import re from typing import Any, Dict, Iterable, List, Optional, Tuple, Union import orjson from django.conf import settings from django.contrib.auth.models import AnonymousUser from django.core.exceptions import ValidationError from django.db import connection from django.http import HttpRequest, HttpResponse from django.utils.html import escape as escape_html from django.utils.translation import ugettext as _ from sqlalchemy import func from sqlalchemy.dialects import postgresql from sqlalchemy.sql import ( ColumnElement, Selectable, alias, and_, column, join, literal, literal_column, not_, or_, select, table, union_all, ) from zerver.context_processors import get_realm_from_request from zerver.decorator import REQ, has_request_variables from zerver.lib.actions import recipient_for_user_profiles from zerver.lib.addressee import get_user_profiles, get_user_profiles_by_ids from zerver.lib.exceptions import ErrorCode, JsonableError from zerver.lib.message import get_first_visible_message_id, messages_for_ids from zerver.lib.narrow import is_web_public_compatible, is_web_public_narrow from zerver.lib.response import json_error, json_success, json_unauthorized from zerver.lib.sqlalchemy_utils import get_sqlalchemy_connection from zerver.lib.streams import ( can_access_stream_history_by_id, can_access_stream_history_by_name, get_public_streams_queryset, get_stream_by_narrow_operand_access_unchecked, get_web_public_streams_queryset, ) from zerver.lib.topic import DB_TOPIC_NAME, MATCH_TOPIC, topic_column_sa, topic_match_sa from zerver.lib.topic_mutes import exclude_topic_mutes from zerver.lib.types import Validator from zerver.lib.utils import statsd from zerver.lib.validator import ( check_bool, check_dict, check_int, check_list, check_required_string, check_string, check_string_or_int, check_string_or_int_list, to_non_negative_int, ) from zerver.models import ( Realm, Recipient, Stream, Subscription, UserMessage, UserProfile, get_active_streams, get_user_by_id_in_realm_including_cross_realm, get_user_including_cross_realm, ) LARGER_THAN_MAX_MESSAGE_ID = 10000000000000000 MAX_MESSAGES_PER_FETCH = 5000 class BadNarrowOperator(JsonableError): code = ErrorCode.BAD_NARROW data_fields = ['desc'] def __init__(self, desc: str) -> None: self.desc: str = desc @staticmethod def msg_format() -> str: return _('Invalid narrow operator: {desc}') # TODO: Should be Select, but sqlalchemy stubs are busted Query = Any # TODO: should be Callable[[ColumnElement], ColumnElement], but sqlalchemy stubs are busted ConditionTransform = Any OptionalNarrowListT = Optional[List[Dict[str, Any]]] # These delimiters will not appear in rendered messages or HTML-escaped topics. TS_START = "" TS_STOP = "" def ts_locs_array( config: ColumnElement, text: ColumnElement, tsquery: ColumnElement, ) -> ColumnElement: options = f"HighlightAll = TRUE, StartSel = {TS_START}, StopSel = {TS_STOP}" delimited = func.ts_headline(config, text, tsquery, options) parts = func.unnest(func.string_to_array(delimited, TS_START)).alias() part = column(parts.name) part_len = func.length(part) - len(TS_STOP) match_pos = func.sum(part_len).over(rows=(None, -1)) + len(TS_STOP) match_len = func.strpos(part, TS_STOP) - 1 return func.array( select([postgresql.array([match_pos, match_len])]) .select_from(parts) .offset(1) .as_scalar(), ) # When you add a new operator to this, also update zerver/lib/narrow.py class NarrowBuilder: ''' Build up a SQLAlchemy query to find messages matching a narrow. ''' # This class has an important security invariant: # # None of these methods ever *add* messages to a query's result. # # That is, the `add_term` method, and its helpers the `by_*` methods, # are passed a Query object representing a query for messages; they may # call some methods on it, and then they return a resulting Query # object. Things these methods may do to the queries they handle # include # * add conditions to filter out rows (i.e., messages), with `query.where` # * add columns for more information on the same message, with `query.column` # * add a join for more information on the same message # # Things they may not do include # * anything that would pull in additional rows, or information on # other messages. def __init__(self, user_profile: Optional[UserProfile], msg_id_column: str, realm: Realm, is_web_public_query: bool=False) -> None: self.user_profile = user_profile self.msg_id_column = msg_id_column self.realm = realm self.is_web_public_query = is_web_public_query def add_term(self, query: Query, term: Dict[str, Any]) -> Query: """ Extend the given query to one narrowed by the given term, and return the result. This method satisfies an important security property: the returned query never includes a message that the given query didn't. In particular, if the given query will only find messages that a given user can legitimately see, then so will the returned query. """ # To maintain the security property, we hold all the `by_*` # methods to the same criterion. See the class's block comment # for details. # We have to be careful here because we're letting users call a method # by name! The prefix 'by_' prevents it from colliding with builtin # Python __magic__ stuff. operator = term['operator'] operand = term['operand'] negated = term.get('negated', False) method_name = 'by_' + operator.replace('-', '_') method = getattr(self, method_name, None) if method is None: raise BadNarrowOperator('unknown operator ' + operator) if negated: maybe_negate = not_ else: maybe_negate = lambda cond: cond return method(query, operand, maybe_negate) def by_has(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: if operand not in ['attachment', 'image', 'link']: raise BadNarrowOperator("unknown 'has' operand " + operand) col_name = 'has_' + operand cond = column(col_name) return query.where(maybe_negate(cond)) def by_in(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: # This operator does not support is_web_public_query. assert not self.is_web_public_query assert self.user_profile is not None if operand == 'home': conditions = exclude_muting_conditions(self.user_profile, []) return query.where(and_(*conditions)) elif operand == 'all': return query raise BadNarrowOperator("unknown 'in' operand " + operand) def by_is(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: # This operator class does not support is_web_public_query. assert not self.is_web_public_query assert self.user_profile is not None if operand == 'private': cond = column("flags").op("&")(UserMessage.flags.is_private.mask) != 0 return query.where(maybe_negate(cond)) elif operand == 'starred': cond = column("flags").op("&")(UserMessage.flags.starred.mask) != 0 return query.where(maybe_negate(cond)) elif operand == 'unread': cond = column("flags").op("&")(UserMessage.flags.read.mask) == 0 return query.where(maybe_negate(cond)) elif operand == 'mentioned': cond1 = column("flags").op("&")(UserMessage.flags.mentioned.mask) != 0 cond2 = column("flags").op("&")(UserMessage.flags.wildcard_mentioned.mask) != 0 cond = or_(cond1, cond2) return query.where(maybe_negate(cond)) elif operand == 'alerted': cond = column("flags").op("&")(UserMessage.flags.has_alert_word.mask) != 0 return query.where(maybe_negate(cond)) raise BadNarrowOperator("unknown 'is' operand " + operand) _alphanum = frozenset( 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789') def _pg_re_escape(self, pattern: str) -> str: """ Escape user input to place in a regex Python's re.escape escapes unicode characters in a way which postgres fails on, '\u03bb' to '\\\u03bb'. This function will correctly escape them for postgres, '\u03bb' to '\\u03bb'. """ s = list(pattern) for i, c in enumerate(s): if c not in self._alphanum: if ord(c) >= 128: # convert the character to hex postgres regex will take # \uXXXX s[i] = f'\\u{ord(c):0>4x}' else: s[i] = '\\' + c return ''.join(s) def by_stream(self, query: Query, operand: Union[str, int], maybe_negate: ConditionTransform) -> Query: try: # Because you can see your own message history for # private streams you are no longer subscribed to, we # need get_stream_by_narrow_operand_access_unchecked here. stream = get_stream_by_narrow_operand_access_unchecked(operand, self.realm) if self.is_web_public_query and not stream.is_web_public: raise BadNarrowOperator('unknown web-public stream ' + str(operand)) except Stream.DoesNotExist: raise BadNarrowOperator('unknown stream ' + str(operand)) if self.realm.is_zephyr_mirror_realm: # MIT users expect narrowing to "social" to also show messages to # /^(un)*social(.d)*$/ (unsocial, ununsocial, social.d, ...). # In `ok_to_include_history`, we assume that a non-negated # `stream` term for a public stream will limit the query to # that specific stream. So it would be a bug to hit this # codepath after relying on this term there. But all streams in # a Zephyr realm are private, so that doesn't happen. assert(not stream.is_public()) m = re.search(r'^(?:un)*(.+?)(?:\.d)*$', stream.name, re.IGNORECASE) # Since the regex has a `.+` in it and "" is invalid as a # stream name, this will always match assert(m is not None) base_stream_name = m.group(1) matching_streams = get_active_streams(self.realm).filter( name__iregex=fr'^(un)*{self._pg_re_escape(base_stream_name)}(\.d)*$') recipient_ids = [matching_stream.recipient_id for matching_stream in matching_streams] cond = column("recipient_id").in_(recipient_ids) return query.where(maybe_negate(cond)) recipient = stream.recipient cond = column("recipient_id") == recipient.id return query.where(maybe_negate(cond)) def by_streams(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: if operand == 'public': # Get all both subscribed and non subscribed public streams # but exclude any private subscribed streams. recipient_queryset = get_public_streams_queryset(self.realm) elif operand == 'web-public': recipient_queryset = get_web_public_streams_queryset(self.realm) else: raise BadNarrowOperator('unknown streams operand ' + operand) recipient_ids = recipient_queryset.values_list("recipient_id", flat=True).order_by('id') cond = column("recipient_id").in_(recipient_ids) return query.where(maybe_negate(cond)) def by_topic(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: if self.realm.is_zephyr_mirror_realm: # MIT users expect narrowing to topic "foo" to also show messages to /^foo(.d)*$/ # (foo, foo.d, foo.d.d, etc) m = re.search(r'^(.*?)(?:\.d)*$', operand, re.IGNORECASE) # Since the regex has a `.