!skills.size())
|
|
|
|
if (!skills.size())
|
|
|
|
skills.push_back(job_skill::STONECRAFT);
|
|
|
|
skills.push_back(job_skill::STONECRAFT);
|
|
|
|
return skills[rand() % skills.size()];
|
|
|
|
return skills[rng.df_trandom(skills.size())];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int getCreatedMetalBars (int32_t idx)
|
|
|
|
int getCreatedMetalBars (int32_t idx)
|
|
|
@ -127,14 +130,14 @@ void selectWord (const df::world_raws::T_language::T_word_table &table, int32_t
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if (table.parts[mode].size())
|
|
|
|
if (table.parts[mode].size())
|
|
|
|
{
|
|
|
|
{
|
|
|
|
int offset = rand() % table.parts[mode].size();
|
|
|
|
int offset = rng.df_trandom(table.parts[mode].size());
|
|
|
|
word = table.words[mode][offset];
|
|
|
|
word = table.words[mode][offset];
|
|
|
|
part = table.parts[mode][offset];
|
|
|
|
part = table.parts[mode][offset];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
else
|
|
|
|
{
|
|
|
|
{
|
|
|
|
word = rand() % world->raws.language.words.size();
|
|
|
|
word = rng.df_trandom(world->raws.language.words.size());
|
|
|
|
part = (df::part_of_speech)(rand() % 9);
|
|
|
|
part = (df::part_of_speech)(rng.df_trandom(9));
|
|
|
|
Core::getInstance().getConsole().printerr("Impoverished Word Selector");
|
|
|
|
Core::getInstance().getConsole().printerr("Impoverished Word Selector");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
@ -147,13 +150,13 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
{
|
|
|
|
{
|
|
|
|
output = df::language_name();
|
|
|
|
output = df::language_name();
|
|
|
|
if (language == -1)
|
|
|
|
if (language == -1)
|
|
|
|
language = rand() % world->raws.language.translations.size();
|
|
|
|
language = rng.df_trandom(world->raws.language.translations.size());
|
|
|
|
output.unknown = mode;
|
|
|
|
output.unknown = mode;
|
|
|
|
output.language = language;
|
|
|
|
output.language = language;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
output.has_name = 1;
|
|
|
|
output.has_name = 1;
|
|
|
|
if (output.language == -1)
|
|
|
|
if (output.language == -1)
|
|
|
|
output.language = rand() % world->raws.language.translations.size();
|
|
|
|
output.language = rng.df_trandom(world->raws.language.translations.size());
|
|
|
|
int r, r2, r3;
|
|
|
|
int r, r2, r3;
|
|
|
|
switch (mode)
|
|
|
|
switch (mode)
|
|
|
|
{
|
|
|
|
{
|
|
|
@ -169,7 +172,7 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
if (mode != 10)
|
|
|
|
if (mode != 10)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 4: case 37: // this is not a typo
|
|
|
|
case 4: case 37: // this is not a typo
|
|
|
|
if (rand() % 2)
|
|
|
|
if (rng.df_trandom(2))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
|
selectWord(table1, output.words[1], output.parts_of_speech[1], 1);
|
|
|
|
selectWord(table1, output.words[1], output.parts_of_speech[1], 1);
|
|
|
@ -183,10 +186,10 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
|
|
case 1: case 13: case 20:
|
|
|
|
case 1: case 13: case 20:
|
|
|
|
r = rand() % 3;
|
|
|
|
r = rng.df_trandom(3);
|
|
|
|
if (r == 0 || r == 1)
|
|
|
|
if (r == 0 || r == 1)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if (rand() % 2)
|
|
|
|
if (rng.df_trandom(2))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
|
selectWord(table1, output.words[1], output.parts_of_speech[1], 1);
|
|
|
|
selectWord(table1, output.words[1], output.parts_of_speech[1], 1);
|
|
|
@ -200,20 +203,20 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
if (r == 1 || r == 2)
|
|
|
|
if (r == 1 || r == 2)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 3: case 8: case 11: // this is not a typo either
|
|
|
|
case 3: case 8: case 11: // this is not a typo either
|
|
|
|
r2 = rand() % 2;
|
|
|
|
r2 = rng.df_trandom(2);
|
|
|
|
if (r2)
|
|
|
|
if (r2)
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
else
|
|
|
|
else
|
|
|
|
selectWord(table2, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
selectWord(table2, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
r3 = rand() % 3;
|
|
|
|
