searxng/searx/engines/startpage.py

# SPDX-License-Identifier: AGPL-3.0-or-later
"""Startpage's language & region selectors are a mess ..

.. _startpage regions:

Startpage regions
=================

In the list of regions there are tags we need to map to common region tags::

  pt-BR_BR --> pt_BR
  zh-CN_CN --> zh_Hans_CN
  zh-TW_TW --> zh_Hant_TW
  zh-TW_HK --> zh_Hant_HK
  en-GB_GB --> en_GB

and there is at least one tag with a three letter language tag (ISO 639-2)::

  fil_PH --> fil_PH

The locale code ``no_NO`` from Startpage does not exists and is mapped to
``nb-NO``::

    babel.core.UnknownLocaleError: unknown locale 'no_NO'

For reference see languages-subtag at iana; ``no`` is the macrolanguage [1]_ and
W3C recommends subtag over macrolanguage [2]_.

.. [1] `iana: language-subtag-registry
   <https://www.iana.org/assignments/language-subtag-registry/language-subtag-registry>`_ ::

      type: language
      Subtag: nb
      Description: Norwegian Bokmål
      Added: 2005-10-16
      Suppress-Script: Latn
      Macrolanguage: no

.. [2]
   Use macrolanguages with care.  Some language subtags have a Scope field set to
   macrolanguage, i.e. this primary language subtag encompasses a number of more
   specific primary language subtags in the registry.  ...  As we recommended for
   the collection subtags mentioned above, in most cases you should try to use
   the more specific subtags ... `W3: The primary language subtag
   <https://www.w3.org/International/questions/qa-choosing-language-tags#langsubtag>`_

.. _startpage languages:

Startpage languages
===================

HTTP ``Accept-Language`` header (``send_accept_language_header``):
  The displayed name in Startpage's settings page depend on the location of the
  IP when ``Accept-Language`` HTTP header is unset.

  Startpage tries to guess user's language and territory from the HTTP
  ``Accept-Language``.  Optional the user can select a search-language (can be
  different to the UI language) and a region filter.

  In :py:obj:`fetch_traits` we use::

    'Accept-Language': "en-US,en;q=0.5",
    ..

  to get uniform names independent from the IP).

.. _startpage categories:

Startpage categories
====================

Startpage's category (for Web-search, News, Videos, ..) is set by
:py:obj:`startpage_categ` in  settings.yml::

  - name: startpage
    engine: startpage
    startpage_categ: web
    ...

.. hint::

  Supported categories are ``web``, ``news`` and ``images``.

"""
# pylint: disable=too-many-statements

import re
import typing as t
from collections import OrderedDict
from datetime import datetime, timedelta
from json import loads
from unicodedata import combining, normalize

import babel.localedata
import dateutil.parser
import lxml.html

from searx.enginelib import EngineCache
from searx.enginelib.traits import EngineTraits
from searx.exceptions import SearxEngineCaptchaException
from searx.locales import region_tag
from searx.network import get  # see https://github.com/searxng/searxng/issues/762
from searx.utils import (
    eval_xpath,
    extr,
    extract_text,
    gen_useragent,
    html_to_text,
    humanize_bytes,
    remove_pua_from_str,
)

# about
about = {
    "website": "https://startpage.com",
    "wikidata_id": "Q2333295",
    "official_api_documentation": None,
    "use_official_api": False,
    "require_api_key": False,
    "results": "HTML",
}

startpage_categ = "web"
"""Startpage's category, visit :ref:`startpage categories`.
"""

# engine dependent config
categories = ["general", "web"]
paging = True
max_page = 18
"""Tested 18 pages maximum (argument ``page``), to be save max is set to 20."""

time_range_support = True
safesearch = True

time_range_dict = {"day": "d", "week": "w", "month": "m", "year": "y"}
safesearch_dict = {0: "none", 1: "moderate", 2: "heavy"}

# search-url
base_url = "https://www.startpage.com"
search_url = base_url + "/sp/search"

# specific xpath variables
# ads xpath //div[@id="results"]/div[@id="sponsored"]//div[@class="result"]
# not ads: div[@class="result"] are the direct children of div[@id="results"]
search_form_xpath = '//form[@id="search"]'
"""XPath of Startpage's origin search form

