whoogle-search/app/filter.py

import cssutils
from bs4 import BeautifulSoup
from bs4.element import ResultSet, Tag
from cryptography.fernet import Fernet
from flask import render_template
import html
import urllib.parse as urlparse
from urllib.parse import parse_qs
import re

from app.models.g_classes import GClasses
from app.request import VALID_PARAMS, MAPS_URL
from app.utils.misc import get_abs_url, read_config_bool
from app.utils.results import (
    BLANK_B64, GOOG_IMG, GOOG_STATIC, G_M_LOGO_URL, LOGO_URL, SITE_ALTS,
    has_ad_content, filter_link_args, append_anon_view, get_site_alt,
)
from app.models.endpoint import Endpoint
from app.models.config import Config


MAPS_ARGS = ['q', 'daddr']

minimal_mode_sections = ['Top stories', 'Images', 'People also ask']
unsupported_g_pages = [
    'support.google.com',
    'accounts.google.com',
    'policies.google.com',
    'google.com/preferences',
    'google.com/intl',
    'advanced_search',
    'tbm=shop',
    'ageverification.google.co.kr'
]

unsupported_g_divs = ['google.com/preferences?hl=', 'ageverification.google.co.kr']


def extract_q(q_str: str, href: str) -> str:
    """Extracts the 'q' element from a result link. This is typically
    either the link to a result's website, or a string.

    Args:
        q_str: The result link to parse
        href: The full url to check for standalone 'q' elements first,
              rather than parsing the whole query string and then checking.

    Returns:
        str: The 'q' element of the link, or an empty string
    """
    return parse_qs(q_str, keep_blank_values=True)['q'][0] if ('&q=' in href or '?q=' in href) else ''


def build_map_url(href: str) -> str:
    """Tries to extract known args that explain the location in the url. If a
    location is found, returns the default url with it. Otherwise, returns the
    url unchanged.

    Args:
        href: The full url to check.

    Returns:
        str: The parsed url, or the url unchanged.
    """
    # parse the url
    parsed_url = parse_qs(href)
    # iterate through the known parameters and try build the url
    for param in MAPS_ARGS:
        if param in parsed_url:
            return MAPS_URL + "?q=" + parsed_url[param][0]

    # query could not be extracted returning unchanged url
    return href


def clean_query(query: str) -> str:
    """Strips the blocked site list from the query, if one is being
    used.

    Args:
        query: The query string

    Returns:
        str: The query string without any "-site:..." filters
    """
    return query[:query.find('-site:')] if '-site:' in query else query


def clean_css(css: str, page_url: str) -> str:
    """Removes all remote URLs from a CSS string.

    Args:
        css: The CSS string

    Returns:
        str: The filtered CSS, with URLs proxied through Whoogle
    """
    sheet = cssutils.parseString(css)
    urls = cssutils.getUrls(sheet)

    for url in urls:
        abs_url = get_abs_url(url, page_url)
        if abs_url.startswith('data:'):
            continue
        css = css.replace(
            url,
            f'{Endpoint.element}?type=image/png&url={abs_url}'
        )

    return css


class Filter:
    # Limit used for determining if a result is a "regular" result or a list
    # type result (such as "people also asked", "related searches", etc)
    RESULT_CHILD_LIMIT = 7

    def __init__(
            self,
            user_key: str,
            config: Config,
            root_url='',
            page_url='',
            query='',
            mobile=False) -> None:
        self.soup = None
        self.config = config
        self.mobile = mobile
        self.user_key = user_key
        self.page_url = page_url
        self.query = query
        self.main_divs = ResultSet('')
        self._elements = 0
        self._av = set()

        self.root_url = root_url[:-1] if root_url.endswith('/') else root_url

    def __getitem__(self, name):
        return getattr(self, name)

