earwigbot/earwigbot/wiki/site.py

# Copyright (C) 2009-2024 Ben Kurtovic <ben.kurtovic@gmail.com>
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

from http.cookiejar import CookieJar
from json import dumps
from logging import NullHandler, getLogger
from os.path import expanduser
from threading import RLock
from time import sleep, time
from urllib.parse import unquote_plus, urlparse

import requests
from requests_oauthlib import OAuth1

from earwigbot import exceptions, importer
from earwigbot.wiki import constants
from earwigbot.wiki.category import Category
from earwigbot.wiki.page import Page
from earwigbot.wiki.user import User

pymysql = importer.new("pymysql")

__all__ = ["Site"]


class Site:
    """
    **EarwigBot: Wiki Toolset: Site**

    Represents a site, with support for API queries and returning
    :py:class:`~earwigbot.wiki.page.Page`,
    :py:class:`~earwigbot.wiki.user.User`,
    and :py:class:`~earwigbot.wiki.category.Category` objects. The constructor
    takes a bunch of arguments and you probably won't need to call it directly,
    rather :py:meth:`wiki.get_site() <earwigbot.wiki.sitesdb.SitesDB.get_site>`
    for returning :py:class:`Site`
    instances, :py:meth:`wiki.add_site()
    <earwigbot.wiki.sitesdb.SitesDB.add_site>` for adding new ones to our
    database, and :py:meth:`wiki.remove_site()
    <earwigbot.wiki.sitesdb.SitesDB.remove_site>` for removing old ones from
    our database, should suffice.

    *Attributes:*

    - :py:attr:`name`:    the site's name (or "wikiid"), like ``"enwiki"``
    - :py:attr:`project`: the site's project name, like ``"wikipedia"``
    - :py:attr:`lang`:    the site's language code, like ``"en"``
    - :py:attr:`domain`:  the site's web domain, like ``"en.wikipedia.org"``
    - :py:attr:`url`:     the site's URL, like ``"https://en.wikipedia.org"``

    *Public methods:*

    - :py:meth:`api_query`:            does an API query with kwargs as params
    - :py:meth:`sql_query`:            does an SQL query and yields its results
    - :py:meth:`get_maxlag`:           returns the internal database lag
    - :py:meth:`get_replag`:           estimates the external database lag
    - :py:meth:`get_token`:            gets a token for a specific API action
    - :py:meth:`namespace_id_to_name`: returns names associated with an NS id
    - :py:meth:`namespace_name_to_id`: returns the ID associated with a NS name
    - :py:meth:`get_page`:             returns a Page for the given title
    - :py:meth:`get_category`:         returns a Category for the given title
    - :py:meth:`get_user`:             returns a User object for the given name
    - :py:meth:`delegate`:             controls when the API or SQL is used
    """

    SERVICE_API = 1
    SERVICE_SQL = 2
    SPECIAL_TOKENS = [
        "createaccount",
        "deleteglobalaccount",
        "login",
        "patrol",
        "rollback",
        "setglobalaccountstatus",
        "userrights",
        "watch",
    ]

    def __init__(
        self,
        name=None,
        project=None,
        lang=None,
        base_url=None,
        article_path=None,
        script_path=None,
        sql=None,
        namespaces=None,
        login=(None, None),
        oauth=None,
        cookiejar=None,
        user_agent=None,
        use_https=True,
        assert_edit=None,
        maxlag=None,
        wait_between_queries=1,
        logger=None,
        search_config=None,
    ):
        """Constructor for new Site instances.

        This probably isn't necessary to call yourself unless you're building a
        Site that's not in your config and you don't want to add it - normally
        all you need is wiki.get_site(name), which creates the Site for you
        based on your config file and the sites database. We accept a bunch of
        kwargs, but the only ones you really "need" are *base_url* and
        *script_path*; this is enough to figure out an API url. *login*, a
        tuple of (username, password), can be used to log in using the legacy
        BotPasswords system; otherwise, a dict of OAuth info should be provided
        to *oauth*. *cookiejar* will be used to store cookies, and we'll use a
        normal CookieJar if none is given.