*` in it, this will always match assert(m is not None) base_topic = m.group(1) # Additionally, MIT users expect the empty instance and # instance "personal" to be the same. if base_topic in ('', 'personal', '(instance "")'): cond = or_( topic_match_sa(""), topic_match_sa(".d"), topic_match_sa(".d.d"), topic_match_sa(".d.d.d"), topic_match_sa(".d.d.d.d"), topic_match_sa("personal"), topic_match_sa("personal.d"), topic_match_sa("personal.d.d"), topic_match_sa("personal.d.d.d"), topic_match_sa("personal.d.d.d.d"), topic_match_sa('(instance "")'), topic_match_sa('(instance "").d'), topic_match_sa('(instance "").d.d'), topic_match_sa('(instance "").d.d.d'), topic_match_sa('(instance "").d.d.d.d'), ) else: # We limit `.d` counts, since postgres has much better # query planning for this than they do for a regular # expression (which would sometimes table scan). cond = or_( topic_match_sa(base_topic), topic_match_sa(base_topic + ".d"), topic_match_sa(base_topic + ".d.d"), topic_match_sa(base_topic + ".d.d.d"), topic_match_sa(base_topic + ".d.d.d.d"), ) return query.where(maybe_negate(cond)) cond = topic_match_sa(operand) return query.where(maybe_negate(cond)) def by_sender(self, query: Query, operand: Union[str, int], maybe_negate: ConditionTransform) -> Query: try: if isinstance(operand, str): sender = get_user_including_cross_realm(operand, self.realm) else: sender = get_user_by_id_in_realm_including_cross_realm(operand, self.realm) except UserProfile.DoesNotExist: raise BadNarrowOperator('unknown user ' + str(operand)) cond = column("sender_id") == literal(sender.id) return query.where(maybe_negate(cond)) def by_near(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: return query def by_id(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: if not str(operand).isdigit(): raise BadNarrowOperator("Invalid message ID") cond = self.msg_id_column == literal(operand) return query.where(maybe_negate(cond)) def by_pm_with(self, query: Query, operand: Union[str, Iterable[int]], maybe_negate: ConditionTransform) -> Query: # This operator does not support is_web_public_query. assert not self.is_web_public_query assert self.user_profile is not None try: if isinstance(operand, str): email_list = operand.split(",") user_profiles = get_user_profiles( emails=email_list, realm=self.realm, ) else: """ This is where we handle passing a list of user IDs for the narrow, which is the preferred/cleaner API. """ user_profiles = get_user_profiles_by_ids( user_ids=operand, realm=self.realm, ) recipient = recipient_for_user_profiles(user_profiles=user_profiles, forwarded_mirror_message=False, forwarder_user_profile=None, sender=self.user_profile, allow_deactivated=True) except (JsonableError, ValidationError): raise BadNarrowOperator('unknown user in ' + str(operand)) # Group DM if recipient.type == Recipient.HUDDLE: cond = column("recipient_id") == recipient.id return query.where(maybe_negate(cond)) # 1:1 PM other_participant = None # Find if another person is in PM for user in user_profiles: if user.id != self.user_profile.id: other_participant = user # PM with another person if other_participant: # We need bidirectional messages PM with another person. # But Recipient.PERSONAL objects only encode the person who # received the message, and not the other participant in # the thread (the sender), we need to do a somewhat # complex query to get messages between these two users # with either of them as the sender. self_recipient_id = self.user_profile.recipient_id cond = or_(and_(column("sender_id") == other_participant.id, column("recipient_id") == self_recipient_id), and_(column("sender_id") == self.user_profile.id, column("recipient_id") == recipient.id)) return query.where(maybe_negate(cond)) # PM with self cond = and_(column("sender_id") == self.user_profile.id, column("recipient_id") == recipient.id) return query.where(maybe_negate(cond)) def by_group_pm_with(self, query: Query, operand: Union[str, int], maybe_negate: ConditionTransform) -> Query: # This operator does not support is_web_public_query. assert not self.is_web_public_query assert self.user_profile is not None try: if isinstance(operand, str): narrow_profile = get_user_including_cross_realm(operand, self.realm) else: narrow_profile = get_user_by_id_in_realm_including_cross_realm(operand, self.realm) except UserProfile.DoesNotExist: raise BadNarrowOperator('unknown user ' + str(operand)) self_recipient_ids = [ recipient_tuple['recipient_id'] for recipient_tuple in Subscription.objects.filter( user_profile=self.user_profile, recipient__type=Recipient.HUDDLE, ).values("recipient_id")] narrow_recipient_ids = [ recipient_tuple['recipient_id'] for recipient_tuple in Subscription.objects.filter( user_profile=narrow_profile, recipient__type=Recipient.HUDDLE, ).values("recipient_id")] recipient_ids = set(self_recipient_ids) & set(narrow_recipient_ids) cond = column("recipient_id").in_(recipient_ids) return query.where(maybe_negate(cond)) def by_search(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: if settings.USING_PGROONGA: return self._by_search_pgroonga(query, operand, maybe_negate) else: return self._by_search_tsearch(query, operand, maybe_negate) def _by_search_pgroonga(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: match_positions_character = func.pgroonga_match_positions_character query_extract_keywords = func.pgroonga_query_extract_keywords operand_escaped = func.escape_html(operand) keywords = query_extract_keywords(operand_escaped) query = query.column(match_positions_character(column("rendered_content"), keywords).label("content_matches")) query = query.column(match_positions_character(func.escape_html(topic_column_sa()), keywords).label("topic_matches")) condition = column("search_pgroonga").op("&@~")(operand_escaped) return query.where(maybe_negate(condition)) def _by_search_tsearch(self, query: Query, operand: str, maybe_negate: ConditionTransform) -> Query: tsquery = func.plainto_tsquery(literal("zulip.english_us_search"), literal(operand)) query = query.column(ts_locs_array(literal("zulip.english_us_search"), column("rendered_content"), tsquery).label("content_matches")) # We HTML-escape the topic in Postgres to avoid doing a server round-trip query = query.column(ts_locs_array(literal("zulip.english_us_search"), func.escape_html(topic_column_sa()), tsquery).label("topic_matches")) # Do quoted string matching. We really want phrase # search here so we can ignore punctuation and do # stemming, but there isn't a standard phrase search # mechanism in Postgres for term in re.findall(r'"[^"]+"|\S+', operand): if term[0] == '"' and term[-1] == '"': term = term[1:-1] term = '%' + connection.ops.prep_for_like_query(term) + '%' cond = or_(column("content").ilike(term), topic_column_sa().ilike(term)) query = query.where(maybe_negate(cond)) cond = column("search_tsvector").op("@@")(tsquery) return query.where(maybe_negate(cond)) def highlight_string(text: str, locs: Iterable[Tuple[int, int]]) -> str: highlight_start = '' highlight_stop = '' pos = 0 result = '' in_tag = False for loc in locs: (offset, length) = loc prefix_start = pos prefix_end = offset match_start = offset match_end = offset + length prefix = text[prefix_start:prefix_end] match = text[match_start:match_end] for character in (prefix + match): if character == '<': in_tag = True elif character == '>': in_tag = False if in_tag: result += prefix result += match else: result += prefix result += highlight_start result += match result += highlight_stop pos = match_end result += text[pos:] return result def get_search_fields(rendered_content: str, topic_name: str, content_matches: Iterable[Tuple[int, int]], topic_matches: Iterable[Tuple[int, int]]) -> Dict[str, str]: return { 'match_content': highlight_string(rendered_content, content_matches), MATCH_TOPIC: highlight_string(escape_html(topic_name), topic_matches), } def narrow_parameter(json: str) -> OptionalNarrowListT: data = orjson.loads(json) if not isinstance(data, list): raise ValueError("argument is not a list") if len(data) == 0: # The "empty narrow" should be None, and not [] return None def convert_term(elem: Union[Dict[str, Any], List[str]]) -> Dict[str, Any]: # We have to support a legacy tuple format. if isinstance(elem, list): if (len(elem) != 2 or any(not isinstance(x, str) for x in elem)): raise ValueError("element is not a string pair") return dict(operator=elem[0], operand=elem[1]) if isinstance(elem, dict): # Make sure to sync this list to frontend also when adding a new operator. # that supports user IDs. Relevant code is located in static/js/message_fetch.js # in handle_operators_supporting_id_based_api function where you will need to update # operators_supporting_id, or operators_supporting_ids array. operators_supporting_id = ['sender', 'group-pm-with', 'stream'] operators_supporting_ids = ['pm-with'] operators_non_empty_operand = {'search'} operator = elem.get('operator', '') if operator in operators_supporting_id: operand_validator: Validator[object] = check_string_or_int elif operator in operators_supporting_ids: operand_validator = check_string_or_int_list elif operator in operators_non_empty_operand: operand_validator = check_required_string else: operand_validator = check_string validator = check_dict( required_keys=[ ('operator', check_string), ('operand', operand_validator), ], optional_keys=[ ('negated', check_bool), ], ) try: validator('elem', elem) except ValidationError as error: raise JsonableError(error.message) # whitelist the fields we care about for now return dict( operator=elem['operator'], operand=elem['operand'], negated=elem.get('negated', False), ) raise ValueError("element is not a dictionary") return list(map(convert_term, data)) def ok_to_include_history(narrow: OptionalNarrowListT, user_profile: Optional[UserProfile], is_web_public_query: bool) -> bool: # There are occasions where we need to find Message rows that # have no corresponding UserMessage row, because the user is # reading a public stream that might include messages that # were sent while the user was not subscribed, but which they are # allowed to see. We have to be very careful about constructing # queries in those situations, so this function should return True # only if we are 100% sure that we're gonna add a clause to the # query that narrows to a particular public stream on the user's realm. # If we screw this up, then we can get into a nasty situation of # polluting our narrow