    @property
    def elements(self):
        return self._elements

    def encrypt_path(self, path, is_element=False) -> str:
        # Encrypts path to avoid plaintext results in logs
        if is_element:
            # Element paths are encrypted separately from text, to allow key
            # regeneration once all items have been served to the user
            enc_path = Fernet(self.user_key).encrypt(path.encode()).decode()
            self._elements += 1
            return enc_path

        return Fernet(self.user_key).encrypt(path.encode()).decode()

    def clean(self, soup) -> BeautifulSoup:
        self.soup = soup
        self.main_divs = self.soup.find('div', {'id': 'main'})
        self.remove_ads()
        self.remove_block_titles()
        self.remove_block_url()
        self.collapse_sections()
        self.update_css()
        self.update_styling()
        self.remove_block_tabs()

        # self.main_divs is only populated for the main page of search results
        # (i.e. not images/news/etc).
        if self.main_divs:
            for div in self.main_divs:
                self.sanitize_div(div)

        for img in [_ for _ in self.soup.find_all('img') if 'src' in _.attrs]:
            self.update_element_src(img, 'image/png')

        for audio in [_ for _ in self.soup.find_all('audio') if 'src' in _.attrs]:
            self.update_element_src(audio, 'audio/mpeg')
            audio['controls'] = ''

        for link in self.soup.find_all('a', href=True):
            self.update_link(link)
            self.add_favicon(link)

        if self.config.alts:
            self.site_alt_swap()

        input_form = self.soup.find('form')
        if input_form is not None:
            input_form['method'] = 'GET' if self.config.get_only else 'POST'
            # Use a relative URI for submissions
            input_form['action'] = 'search'

        # Ensure no extra scripts passed through
        for script in self.soup('script'):
            script.decompose()

        # Update default footer and header
        footer = self.soup.find('footer')
        if footer:
            # Remove divs that have multiple links beyond just page navigation
            [_.decompose() for _ in footer.find_all('div', recursive=False)
             if len(_.find_all('a', href=True)) > 3]
            for link in footer.find_all('a', href=True):
                link['href'] = f'{link["href"]}&preferences={self.config.preferences}'

        header = self.soup.find('header')
        if header:
            header.decompose()
        self.remove_site_blocks(self.soup)
        return self.soup

    def sanitize_div(self, div) -> None:
        """Removes escaped script and iframe tags from results

        Returns:
            None (The soup object is modified directly)
        """
        if not div:
            return

        for d in div.find_all('div', recursive=True):
            d_text = d.find(text=True, recursive=False)

            # Ensure we're working with tags that contain text content
            if not d_text or not d.string:
                continue

            d.string = html.unescape(d_text)
            div_soup = BeautifulSoup(d.string, 'html.parser')

            # Remove all valid script or iframe tags in the div
            for script in div_soup.find_all('script'):
                script.decompose()

            for iframe in div_soup.find_all('iframe'):
                iframe.decompose()

            d.string = str(div_soup)

    def add_favicon(self, link) -> None:
        """Adds icons for each returned result, using the result site's favicon

        Returns:
            None (The soup object is modified directly)
        """
        # Skip empty, parentless, or internal links
        show_favicons = read_config_bool('WHOOGLE_SHOW_FAVICONS', True)
        is_valid_link = link and link.parent and link['href'].startswith('http')
        if not show_favicons or not is_valid_link:
            return

        parent = link.parent
        is_result_div = False

        # Check each parent to make sure that the div doesn't already have a
        # favicon attached, and that the div is a result div
        while parent:
            p_cls = parent.attrs.get('class') or []
            if 'has-favicon' in p_cls or GClasses.scroller_class in p_cls:
                return
            elif GClasses.result_class_a not in p_cls:
                parent = parent.parent
            else:
                is_result_div = True
                break

        if not is_result_div:
            return

        # Construct the html for inserting the icon into the parent div
        parsed = urlparse.urlparse(link['href'])
        favicon = self.encrypt_path(
            f'{parsed.scheme}://{parsed.netloc}/favicon.ico',
            is_element=True)
        src = f'{self.root_url}/{Endpoint.element}?url={favicon}' + \
            '&type=image/x-icon'
        html = f'<img class="site-favicon" src="{src}">'

        favicon = BeautifulSoup(html, 'html.parser')
        link.parent.insert(0, favicon)