        First, we'll store the given arguments as attributes, then set up our
        requests session. We'll load any of the attributes that weren't given
        from the API, and then log in if a username/pass was given and we
        aren't already logged in.
        """
        # Attributes referring to site information, filled in by an API query
        # if they are missing (and an API url can be determined):
        self._name = name
        self._project = project
        self._lang = lang
        self._base_url = base_url
        self._article_path = article_path
        self._script_path = script_path
        self._namespaces = namespaces

        # Attributes used for API queries:
        self._use_https = use_https
        self._assert_edit = assert_edit
        self._maxlag = maxlag
        self._wait_between_queries = wait_between_queries
        self._max_retries = 6
        self._last_query_time = 0
        self._tokens = {}
        self._api_lock = RLock()
        self._api_info_cache = {"maxlag": 0, "lastcheck": 0}

        # Attributes used for SQL queries:
        if sql:
            self._sql_data = sql
        else:
            self._sql_data = {}
        self._sql_conn = None
        self._sql_lock = RLock()
        self._sql_info_cache = {"replag": 0, "lastcheck": 0, "usable": None}

        # Attribute used in copyright violation checks (see CopyrightMixIn):
        if search_config:
            self._search_config = search_config
        else:
            self._search_config = {}

        # Set up cookiejar and requests session for making API queries:
        if cookiejar is not None:
            self._cookiejar = cookiejar
        else:
            self._cookiejar = CookieJar()
        self._last_cookiejar_save = None
        if not user_agent:
            user_agent = constants.USER_AGENT  # Set default UA
        self._oauth = oauth
        self._session = requests.Session()
        self._session.cookies = self._cookiejar
        self._session.headers["User-Agent"] = user_agent
        if oauth:
            self._session.auth = OAuth1(
                oauth["consumer_token"],
                oauth["consumer_secret"],
                oauth["access_token"],
                oauth["access_secret"],
            )

        # Set up our internal logger:
        if logger:
            self._logger = logger
        else:  # Just set up a null logger to eat up our messages:
            self._logger = getLogger("earwigbot.wiki")
            self._logger.addHandler(NullHandler())

        # Get all of the above attributes that were not specified as arguments:
        self._load_attributes()

        # If we have a name/pass and the API says we're not logged in, log in:
        self._login_info = name, password = login
        if not self._oauth and name and password:
            logged_in_as = self._get_username_from_cookies()
            if not logged_in_as or name.replace("_", " ") != logged_in_as:
                self._login(login)

    def __repr__(self):
        """Return the canonical string representation of the Site."""
        res = ", ".join(
            (
                "Site(name={_name!r}",
                "project={_project!r}",
                "lang={_lang!r}",
                "base_url={_base_url!r}",
                "article_path={_article_path!r}",
                "script_path={_script_path!r}",
                "use_https={_use_https!r}",
                "assert_edit={_assert_edit!r}",
                "maxlag={_maxlag!r}",
                "sql={_sql_data!r}",
                "login={0}",
                "oauth={1}",
                "user_agent={3!r}",
                "cookiejar={2})",
            )
        )
        name, password = self._login_info
        login = "({}, {})".format(repr(name), "hidden" if password else None)
        oauth = "hidden" if self._oauth else None
        cookies = self._cookiejar.__class__.__name__
        if hasattr(self._cookiejar, "filename"):
            cookies += "({!r})".format(getattr(self._cookiejar, "filename"))
        else:
            cookies += "()"
        agent = self.user_agent
        return res.format(login, oauth, cookies, agent, **self.__dict__)

    def __str__(self):
        """Return a nice string representation of the Site."""
        res = "<Site {0} ({1}:{2}) at {3}>"
        return res.format(self.name, self.project, self.lang, self.domain)

    def _unicodeify(self, value, encoding="utf8"):
        """Return input as unicode if it's not unicode to begin with."""
        if isinstance(value, str):
            return value
        return str(value, encoding)

    def _api_query(
        self,
        params,
        tries=0,
        wait=5,
        ignore_maxlag=False,
        no_assert=False,
        ae_retry=True,
    ):
        """Do an API query with *params* as a dict of parameters.

        See the documentation for :py:meth:`api_query` for full implementation
        details. *tries*, *wait*, and *ignore_maxlag* are for maxlag;
        *no_assert* and *ae_retry* are for AssertEdit.
        """
        since_last_query = time() - self._last_query_time  # Throttling support
        if since_last_query < self._wait_between_queries:
            wait_time = self._wait_between_queries - since_last_query
            log = f"Throttled: waiting {round(wait_time, 2)} seconds"
            self._logger.debug(log)
            sleep(wait_time)
        self._last_query_time = time()

