mirror of
https://github.com/Mikaela/Limnoria.git
synced 2024-11-29 06:09:23 +01:00
callbacks: Make snarfers' output paginatable with @more
by moving the 'smart' reply() method from NestedCommandsIrcProxy to ReplyIrcProxy. There is no reason only commands should have a paginated output and not snarfers defined in PluginRegexp.
This commit is contained in:
parent
24ca278b93
commit
3c1c4a69e9
@ -29,6 +29,7 @@
|
||||
###
|
||||
|
||||
import os
|
||||
import copy
|
||||
import json
|
||||
import functools
|
||||
import contextlib
|
||||
@ -50,6 +51,7 @@ from .test_data import (
|
||||
OUTBOX_DATA,
|
||||
STATUS_URL,
|
||||
STATUS_DATA,
|
||||
STATUS_VALUE,
|
||||
STATUS_WITH_PHOTO_URL,
|
||||
STATUS_WITH_PHOTO_DATA,
|
||||
OUTBOX_FIRSTPAGE_URL,
|
||||
@ -450,6 +452,36 @@ class NetworklessFediverseTestCase(BaseFediverseTestCase):
|
||||
+ "@ FirstAuthor I am replying to you",
|
||||
)
|
||||
|
||||
def testStatusUrlSnarferMore(self):
|
||||
# Actually this is a test for src/callbacks.py, but it's easier to
|
||||
# stick it here.
|
||||
status_value = copy.deepcopy(STATUS_VALUE)
|
||||
str_long = "l" + ("o" * 400) + "ng"
|
||||
str_message = "mess" + ("a" * 400) + "ge"
|
||||
status_obj = status_value["object"]
|
||||
status_obj["content"] = status_obj["content"].replace(
|
||||
"to you", "to you with a " + str_long + " " + str_message
|
||||
)
|
||||
with conf.supybot.plugins.Fediverse.snarfers.status.context(True):
|
||||
expected_requests = [
|
||||
(STATUS_URL, json.dumps(status_value).encode()),
|
||||
(ACTOR_URL, ACTOR_DATA),
|
||||
]
|
||||
|
||||
with self.mockWebfingerSupport(True), self.mockRequests(
|
||||
expected_requests
|
||||
):
|
||||
self.assertSnarfResponse(
|
||||
"aaa https://example.org/users/someuser/statuses/1234 bbb",
|
||||
"\x02someuser\x02 (@someuser@example.org): "
|
||||
+ "@ FirstAuthor I am replying to you with a "
|
||||
+ " \x02(2 more messages)\x02",
|
||||
)
|
||||
|
||||
self.assertNoResponse(" ")
|
||||
self.assertResponse("more", str_long + " \x02(1 more message)\x02")
|
||||
self.assertResponse("more", str_message)
|
||||
|
||||
def testStatusUrlSnarferErrors(self):
|
||||
with conf.supybot.plugins.Fediverse.snarfers.status.context(True):
|
||||
expected_requests = [(STATUS_URL, utils.web.Error("blah"))]
|
||||
|
549
src/callbacks.py
549
src/callbacks.py
@ -627,6 +627,7 @@ class ReplyIrcProxy(RichReplyMethods):
|
||||
"""This class is a thin wrapper around an irclib.Irc object that gives it
|
||||
the reply() and error() methods (as well as everything in RichReplyMethods,
|
||||
based on those two)."""
|
||||
_mores = ircutils.IrcDict()
|
||||
def __init__(self, irc, msg):
|
||||
self.irc = irc
|
||||
self.msg = msg
|
||||
@ -660,24 +661,295 @@ class ReplyIrcProxy(RichReplyMethods):
|
||||
self.irc.queueMsg(m)
|
||||
return m
|
||||
|
||||
def _defaultPrefixNick(self, msg):
|
||||
if msg.channel:
|
||||
return conf.get(conf.supybot.reply.withNickPrefix,
|
||||
channel=msg.channel, network=self.irc.network)
|
||||
else:
|
||||
return conf.supybot.reply.withNickPrefix()
|
||||
|
||||
def reply(self, s, msg=None, **kwargs):
|
||||
"""
|
||||
Keyword arguments:
|
||||
|
||||
:arg bool noLengthCheck:
|
||||
True if the length shouldn't be checked (used for 'more' handling)
|
||||
|
||||
:arg bool prefixNick:
|
||||
False if the nick shouldn't be prefixed to the reply.
