Add option to disable auto member chunking.
This commit is contained in:
@ -63,6 +63,7 @@ class ConnectionState:
|
||||
self.syncer = syncer
|
||||
self.is_bot = None
|
||||
self.shard_count = None
|
||||
self._fetch_offline = options.get('fetch_offline_members', True)
|
||||
self._listeners = []
|
||||
self.clear()
|
||||
|
||||
@ -197,16 +198,7 @@ class ConnectionState:
|
||||
yield self.receive_chunk(guild.id)
|
||||
|
||||
@asyncio.coroutine
|
||||
def _delay_ready(self):
|
||||
launch = self._ready_state.launch
|
||||
while not launch.is_set():
|
||||
# this snippet of code is basically waiting 2 seconds
|
||||
# until the last GUILD_CREATE was sent
|
||||
launch.set()
|
||||
yield from asyncio.sleep(2, loop=self.loop)
|
||||
|
||||
guilds = self._ready_state.guilds
|
||||
|
||||
def request_offline_members(self, guilds):
|
||||
# get all the chunks
|
||||
chunks = []
|
||||
for guild in guilds:
|
||||
@ -224,6 +216,22 @@ class ConnectionState:
|
||||
except asyncio.TimeoutError:
|
||||
log.info('Somehow timed out waiting for chunks.')
|
||||
|
||||
@asyncio.coroutine
|
||||
def _delay_ready(self):
|
||||
launch = self._ready_state.launch
|
||||
|
||||
# only real bots wait for GUILD_CREATE streaming
|
||||
if self.is_bot:
|
||||
while not launch.is_set():
|
||||
# this snippet of code is basically waiting 2 seconds
|
||||
# until the last GUILD_CREATE was sent
|
||||
launch.set()
|
||||
yield from asyncio.sleep(2, loop=self.loop)
|
||||
|
||||
guilds = self._ready_state.guilds
|
||||
if self._fetch_offline:
|
||||
yield from self.request_offline_members(guilds)
|
||||
|
||||
# remove the state
|
||||
try:
|
||||
del self._ready_state
|
||||
@ -260,6 +268,7 @@ class ConnectionState:
|
||||
factory, _ = _channel_factory(pm['type'])
|
||||
self._add_private_channel(factory(me=self.user, data=pm, state=self))
|
||||
|
||||
self.dispatch('connect')
|
||||
compat.create_task(self._delay_ready(), loop=self.loop)
|
||||
|
||||
def parse_resumed(self, data):
|
||||
@ -477,8 +486,8 @@ class ConnectionState:
|
||||
|
||||
@asyncio.coroutine
|
||||
def _chunk_and_dispatch(self, guild, unavailable):
|
||||
yield from self.chunker(guild)
|
||||
chunks = list(self.chunks_needed(guild))
|
||||
yield from self.chunker(guild)
|
||||
if chunks:
|
||||
try:
|
||||
yield from asyncio.wait(chunks, timeout=len(chunks), loop=self.loop)
|
||||
@ -518,9 +527,10 @@ class ConnectionState:
|
||||
return
|
||||
|
||||
# since we're not waiting for 'useful' READY we'll just
|
||||
# do the chunk request here
|
||||
compat.create_task(self._chunk_and_dispatch(guild, unavailable), loop=self.loop)
|
||||
return
|
||||
# do the chunk request here if wanted
|
||||
if self._fetch_offline:
|
||||
compat.create_task(self._chunk_and_dispatch(guild, unavailable), loop=self.loop)
|
||||
return
|
||||
|
||||
# Dispatch available if newly available
|
||||
if unavailable == False:
|
||||
@ -740,6 +750,25 @@ class AutoShardedConnectionState(ConnectionState):
|
||||
self._ready_state = ReadyState(launch=asyncio.Event(), guilds=[])
|
||||
self._ready_task = None
|
||||
|
||||
@asyncio.coroutine
|
||||
def request_offline_members(self, guilds, *, shard_id):
|
||||
# get all the chunks
|
||||
chunks = []
|
||||
for guild in guilds:
|
||||
chunks.extend(self.chunks_needed(guild))
|
||||
|
||||
# we only want to request ~75 guilds per chunk request.
|
||||
splits = [guilds[i:i + 75] for i in range(0, len(guilds), 75)]
|
||||
for split in splits:
|
||||
yield from self.chunker(split, shard_id=shard_id)
|
||||
|
||||
# wait for the chunks
|
||||
if chunks:
|
||||
try:
|
||||
yield from asyncio.wait(chunks, timeout=len(chunks) * 30.0, loop=self.loop)
|
||||
except asyncio.TimeoutError:
|
||||
log.info('Somehow timed out waiting for chunks.')
|
||||
|
||||
@asyncio.coroutine
|
||||
def _delay_ready(self):
|
||||
launch = self._ready_state.launch
|
||||
@ -749,30 +778,14 @@ class AutoShardedConnectionState(ConnectionState):
|
||||
launch.set()
|
||||
yield from asyncio.sleep(2.0 * self.shard_count, loop=self.loop)
|
||||
|
||||
guilds = sorted(self._ready_state.guilds, key=lambda g: g.shard_id)
|
||||
|
||||
# we only want to request ~75 guilds per chunk request.
|
||||
# we also want to split the chunks per shard_id
|
||||
for shard_id, sub_guilds in itertools.groupby(guilds, key=lambda g: g.shard_id):
|
||||
sub_guilds = list(sub_guilds)
|
||||
if self._fetch_offline:
|
||||
guilds = sorted(self._ready_state.guilds, key=lambda g: g.shard_id)
|
||||
|
||||
# split chunks by shard ID
|
||||
chunks = []
|
||||
for guild in sub_guilds:
|
||||
chunks.extend(self.chunks_needed(guild))
|
||||
|
||||
splits = [sub_guilds[i:i + 75] for i in range(0, len(sub_guilds), 75)]
|
||||
for split in splits:
|
||||
yield from self.chunker(split, shard_id=shard_id)
|
||||
|
||||
# wait for the chunks
|
||||
if chunks:
|
||||
try:
|
||||
yield from asyncio.wait(chunks, timeout=len(chunks) * 30.0, loop=self.loop)
|
||||
except asyncio.TimeoutError:
|
||||
log.info('Somehow timed out waiting for chunks for %s shard_id' % shard_id)
|
||||
|
||||
self.dispatch('shard_ready', shard_id)
|
||||
for shard_id, sub_guilds in itertools.groupby(guilds, key=lambda g: g.shard_id):
|
||||
sub_guilds = list(sub_guilds)
|
||||
yield from self.request_offline_members(sub_guilds, shard_id=shard_id)
|
||||
self.dispatch('shard_ready', shard_id)
|
||||
|
||||
# remove the state
|
||||
try:
|
||||
@ -782,6 +795,9 @@ class AutoShardedConnectionState(ConnectionState):
|
||||
|
||||
# regular users cannot shard so we won't worry about it here.
|
||||
|
||||
# clear the current task
|
||||
self._ready_task = None
|
||||
|
||||
# dispatch the event
|
||||
self.dispatch('ready')
|
||||
|
||||
@ -801,5 +817,6 @@ class AutoShardedConnectionState(ConnectionState):
|
||||
factory, _ = _channel_factory(pm['type'])
|
||||
self._add_private_channel(factory(me=self.user, data=pm, state=self))
|
||||
|
||||
self.dispatch('connect')
|
||||
if self._ready_task is None:
|
||||
self._ready_task = compat.create_task(self._delay_ready(), loop=self.loop)
|
||||
|
Reference in New Issue
Block a user