        # Update all parents to indicate that a favicon has been attached
        parent = link.parent
        while parent:
            p_cls = parent.get('class') or []
            p_cls.append('has-favicon')
            parent['class'] = p_cls
            parent = parent.parent

            if GClasses.result_class_a in p_cls:
                break

    def remove_site_blocks(self, soup) -> None:
        if not self.config.block or not soup.body:
            return
        search_string = ' '.join(['-site:' +
                                 _ for _ in self.config.block.split(',')])
        selected = soup.body.findAll(text=re.compile(search_string))

        for result in selected:
            result.string.replace_with(result.string.replace(
                                       search_string, ''))

    def remove_ads(self) -> None:
        """Removes ads found in the list of search result divs

        Returns:
            None (The soup object is modified directly)
        """
        if not self.main_divs:
            return

        for div in [_ for _ in self.main_divs.find_all('div', recursive=True)]:
            div_ads = [_ for _ in div.find_all('span', recursive=True)
                       if has_ad_content(_.text)]
            _ = div.decompose() if len(div_ads) else None

    def remove_block_titles(self) -> None:
        if not self.main_divs or not self.config.block_title:
            return
        block_title = re.compile(self.config.block_title)
        for div in [_ for _ in self.main_divs.find_all('div', recursive=True)]:
            block_divs = [_ for _ in div.find_all('h3', recursive=True)
                          if block_title.search(_.text) is not None]
            _ = div.decompose() if len(block_divs) else None

    def remove_block_url(self) -> None:
        if not self.main_divs or not self.config.block_url:
            return
        block_url = re.compile(self.config.block_url)
        for div in [_ for _ in self.main_divs.find_all('div', recursive=True)]:
            block_divs = [_ for _ in div.find_all('a', recursive=True)
                          if block_url.search(_.attrs['href']) is not None]
            _ = div.decompose() if len(block_divs) else None

    def remove_block_tabs(self) -> None:
        if self.main_divs:
            for div in self.main_divs.find_all(
                'div',
                attrs={'class': f'{GClasses.main_tbm_tab}'}
            ):
                _ = div.decompose()
        else:
            # when in images tab
            for div in self.soup.find_all(
                'div',
                attrs={'class': f'{GClasses.images_tbm_tab}'}
            ):
                _ = div.decompose()

    def collapse_sections(self) -> None:
        """Collapses long result sections ("people also asked", "related
         searches", etc) into "details" elements

        These sections are typically the only sections in the results page that
        have more than ~5 child divs within a primary result div.

        Returns:
            None (The soup object is modified directly)
        """
        minimal_mode = read_config_bool('WHOOGLE_MINIMAL')

        def pull_child_divs(result_div: BeautifulSoup):
            try:
                return result_div.findChildren(
                    'div', recursive=False
                )[0].findChildren(
                    'div', recursive=False)
            except IndexError:
                return []

        if not self.main_divs:
            return

        # Loop through results and check for the number of child divs in each
        for result in self.main_divs.find_all():
            result_children = pull_child_divs(result)
            if minimal_mode:
                if any(f">{x}</span" in str(s) for s in result_children
                   for x in minimal_mode_sections):
                    result.decompose()
                    continue
                for s in result_children:
                    if ('Twitter ›' in str(s)):
                        result.decompose()
                        continue
                if len(result_children) < self.RESULT_CHILD_LIMIT:
                    continue
            else:
                if len(result_children) < self.RESULT_CHILD_LIMIT:
                    continue