|
||||
|
||||
:arg bool action:
|
||||
True if the reply should be an action.
|
||||
|
||||
:arg bool private:
|
||||
True if the reply should be in private.
|
||||
|
||||
:arg bool notice:
|
||||
True if the reply should be noticed when the bot is configured
|
||||
to do so.
|
||||
|
||||
:arg str to:
|
||||
The nick or channel the reply should go to.
|
||||
Defaults to msg.args[0] (or msg.nick if private)
|
||||
|
||||
:arg bool sendImmediately:
|
||||
True if the reply should use sendMsg() which
|
||||
bypasses conf.supybot.protocols.irc.throttleTime
|
||||
and gets sent before any queued messages
|
||||
"""
|
||||
if msg is None:
|
||||
msg = self.msg
|
||||
assert not isinstance(s, ircmsgs.IrcMsg), \
|
||||
'Old code alert: there is no longer a "msg" argument to reply.'
|
||||
kwargs.pop('noLengthCheck', None)
|
||||
m = _makeReply(self, msg, s, **kwargs)
|
||||
self.irc.queueMsg(m)
|
||||
return m
|
||||
if 'target' not in kwargs:
|
||||
target = kwargs.get('private', False) and kwargs.get('to', None) \
|
||||
or msg.args[0]
|
||||
if 'prefixNick' not in kwargs:
|
||||
kwargs['prefixNick'] = self._defaultPrefixNick(msg)
|
||||
self._sendReply(s, target=target, msg=msg, **kwargs)
|
||||
|
||||
def __getattr__(self, attr):
|
||||
return getattr(self.irc, attr)
|
||||
|
||||
def _replyOverhead(self, target, targetNick, prefixNick):
|
||||
"""Returns the number of bytes added to a PRIVMSG payload, either by
|
||||
Limnoria itself or by the server.
|
||||
Ignores tag bytes, as they are accounted for separatly."""
|
||||
overhead = (
|
||||
len(':')
|
||||
+ len(self.irc.prefix.encode())
|
||||
+ len(' PRIVMSG ')
|
||||
+ len(target.encode())
|
||||
+ len(' :')
|
||||
+ len('\r\n')
|
||||
)
|
||||
if prefixNick and targetNick is not None:
|
||||
overhead += len(targetNick) + len(': ')
|
||||
return overhead
|
||||
|
||||
def _sendReply(self, s, target, msg, sendImmediately=False,
|
||||
noLengthCheck=False, **kwargs):
|
||||
if sendImmediately:
|
||||
sendMsg = self.irc.sendMsg
|
||||
else:
|
||||
sendMsg = self.irc.queueMsg
|
||||
|
||||
if isinstance(self.irc, self.__class__):
|
||||
s = s[:conf.supybot.reply.maximumLength()]
|
||||
return self.irc.reply(s,
|
||||
noLengthCheck=noLengthCheck,
|
||||
**kwargs)
|
||||
elif noLengthCheck:
|
||||
# noLengthCheck only matters to NestedCommandsIrcProxy, so
|
||||
# it's not used here. Just in case you were wondering.
|
||||
m = _makeReply(self, msg, s, **kwargs)
|
||||
sendMsg(m)
|
||||
return m
|
||||
else:
|
||||
s = ircutils.safeArgument(s)
|
||||
allowedLength = conf.get(conf.supybot.reply.mores.length,
|
||||
channel=target, network=self.irc.network)
|
||||
if not allowedLength: # 0 indicates this.