        url, params = self._build_api_query(params, ignore_maxlag, no_assert)
        if "lgpassword" in params:
            self._logger.debug(f"{url} -> <hidden>")
        else:
            data = dumps(params)
            if len(data) > 1000:
                self._logger.debug(f"{url} -> {data[:997]}...")
            else:
                self._logger.debug(f"{url} -> {data}")

        try:
            response = self._session.post(url, data=params)
            response.raise_for_status()
        except requests.RequestException as exc:
            raise exceptions.APIError(f"API query failed: {exc}")

        return self._handle_api_result(response, params, tries, wait, ae_retry)

    def _request_csrf_token(self, params):
        """If possible, add a request for a CSRF token to an API query."""
        if params.get("action") == "query":
            if params.get("meta"):
                if "tokens" not in params["meta"].split("|"):
                    params["meta"] += "|tokens"
            else:
                params["meta"] = "tokens"
            if params.get("type"):
                if "csrf" not in params["type"].split("|"):
                    params["type"] += "|csrf"

    def _build_api_query(self, params, ignore_maxlag, no_assert):
        """Given API query params, return the URL to query and POST data."""
        if not self._base_url or self._script_path is None:
            e = "Tried to do an API query, but no API URL is known."
            raise exceptions.APIError(e)

        url = self.url + self._script_path + "/api.php"
        params["format"] = "json"  # This is the only format we understand
        if self._assert_edit and not no_assert:
            # If requested, ensure that we're logged in
            params["assert"] = self._assert_edit
        if self._maxlag and not ignore_maxlag:
            # If requested, don't overload the servers:
            params["maxlag"] = self._maxlag
        if "csrf" not in self._tokens:
            # If we don't have a CSRF token, try to fetch one:
            self._request_csrf_token(params)
        return url, params

    def _handle_api_result(self, response, params, tries, wait, ae_retry):
        """Given an API query response, attempt to return useful data."""
        try:
            res = response.json()
        except ValueError:
            e = "API query failed: JSON could not be decoded."
            raise exceptions.APIError(e)

        if "warnings" in res:
            for name, value in res["warnings"].items():
                try:
                    warning = value["warnings"]
                except KeyError:
                    try:
                        warning = value["*"]
                    except KeyError:
                        warning = value
                self._logger.warning("API warning: %s: %s", name, warning)

        if self._should_save_cookiejar():
            self._save_cookiejar()

        try:
            code = res["error"]["code"]
            info = res["error"]["info"]
        except (TypeError, KeyError):  # If there's no error code/info, return
            if "query" in res and "tokens" in res["query"]:
                for name, token in res["query"]["tokens"].items():
                    self._tokens[name.split("token")[0]] = token
            return res

        if code == "maxlag":  # We've been throttled by the server
            if tries >= self._max_retries:
                e = "Maximum number of retries reached ({0})."
                raise exceptions.APIError(e.format(self._max_retries))
            tries += 1
            msg = 'Server says "{0}"; retrying in {1} seconds ({2}/{3})'
            self._logger.info(msg.format(info, wait, tries, self._max_retries))
            sleep(wait)
            return self._api_query(params, tries, wait * 2, ae_retry=ae_retry)
        elif code in ["assertuserfailed", "assertbotfailed"]:  # AssertEdit
            if ae_retry and all(self._login_info) and not self._oauth:
                # Try to log in if we got logged out:
                self._login(self._login_info)
                if "token" in params:  # Fetch a new one; this is invalid now
                    params["token"] = self.get_token(params["action"])
                return self._api_query(params, tries, wait, ae_retry=False)
            if not all(self._login_info) and not self._oauth:
                e = "Assertion failed, and no login info was provided."
            elif code == "assertbotfailed":
                e = "Bot assertion failed: we don't have a bot flag!"
            else:
                e = "User assertion failed due to an unknown issue. Cookie or OAuth problem?"
            raise exceptions.PermissionsError("AssertEdit: " + e)
        else:  # Some unknown error occurred
            e = 'API query failed: got error "{0}"; server says: "{1}".'
            error = exceptions.APIError(e.format(code, info))
            error.code, error.info = code, info
            raise error

    def _load_attributes(self, force=False):
        """Load data about our Site from the API.

        This function is called by __init__() when one of the site attributes
        was not given as a keyword argument. We'll do an API query to get the
        missing data, but only if there actually *is* missing data.