            # Find and decompose the first element with an inner HTML text val.
            # This typically extracts the title of the section (i.e. "Related
            # Searches", "People also ask", etc)
            # If there are more than one child tags with text
            # parenthesize the rest except the first
            label = 'Collapsed Results'
            subtitle = None
            for elem in result_children:
                if elem.text:
                    content = list(elem.strings)
                    label = content[0]
                    if len(content) > 1:
                        subtitle = '<span> (' + \
                            ''.join(content[1:]) + ')</span>'
                    elem.decompose()
                    break

            # Create the new details element to wrap around the result's
            # first parent
            parent = None
            idx = 0
            while not parent and idx < len(result_children):
                parent = result_children[idx].parent
                idx += 1

            details = BeautifulSoup(features='html.parser').new_tag('details')
            summary = BeautifulSoup(features='html.parser').new_tag('summary')
            summary.string = label

            if subtitle:
                soup = BeautifulSoup(subtitle, 'html.parser')
                summary.append(soup)

            details.append(summary)

            if parent and not minimal_mode:
                parent.wrap(details)
            elif parent and minimal_mode:
                # Remove parent element from document if "minimal mode" is
                # enabled
                parent.decompose()

    def update_element_src(self, element: Tag, mime: str, attr='src') -> None:
        """Encrypts the original src of an element and rewrites the element src
        to use the "/element?src=" pass-through.

        Returns:
            None (The soup element is modified directly)

        """
        src = element[attr].split(' ')[0]

        if src.startswith('//'):
            src = 'https:' + src
        elif src.startswith('data:'):
            return

        if src.startswith(LOGO_URL):
            # Re-brand with Whoogle logo
            element.replace_with(BeautifulSoup(
                render_template('logo.html'),
                features='html.parser'))
            return
        elif src.startswith(G_M_LOGO_URL):
            # Re-brand with single-letter Whoogle logo
            element['src'] = 'static/img/favicon/apple-icon.png'
            element.parent['href'] = 'home'
            return
        elif src.startswith(GOOG_IMG) or GOOG_STATIC in src:
            element['src'] = BLANK_B64
            return

        element[attr] = f'{self.root_url}/{Endpoint.element}?url=' + (
            self.encrypt_path(
                src,
                is_element=True
            ) + '&type=' + urlparse.quote(mime)
        )

    def update_css(self) -> None:
        """Updates URLs used in inline styles to be proxied by Whoogle
        using the /element endpoint.

        Returns:
            None (The soup element is modified directly)

        """
        # Filter all <style> tags
        for style in self.soup.find_all('style'):
            style.string = clean_css(style.string, self.page_url)

        # TODO: Convert remote stylesheets to style tags and proxy all
        # remote requests
        # for link in soup.find_all('link', attrs={'rel': 'stylesheet'}):
            # print(link)

    def update_styling(self) -> None:
        # Update CSS classes for result divs
        soup = GClasses.replace_css_classes(self.soup)

        # Remove unnecessary button(s)
        for button in self.soup.find_all('button'):
            button.decompose()

        # Remove svg logos
        for svg in self.soup.find_all('svg'):
            svg.decompose()

        # Update logo
        logo = self.soup.find('a', {'class': 'l'})
        if logo and self.mobile:
            logo['style'] = ('display:flex; justify-content:center; '
                             'align-items:center; color:#685e79; '
                             'font-size:18px; ')

        # Fix search bar length on mobile
        try:
            search_bar = self.soup.find('header').find('form').find('div')
            search_bar['style'] = 'width: 100%;'
        except AttributeError:
            pass

        # Fix body max width on images tab
        style = self.soup.find('style')
        div = self.soup.find('div', attrs={
            'class': f'{GClasses.images_tbm_tab}'})
        if style and div and not self.mobile:
            css = style.string
            css_html_tag = (
                'html{'
                'font-family: Roboto, Helvetica Neue, Arial, sans-serif;'
                'font-size: 14px;'
                'line-height: 20px;'
                'text-size-adjust: 100%;'
                'word-wrap: break-word;'
                '}'
            )
            css = f"{css_html_tag}{css}"
            css = re.sub('body{(.*?)}',
                         'body{padding:0 8px;margin:0 auto;max-width:736px;}',
                         css)
            style.string = css

    def update_link(self, link: Tag) -> None:
        """Update internal link paths with encrypted path, otherwise remove
        unnecessary redirects and/or marketing params from the url