r3 = rng.df_trandom(3);
|
|
|
|
if (rand() % 50)
|
|
|
|
if (rng.df_trandom(50))
|
|
|
|
r3 = rand() % 2;
|
|
|
|
r3 = rng.df_trandom(2);
|
|
|
|
switch (r3)
|
|
|
|
switch (r3)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
case 2:
|
|
|
|
case 2:
|
|
|
|
if (r3 == 2)
|
|
|
|
if (r3 == 2)
|
|
|
|
r2 = rand() % 2;
|
|
|
|
r2 = rng.df_trandom(2);
|
|
|
|
if (r2)
|
|
|
|
if (r2)
|
|
|
|
selectWord(table2, output.words[6], output.parts_of_speech[6], 5);
|
|
|
|
selectWord(table2, output.words[6], output.parts_of_speech[6], 5);
|
|
|
|
else
|
|
|
|
else
|
|
|
@ -226,14 +229,14 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
selectWord(table1, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
selectWord(table1, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
else
|
|
|
|
else
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
if (!(rand() % 100))
|
|
|
|
if (!(rng.df_trandom(100)))
|
|
|
|
selectWord(table1, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
selectWord(table1, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (rand() % 100)
|
|
|
|
if (rng.df_trandom(100))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if (rand() % 2)
|
|
|
|
if (rng.df_trandom(2))
|
|
|
|
selectWord(table1, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
selectWord(table1, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
else
|
|
|
|
else
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
@ -246,9 +249,9 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
case 21: case 22: case 23: case 24: case 25: case 26: case 27: case 28: case 29: case 30:
|
|
|
|
case 21: case 22: case 23: case 24: case 25: case 26: case 27: case 28: case 29: case 30:
|
|
|
|
case 31: case 32: case 33: case 34: case 35: case 36: case 38: case 39:
|
|
|
|
case 31: case 32: case 33: case 34: case 35: case 36: case 38: case 39:
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
r3 = rand() % 3;
|
|
|
|
r3 = rng.df_trandom(3);
|
|
|
|
if (rand() % 50)
|
|
|
|
if (rng.df_trandom(50))
|
|
|
|
r3 = rand() % 2;
|
|
|
|
r3 = rng.df_trandom(2);
|
|
|
|
switch (r3)
|
|
|
|
switch (r3)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
@ -258,16 +261,16 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
case 1:
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
if (!(rand() % 100))
|
|
|
|
if (!(rng.df_trandom(100)))
|
|
|
|
selectWord(table2, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
selectWord(table2, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (rand() % 100)
|
|
|
|
if (rng.df_trandom(100))
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
|
|
case 7:
|
|
|
|
case 7:
|
|
|
|
r = rand() % 3;
|
|
|
|
r = rng.df_trandom(3);
|
|
|
|
if (r == 0 || r == 1)
|
|
|
|
if (r == 0 || r == 1)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
|
selectWord(table2, output.words[0], output.parts_of_speech[0], 0);
|
|
|
@ -275,14 +278,14 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (r == 1 || r == 2)
|
|
|
|
if (r == 1 || r == 2)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
r2 = rand() % 2;
|
|
|
|
r2 = rng.df_trandom(2);
|
|
|
|
if (r == 2 || r2 == 1)
|
|
|
|
if (r == 2 || r2 == 1)
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
selectWord(table1, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
else
|
|
|
|
else
|
|
|
|
selectWord(table2, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
selectWord(table2, output.words[5], output.parts_of_speech[5], 2);
|
|
|
|
r3 = rand() % 3;
|
|
|
|
r3 = rng.df_trandom(3);
|
|
|
|
if (rand() % 50)
|
|
|
|
if (rng.df_trandom(50))
|
|
|
|
r3 = rand() % 2;
|
|
|
|
r3 = rng.df_trandom(2);
|
|
|
|
switch (r3)
|
|
|
|
switch (r3)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
@ -292,12 +295,12 @@ void generateName(df::language_name &output, int language, int mode, const df::w
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
case 1:
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
selectWord(table2, output.words[2], output.parts_of_speech[2], 3);
|
|
|
|
if (!(rand() % 100))
|
|
|
|
if (!(rng.df_trandom(100)))
|
|
|
|