        Additionally, you can call this with *force* set to True to forcibly
        reload all attributes.
        """
        # All attributes to be loaded, except _namespaces, which is a special
        # case because it requires additional params in the API query:
        attrs = [
            self._name,
            self._project,
            self._lang,
            self._base_url,
            self._article_path,
            self._script_path,
        ]

        params = {"action": "query", "meta": "siteinfo", "siprop": "general"}

        if not self._namespaces or force:
            params["siprop"] += "|namespaces|namespacealiases"
            with self._api_lock:
                result = self._api_query(params, no_assert=True)
            self._load_namespaces(result)
        elif all(attrs):  # Everything is already specified and we're not told
            return  # to force a reload, so do nothing
        else:  # We're only loading attributes other than _namespaces
            with self._api_lock:
                result = self._api_query(params, no_assert=True)

        res = result["query"]["general"]
        self._name = res["wikiid"]
        self._project = res["sitename"].lower()
        self._lang = res["lang"]
        self._base_url = res["server"]
        self._article_path = res["articlepath"]
        self._script_path = res["scriptpath"]

    def _load_namespaces(self, result):
        """Fill self._namespaces with a dict of namespace IDs and names.

        Called by _load_attributes() with API data as *result* when
        self._namespaces was not given as an kwarg to __init__().
        """
        self._namespaces = {}

        for namespace in result["query"]["namespaces"].values():
            ns_id = namespace["id"]
            name = namespace["*"]
            try:
                canonical = namespace["canonical"]
            except KeyError:
                self._namespaces[ns_id] = [name]
            else:
                if name != canonical:
                    self._namespaces[ns_id] = [name, canonical]
                else:
                    self._namespaces[ns_id] = [name]

        for namespace in result["query"]["namespacealiases"]:
            ns_id = namespace["id"]
            alias = namespace["*"]
            self._namespaces[ns_id].append(alias)

    def _get_cookie(self, name, domain):
        """Return the named cookie unless it is expired or doesn't exist."""
        for cookie in self._cookiejar:
            if cookie.name == name and cookie.domain == domain:
                if cookie.is_expired():
                    break
                return cookie

    def _get_username_from_cookies(self):
        """Try to return our username based solely on cookies.

        First, we'll look for a cookie named self._name + "Token", like
        "enwikiToken". If it exists and isn't expired, we'll assume it's valid
        and try to return the value of the cookie self._name + "UserName" (like
        "enwikiUserName"). This should work fine on wikis without single-user
        login.

        If `enwikiToken` doesn't exist, we'll try to find a cookie named
        `centralauth_Token`. If this exists and is not expired, we'll try to
        return the value of `centralauth_User`.

        If we didn't get any matches, we'll return None. Our goal here isn't to
        return the most likely username, or what we *want* our username to be
        (for that, we'd do self._login_info[0]), but rather to get our current
        username without an unnecessary ?action=query&meta=userinfo API query.
        """
        name = "".join((self._name, "Token"))
        cookie = self._get_cookie(name, self.domain)

        if cookie:
            name = "".join((self._name, "UserName"))
            user_name = self._get_cookie(name, self.domain)
            if user_name:
                return unquote_plus(user_name.value)

        for cookie in self._cookiejar:
            if cookie.name != "centralauth_Token" or cookie.is_expired():
                continue
            base = cookie.domain
            if base.startswith(".") and not cookie.domain_initial_dot:
                base = base[1:]
            if self.domain.endswith(base):
                user_name = self._get_cookie("centralauth_User", cookie.domain)
                if user_name:
                    return unquote_plus(user_name.value)

    def _get_username_from_api(self):
        """Do a simple API query to get our username and return it.

        This is a reliable way to make sure we are actually logged in, because
        it doesn't deal with annoying cookie logic, but it results in an API
        query that is unnecessary in some cases.

        Called by _get_username() (in turn called by get_user() with no
        username argument) when cookie lookup fails, probably indicating that
        we are logged out.
        """
        result = self.api_query(action="query", meta="userinfo")
        return result["query"]["userinfo"]["name"]

    def _get_username(self):
        """Return the name of the current user, whether logged in or not.

        First, we'll try to deduce it solely from cookies, to avoid an
        unnecessary API query. For the cookie-detection method, see
        _get_username_from_cookies()'s docs.

        If our username isn't in cookies, then we're either using OAuth or
        we're probably not logged in, or something fishy is going on (like
        forced logout). If we're using OAuth and a username was configured,
        assume it is accurate and use it. Otherwise, do a single API query for
        our username (or IP address) and return that.
        """
        name = self._get_username_from_cookies()
        if name:
            return name
        if self._oauth and self._login_info[0]:
            return self._login_info[0]
        return self._get_username_from_api()

    def _should_save_cookiejar(self):
        """Return a bool indicating whether we should save the cookiejar.