.. code: html

    <form action="/sp/search" method="post">
      <input type="text" name="query"  value="" ..>
      <input type="hidden" name="t" value="device">
      <input type="hidden" name="lui" value="english">
      <input type="hidden" name="sc" value="Q7Mt5TRqowKB00">
      <input type="hidden" name="cat" value="web">
      <input type="hidden" class="abp" id="abp-input" name="abp" value="1">
    </form>
"""


CACHE: EngineCache
"""Persistent (SQLite) key/value cache that deletes its values after ``expire``
seconds."""


def init(_):
    global CACHE  # pylint: disable=global-statement

    # hint: all three startpage engines (WEB, Images & News) can/should use the
    # same sc_code ..
    CACHE = EngineCache("startpage")


sc_code_cache_sec = 3600
"""Time in seconds the sc-code is cached in memory :py:obj:`get_sc_code`."""


def get_sc_code(params):
    """Get an actual ``sc`` argument from Startpage's search form (HTML page).

    Startpage puts a ``sc`` argument on every HTML :py:obj:`search form
    <search_form_xpath>`.  Without this argument Startpage considers the request
    is from a bot.  We do not know what is encoded in the value of the ``sc``
    argument, but it seems to be a kind of a *timestamp*.

    Startpage's search form generates a new sc-code on each request.  This
    function scrapes a new sc-code from Startpage's home page every
    :py:obj:`sc_code_cache_sec` seconds."""

    sc_code = CACHE.get("SC_CODE")
    if sc_code:
        logger.debug("get_sc_code: using cached value: %s", sc_code)
        return sc_code

    get_sc_url = base_url + "/"
    logger.debug("get_sc_code: querying new sc timestamp @ %s", get_sc_url)

    headers = {**params["headers"]}
    logger.debug("get_sc_code: request headers: %s", headers)
    resp = get(get_sc_url, headers=headers)

    # ?? x = network.get('https://www.startpage.com/sp/cdn/images/filter-chevron.svg', headers=headers)
    # ?? https://www.startpage.com/sp/cdn/images/filter-chevron.svg
    # ?? ping-back URL: https://www.startpage.com/sp/pb?sc=TLsB0oITjZ8F21

    if str(resp.url).startswith("https://www.startpage.com/sp/captcha"):
        raise SearxEngineCaptchaException(
            message="get_sc_code: got redirected to https://www.startpage.com/sp/captcha",
        )

    dom = lxml.html.fromstring(resp.text)

    try:
        sc_code = eval_xpath(dom, search_form_xpath + '//input[@name="sc"]/@value')[0]
    except IndexError as exc:
        logger.debug("suspend startpage API --> https://github.com/searxng/searxng/pull/695")
        raise SearxEngineCaptchaException(
            message="get_sc_code: [PR-695] querying new sc timestamp failed! (%s)" % resp.url,
        ) from exc

    sc_code = str(sc_code)
    logger.debug("get_sc_code: new value is: %s", sc_code)
    CACHE.set(key="SC_CODE", value=sc_code, expire=sc_code_cache_sec)
    return sc_code


def request(query, params):
    """Assemble a Startpage request.

    To avoid CAPTCHAs we need to send a well formed HTTP POST request with a
    cookie. We need to form a request that is identical to the request built by
    Startpage's search form:

    - in the cookie the **region** is selected
    - in the HTTP POST data the **language** is selected

    Additionally the arguments form Startpage's search form needs to be set in
    HTML POST data / compare ``<input>`` elements: :py:obj:`search_form_xpath`.
    """
    engine_region = traits.get_region(params["searxng_locale"], "en-US")
    engine_language = traits.get_language(params["searxng_locale"], "en")

    params["headers"]["Origin"] = base_url
    params["headers"]["Referer"] = base_url + "/"

    # Build form data
    args = {
        "query": query,
        "cat": startpage_categ,
        "t": "device",
        "sc": get_sc_code(params),
        "with_date": time_range_dict.get(params["time_range"], ""),
        "abd": "1",
        "abe": "1",
        "qsr": "all",
        "qadf": safesearch_dict[params["safesearch"]],
    }

    if engine_language:
        args["language"] = engine_language
        args["lui"] = engine_language

    if params["pageno"] > 1:
        args["page"] = params["pageno"]
        args["segment"] = "startpage.udog"