|
||||
allowedLength = 512 - self._replyOverhead(
|
||||
target, msg.nick, prefixNick=kwargs['prefixNick'])
|
||||
maximumMores = conf.get(conf.supybot.reply.mores.maximum,
|
||||
channel=target, network=self.irc.network)
|
||||
maximumLength = allowedLength * maximumMores
|
||||
if len(s) > maximumLength:
|
||||
log.warning('Truncating to %s bytes from %s bytes.',
|
||||
maximumLength, len(s))
|
||||
s = s[:maximumLength]
|
||||
s_size = len(s.encode()) if minisix.PY3 else len(s)
|
||||
if s_size <= allowedLength or \
|
||||
not conf.get(conf.supybot.reply.mores,
|
||||
channel=target, network=self.irc.network):
|
||||
# There's no need for action=self.action here because
|
||||
# action implies noLengthCheck, which has already been
|
||||
# handled. Let's stick an assert in here just in case.
|
||||
assert not kwargs.get('action')
|
||||
m = _makeReply(self, msg, s, **kwargs)
|
||||
sendMsg(m)
|
||||
return m
|
||||
# The '(XX more messages)' may have not the same
|
||||
# length in the current locale
|
||||
allowedLength -= len(_('(XX more messages)')) + 1 # bold
|
||||
chunks = ircutils.wrap(s, allowedLength)
|
||||
|
||||
# Last messages to display at the beginning of the list
|
||||
# (which is used like a stack)
|
||||
chunks.reverse()
|
||||
|
||||
instant = conf.get(conf.supybot.reply.mores.instant,
|
||||
channel=target, network=self.irc.network)
|
||||
|
||||
# Big complex loop ahead, with lots of cases and opportunities for
|
||||
# off-by-one errors. Here is the meaning of each of the variables
|
||||
#
|
||||
# * 'i' is the number of chunks after the current one
|
||||
#
|
||||
# * 'is_first' is True when the message is the very first message
|
||||
# (so last iteration of the loop)
|
||||
#
|
||||
# * 'is_last' is True when the message is the very last (so first
|
||||
# iteration of the loop)
|
||||
#
|
||||
# * 'is_instant' is True when the message is in one of the messages
|
||||
# sent immediately when the command is called, ie. without
|
||||
# calling @misc more. (when supybot.reply.mores.instant is 1,
|
||||
# which is the default, this is equivalent to 'is_first')
|
||||
#
|
||||
# * 'is_last_instant' is True when the message is the last of the
|
||||
# instant message (so the first iteration of the loop with an
|
||||
# instant message).
|
||||
#
|
||||
# We need all this complexity because pagination is hard, and we
|
||||
# want:
|
||||
#
|
||||
# * the '(XX more messages)' suffix on the last instant message,
|
||||
# and every other message (mandatory, it's a great feature),
|
||||
# but not on the other instant messages (mandatory when
|
||||
# multiline is enabled, but very nice to have in general)
|
||||
# * the nick prefix on the first message and every other message
|
||||
# that isn't instant (mandatory), but not on the other instant
|
||||
# messages (also mandatory only when multiline is enabled)
|
||||
msgs = []
|
||||
for (i, chunk) in enumerate(chunks):
|
||||
is_first = i == len(chunks) - 1
|
||||
is_last = i == 0
|
||||
is_instant = len(chunks) - i <= instant
|
||||
is_last_instant = len(chunks) - i == instant
|
||||
if is_last:
|
||||
# last message, no suffix to add
|
||||
pass
|
||||
elif is_instant and not is_last_instant:
|
||||
# one of the first messages, and the next one will
|
||||
# also be sent immediately, so no suffix
|
||||
pass
|
||||
else:
|
||||
if i == 1:
|
||||
more = _('more message')
|
||||
else:
|
||||
more = _('more messages')
|
||||
n = ircutils.bold('(%i %s)' % (len(msgs), more))
|
||||
chunk = '%s %s' % (chunk, n)
|
||||
|
||||
if is_instant and not is_first:
|
||||
d = kwargs.copy()
|
||||
d['prefixNick'] = False
|
||||
msgs.append(_makeReply(self, msg, chunk, **d))
|
||||
else:
|
||||
msgs.append(_makeReply(self, msg, chunk, **kwargs))
|
||||
|
||||
instant_messages = []
|
||||
|
||||
while instant > 0 and msgs:
|
||||
instant -= 1
|
||||
response = msgs.pop()
|
||||
instant_messages.append(response)