selectWord(table2, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
selectWord(table2, output.words[3], output.parts_of_speech[3], 3);
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (rand() % 100)
|
|
|
|
if (rng.df_trandom(100))
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
selectWord(table2, output.words[4], output.parts_of_speech[4], 4);
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
@ -610,7 +613,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
out.printerr("No units are eligible to enter a mood!\n");
|
|
|
|
out.printerr("No units are eligible to enter a mood!\n");
|
|
|
|
return CR_FAILURE;
|
|
|
|
return CR_FAILURE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
unit = moodable_units[tickets[rand() % tickets.size()]];
|
|
|
|
unit = moodable_units[tickets[rng.df_trandom(tickets.size())]];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
df::unit_soul *soul = unit->status.current_soul;
|
|
|
|
df::unit_soul *soul = unit->status.current_soul;
|
|
|
|
|
|
|
|
|
|
|
@ -626,9 +629,9 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
// If no mood type was specified, pick one randomly
|
|
|
|
// If no mood type was specified, pick one randomly
|
|
|
|
if (type == mood_type::None)
|
|
|
|
if (type == mood_type::None)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if (rand() % 100 > unit->status.happiness)
|
|
|
|
if (rng.df_trandom(100) > unit->status.happiness)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
switch (rand() % 2)
|
|
|
|
switch (rng.df_trandom(2))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0: type = mood_type::Fell; break;
|
|
|
|
case 0: type = mood_type::Fell; break;
|
|
|
|
case 1: type = mood_type::Macabre; break;
|
|
|
|
case 1: type = mood_type::Macabre; break;
|
|
|
@ -636,7 +639,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
else
|
|
|
|
{
|
|
|
|
{
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0: type = mood_type::Fey; break;
|
|
|
|
case 0: type = mood_type::Fey; break;
|
|
|
|
case 1: type = mood_type::Secretive; break;
|
|
|
|
case 1: type = mood_type::Secretive; break;
|
|
|
@ -771,9 +774,9 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
// The dwarf will want 1-3 of the base material
|
|
|
|
// The dwarf will want 1-3 of the base material
|
|
|
|
int base_item_count = 1 + (rand() % 3);
|
|
|
|
int base_item_count = 1 + rng.df_trandom(3);
|
|
|
|
// Gem Cutters and Gem Setters have a 50% chance of using only one base item
|
|
|
|
// Gem Cutters and Gem Setters have a 50% chance of using only one base item
|
|
|
|
if (((skill == job_skill::CUTGEM) || (skill == job_skill::ENCRUSTGEM)) && (rand() % 2))
|
|
|
|
if (((skill == job_skill::CUTGEM) || (skill == job_skill::ENCRUSTGEM)) && (rng.df_trandom(2)))
|
|
|
|
base_item_count = 1;
|
|
|
|
base_item_count = 1;
|
|
|
|
|
|
|
|
|
|
|
|
// Choose the base material
|
|
|
|
// Choose the base material
|
|
|
@ -789,7 +792,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (job->job_type == job_type::StrangeMoodBrooding)
|
|
|
|
else if (job->job_type == job_type::StrangeMoodBrooding)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
job->job_items.push_back(item = new df::job_item());
|
|
|
|
job->job_items.push_back(item = new df::job_item());
|
|
|
@ -930,7 +933,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (!found_pref)
|
|
|
|
if (!found_pref)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
item->flags2.bits.silk = true;
|
|
|
|
item->flags2.bits.silk = true;
|
|
|
@ -1000,7 +1003,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (mats.size())
|
|
|
|
if (mats.size())
|
|
|
|
item->mat_index = mats[rand() & mats.size()];
|
|
|
|
item->mat_index = mats[rng.df_trandom(mats.size())];
|
|
|
|
item->quantity = base_item_count;
|
|
|
|
item->quantity = base_item_count;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
break;
|
|
|
@ -1042,7 +1045,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
mats.push_back(builtin_mats::GLASS_CLEAR);
|
|
|
|
mats.push_back(builtin_mats::GLASS_CLEAR);
|
|
|
|
if (have_glass[2])
|
|
|
|