    # Build cookie
    lang_homepage = "en"
    cookie = OrderedDict()
    cookie["date_time"] = "world"
    cookie["disable_family_filter"] = safesearch_dict[params["safesearch"]]
    cookie["disable_open_in_new_window"] = "0"
    cookie["enable_post_method"] = "1"  # hint: POST
    cookie["enable_proxy_safety_suggest"] = "1"
    cookie["enable_stay_control"] = "1"
    cookie["instant_answers"] = "1"
    cookie["lang_homepage"] = "s/device/%s/" % lang_homepage
    cookie["num_of_results"] = "10"
    cookie["suggestions"] = "1"
    cookie["wt_unit"] = "celsius"

    if engine_language:
        cookie["language"] = engine_language
        cookie["language_ui"] = engine_language

    if engine_region:
        cookie["search_results_region"] = engine_region

    params["cookies"]["preferences"] = "N1N".join(["%sEEE%s" % x for x in cookie.items()])
    logger.debug("cookie preferences: %s", params["cookies"]["preferences"])

    logger.debug("data: %s", args)
    params["data"] = args
    params["method"] = "POST"
    params["url"] = search_url

    return params


def _parse_published_date(content: str) -> tuple[str, datetime | None]:
    published_date = None

    # check if search result starts with something like: "2 Sep 2014 ... "
    if re.match(r"^([1-9]|[1-2][0-9]|3[0-1]) [A-Z][a-z]{2} [0-9]{4} \.\.\. ", content):
        date_pos = content.find("...") + 4
        date_string = content[0 : date_pos - 5]
        # fix content string
        content = content[date_pos:]

        try:
            published_date = dateutil.parser.parse(date_string, dayfirst=True)
        except ValueError:
            pass

    # check if search result starts with something like: "5 days ago ... "
    elif re.match(r"^[0-9]+ days? ago \.\.\. ", content):
        date_pos = content.find("...") + 4
        date_string = content[0 : date_pos - 5]

        # calculate datetime
        published_date = datetime.now() - timedelta(days=int(re.match(r"\d+", date_string).group()))  # type: ignore

        # fix content string
        content = content[date_pos:]

    return content, published_date


def _get_web_result(result):
    content = html_to_text(result.get("description"))
    content, publishedDate = _parse_published_date(content)

    return {
        "url": result["clickUrl"],
        "title": html_to_text(result["title"]),
        "content": content,
        "publishedDate": publishedDate,
    }


def _get_news_result(result):

    title = remove_pua_from_str(html_to_text(result["title"]))
    content = remove_pua_from_str(html_to_text(result.get("description")))

    publishedDate = None
    if result.get("date"):
        try:
            publishedDate = datetime.fromtimestamp(int(result["date"]) / 1000)
        except (TypeError, ValueError):
            pass

    thumbnailUrl = None
    if result.get("thumbnailUrl"):
        thumbnailUrl = base_url + result["thumbnailUrl"]

    return {
        "url": result["clickUrl"],
        "title": title,
        "content": content,
        "publishedDate": publishedDate,
        "thumbnail": thumbnailUrl,
    }


def _get_image_result(result) -> dict[str, t.Any] | None:
    url = result.get("altClickUrl")
    if not url:
        return None

    thumbnailUrl = None
    if result.get("thumbnailUrl"):
        thumbnailUrl = base_url + result["thumbnailUrl"]

    resolution = None
    if result.get("width") and result.get("height"):
        resolution = f"{result['width']}x{result['height']}"

    filesize = None
    if result.get("filesize"):
        size_str = "".join(filter(str.isdigit, result["filesize"]))
        filesize = humanize_bytes(int(size_str))

    return {
        "template": "images.html",
        "url": url,
        "title": html_to_text(result["title"]),
        "content": "",
        "img_src": result.get("rawImageUrl"),
        "thumbnail_src": thumbnailUrl,
        "resolution": resolution,
        "img_format": result.get("format"),
        "filesize": filesize,
    }


def response(resp):
    categ = startpage_categ.capitalize()
    results_raw = "{" + extr(resp.text, f"React.createElement(UIStartpage.AppSerp{categ}, {{", "}})") + "}}"

    if resp.headers.get("Location", "").startswith("https://www.startpage.com/sp/captcha"):
        raise SearxEngineCaptchaException()

    results_json = loads(results_raw)
    results_obj = results_json.get("render", {}).get("presenter", {}).get("regions", {})