        This is True if we haven't saved the cookiejar yet this session, or if
        our last save was over a day ago.
        """
        max_staleness = 60 * 60 * 24  # 1 day
        if not self._last_cookiejar_save:
            return True
        return time() - self._last_cookiejar_save > max_staleness

    def _save_cookiejar(self):
        """Try to save our cookiejar after doing a (normal) login or logout.

        Calls the standard .save() method with no filename. Don't fret if our
        cookiejar doesn't support saving (CookieJar raises AttributeError,
        FileCookieJar raises NotImplementedError) or no default filename was
        given (LWPCookieJar and MozillaCookieJar raise ValueError).
        """
        if hasattr(self._cookiejar, "save"):
            try:
                getattr(self._cookiejar, "save")()
            except (NotImplementedError, ValueError):
                pass
        self._last_cookiejar_save = time()

    def _login(self, login):
        """Safely login through the API.

        Normally, this is called by __init__() if a username and password have
        been provided and no valid login cookies were found. The only other
        time it needs to be called is when those cookies expire, which is done
        automatically by api_query() if a query fails.

        *login* is a (username, password) tuple.

        Raises LoginError on login errors (duh), like bad passwords and
        nonexistent usernames.
        """
        self._tokens.clear()
        name, password = login

        params = {"action": "query", "meta": "tokens", "type": "login"}
        with self._api_lock:
            result = self._api_query(params, no_assert=True)
        try:
            token = result["query"]["tokens"]["logintoken"]
        except KeyError:
            raise exceptions.LoginError("Couldn't get login token")

        params = {
            "action": "login",
            "lgname": name,
            "lgpassword": password,
            "lgtoken": token,
        }
        with self._api_lock:
            result = self._api_query(params, no_assert=True)

        res = result["login"]["result"]
        if res == "Success":
            self._tokens.clear()
            self._save_cookiejar()
            return
        if res == "Illegal":
            e = "The provided username is illegal."
        elif res == "NotExists":
            e = "The provided username does not exist."
        elif res == "EmptyPass":
            e = "No password was given."
        elif res == "WrongPass" or res == "WrongPluginPass":
            e = "The given password is incorrect."
        else:
            e = f"Couldn't login; server says '{res}'."
        raise exceptions.LoginError(e)

    def _logout(self):
        """Safely logout through the API.

        We'll do a simple API request (api.php?action=logout), clear our
        cookiejar (which probably contains now-invalidated cookies) and try to
        save it, if it supports that sort of thing.
        """
        self.api_query(action="logout")
        self._cookiejar.clear()
        self._save_cookiejar()

    def _sql_connect(self, **kwargs):
        """Attempt to establish a connection with this site's SQL database.

        pymysql.connect() will be called with self._sql_data as its kwargs.
        Any kwargs given to this function will be passed to connect() and will
        have precedence over the config file.

        Will raise SQLError() if the module "pymysql" is not available. pymysql
        may raise its own exceptions (e.g. pymysql.InterfaceError) if it cannot
        establish a connection.
        """
        args = self._sql_data
        for key, value in kwargs.items():
            args[key] = value
        if (
            "read_default_file" not in args
            and "user" not in args
            and "passwd" not in args
        ):
            args["read_default_file"] = expanduser("~/.my.cnf")
        elif "read_default_file" in args:
            args["read_default_file"] = expanduser(args["read_default_file"])
        if "autoping" not in args:
            args["autoping"] = True
        if "autoreconnect" not in args:
            args["autoreconnect"] = True

        try:
            self._sql_conn = pymysql.connect(**args)
        except ImportError:
            e = "SQL querying requires the 'pymysql' package: https://pymysql.readthedocs.io/"
            raise exceptions.SQLError(e)

    def _get_service_order(self):
        """Return a preferred order for using services (e.g. the API and SQL).

