mirror of
https://github.com/searxng/searxng.git
synced 2024-11-23 11:21:00 +00:00
4a28b593c2
The 'scrap_img_by_id' function didn't return any longer anything useful. This fix allows the google images engine to present the full source image instead of only the thumbnail. The function scrap_img_by_id() is rpelaced by a fully rewrite to parse image URLs by a regular expression. The new function parse_urls_img_from_js(dom) returns a mapping of data-id to image URL. Closes: https://github.com/searxng/searxng/issues/909 Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
238 lines
7.8 KiB
Python
238 lines
7.8 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
# lint: pylint
|
|
"""This is the implementation of the google images engine.
|
|
|
|
.. admonition:: Content-Security-Policy (CSP)
|
|
|
|
This engine needs to allow images from the `data URLs`_ (prefixed with the
|
|
``data:`` scheme)::
|
|
|
|
Header set Content-Security-Policy "img-src 'self' data: ;"
|
|
|
|
.. _data URLs:
|
|
https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/Data_URIs
|
|
"""
|
|
|
|
import re
|
|
from urllib.parse import urlencode, unquote
|
|
from lxml import html
|
|
|
|
from searx.utils import (
|
|
eval_xpath,
|
|
eval_xpath_list,
|
|
eval_xpath_getindex,
|
|
extract_text,
|
|
)
|
|
|
|
from searx.engines.google import (
|
|
get_lang_info,
|
|
time_range_dict,
|
|
detect_google_sorry,
|
|
)
|
|
|
|
# pylint: disable=unused-import
|
|
from searx.engines.google import supported_languages_url, _fetch_supported_languages
|
|
|
|
# pylint: enable=unused-import
|
|
|
|
# about
|
|
about = {
|
|
"website": 'https://images.google.com',
|
|
"wikidata_id": 'Q521550',
|
|
"official_api_documentation": 'https://developers.google.com/custom-search',
|
|
"use_official_api": False,
|
|
"require_api_key": False,
|
|
"results": 'HTML',
|
|
}
|
|
|
|
# engine dependent config
|
|
categories = ['images', 'web']
|
|
paging = False
|
|
use_locale_domain = True
|
|
time_range_support = True
|
|
safesearch = True
|
|
|
|
filter_mapping = {0: 'images', 1: 'active', 2: 'active'}
|
|
|
|
|
|
def scrap_out_thumbs(dom):
|
|
"""Scrap out thumbnail data from <script> tags."""
|
|
ret_val = {}
|
|
for script in eval_xpath(dom, '//script[contains(., "_setImgSrc(")]'):
|
|
_script = script.text
|
|
# _setImgSrc('0','data:image\/jpeg;base64,\/9j\/4AAQSkZJR ....');
|
|
_thumb_no, _img_data = _script[len("_setImgSrc(") : -2].split(",", 1)
|
|
_thumb_no = _thumb_no.replace("'", "")
|
|
_img_data = _img_data.replace("'", "")
|
|
_img_data = _img_data.replace(r"\/", r"/")
|
|
ret_val[_thumb_no] = _img_data.replace(r"\x3d", "=")
|
|
return ret_val
|
|
|
|
|
|
# [0, "-H96xjSoW5DsgM", ["https://encrypted-tbn0.gstatic.com/images?q...", 155, 324]
|
|
# , ["https://assets.cdn.moviepilot.de/files/d3bf..", 576, 1200],
|
|
_RE_JS_IMAGE_URL = re.compile(
|
|
r'"'
|
|
r'([^"]*)' # -H96xjSoW5DsgM
|
|
r'",\s*\["'
|
|
r'https://[^\.]*\.gstatic.com/images[^"]*' # https://encrypted-tbn0.gstatic.com/images?q...
|
|
r'[^\[]*\["'
|
|
r'(https?://[^"]*)' # https://assets.cdn.moviepilot.de/files/d3bf...
|
|
)
|
|
|
|
|
|
def parse_urls_img_from_js(dom):
|
|
|
|
# There are two HTML script tags starting with a JS function
|
|
# 'AF_initDataCallback(...)'
|
|
#
|
|
# <script nonce="zscm+Ab/JzBk1Qd4GY6wGQ">
|
|
# AF_initDataCallback({key: 'ds:0', hash: '1', data:[], sideChannel: {}});
|
|
# </script>
|
|
# <script nonce="zscm+Ab/JzBk1Qd4GY6wGQ">
|
|
# AF_initDataCallback({key: 'ds:1', hash: '2', data:[null,[[["online_chips",[["the big",
|
|
# ["https://encrypted-tbn0.gstatic.com/images?q...",null,null,true,[null,0],f
|
|
# ...
|
|
# </script>
|
|
#
|
|
# The second script contains the URLs of the images.
|
|
|
|
# The AF_initDataCallback(..) is called with very large dictionary, that
|
|
# looks like JSON but it is not JSON since it contains JS variables and
|
|
# constants like 'null' (we can't use a JSON parser for).
|
|
#
|
|
# The alternative is to parse the entire <script> and find all image URLs by
|
|
# a regular expression.
|
|
|
|
img_src_script = eval_xpath_getindex(dom, '//script[contains(., "AF_initDataCallback({key: ")]', 1).text
|
|
data_id_to_img_url = {}
|
|
for data_id, url in _RE_JS_IMAGE_URL.findall(img_src_script):
|
|
data_id_to_img_url[data_id] = url
|
|
return data_id_to_img_url
|
|
|
|
|
|
def get_img_url_by_data_id(data_id_to_img_url, img_node):
|
|
"""Get full image URL by @data-id from parent element."""