    results = []
    for results_categ in results_obj.get("mainline", []):
        for item in results_categ.get("results", []):
            if results_categ["display_type"] == "web-google":
                results.append(_get_web_result(item))
            elif results_categ["display_type"] == "news-bing":
                results.append(_get_news_result(item))
            elif "images" in results_categ["display_type"]:
                item = _get_image_result(item)
                if item:
                    results.append(item)

    return results


def fetch_traits(engine_traits: EngineTraits):
    """Fetch :ref:`languages <startpage languages>` and :ref:`regions <startpage
    regions>` from Startpage."""
    # pylint: disable=too-many-branches

    headers = {
        "User-Agent": gen_useragent(),
        "Accept-Language": "en-US,en;q=0.5",  # bing needs to set the English language
    }

    resp = get(
        "https://www.startpage.com/do/settings",
        headers=headers,
        timeout=5,
    )
    if not resp.ok:
        raise RuntimeError("Response from Startpage is not OK.")

    dom = lxml.html.fromstring(resp.text)

    # regions

    sp_region_names = []
    for option in dom.xpath('//form[@name="settings"]//select[@name="search_results_region"]/option'):
        sp_region_names.append(option.get("value"))

    for eng_tag in sp_region_names:
        if eng_tag == "all":
            continue
        babel_region_tag = {"no_NO": "nb_NO"}.get(eng_tag, eng_tag)  # norway

        if "-" in babel_region_tag:  # pyright: ignore[reportOperatorIssue]
            l, r = babel_region_tag.split("-")
            r = r.split("_")[-1]
            sxng_tag = region_tag(babel.Locale.parse(l + "_" + r, sep="_"))

        else:
            try:
                sxng_tag = region_tag(babel.Locale.parse(babel_region_tag, sep="_"))

            except babel.UnknownLocaleError:
                print("IGNORE: can't determine babel locale of startpage's locale %s" % eng_tag)
                continue

        conflict = engine_traits.regions.get(sxng_tag)
        if conflict:
            if conflict != eng_tag:
                print("CONFLICT: babel %s --> %s, %s" % (sxng_tag, conflict, eng_tag))
            continue
        engine_traits.regions[sxng_tag] = eng_tag

    # languages

    catalog_engine2code = {name.lower(): lang_code for lang_code, name in babel.Locale("en").languages.items()}

    # get the native name of every language known by babel

    for lang_code in filter(
        lambda lang_code: lang_code.find("_") == -1,
        babel.localedata.locale_identifiers(),
    ):
        native_name = babel.Locale(lang_code).get_language_name()
        if not native_name:
            print(f"IGNORE: language name of startpage's language {lang_code} is unknown by babel")
            continue
        native_name = native_name.lower()
        # add native name exactly as it is
        catalog_engine2code[native_name] = lang_code

        # add "normalized" language name (i.e. français becomes francais and español becomes espanol)
        unaccented_name = "".join(filter(lambda c: not combining(c), normalize("NFKD", native_name)))
        if len(unaccented_name) == len(unaccented_name.encode()):
            # add only if result is ascii (otherwise "normalization" didn't work)
            catalog_engine2code[unaccented_name] = lang_code

    # values that can't be determined by babel's languages names

    catalog_engine2code.update(
        {
            # Brazilian Portuguese
            "brazilian": "pt_BR",
            # traditional chinese used in ..
            "fantizhengwen": "zh_Hant",
            # Korean alphabet
            "hangul": "ko",
            # Malayalam is one of 22 scheduled languages of India.
            "malayam": "ml",
            "norsk": "nb",
            "sinhalese": "si",
        }
    )

    skip_eng_tags = {
        "english_uk",  # SearXNG lang 'en' already maps to 'english'
    }

    for option in dom.xpath('//form[@name="settings"]//select[@name="language"]/option'):
        eng_tag = option.get("value")
        if eng_tag in skip_eng_tags:
            continue
        name = extract_text(option).lower()  # type: ignore

        sxng_tag = catalog_engine2code.get(eng_tag)
        if sxng_tag is None:
            sxng_tag = catalog_engine2code.get(name)
        if sxng_tag is None:
            # silently ignore unknown languages
            continue

        conflict = engine_traits.languages.get(sxng_tag)
        if conflict:
            if conflict != eng_tag:
                print("CONFLICT: babel %s --> %s, %s" % (sxng_tag, conflict, eng_tag))
            continue
        engine_traits.languages[sxng_tag] = eng_tag