results with messages from other realms. # For web-public queries, we are always returning history. The # analogues of the below stream access checks for whether streams # have is_web_public set and banning is operators in this code # path are done directly in NarrowBuilder. if is_web_public_query: assert user_profile is None return True assert user_profile is not None include_history = False if narrow is not None: for term in narrow: if term['operator'] == "stream" and not term.get('negated', False): operand: Union[str, int] = term['operand'] if isinstance(operand, str): include_history = can_access_stream_history_by_name(user_profile, operand) else: include_history = can_access_stream_history_by_id(user_profile, operand) elif (term['operator'] == "streams" and term['operand'] == "public" and not term.get('negated', False) and user_profile.can_access_public_streams()): include_history = True # Disable historical messages if the user is narrowing on anything # that's a property on the UserMessage table. There cannot be # historical messages in these cases anyway. for term in narrow: if term['operator'] == "is": include_history = False return include_history def get_stream_from_narrow_access_unchecked(narrow: OptionalNarrowListT, realm: Realm) -> Optional[Stream]: if narrow is not None: for term in narrow: if term['operator'] == 'stream': return get_stream_by_narrow_operand_access_unchecked(term['operand'], realm) return None def exclude_muting_conditions(user_profile: UserProfile, narrow: OptionalNarrowListT) -> List[Selectable]: conditions = [] stream_id = None try: # Note: It is okay here to not check access to stream # because we are only using the stream id to exclude data, # not to include results. stream = get_stream_from_narrow_access_unchecked(narrow, user_profile.realm) if stream is not None: stream_id = stream.id except Stream.DoesNotExist: pass if stream_id is None: rows = Subscription.objects.filter( user_profile=user_profile, active=True, is_muted=True, recipient__type=Recipient.STREAM, ).values('recipient_id') muted_recipient_ids = [row['recipient_id'] for row in rows] if len(muted_recipient_ids) > 0: # Only add the condition if we have muted streams to simplify/avoid warnings. condition = not_(column("recipient_id").in_(muted_recipient_ids)) conditions.append(condition) conditions = exclude_topic_mutes(conditions, user_profile, stream_id) return conditions def get_base_query_for_search(user_profile: Optional[UserProfile], need_message: bool, need_user_message: bool) -> Tuple[Query, ColumnElement]: # Handle the simple case where user_message isn't involved first. if not need_user_message: assert(need_message) query = select([column("id").label("message_id")], None, table("zerver_message")) inner_msg_id_col = literal_column("zerver_message.id") return (query, inner_msg_id_col) assert user_profile is not None if need_message: query = select([column("message_id"), column("flags")], column("user_profile_id") == literal(user_profile.id), join(table("zerver_usermessage"), table("zerver_message"), literal_column("zerver_usermessage.message_id") == literal_column("zerver_message.id"))) inner_msg_id_col = column("message_id") return (query, inner_msg_id_col) query = select([column("message_id"), column("flags")], column("user_profile_id") == literal(user_profile.id), table("zerver_usermessage")) inner_msg_id_col = column("message_id") return (query, inner_msg_id_col) def add_narrow_conditions(user_profile: Optional[UserProfile], inner_msg_id_col: ColumnElement, query: Query, narrow: OptionalNarrowListT, is_web_public_query: bool, realm: Realm) -> Tuple[Query, bool]: is_search = False # for now if narrow is None: return (query, is_search) # Build the query for the narrow builder = NarrowBuilder(user_profile, inner_msg_id_col, realm, is_web_public_query) search_operands = [] # As we loop through terms, builder does most of the work to extend # our query, but we need to collect the search operands and handle # them after the loop. for term in narrow: if term['operator'] == 'search': search_operands.append(term['operand']) else: query = builder.add_term(query, term) if search_operands: is_search = True query = query.column(topic_column_sa()).column(column("rendered_content")) search_term = dict( operator='search', operand=' '.join(search_operands), ) query = builder.add_term(query, search_term) return (query, is_search) def find_first_unread_anchor(sa_conn: Any, user_profile: Optional[UserProfile], narrow: OptionalNarrowListT) -> int: # For anonymous web users, all messages are treated as read, and so # always return LARGER_THAN_MAX_MESSAGE_ID. if user_profile is None: return LARGER_THAN_MAX_MESSAGE_ID # We always need UserMessage in our query, because it has the unread # flag for the user. need_user_message = True # Because we will need to call exclude_muting_conditions, unless # the user hasn't muted anything, we will need to include Message # in our query. It may be worth eventually adding an optimization # for the case of a user who hasn't muted anything to avoid the # join in that case, but it's low priority. need_message = True query, inner_msg_id_col = get_base_query_for_search( user_profile=user_profile, need_message=need_message, need_user_message=need_user_message, ) query, is_search = add_narrow_conditions( user_profile=user_profile, inner_msg_id_col=inner_msg_id_col, query=query, narrow=narrow, is_web_public_query=False, realm=user_profile.realm, ) condition = column("flags").op("&")(UserMessage.flags.read.mask) == 0 # We exclude messages on muted topics when finding the first unread # message in this narrow muting_conditions = exclude_muting_conditions(user_profile, narrow) if muting_conditions: condition = and_(condition, *muting_conditions) first_unread_query = query.where(condition) first_unread_query = first_unread_query.order_by(inner_msg_id_col.asc()).limit(1) first_unread_result = list(sa_conn.execute(first_unread_query).fetchall()) if len(first_unread_result) > 0: anchor = first_unread_result[0][0] else: anchor = LARGER_THAN_MAX_MESSAGE_ID return anchor def parse_anchor_value(anchor_val: Optional[str], use_first_unread_anchor: bool) -> Optional[int]: """Given the anchor and use_first_unread_anchor parameters passed by the client, computes what anchor value the client requested, handling backwards-compatibility and the various string-valued fields. We encode use_first_unread_anchor as anchor=None. """ if use_first_unread_anchor: # Backwards-compatibility: Before we added support for the # special string-typed anchor values, clients would pass # anchor=None and use_first_unread_anchor=True to indicate # what is now expressed as anchor="first_unread". return None if anchor_val is None: # Throw an exception if neither an anchor argument not # use_first_unread_anchor was specified. raise JsonableError(_("Missing 'anchor' argument.")) if anchor_val == "oldest": return 0 if anchor_val == "newest": return LARGER_THAN_MAX_MESSAGE_ID if anchor_val == "first_unread": return None try: # We don't use `.isnumeric()` to support negative numbers for # anchor. We don't recommend it in the API (if you want the # very first message, use 0 or 1), but it used to be supported # and was used by the webapp, so we need to continue # supporting it for backwards-compatibility anchor = int(anchor_val) if anchor < 0: return 0 return anchor except ValueError: raise JsonableError(_("Invalid anchor")) @has_request_variables def get_messages_backend(request: HttpRequest, maybe_user_profile: Union[UserProfile, AnonymousUser], anchor_val: Optional[str]=REQ( 'anchor', str_validator=check_string, default=None), num_before: int=REQ(converter=to_non_negative_int), num_after: int=REQ(converter=to_non_negative_int), narrow: OptionalNarrowListT=REQ('narrow', converter=narrow_parameter, default=None), use_first_unread_anchor_val: bool=REQ('use_first_unread_anchor', validator=check_bool, default=False), client_gravatar: bool=REQ(validator=check_bool, default=False), apply_markdown: bool=REQ(validator=check_bool, default=True)) -> HttpResponse: anchor = parse_anchor_value(anchor_val, use_first_unread_anchor_val) if num_before + num_after > MAX_MESSAGES_PER_FETCH: return json_error(_("Too many messages requested (maximum {}).").format( MAX_MESSAGES_PER_FETCH, )) if not maybe_user_profile.is_authenticated: # If user is not authenticated, clients must include # `streams:web-public` in their narrow query to indicate this # is a web-public query. This helps differentiate between # cases of web-public queries (where we should return the # web-public results only) and clients with buggy # authentication code (where we should return an auth error). if not is_web_public_narrow(narrow): return json_unauthorized() assert narrow is not None if not is_web_public_compatible(narrow): return json_unauthorized() realm = get_realm_from_request(request) if realm is None: return json_error(_("Invalid subdomain.")) # We use None to indicate unauthenticated requests as it's more # readable than using AnonymousUser, and the lack of Django # stubs means that mypy can't check AnonymousUser well. user_profile: Optional[UserProfile] = None is_web_public_query = True else: assert isinstance(maybe_user_profile, UserProfile) user_profile = maybe_user_profile assert user_profile is not None realm = user_profile.realm is_web_public_query = False assert realm is not None if is_web_public_query or \ realm.email_address_visibility != Realm.EMAIL_ADDRESS_VISIBILITY_EVERYONE: # If email addresses are only available to administrators, # clients cannot compute gravatars, so we force-set it to false. client_gravatar = False include_history = ok_to_include_history(narrow, user_profile, is_web_public_query) if include_history: # The initial query in this case doesn't use `zerver_usermessage`, # and isn't yet limited to messages the user is entitled to see! # # This is OK only because we've made sure this is a narrow that # will cause us to limit the query appropriately elsewhere. # See `ok_to_include_history` for details. # # Note that is_web_public_query=True goes here, since # include_history is semantically correct for is_web_public_query. need_message = True need_user_message = False elif narrow is None: # We need to limit to messages the user has received, but we don't actually # need any fields from Message need_message = False need_user_message = True else: need_message = True need_user_message = True