        A list is returned, starting with the most preferred service first and
        ending with the least preferred one. Currently, there are only two
        services. SERVICE_API will always be included since the API is expected
        to be always usable. In normal circumstances, self.SERVICE_SQL will be
        first (with the API second), since using SQL directly is easier on the
        servers than making web queries with the API. self.SERVICE_SQL will be
        second if replag is greater than three minutes (a cached value updated
        every two minutes at most), *unless* API lag is also very high.
        self.SERVICE_SQL will not be included in the list if we cannot form a
        proper SQL connection.
        """
        now = time()
        if now - self._sql_info_cache["lastcheck"] > 120:
            self._sql_info_cache["lastcheck"] = now
            try:
                try:
                    self._sql_info_cache["replag"] = sqllag = self.get_replag()
                except pymysql.Error as exc:
                    raise exceptions.SQLError(str(exc))
            except (exceptions.SQLError, ImportError):
                self._sql_info_cache["usable"] = False
                return [self.SERVICE_API]
            self._sql_info_cache["usable"] = True
        else:
            if not self._sql_info_cache["usable"]:
                return [self.SERVICE_API]
            sqllag = self._sql_info_cache["replag"]

        if sqllag > 300:
            if not self._maxlag:
                return [self.SERVICE_API, self.SERVICE_SQL]
            if now - self._api_info_cache["lastcheck"] > 300:
                self._api_info_cache["lastcheck"] = now
                try:
                    self._api_info_cache["maxlag"] = apilag = self.get_maxlag()
                except exceptions.APIError:
                    self._api_info_cache["maxlag"] = apilag = 0
            else:
                apilag = self._api_info_cache["maxlag"]
            if apilag > self._maxlag:
                return [self.SERVICE_SQL, self.SERVICE_API]
            return [self.SERVICE_API, self.SERVICE_SQL]

        return [self.SERVICE_SQL, self.SERVICE_API]

    @property
    def name(self):
        """The Site's name (or "wikiid" in the API), like ``"enwiki"``."""
        return self._name

    @property
    def project(self):
        """The Site's project name in lowercase, like ``"wikipedia"``."""
        return self._project

    @property
    def lang(self):
        """The Site's language code, like ``"en"`` or ``"es"``."""
        return self._lang

    @property
    def domain(self):
        """The Site's web domain, like ``"en.wikipedia.org"``."""
        return urlparse(self._base_url).netloc

    @property
    def url(self):
        """The Site's full base URL, like ``"https://en.wikipedia.org"``."""
        url = self._base_url
        if url.startswith("//"):  # Protocol-relative URLs from 1.18
            if self._use_https:
                url = "https:" + url
            else:
                url = "http:" + url
        return url

    @property
    def user_agent(self):
        """The User-Agent header sent to the API by the requests session."""
        return self._session.headers["User-Agent"]

    def api_query(self, **kwargs):
        """Do an API query with `kwargs` as the parameters.

        This will first attempt to construct an API url from
        :py:attr:`self._base_url` and :py:attr:`self._script_path`. We need
        both of these, or else we'll raise
        :py:exc:`~earwigbot.exceptions.APIError`. If
        :py:attr:`self._base_url` is protocol-relative (introduced in MediaWiki
        1.18), we'll choose HTTPS only if :py:attr:`self._user_https` is
        ``True``, otherwise HTTP.

        We'll encode the given params, adding ``format=json`` along the way, as
        well as ``&assert=`` and ``&maxlag=`` based on
        :py:attr:`self._assert_edit` and :py:attr:`_maxlag` respectively.
        Additionally, we'll sleep a bit if the last query was made fewer than
        :py:attr:`self._wait_between_queries` seconds ago. The request is made
        through :py:attr:`self._session`, which has cookie support
        (:py:attr:`self._cookiejar`) and a ``User-Agent``
        (:py:const:`earwigbot.wiki.constants.USER_AGENT`).

        Assuming everything went well, we'll gunzip the data (if compressed),
        load it as a JSON object, and return it.

        If our request failed for some reason, we'll raise
        :py:exc:`~earwigbot.exceptions.APIError` with details. If that
        reason was due to maxlag, we'll sleep for a bit and then repeat the
        query until we exceed :py:attr:`self._max_retries`.

        There is helpful MediaWiki API documentation at `MediaWiki.org
        <https://www.mediawiki.org/wiki/API>`_.
        """
        with self._api_lock:
            return self._api_query(kwargs)

    def sql_query(
        self,
        query,
        params=(),
        plain_query=False,
        dict_cursor=False,
        cursor_class=None,
        buffsize=1024,
    ):
        """Do an SQL query and yield its results.

        If *plain_query* is ``True``, we will force an unparameterized query.
        Specifying both *params* and *plain_query* will cause an error. If
        *dict_cursor* is ``True``, we will use
        :py:class:`pymysql.cursors.DictCursor` as our cursor, otherwise the
        default :py:class:`pymysql.cursors.Cursor`. If *cursor_class* is given,
        it will override this option.

