mirror of
https://github.com/LmeSzinc/StarRailCopilot.git
synced 2024-11-25 10:01:10 +00:00
155 lines
5.4 KiB
Python
155 lines
5.4 KiB
Python
from datetime import timedelta
|
|
|
|
from module.base.base import ModuleBase
|
|
from module.base.timer import Timer
|
|
from module.base.utils import crop
|
|
from module.config.stored.classes import now
|
|
from module.config.utils import DEFAULT_TIME
|
|
from module.logger import logger
|
|
from module.ocr.ocr import DigitCounter
|
|
from tasks.base.ui import UI
|
|
from tasks.dungeon.assets.assets_dungeon_state import OCR_SIMUNI_POINT, OCR_SIMUNI_POINT_OFFSET, OCR_STAMINA
|
|
from tasks.dungeon.keywords import DungeonList
|
|
|
|
|
|
class OcrSimUniPoint(DigitCounter):
|
|
def after_process(self, result):
|
|
result = super().after_process(result)
|
|
result = result.replace('O', '0').replace('o', '0')
|
|
return result
|
|
|
|
|
|
class DungeonState(UI):
|
|
def dungeon_get_simuni_point(self, image=None) -> int:
|
|
"""
|
|
Page:
|
|
in: page_guide, Survival_Index, Simulated_Universe
|
|
"""
|
|
logger.info('Get simulated universe points')
|
|
if image is None:
|
|
image = self.device.image
|
|
|
|
_ = OCR_SIMUNI_POINT_OFFSET.match_template(image)
|
|
OCR_SIMUNI_POINT.load_offset(OCR_SIMUNI_POINT_OFFSET)
|
|
area = (
|
|
OCR_SIMUNI_POINT.area[0],
|
|
OCR_SIMUNI_POINT.button[1],
|
|
OCR_SIMUNI_POINT.area[2],
|
|
OCR_SIMUNI_POINT.button[3],
|
|
)
|
|
|
|
ocr = OcrSimUniPoint(OCR_SIMUNI_POINT)
|
|
value, _, total = ocr.ocr_single_line(crop(image, area), direct_ocr=True)
|
|
if total and value <= total:
|
|
logger.attr('SimulatedUniverse', f'{value}/{total}')
|
|
self.config.stored.SimulatedUniverse.set(value, total)
|
|
return value
|
|
else:
|
|
logger.warning(f'Invalid SimulatedUniverse points: {value}/{total}')
|
|
return 0
|
|
|
|
def dungeon_update_stamina(self, image=None, skip_first_screenshot=True):
|
|
"""
|
|
Returns:
|
|
bool: If success
|
|
|
|
Pages:
|
|
in: page_guild, Survival_Index, Simulated_Universe
|
|
or page_rogue, LEVEL_CONFIRM
|
|
or rogue, REWARD_CLOSE
|
|
"""
|
|
ocr = DigitCounter(OCR_STAMINA)
|
|
timeout = Timer(1, count=2).start()
|
|
if image is None:
|
|
image = self.device.image
|
|
use_cached_image = False
|
|
else:
|
|
skip_first_screenshot = True
|
|
use_cached_image = True
|
|
|
|
while 1:
|
|
if skip_first_screenshot:
|
|
skip_first_screenshot = False
|
|
else:
|
|
self.device.screenshot()
|
|
image = self.device.image
|
|
|
|
stamina = (0, 0, 0)
|
|
immersifier = (0, 0, 0)
|
|
|
|
if timeout.reached():
|
|
logger.warning('dungeon_update_stamina() timeout')
|
|
return False
|
|
|
|
for row in ocr.detect_and_ocr(image):
|
|
if row.ocr_text.isdigit():
|
|
continue
|
|
if row.ocr_text == '+':
|
|
continue
|
|
if not ocr.is_format_matched(row.ocr_text):
|
|
continue
|
|
data = ocr.format_result(row.ocr_text)
|
|
if data[2] == self.config.stored.TrailblazePower.FIXED_TOTAL:
|
|
stamina = data
|
|
if data[2] == self.config.stored.Immersifier.FIXED_TOTAL:
|
|
immersifier = data
|
|
|
|
if stamina[2] > 0 and immersifier[2] > 0:
|
|
break
|
|
if use_cached_image:
|
|
logger.info('dungeon_update_stamina() ended')
|
|
return
|
|
|
|
stamina = stamina[0]
|
|
immersifier = immersifier[0]
|
|
logger.attr('TrailblazePower', stamina)
|
|
logger.attr('Imersifier', immersifier)
|
|
with self.config.multi_set():
|
|
self.config.stored.TrailblazePower.value = stamina
|
|
self.config.stored.Immersifier.value = immersifier
|
|
return True
|
|
|
|
def dungeon_update_simuni(self):
|
|
"""
|
|
Update rogue weekly points, stamina, immersifier
|
|
Run in a new thread to be faster as data is not used immediately
|
|
|
|
Page:
|
|
in: page_guide, Survival_Index, Simulated_Universe
|
|
"""
|
|
logger.info('dungeon_update_simuni')
|
|
|
|
def func(image):
|
|
logger.info('Update thread start')
|
|
with self.config.multi_set():
|
|
self.dungeon_get_simuni_point(image)
|
|
self.dungeon_update_stamina(image)
|
|
|
|
ModuleBase.worker.submit(func, self.device.image)
|
|
|
|
def dungeon_stamina_delay(self, dungeon: DungeonList):
|
|
"""
|
|
Delay tasks that use stamina
|
|
"""
|
|
if dungeon.is_Simulated_Universe:
|
|
limit = 80
|
|
elif dungeon.is_Cavern_of_Corrosion:
|
|
limit = 80
|
|
elif dungeon.is_Echo_of_War:
|
|
limit = 30
|
|
else:
|
|
limit = 60
|
|
# Recover 1 trailbaze power each 6 minutes
|
|
current = self.config.stored.TrailblazePower.value
|
|
cover = max(limit - current, 0) * 6
|
|
future = now() + timedelta(minutes=cover)
|
|
logger.info(f'Currently has {current} need {cover} minutes to reach {limit}')
|
|
|
|
tasks = ['Dungeon', 'Weekly']
|
|
with self.config.multi_set():
|
|
for task in tasks:
|
|
next_run = self.config.cross_get(keys=f'{task}.Scheduler.NextRun', default=DEFAULT_TIME)
|
|
if future > next_run:
|
|
logger.info(f"Delay task `{task}` to {future}")
|
|
self.config.cross_set(keys=f'{task}.Scheduler.NextRun', value=future)
|