|
||||
# XXX We should somehow allow these to be returned, but
|
||||
# until someone complains, we'll be fine :) We
|
||||
# can't return from here, though, for obvious
|
||||
# reasons.
|
||||
# return m
|
||||
|
||||
if conf.supybot.protocols.irc.experimentalExtensions() \
|
||||
and 'draft/multiline' in self.state.capabilities_ack \
|
||||
and len(instant_messages) > 1:
|
||||
# More than one message to send now, and we are allowed to use
|
||||
# multiline batches, so let's do it
|
||||
self.queueMultilineBatches(
|
||||
instant_messages, target, msg.nick, concat=True,
|
||||
allowedLength=allowedLength, sendImmediately=sendImmediately)
|
||||
else:
|
||||
for instant_msg in instant_messages:
|
||||
sendMsg(instant_msg)
|
||||
|
||||
if not msgs:
|
||||
return
|
||||
prefix = msg.prefix
|
||||
if target and ircutils.isNick(target):
|
||||
try:
|
||||
state = self.getRealIrc().state
|
||||
prefix = state.nickToHostmask(target)
|
||||
except KeyError:
|
||||
pass # We'll leave it as it is.
|
||||
mask = prefix.split('!', 1)[1]
|
||||
self._mores[mask] = msgs
|
||||
public = bool(self.msg.channel)
|
||||
private = kwargs.get('private', False) or not public
|
||||
self._mores[msg.nick] = (private, msgs)
|
||||
return response
|
||||
|
||||
def queueMultilineBatches(self, msgs, target, targetNick, concat,
|
||||
allowedLength=0, sendImmediately=False):
|
||||
"""Queues the msgs passed as argument in batches using draft/multiline
|
||||
batches.
|
||||
|
||||
This errors if experimentalExtensions is disabled or draft/multiline
|
||||
was not negotiated."""
|
||||
assert conf.supybot.protocols.irc.experimentalExtensions()
|
||||
assert 'draft/multiline' in self.state.capabilities_ack
|
||||
|
||||
if not allowedLength: # 0 indicates this.
|
||||
# We're only interested in the overhead outside the payload,
|
||||
# regardless of the entire payload (nick prefix included),
|
||||
# so prefixNick=False
|
||||
allowedLength = 512 - self._replyOverhead(
|
||||
target, targetNick, prefixNick=False)
|
||||
|
||||
multiline_cap_values = ircutils.parseCapabilityKeyValue(
|
||||
self.state.capabilities_ls['draft/multiline'])
|
||||
|
||||
# All the messages in instant_messages are to be sent
|
||||
# immediately, in multiline batches.
|
||||
max_bytes_per_batch = int(multiline_cap_values['max-bytes'])