        *buffsize* is the size of each memory-buffered group of results, to
        reduce the number of conversations with the database; it is passed to
        :py:meth:`cursor.fetchmany() <pymysql.cursors.Cursor.fetchmany>`. If
        set to ``0```, all results will be buffered in memory at once (this
        uses :py:meth:`fetchall() <pymysql.cursors.Cursor.fetchall>`). If set
        to ``1``, it is equivalent to using
        :py:meth:`fetchone() <pymysql.cursors.Cursor.fetchone>`.

        Example usage::

            >>> query = "SELECT user_id, user_registration FROM user WHERE user_name = ?"
            >>> params = ("The Earwig",)
            >>> result1 = site.sql_query(query, params)
            >>> result2 = site.sql_query(query, params, dict_cursor=True)
            >>> for row in result1: print row
            (7418060L, '20080703215134')
            >>> for row in result2: print row
            {'user_id': 7418060L, 'user_registration': '20080703215134'}

        This may raise :py:exc:`~earwigbot.exceptions.SQLError` or one of
        pymysql's exceptions (:py:exc:`pymysql.ProgrammingError`,
        :py:exc:`pymysql.InterfaceError`, ...) if there were problems with the
        query.

        See :py:meth:`_sql_connect` for information on how a connection is
        acquired. Also relevant is `pymysql's documentation
        <https://pymysql.readthedocs.io/>`_ for details on that package.
        """
        if not cursor_class:
            if dict_cursor:
                cursor_class = pymysql.cursors.DictCursor
            else:
                cursor_class = pymysql.cursors.Cursor
        klass = cursor_class

        with self._sql_lock:
            if not self._sql_conn:
                self._sql_connect()
            with self._sql_conn.cursor(klass) as cur:
                cur.execute(query, params, plain_query)
                if buffsize:
                    while True:
                        group = cur.fetchmany(buffsize)
                        if not group:
                            return
                        for result in group:
                            yield result
                for result in cur.fetchall():
                    yield result

    def get_maxlag(self, showall=False):
        """Return the internal database replication lag in seconds.

        In a typical setup, this function returns the replication lag *within*
        the WMF's cluster, *not* external replication lag affecting the
        Toolserver (see :py:meth:`get_replag` for that). This is useful when
        combined with the ``maxlag`` API query param (added by config), in
        which queries will be halted and retried if the lag is too high,
        usually above five seconds.

        With *showall*, will return a list of the lag for all servers in the
        cluster, not just the one with the highest lag.
        """
        params = {"action": "query", "meta": "siteinfo", "siprop": "dbrepllag"}
        if showall:
            params["sishowalldb"] = 1
        with self._api_lock:
            result = self._api_query(params, ignore_maxlag=True)
        if showall:
            return [server["lag"] for server in result["query"]["dbrepllag"]]
        return result["query"]["dbrepllag"][0]["lag"]

    def get_replag(self):
        """Return the estimated external database replication lag in seconds.

        Requires SQL access. This function only makes sense on a replicated
        database (e.g. the Wikimedia Toolserver) and on a wiki that receives a
        large number of edits (ideally, at least one per second), or the result
        may be larger than expected, since it works by subtracting the current
        time from the timestamp of the latest recent changes event.

        This may raise :py:exc:`~earwigbot.exceptions.SQLError` or one of
        pymysql's exceptions (:py:exc:`pymysql.ProgrammingError`,
        :py:exc:`pymysql.InterfaceError`, ...) if there were problems.
        """
        query = """SELECT UNIX_TIMESTAMP() - UNIX_TIMESTAMP(rc_timestamp) FROM
                   recentchanges ORDER BY rc_timestamp DESC LIMIT 1"""
        result = list(self.sql_query(query))
        return int(result[0][0])

    def get_token(self, action=None, force=False):
        """Return a token for a data-modifying API action.

        In general, this will be a CSRF token, unless *action* is in a special
        list of non-CSRF tokens. Tokens are cached for the session (until
        :meth:`_login` is called again); set *force* to ``True`` to force a new
        token to be fetched.

        Raises :exc:`.APIError` if there was an API issue.
        """
        if action not in self.SPECIAL_TOKENS:
            action = "csrf"
        if action in self._tokens and not force:
            return self._tokens[action]

        res = self.api_query(action="query", meta="tokens", type=action)
        if action not in self._tokens:
            err = "Tried to fetch a {0} token, but API returned: {1}"
            raise exceptions.APIError(err.format(action, res))
        return self._tokens[action]

    def namespace_id_to_name(self, ns_id, all=False):
        """Given a namespace ID, returns associated namespace names.