query, inner_msg_id_col = get_base_query_for_search( user_profile=user_profile, need_message=need_message, need_user_message=need_user_message, ) query, is_search = add_narrow_conditions( user_profile=user_profile, inner_msg_id_col=inner_msg_id_col, query=query, narrow=narrow, realm=realm, is_web_public_query=is_web_public_query, ) if narrow is not None: # Add some metadata to our logging data for narrows verbose_operators = [] for term in narrow: if term['operator'] == "is": verbose_operators.append("is:" + term['operand']) else: verbose_operators.append(term['operator']) request._log_data['extra'] = "[{}]".format(",".join(verbose_operators)) sa_conn = get_sqlalchemy_connection() if anchor is None: # `anchor=None` corresponds to the anchor="first_unread" parameter. anchor = find_first_unread_anchor( sa_conn, user_profile, narrow, ) anchored_to_left = (anchor == 0) # Set value that will be used to short circuit the after_query # altogether and avoid needless conditions in the before_query. anchored_to_right = (anchor >= LARGER_THAN_MAX_MESSAGE_ID) if anchored_to_right: num_after = 0 first_visible_message_id = get_first_visible_message_id(realm) query = limit_query_to_range( query=query, num_before=num_before, num_after=num_after, anchor=anchor, anchored_to_left=anchored_to_left, anchored_to_right=anchored_to_right, id_col=inner_msg_id_col, first_visible_message_id=first_visible_message_id, ) main_query = alias(query) query = select(main_query.c, None, main_query).order_by(column("message_id").asc()) # This is a hack to tag the query we use for testing query = query.prefix_with("/* get_messages */") rows = list(sa_conn.execute(query).fetchall()) query_info = post_process_limited_query( rows=rows, num_before=num_before, num_after=num_after, anchor=anchor, anchored_to_left=anchored_to_left, anchored_to_right=anchored_to_right, first_visible_message_id=first_visible_message_id, ) rows = query_info['rows'] # The following is a little messy, but ensures that the code paths # are similar regardless of the value of include_history. The # 'user_messages' dictionary maps each message to the user's # UserMessage object for that message, which we will attach to the # rendered message dict before returning it. We attempt to # bulk-fetch rendered message dicts from remote cache using the # 'messages' list. message_ids: List[int] = [] user_message_flags: Dict[int, List[str]] = {} if is_web_public_query: # For web-public users, we treat all historical messages as read. for row in rows: message_id = row[0] message_ids.append(message_id) user_message_flags[message_id] = ["read"] elif include_history: assert user_profile is not None message_ids = [row[0] for row in rows] # TODO: This could be done with an outer join instead of two queries um_rows = UserMessage.objects.filter(user_profile=user_profile, message__id__in=message_ids) user_message_flags = {um.message_id: um.flags_list() for um in um_rows} for message_id in message_ids: if message_id not in user_message_flags: user_message_flags[message_id] = ["read", "historical"] else: for row in rows: message_id = row[0] flags = row[1] user_message_flags[message_id] = UserMessage.flags_list_for_flags(flags) message_ids.append(message_id) search_fields: Dict[int, Dict[str, str]] = {} if is_search: for row in rows: message_id = row[0] (topic_name, rendered_content, content_matches, topic_matches) = row[-4:] try: search_fields[message_id] = get_search_fields(rendered_content, topic_name, content_matches, topic_matches) except UnicodeDecodeError as err: # nocoverage # No coverage for this block since it should be # impossible, and we plan to remove it once we've # debugged the case that makes it happen. raise Exception(str(err), message_id, narrow) message_list = messages_for_ids( message_ids=message_ids, user_message_flags=user_message_flags, search_fields=search_fields, apply_markdown=apply_markdown, client_gravatar=client_gravatar, allow_edit_history=realm.allow_edit_history, ) statsd.incr('loaded_old_messages', len(message_list)) ret = dict( messages=message_list, result='success', msg='', found_anchor=query_info['found_anchor'], found_oldest=query_info['found_oldest'], found_newest=query_info['found_newest'], history_limited=query_info['history_limited'], anchor=anchor, ) return json_success(ret) def limit_query_to_range(query: Query, num_before: int, num_after: int, anchor: int, anchored_to_left: bool, anchored_to_right: bool, id_col: ColumnElement, first_visible_message_id: int) -> Query: ''' This code is actually generic enough that we could move it to a library, but our only caller for now is message search. ''' need_before_query = (not anchored_to_left) and (num_before > 0) need_after_query = (not anchored_to_right) and (num_after > 0) need_both_sides = need_before_query and need_after_query # The semantics of our flags are as follows: # # num_after = number of rows < anchor # num_after = number of rows > anchor # # But we also want the row where id == anchor (if it exists), # and we don't want to union up to 3 queries. So in some cases # we do things like `after_limit = num_after + 1` to grab the # anchor row in the "after" query. # # Note that in some cases, if the anchor row isn't found, we # actually may