|
|
|
|
data_id = eval_xpath_getindex(img_node, '../../../@data-id', 0)
|
|
img_url = data_id_to_img_url.get(data_id, '')
|
|
img_url = unquote(img_url.replace(r'\u00', r'%'))
|
|
|
|
return img_url
|
|
|
|
|
|
def request(query, params):
|
|
"""Google-Video search request"""
|
|
|
|
lang_info = get_lang_info(params, supported_languages, language_aliases, False)
|
|
logger.debug("HTTP header Accept-Language --> %s", lang_info['headers']['Accept-Language'])
|
|
|
|
query_url = (
|
|
'https://'
|
|
+ lang_info['subdomain']
|
|
+ '/search'
|
|
+ "?"
|
|
+ urlencode(
|
|
{
|
|
'q': query,
|
|
'tbm': "isch",
|
|
**lang_info['params'],
|
|
'ie': "utf8",
|
|
'oe': "utf8",
|
|
'num': 30,
|
|
}
|
|
)
|
|
)
|
|
|
|
if params['time_range'] in time_range_dict:
|
|
query_url += '&' + urlencode({'tbs': 'qdr:' + time_range_dict[params['time_range']]})
|
|
if params['safesearch']:
|
|
query_url += '&' + urlencode({'safe': filter_mapping[params['safesearch']]})
|
|
params['url'] = query_url
|
|
|
|
params['headers'].update(lang_info['headers'])
|
|
params['headers']['Accept'] = 'text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8'
|
|
return params
|
|
|
|
|
|
def response(resp):
|
|
"""Get response from google's search request"""
|
|
results = []
|
|
|
|
detect_google_sorry(resp)
|
|
|
|
# convert the text to dom
|
|
dom = html.fromstring(resp.text)
|
|
img_bas64_map = scrap_out_thumbs(dom)
|
|
data_id_to_img_url = parse_urls_img_from_js(dom)
|
|
|
|
# parse results
|
|
#
|
|
# root element::
|
|
# <div id="islmp" ..>
|
|
# result div per image::
|
|
# <div jsmodel="tTXmib"> / <div jsaction="..." data-id="..."
|
|
# The data-id matches to a item in a json-data structure in::
|
|
# <script nonce="I+vqelcy/01CKiBJi5Z1Ow">AF_initDataCallback({key: 'ds:1', ... data:function(){return [ ...
|
|
# In this structure the link to the origin PNG, JPG or whatever is given
|
|
# first link per image-div contains a <img> with the data-iid for bas64 encoded image data::
|
|
# <img class="rg_i Q4LuWd" data-iid="0"
|
|
# second link per image-div is the target link::
|
|
# <a class="VFACy kGQAp" href="https://en.wikipedia.org/wiki/The_Sacrament_of_the_Last_Supper">
|
|
# the second link also contains two div tags with the *description* and *publisher*::
|
|
# <div class="WGvvNb">The Sacrament of the Last Supper ...</div>
|
|
# <div class="fxgdke">en.wikipedia.org</div>
|
|
|
|
root = eval_xpath(dom, '//div[@id="islmp"]')
|
|
if not root:
|
|
logger.error("did not find root element id='islmp'")
|
|
return results
|
|
|
|
root = root[0]
|
|
for img_node in eval_xpath_list(root, './/img[contains(@class, "rg_i")]'):
|
|
|
|
img_alt = eval_xpath_getindex(img_node, '@alt', 0)
|
|
|
|
img_base64_id = eval_xpath(img_node, '@data-iid')
|
|
if img_base64_id:
|
|
img_base64_id = img_base64_id[0]
|
|
thumbnail_src = img_bas64_map[img_base64_id]
|
|
else:
|
|
thumbnail_src = eval_xpath(img_node, '@src')
|
|
if not thumbnail_src:
|
|
thumbnail_src = eval_xpath(img_node, '@data-src')
|
|
if thumbnail_src:
|
|
thumbnail_src = thumbnail_src[0]
|
|
else:
|
|
thumbnail_src = ''
|
|
|
|
link_node = eval_xpath_getindex(img_node, '../../../a[2]', 0)
|
|
url = eval_xpath_getindex(link_node, '@href', 0, None)
|
|
if url is None:
|
|
logger.error("missing @href in node: %s", html.tostring(link_node))
|
|
continue
|
|
|
|
pub_nodes = eval_xpath(link_node, './div/div')
|
|
pub_descr = img_alt
|
|
pub_source = ''
|
|
if pub_nodes:
|
|
pub_descr = extract_text(pub_nodes[0])
|
|
pub_source = extract_text(pub_nodes[1])
|
|
|
|
src_url = get_img_url_by_data_id(data_id_to_img_url, img_node)
|
|
if not src_url:
|
|
src_url = thumbnail_src
|
|
|
|
results.append(
|
|
{
|
|
'url': url,
|
|
'title': img_alt,
|
|
'content': pub_descr,
|
|
'source': pub_source,
|
|
'img_src': src_url,
|
|
'thumbnail_src': thumbnail_src,
|
|
'template': 'images.html',
|
|
}
|
|
)
|
|
|
|
return results
|