|
||||
|
||||
# We have to honor max_bytes_per_batch, but I don't want to
|
||||
# encode messages again here just to have their length, so
|
||||
# let's assume they all have the maximum length.
|
||||
# It's not optimal, but close enough and simplifies the code.
|
||||
messages_per_batch = max_bytes_per_batch // allowedLength
|
||||
|
||||
# "Clients MUST NOT send tags other than draft/multiline-concat and
|
||||
# batch on messages within the batch. In particular, all client-only
|
||||
# tags associated with the message must be sent attached to the initial
|
||||
# BATCH command."
|
||||
# -- <https://ircv3.net/specs/extensions/multiline>
|
||||
# So we copy the tags of the first message, discard the tags of all
|
||||
# other messages, and apply the tags to the opening BATCH
|
||||
server_tags = msgs[0].server_tags
|
||||
|
||||
for batch_msgs in utils.iter.grouper(msgs, messages_per_batch):
|
||||
# TODO: should use sendBatch instead of queueBatch if
|
||||
# sendImmediately is True
|
||||
batch_name = ircutils.makeLabel()
|
||||
batch = []
|
||||
batch.append(ircmsgs.IrcMsg(command='BATCH',
|
||||
args=('+' + batch_name, 'draft/multiline', target),
|
||||
server_tags=server_tags))
|
||||
|
||||
for (i, batch_msg) in enumerate(batch_msgs):
|
||||
if batch_msg is None:
|
||||
continue # 'grouper' generates None at the end
|
||||
assert 'batch' not in batch_msg.server_tags
|
||||
|
||||
# Discard the existing tags, and add the batch ones.
|
||||
batch_msg.server_tags = {'batch': batch_name}
|
||||
if concat and i > 0:
|
||||
# Tell clients not to add a newline after this
|
||||
batch_msg.server_tags['draft/multiline-concat'] = None
|
||||
batch.append(batch_msg)
|
||||
|
||||
batch.append(ircmsgs.IrcMsg(
|
||||
command='BATCH', args=('-' + batch_name,)))
|
||||
|
||||
self.queueBatch(batch)
|
||||
|
||||
SimpleProxy = ReplyIrcProxy # Backwards-compatibility
|
||||
|
||||
class NestedCommandsIrcProxy(ReplyIrcProxy):
|
||||
"A proxy object to allow proper nesting of commands (even threaded ones)."
|
||||
_mores = ircutils.IrcDict()
|
||||
def __init__(self, irc, msg, args, nested=0):
|
||||
assert isinstance(args, list), 'Args should be a list, not a string.'
|
||||
super(NestedCommandsIrcProxy, self).__init__(irc, msg)
|
||||
@ -721,11 +993,7 @@ class NestedCommandsIrcProxy(ReplyIrcProxy):
|
||||
self.notice = None
|
||||
self.private = None
|
||||
self.noLengthCheck = None
|
||||
if self.msg.channel:
|
||||
self.prefixNick = conf.get(conf.supybot.reply.withNickPrefix,
|
||||
channel=self.msg.channel, network=self.irc.network)
|
||||
else:
|
||||
self.prefixNick = conf.supybot.reply.withNickPrefix()
|
||||
self.prefixNick = self._defaultPrefixNick(self.msg)
|
||||
|
||||
def evalArgs(self, withClass=None):
|
||||
while self.counter < len(self.args):
|
||||
@ -900,34 +1168,6 @@ class NestedCommandsIrcProxy(ReplyIrcProxy):
|
||||
def reply(self, s, noLengthCheck=False, prefixNick=None, action=None,
|
||||
private=None, notice=None, to=None, msg=None,
|
||||
sendImmediately=False, stripCtcp=True):
|
||||
"""
|
||||
Keyword arguments:
|
||||
|
||||
:arg bool noLengthCheck:
|
||||
True if the length shouldn't be checked (used for 'more' handling)
|
||||
|
||||
:arg bool prefixNick:
|
||||
False if the nick shouldn't be prefixed to the reply.
|
||||
|
||||
:arg bool action:
|
||||
True if the reply should be an action.
|
||||
|
||||
:arg bool private:
|
||||
True if the reply should be in private.
|
||||
|
||||
:arg bool notice:
|
||||
True if the reply should be noticed when the bot is configured
|
||||
to do so.
|
||||
|
||||
:arg str to:
|
||||
The nick or channel the reply should go to.