        Args:
            link: A bs4 Tag element to inspect and update

        Returns:
            None (the tag is updated directly)

        """
        parsed_link = urlparse.urlparse(link['href'])
        if '/url?q=' in link['href']:
            link_netloc = extract_q(parsed_link.query, link['href'])
        else:
            link_netloc = parsed_link.netloc

        # Remove any elements that direct to unsupported Google pages
        if any(url in link_netloc for url in unsupported_g_pages):
            # FIXME: The "Shopping" tab requires further filtering (see #136)
            # Temporarily removing all links to that tab for now.

            # Replaces the /url google unsupported link to the direct url
            link['href'] = link_netloc
            parent = link.parent

            if any(divlink in link_netloc for divlink in unsupported_g_divs):
                # Handle case where a search is performed in a different
                # language than what is configured. This usually returns a
                # div with the same classes as normal search results, but with
                # a link to configure language preferences through Google.
                # Since we want all language config done through Whoogle, we
                # can safely decompose this element.
                while parent:
                    p_cls = parent.attrs.get('class') or []
                    if f'{GClasses.result_class_a}' in p_cls:
                        parent.decompose()
                        break
                    parent = parent.parent
            else:
                # Remove cases where google links appear in the footer
                while parent:
                    p_cls = parent.attrs.get('class') or []
                    if parent.name == 'footer' or f'{GClasses.footer}' in p_cls:
                        link.decompose()
                    parent = parent.parent

            if link.decomposed:
                return

        # Replace href with only the intended destination (no "utm" type tags)
        href = link['href'].replace('https://www.google.com', '')
        result_link = urlparse.urlparse(href)
        q = extract_q(result_link.query, href)

        if q.startswith('/') and q not in self.query and 'spell=1' not in href:
            # Internal google links (i.e. mail, maps, etc) should still
            # be forwarded to Google
            link['href'] = 'https://google.com' + q
        elif q.startswith('https://accounts.google.com'):
            # Remove Sign-in link
            link.decompose()
            return
        elif '/search?q=' in href:
            # "li:1" implies the query should be interpreted verbatim,
            # which is accomplished by wrapping the query in double quotes
            if 'li:1' in href:
                q = '"' + q + '"'
            new_search = 'search?q=' + self.encrypt_path(q)

            query_params = parse_qs(urlparse.urlparse(href).query)
            for param in VALID_PARAMS:
                if param not in query_params:
                    continue
                param_val = query_params[param][0]
                new_search += '&' + param + '=' + param_val
            link['href'] = new_search
        elif 'url?q=' in href:
            # Strip unneeded arguments
            link['href'] = filter_link_args(q)

            # Add alternate viewing options for results,
            # if the result doesn't already have an AV link
            netloc = urlparse.urlparse(link['href']).netloc
            if self.config.anon_view and netloc not in self._av:
                self._av.add(netloc)
                append_anon_view(link, self.config)

        else:
            if href.startswith(MAPS_URL):
                # Maps links don't work if a site filter is applied
                link['href'] = build_map_url(link['href'])
            elif (href.startswith('/?') or href.startswith('/search?') or
                  href.startswith('/imgres?')):
                # make sure that tags can be clicked as relative URLs
                link['href'] = href[1:]
            elif href.startswith('/intl/'):
                # do nothing, keep original URL for ToS
                pass
            elif href.startswith('/preferences'):
                # there is no config specific URL, remove this
                link.decompose()
                return
            else:
                link['href'] = href

        if self.config.new_tab and (
            link["href"].startswith("http")
            or link["href"].startswith("imgres?")
        ):
            link["target"] = "_blank"