if (have_glass[2])
|
|
|
|
mats.push_back(builtin_mats::GLASS_CRYSTAL);
|
|
|
|
mats.push_back(builtin_mats::GLASS_CRYSTAL);
|
|
|
|
item->mat_type = mats[rand() % mats.size()];
|
|
|
|
item->mat_type = mats[rng.df_trandom(mats.size())];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
item->quantity = base_item_count;
|
|
|
|
item->quantity = base_item_count;
|
|
|
|
break;
|
|
|
|
break;
|
|
|
@ -1096,9 +1099,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
(job->job_type == job_type::StrangeMoodFell)
|
|
|
|
(job->job_type == job_type::StrangeMoodFell)
|
|
|
|
))
|
|
|
|
))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
int extra_items = (ui->tasks.num_artifacts * 20 + moodable_units.size()) / 20;
|
|
|
|
int extra_items = std::min(rng.df_trandom((ui->tasks.num_artifacts * 20 + moodable_units.size()) / 20 + 1), 7);
|
|
|
|
if (extra_items > 0)
|
|
|
|
|
|
|
|
extra_items = std::min(rand() % (extra_items + 1), 7);
|
|
|
|
|
|
|
|
df::item_type avoid_type = item_type::NONE;
|
|
|
|
df::item_type avoid_type = item_type::NONE;
|
|
|
|
int avoid_glass = 0;
|
|
|
|
int avoid_glass = 0;
|
|
|
|
switch (skill)
|
|
|
|
switch (skill)
|
|
|
@ -1137,9 +1138,9 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (size_t i = 0; i < extra_items; i++)
|
|
|
|
for (size_t i = 0; i < extra_items; i++)
|
|
|
|
{
|
|
|
|
{
|
|
|
|
if ((job->job_type == job_type::StrangeMoodBrooding) && (rand() % 2))
|
|
|
|
if ((job->job_type == job_type::StrangeMoodBrooding) && (rng.df_trandom(2)))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
job->job_items.push_back(item = new df::job_item());
|
|
|
|
job->job_items.push_back(item = new df::job_item());
|
|
|
@ -1170,7 +1171,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
item_type = item_type::NONE;
|
|
|
|
item_type = item_type::NONE;
|
|
|
|
mat_type = -1;
|
|
|
|
mat_type = -1;
|
|
|
|
flags2.whole = 0;
|
|
|
|
flags2.whole = 0;
|
|
|
|
switch (rand() % 10)
|
|
|
|
switch (rng.df_trandom(10))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
item_type = item_type::WOOD;
|
|
|
|
item_type = item_type::WOOD;
|
|
|
@ -1207,7 +1208,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
case 8:
|
|
|
|
case 8:
|
|
|
|
item_type = item_type::CLOTH;
|
|
|
|
item_type = item_type::CLOTH;
|
|
|
|
mat_type = -1;
|
|
|
|
mat_type = -1;
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
flags2.bits.plant = true;
|
|
|
|
flags2.bits.plant = true;
|
|
|
@ -1222,7 +1223,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
break;
|
|
|
|
break;
|
|
|
|
case 9:
|
|
|
|
case 9:
|
|
|
|
item_type = item_type::ROUGH;
|
|
|
|
item_type = item_type::ROUGH;
|
|
|
|
switch (rand() % 3)
|
|
|
|
switch (rng.df_trandom(3))
|
|
|
|
{
|
|
|
|
{
|
|
|
|
case 0:
|
|
|
|
case 0:
|
|
|
|
mat_type = builtin_mats::GLASS_GREEN;
|
|
|
|
mat_type = builtin_mats::GLASS_GREEN;
|
|
|
@ -1281,7 +1282,7 @@ command_result df_strangemood (color_ostream &out, vector <string> & parameters)
|
|
|
|
else
|
|
|
|
else
|
|
|
|
{
|
|
|
|
{
|
|
|
|
generateName(unit->status.artifact_name, unit->name.language, 1, world->raws.language.word_table[0][1], world->raws.language.word_table[1][1]);
|
|
|
|
generateName(unit->status.artifact_name, unit->name.language, 1, world->raws.language.word_table[0][1], world->raws.language.word_table[1][1]);
|
|
|
|
if (!(rand() % 100))
|
|
|
|
if (!rng.df_trandom(100))
|
|
|
|
unit->status.artifact_name = unit->name;
|
|
|
|
unit->status.artifact_name = unit->name;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
unit->unk_18e = 0;
|
|
|
|
unit->unk_18e = 0;
|
|
|
@ -1302,6 +1303,7 @@ DFhackCExport command_result plugin_init (color_ostream &out, std::vector<Plugin
|
|
|
|
" Skill name must be lowercase and without spaces.\n"
|
|
|
|
" Skill name must be lowercase and without spaces.\n"
|
|
|
|
" Example: miner, gemcutter, metalcrafter, bonecarver, mason\n"
|
|
|
|
" Example: miner, gemcutter, metalcrafter, bonecarver, mason\n"
|
|
|
|
));
|
|
|
|
));
|
|
|
|
|
|
|
|
rng.init();
|
|
|
|
|
|
|
|
|
|
|
|
return CR_OK;
|
|
|
|
return CR_OK;
|
|
|
|
}
|
|
|
|
}
|
|
|
|