|
||||
Defaults to msg.args[0] (or msg.nick if private)
|
||||
|
||||
:arg bool sendImmediately:
|
||||
True if the reply should use sendMsg() which
|
||||
bypasses conf.supybot.protocols.irc.throttleTime
|
||||
and gets sent before any queued messages
|
||||
"""
|
||||
# These use and or or based on whether or not they default to True or
|
||||
# False. Those that default to True use and; those that default to
|
||||
# False use or.
|
||||
@ -981,241 +1221,6 @@ class NestedCommandsIrcProxy(ReplyIrcProxy):
|
||||
self.args[self.counter] = s
|
||||
self.evalArgs()
|
||||
|
||||
def _replyOverhead(self, target, targetNick, prefixNick):
|
||||
"""Returns the number of bytes added to a PRIVMSG payload, either by
|
||||
Limnoria itself or by the server.
|
||||
Ignores tag bytes, as they are accounted for separatly."""
|
||||
overhead = (
|
||||
len(':')
|
||||
+ len(self.irc.prefix.encode())
|
||||
+ len(' PRIVMSG ')
|
||||
+ len(target.encode())
|
||||
+ len(' :')
|
||||
+ len('\r\n')
|
||||
)
|
||||
if prefixNick and targetNick is not None:
|
||||
overhead += len(targetNick) + len(': ')
|
||||
return overhead
|
||||
|
||||
def _sendReply(self, s, target, msg, sendImmediately,
|
||||
noLengthCheck, **kwargs):
|
||||
if sendImmediately:
|
||||
sendMsg = self.irc.sendMsg
|
||||
else:
|
||||
sendMsg = self.irc.queueMsg
|
||||
|
||||
if isinstance(self.irc, self.__class__):
|
||||
s = s[:conf.supybot.reply.maximumLength()]
|
||||
return self.irc.reply(s,
|
||||
noLengthCheck=self.noLengthCheck,
|
||||
**kwargs)
|
||||
elif noLengthCheck:
|
||||
# noLengthCheck only matters to NestedCommandsIrcProxy, so
|
||||
# it's not used here. Just in case you were wondering.
|
||||
m = _makeReply(self, msg, s, **kwargs)
|
||||
sendMsg(m)
|
||||
return m
|
||||
else:
|
||||
s = ircutils.safeArgument(s)
|
||||
allowedLength = conf.get(conf.supybot.reply.mores.length,
|
||||
channel=target, network=self.irc.network)
|
||||
if not allowedLength: # 0 indicates this.
|
||||
allowedLength = 512 - self._replyOverhead(
|
||||
target, msg.nick, prefixNick=kwargs['prefixNick'])
|
||||
maximumMores = conf.get(conf.supybot.reply.mores.maximum,
|
||||
channel=target, network=self.irc.network)
|
||||
maximumLength = allowedLength * maximumMores
|
||||
if len(s) > maximumLength:
|
||||
log.warning('Truncating to %s bytes from %s bytes.',
|
||||
maximumLength, len(s))
|
||||
s = s[:maximumLength]
|
||||
s_size = len(s.encode()) if minisix.PY3 else len(s)
|
||||
if s_size <= allowedLength or \
|
||||
not conf.get(conf.supybot.reply.mores,
|
||||
channel=target, network=self.irc.network):