    def site_alt_swap(self) -> None:
        """Replaces link locations and page elements if "alts" config
        is enabled
        """
        for site, alt in SITE_ALTS.items():
            if site != "medium.com" and alt != "":
                # Ignore medium.com replacements since these are handled
                # specifically in the link description replacement, and medium
                # results are never given their own "card" result where this
                # replacement would make sense.
                # Also ignore if the alt is empty, since this is used to indicate
                # that the alt is not enabled.
                for div in self.soup.find_all('div', text=re.compile(site)):
                    # Use the number of words in the div string to determine if the
                    # string is a result description (shouldn't replace domains used
                    # in desc text).
                    if len(div.string.split(' ')) == 1:
                        div.string = div.string.replace(site, alt)

            for link in self.soup.find_all('a', href=True):
                # Search and replace all link descriptions
                # with alternative location
                link['href'] = get_site_alt(link['href'])
                link_desc = link.find_all(
                    text=re.compile('|'.join(SITE_ALTS.keys())))
                if len(link_desc) == 0:
                    continue

                # Replace link description
                link_desc = link_desc[0]
                if site not in link_desc or not alt:
                    continue

                new_desc = BeautifulSoup(features='html.parser').new_tag('div')
                link_str = str(link_desc)

                # Medium links should be handled differently, since 'medium.com'
                # is a common substring of domain names, but shouldn't be
                # replaced (i.e. 'philomedium.com' should stay as it is).
                if 'medium.com' in link_str:
                    if link_str.startswith('medium.com') or '.medium.com' in link_str:
                        link_str = SITE_ALTS['medium.com'] + link_str[
                            link_str.find('medium.com') + len('medium.com'):]
                    new_desc.string = link_str
                else:
                    new_desc.string = link_str.replace(site, alt)

                link_desc.replace_with(new_desc)

    def view_image(self, soup) -> BeautifulSoup:
        """Replaces the soup with a new one that handles mobile results and
        adds the link of the image full res to the results.

        Args:
            soup: A BeautifulSoup object containing the image mobile results.

        Returns:
            BeautifulSoup: The new BeautifulSoup object
        """

        # get some tags that are unchanged between mobile and pc versions
        cor_suggested = soup.find_all('table', attrs={'class': "By0U9"})
        next_pages = soup.find('table', attrs={'class': "uZgmoc"})

        results = []
        # find results div
        results_div = soup.find('div', attrs={'class': "nQvrDb"})
        # find all the results (if any)
        results_all = []
        if results_div:
            results_all = results_div.find_all('div', attrs={'class': "lIMUZd"})

        for item in results_all:
            urls = item.find('a')['href'].split('&imgrefurl=')

            # Skip urls that are not two-element lists
            if len(urls) != 2:
                continue

            img_url = urlparse.unquote(urls[0].replace(
                f'/{Endpoint.imgres}?imgurl=', ''))

            try:
                # Try to strip out only the necessary part of the web page link
                web_page = urlparse.unquote(urls[1].split('&')[0])
            except IndexError:
                web_page = urlparse.unquote(urls[1])

            img_tbn = urlparse.unquote(item.find('a').find('img')['src'])

            results.append({
                'domain': urlparse.urlparse(web_page).netloc,
                'img_url': img_url,
                'web_page': web_page,
                'img_tbn': img_tbn
            })

        soup = BeautifulSoup(render_template('imageresults.html',
                                             length=len(results),
                                             results=results,
                                             view_label="View Image"),
                             features='html.parser')

        # replace correction suggested by google object if exists
        if len(cor_suggested):
            soup.find_all(
                'table',
                attrs={'class': "By0U9"}
            )[0].replaceWith(cor_suggested[0])
        # replace next page object at the bottom of the page
        soup.find_all('table',
                      attrs={'class': "uZgmoc"})[0].replaceWith(next_pages)
        return soup