        If *all* is ``False`` (default), we'll return the first name in the
        list, which is usually the localized version. Otherwise, we'll return
        the entire list, which includes the canonical name. For example, this
        returns ``u"Wikipedia"`` if *ns_id* = ``4`` and *all* is ``False`` on
        ``enwiki``; returns ``[u"Wikipedia", u"Project", u"WP"]`` if *ns_id* =
        ``4`` and *all* is ``True``.

        Raises :py:exc:`~earwigbot.exceptions.NamespaceNotFoundError` if the ID
        is not found.
        """
        try:
            if all:
                return self._namespaces[ns_id]
            else:
                return self._namespaces[ns_id][0]
        except KeyError:
            e = f"There is no namespace with id {ns_id}."
            raise exceptions.NamespaceNotFoundError(e)

    def namespace_name_to_id(self, name):
        """Given a namespace name, returns the associated ID.

        Like :py:meth:`namespace_id_to_name`, but reversed. Case is ignored,
        because namespaces are assumed to be case-insensitive.

        Raises :py:exc:`~earwigbot.exceptions.NamespaceNotFoundError` if the
        name is not found.
        """
        lname = name.lower()
        for ns_id, names in self._namespaces.items():
            lnames = [n.lower() for n in names]  # Be case-insensitive
            if lname in lnames:
                return ns_id

        e = f"There is no namespace with name '{name}'."
        raise exceptions.NamespaceNotFoundError(e)

    def get_page(self, title, follow_redirects=False, pageid=None):
        """Return a :py:class:`Page` object for the given title.

        *follow_redirects* is passed directly to
        :py:class:`~earwigbot.wiki.page.Page`'s constructor. Also, this will
        return a :py:class:`~earwigbot.wiki.category.Category` object instead
        if the given title is in the category namespace. As
        :py:class:`~earwigbot.wiki.category.Category` is a subclass of
        :py:class:`~earwigbot.wiki.page.Page`, this should not cause problems.

        Note that this doesn't do any direct checks for existence or
        redirect-following: :py:class:`~earwigbot.wiki.page.Page`'s methods
        provide that.
        """
        title = self._unicodeify(title)
        prefixes = self.namespace_id_to_name(constants.NS_CATEGORY, all=True)
        prefix = title.split(":", 1)[0]
        if prefix != title:  # Avoid a page that is simply "Category"
            if prefix in prefixes:
                return Category(self, title, follow_redirects, pageid, self._logger)
        return Page(self, title, follow_redirects, pageid, self._logger)

    def get_category(self, catname, follow_redirects=False, pageid=None):
        """Return a :py:class:`Category` object for the given category name.

        *catname* should be given *without* a namespace prefix. This method is
        really just shorthand for :py:meth:`get_page("Category:" + catname)
        <get_page>`.
        """
        catname = self._unicodeify(catname)
        prefix = self.namespace_id_to_name(constants.NS_CATEGORY)
        pagename = ":".join((prefix, catname))
        return Category(self, pagename, follow_redirects, pageid, self._logger)

    def get_user(self, username=None):
        """Return a :py:class:`User` object for the given username.

        If *username* is left as ``None``, then a
        :py:class:`~earwigbot.wiki.user.User` object representing the currently
        logged-in (or anonymous!) user is returned.
        """
        if username:
            username = self._unicodeify(username)
        else:
            username = self._get_username()
        return User(self, username, self._logger)

    def delegate(self, services, args=None, kwargs=None):
        """Delegate a task to either the API or SQL depending on conditions.

        *services* should be a dictionary in which the key is the service name
        (:py:attr:`self.SERVICE_API <SERVICE_API>` or
        :py:attr:`self.SERVICE_SQL <SERVICE_SQL>`), and the value is the
        function to call for this service. All functions will be passed the
        same arguments the tuple *args* and the dict *kwargs*, which are both
        empty by default. The service order is determined by
        :py:meth:`_get_service_order`.

        Not every service needs an entry in the dictionary. Will raise
        :py:exc:`~earwigbot.exceptions.NoServiceError` if an appropriate
        service cannot be found.
        """
        if not args:
            args = ()
        if not kwargs:
            kwargs = {}

        order = self._get_service_order()
        for srv in order:
            if srv in services:
                try:
                    return services[srv](*args, **kwargs)
                except exceptions.ServiceError:
                    continue
        raise exceptions.NoServiceError(services)