|
||||
# There's no need for action=self.action here because
|
||||
# action implies noLengthCheck, which has already been
|
||||
# handled. Let's stick an assert in here just in case.
|
||||
assert not kwargs.get('action')
|
||||
m = _makeReply(self, msg, s, **kwargs)
|
||||
sendMsg(m)
|
||||
return m
|
||||
# The '(XX more messages)' may have not the same
|
||||
# length in the current locale
|
||||
allowedLength -= len(_('(XX more messages)')) + 1 # bold
|
||||
chunks = ircutils.wrap(s, allowedLength)
|
||||
|
||||
# Last messages to display at the beginning of the list
|
||||
# (which is used like a stack)
|
||||
chunks.reverse()
|
||||
|
||||
instant = conf.get(conf.supybot.reply.mores.instant,
|
||||
channel=target, network=self.irc.network)
|
||||
|
||||
# Big complex loop ahead, with lots of cases and opportunities for
|
||||
# off-by-one errors. Here is the meaning of each of the variables
|
||||
#
|
||||
# * 'i' is the number of chunks after the current one
|
||||
#
|
||||
# * 'is_first' is True when the message is the very first message
|
||||
# (so last iteration of the loop)
|
||||
#
|
||||
# * 'is_last' is True when the message is the very last (so first
|
||||
# iteration of the loop)
|
||||
#
|
||||
# * 'is_instant' is True when the message is in one of the messages
|
||||
# sent immediately when the command is called, ie. without
|
||||
# calling @misc more. (when supybot.reply.mores.instant is 1,
|
||||
# which is the default, this is equivalent to 'is_first')
|
||||
#
|
||||
# * 'is_last_instant' is True when the message is the last of the
|
||||
# instant message (so the first iteration of the loop with an
|
||||
# instant message).
|
||||
#
|
||||
# We need all this complexity because pagination is hard, and we
|
||||
# want:
|
||||
#
|
||||
# * the '(XX more messages)' suffix on the last instant message,
|
||||
# and every other message (mandatory, it's a great feature),
|
||||
# but not on the other instant messages (mandatory when
|
||||
# multiline is enabled, but very nice to have in general)
|
||||
# * the nick prefix on the first message and every other message
|
||||
# that isn't instant (mandatory), but not on the other instant
|
||||
# messages (also mandatory only when multiline is enabled)
|
||||
msgs = []
|
||||
for (i, chunk) in enumerate(chunks):
|
||||
is_first = i == len(chunks) - 1
|
||||
is_last = i == 0
|
||||
is_instant = len(chunks) - i <= instant
|
||||
is_last_instant = len(chunks) - i == instant
|
||||
if is_last:
|
||||
# last message, no suffix to add
|
||||
pass
|
||||
elif is_instant and not is_last_instant:
|
||||
# one of the first messages, and the next one will
|
||||
# also be sent immediately, so no suffix
|
||||
pass
|
||||
else:
|
||||
if i == 1:
|
||||
more = _('more message')
|
||||
else:
|
||||
more = _('more messages')
|
||||
n = ircutils.bold('(%i %s)' % (len(msgs), more))
|
||||
chunk = '%s %s' % (chunk, n)
|
||||
|
||||
if is_instant and not is_first:
|
||||
d = kwargs.copy()
|
||||
d['prefixNick'] = False
|
||||
msgs.append(_makeReply(self, msg, chunk, **d))
|
||||
else:
|
||||
msgs.append(_makeReply(self, msg, chunk, **kwargs))
|
||||
|
||||
instant_messages = []
|
||||
|
||||
while instant > 0 and msgs:
|
||||
instant -= 1
|
||||
response = msgs.pop()
|
||||
instant_messages.append(response)
|
||||
# XXX We should somehow allow these to be returned, but
|
||||
# until someone complains, we'll be fine :) We
|
||||
# can't return from here, though, for obvious
|
||||
# reasons.
|
||||
# return m
|
||||
|
||||
if conf.supybot.protocols.irc.experimentalExtensions() \
|
||||
and 'draft/multiline' in self.state.capabilities_ack \
|
||||
and len(instant_messages) > 1:
|
||||
# More than one message to send now, and we are allowed to use
|
||||
# multiline batches, so let's do it
|
||||
self.queueMultilineBatches(
|
||||
instant_messages, target, msg.nick, concat=True,
|
||||
allowedLength=allowedLength, sendImmediately=sendImmediately)
|
||||
else:
|
||||
for instant_msg in instant_messages:
|
||||
sendMsg(instant_msg)
|
||||
|
||||
if not msgs:
|
||||
return
|
||||
prefix = msg.prefix
|
||||
to = kwargs['to']
|
||||
if to and ircutils.isNick(to):
|
||||
try:
|
||||
state = self.getRealIrc().state
|
||||
prefix = state.nickToHostmask(to)
|
||||
except KeyError:
|
||||
pass # We'll leave it as it is.