fetch an extra row at one of the extremes. if need_both_sides: before_anchor = anchor - 1 after_anchor = max(anchor, first_visible_message_id) before_limit = num_before after_limit = num_after + 1 elif need_before_query: before_anchor = anchor before_limit = num_before if not anchored_to_right: before_limit += 1 elif need_after_query: after_anchor = max(anchor, first_visible_message_id) after_limit = num_after + 1 if need_before_query: before_query = query if not anchored_to_right: before_query = before_query.where(id_col <= before_anchor) before_query = before_query.order_by(id_col.desc()) before_query = before_query.limit(before_limit) if need_after_query: after_query = query if not anchored_to_left: after_query = after_query.where(id_col >= after_anchor) after_query = after_query.order_by(id_col.asc()) after_query = after_query.limit(after_limit) if need_both_sides: query = union_all(before_query.self_group(), after_query.self_group()) elif need_before_query: query = before_query elif need_after_query: query = after_query else: # If we don't have either a before_query or after_query, it's because # some combination of num_before/num_after/anchor are zero or # use_first_unread_anchor logic found no unread messages. # # The most likely reason is somebody is doing an id search, so searching # for something like `message_id = 42` is exactly what we want. In other # cases, which could possibly be buggy API clients, at least we will # return at most one row here. query = query.where(id_col == anchor) return query def post_process_limited_query(rows: List[Any], num_before: int, num_after: int, anchor: int, anchored_to_left: bool, anchored_to_right: bool, first_visible_message_id: int) -> Dict[str, Any]: # Our queries may have fetched extra rows if they added # "headroom" to the limits, but we want to truncate those # rows. # # Also, in cases where we had non-zero values of num_before or # num_after, we want to know found_oldest and found_newest, so # that the clients will know that they got complete results. if first_visible_message_id > 0: visible_rows = [r for r in rows if r[0] >= first_visible_message_id] else: visible_rows = rows rows_limited = len(visible_rows) != len(rows) if anchored_to_right: num_after = 0 before_rows = visible_rows[:] anchor_rows: List[Any] = [] after_rows: List[Any] = [] else: before_rows = [r for r in visible_rows if r[0] < anchor] anchor_rows = [r for r in visible_rows if r[0] == anchor] after_rows = [r for r in visible_rows if r[0] > anchor] if num_before: before_rows = before_rows[-1 * num_before:] if num_after: after_rows = after_rows[:num_after] visible_rows = before_rows + anchor_rows + after_rows found_anchor = len(anchor_rows) == 1 found_oldest = anchored_to_left or (len(before_rows) < num_before) found_newest = anchored_to_right or (len(after_rows) < num_after) # BUG: history_limited is incorrect False in the event that we had # to bump `anchor` up due to first_visible_message_id, and there # were actually older messages. This may be a rare event in the # context where history_limited is relevant, because it can only # happen in one-sided queries with no num_before (see tests tagged # BUG in PostProcessTest for examples), and we don't generally do # those from the UI, so this might be OK for now. # # The correct fix for this probably involves e.g. making a # `before_query` when we increase `anchor` just to confirm whether # messages were hidden. history_limited = rows_limited and found_oldest return dict( rows=visible_rows, found_anchor=found_anchor, found_newest=found_newest, found_oldest=found_oldest, history_limited=history_limited, ) @has_request_variables def messages_in_narrow_backend(request: HttpRequest, user_profile: UserProfile, msg_ids: List[int]=REQ(validator=check_list(check_int)), narrow: OptionalNarrowListT=REQ(converter=narrow_parameter), ) -> HttpResponse: first_visible_message_id = get_first_visible_message_id(user_profile.realm) msg_ids = [message_id for message_id in msg_ids if message_id >= first_visible_message_id] # This query is limited to messages the user has access to because they # actually received them, as reflected in `zerver_usermessage`. query = select([column("message_id"), topic_column_sa(), column("rendered_content")], and_(column("user_profile_id") == literal(user_profile.id), column("message_id").in_(msg_ids)), join(table("zerver_usermessage"), table("zerver_message"), literal_column("zerver_usermessage.message_id") == literal_column("zerver_message.id"))) builder = NarrowBuilder(user_profile, column("message_id"), user_profile.realm) if narrow is not None: for term in narrow: query = builder.add_term(query, term) sa_conn = get_sqlalchemy_connection() query_result = list(sa_conn.execute(query).fetchall()) search_fields = {} for row in query_result: message_id = row['message_id'] topic_name = row[DB_TOPIC_NAME] rendered_content = row['rendered_content'] if 'content_matches' in row: content_matches = row['content_matches'] topic_matches = row['topic_matches'] else: content_matches = topic_matches = [] search_fields[str(message_id)] = get_search_fields( rendered_content, topic_name, content_matches, topic_matches, ) return json_success({"messages": search_fields})