|
||||
mask = prefix.split('!', 1)[1]
|
||||
self._mores[mask] = msgs
|
||||
public = bool(self.msg.channel)
|
||||
private = kwargs['private'] or not public
|
||||
self._mores[msg.nick] = (private, msgs)
|
||||
return response
|
||||
|
||||
def queueMultilineBatches(self, msgs, target, targetNick, concat,
|
||||
allowedLength=0, sendImmediately=False):
|
||||
"""Queues the msgs passed as argument in batches using draft/multiline
|
||||
batches.
|
||||
|
||||
This errors if experimentalExtensions is disabled or draft/multiline
|
||||
was not negotiated."""
|
||||
assert conf.supybot.protocols.irc.experimentalExtensions()
|
||||
assert 'draft/multiline' in self.state.capabilities_ack
|
||||
|
||||
if not allowedLength: # 0 indicates this.
|
||||
# We're only interested in the overhead outside the payload,
|
||||
# regardless of the entire payload (nick prefix included),
|
||||
# so prefixNick=False
|
||||
allowedLength = 512 - self._replyOverhead(
|
||||
target, targetNick, prefixNick=False)
|
||||
|
||||
multiline_cap_values = ircutils.parseCapabilityKeyValue(
|
||||
self.state.capabilities_ls['draft/multiline'])
|
||||
|
||||
# All the messages in instant_messages are to be sent
|
||||
# immediately, in multiline batches.
|
||||
max_bytes_per_batch = int(multiline_cap_values['max-bytes'])
|
||||
|
||||
# We have to honor max_bytes_per_batch, but I don't want to
|
||||
# encode messages again here just to have their length, so
|
||||
# let's assume they all have the maximum length.
|
||||
# It's not optimal, but close enough and simplifies the code.
|
||||
messages_per_batch = max_bytes_per_batch // allowedLength
|
||||
|
||||
# "Clients MUST NOT send tags other than draft/multiline-concat and
|
||||
# batch on messages within the batch. In particular, all client-only
|
||||
# tags associated with the message must be sent attached to the initial
|
||||
# BATCH command."
|
||||
# -- <https://ircv3.net/specs/extensions/multiline>
|
||||
# So we copy the tags of the first message, discard the tags of all
|
||||
# other messages, and apply the tags to the opening BATCH
|
||||
server_tags = msgs[0].server_tags
|
||||
|
||||
for batch_msgs in utils.iter.grouper(msgs, messages_per_batch):
|
||||
# TODO: should use sendBatch instead of queueBatch if
|
||||
# sendImmediately is True
|
||||
batch_name = ircutils.makeLabel()
|
||||
batch = []
|
||||
batch.append(ircmsgs.IrcMsg(command='BATCH',
|
||||
args=('+' + batch_name, 'draft/multiline', target),
|
||||
server_tags=server_tags))
|
||||
|
||||
for (i, batch_msg) in enumerate(batch_msgs):
|
||||
if batch_msg is None:
|
||||
continue # 'grouper' generates None at the end
|
||||
assert 'batch' not in batch_msg.server_tags
|
||||
|
||||
# Discard the existing tags, and add the batch ones.
|
||||
batch_msg.server_tags = {'batch': batch_name}
|
||||
if concat and i > 0:
|
||||
# Tell clients not to add a newline after this
|
||||
batch_msg.server_tags['draft/multiline-concat'] = None
|
||||
batch.append(batch_msg)
|
||||
|
||||
batch.append(ircmsgs.IrcMsg(
|
||||
command='BATCH', args=('-' + batch_name,)))
|
||||
|
||||
self.queueBatch(batch)
|
||||
|
||||
def noReply(self, msg=None):
|
||||
if msg is None:
|
||||
msg = self.msg
|
||||
|
Loading…
Reference in New Issue
Block a user